Time Series Analysis of Daily Closing Stock Price Index
description
Transcript of Time Series Analysis of Daily Closing Stock Price Index
Time series analysis of
Daily Closing Stock Price Indices of
Standard amp Poorrsquos 500 (SampP 500)
Course Instructor Prof Samarjit Das
Report by Mohar Sen QE1201
Data
Data analyzed ndash Unadjusted closing price of SampP 500
Vintage ndash 03011950 ndash 09122013
Source ndash Yahoo Finance
Number of Observations ndash 16088
Nature ndash Non Seasonal (Daily data)
Variables
Level data ndash series P
First difference ndash Series DP
Methodology
Stationarity Analysis
bull Raw data plot
bull Unit root test
bull Removal of non-stationarity
bull Confirmation of stationarity
Conditional mean
bull Box-Jenkins analysis
bull Estimation
bull Information criteria ndash finalizing specification
bull Residual diagnostics
Volatility analysis
bull ARCH test
bull ARCHGARCH model fitting
bull Information criteria ndash finalizing specification
bull Residual diagnostics
Conclusion
bull Final specification of the model
bull Recommendation if any
Plot of Raw Data
0
400
800
1200
1600
2000
55 60 65 70 75 80 85 90 95 00 05 10
P
Observations
bull Uneven trendbull Non-Stationary data
Checking for Trend
Observations
Both trend and intercept terms are significant
Therefore we proceed to test for unit root using ADF test with trend and intercept
Dependent Variable P
Method Least Squares
Date 121113 Time 1132
Sample 1031950 12092013
Included observations 16088
Variable Coefficient Std Error t-Statistic Prob
C -3072818 3871625 -7936765 00000
TREND 0092369 0000417 2215926 00000
R-squared 0753242 Mean dependent var 4356903
Adjusted R-squared 0753226 SD dependent var 4942939
SE of regression 2455470 Akaike info criterion 1384498
Sum squared resid 970E+08 Schwarz criterion 1384593
Log likelihood -1113670 Hannan-Quinn criter 1384529
F-statistic 4910329 Durbin-Watson stat 0000960
Prob(F-statistic) 0000000
ADF test (series p)
Observations
P-Valuegt005
Null hypothesis rejected Unit root present (as expected)
First difference calculated dp=p-p(-1)
Null Hypothesis P has a unit root
Exogenous Constant Linear Trend
Lag Length 21 (Automatic - based on SIC maxlag=42)
t-Statistic Prob
Augmented Dickey-Fuller test statistic -0874981 09572
Test critical values 1 level -3958605
5 level -3410082
10 level -3126769
MacKinnon (1996) one-sided p-values
Variable Coefficient Std Error t-Statistic Prob
P(-1) -0000213 0000243 -0874981 03816
D(P(-1)) -0065128 0007889 -8255719 00000
D(P(-2)) -0042691 0007907 -5399168 00000
D(P(-3)) -0017087 0007914 -2159080 00309
D(P(-4)) -0009753 0007906 -1233615 02174
D(P(-5)) -0038168 0007906 -4827940 00000
D(P(-6)) -0012548 0007909 -1586680 01126
D(P(-7)) -0030904 0007907 -3908456 00001
D(P(-8)) 0007349 0007909 0929223 03528
D(P(-9)) -0004196 0007908 -0530685 05956
D(P(-10)) 0032448 0007898 4108179 00000
D(P(-11)) -0011639 0007902 -1472910 01408
D(P(-12)) 0048758 0007900 6172055 00000
D(P(-13)) 0021209 0007909 2681615 00073
D(P(-14)) -0019073 0007911 -2410950 00159
D(P(-15)) -0024191 0007909 -3058820 00022
D(P(-16)) 0027656 0007911 3496115 00005
D(P(-17)) 0018294 0007908 2313352 00207
D(P(-18)) -0047570 0007910 -6014067 00000
D(P(-19)) 0008477 0007918 1070654 02843
D(P(-20)) 817E-05 0007911 0010329 09918
D(P(-21)) -0046335 0007896 -5868435 00000
C -0145949 0140840 -1036276 03001
TREND(1031950) 463E-05 259E-05 1790536 00734
R-squared 0020029 Mean dependent var 0111498
Adjusted R-squared 0018624 SD dependent var 7611029
SE of regression 7539821 Akaike info criterion 6879767
Sum squared resid 9119701 Schwarz criterion 6891246
Log likelihood -5524117 Hannan-Quinn criter 6883563
F-statistic 1425548 Durbin-Watson stat 1999061
Prob(F-statistic) 0000000
Plot of Series dp (after first differencing of p)
Observations
bull Apparently stationarybull Conditionally
heteroskedasticbull Unit Root test required
for confirmation of stationarity
-120
-80
-40
0
40
80
120
55 60 65 70 75 80 85 90 95 00 05 10
DP
ADF test (series dp)
Observations
P-Valueltlt005 =gt Null hypothesis accepted ndash no unit root in the series dp
It is established that series dp is stationary We try to model it as an ARMA(pq) process by Box Jenkinrsquosmethod
Null Hypothesis P has a unit root
Exogenous Constant Linear Trend
Lag Length 20 (Automatic - based on SIC maxlag=42)
t-Statistic Prob
Augmented Dickey-Fuller test statistic -2906799 00000
Test critical values 1 level -3958605
5 level -3410082
10 level -3126769
MacKinnon (1996) one-sided p-values
Variable Coefficient Std Error t-Statistic Prob
DP(-1) -1208127 0041562 -2906799 00000
D(DP(-1)) 0142840 0040571 3520791 00004
D(DP(-2)) 0099996 0039534 2529330 00114
D(DP(-3)) 0082756 0038448 2152420 00314
D(DP(-4)) 0072855 0037418 1947064 00515
D(DP(-5)) 0034538 0036323 0950846 03417
D(DP(-6)) 0021837 0035136 0621521 05343
D(DP(-7)) -0009215 0033957 -0271356 07861
D(DP(-8)) -0002010 0032778 -0061307 09511
D(DP(-9)) -0006353 0031506 -0201653 08402
D(DP(-10)) 0025942 0030062 0862927 03882
D(DP(-11)) 0014150 0028576 0495183 06205
D(DP(-12)) 0062754 0026917 2331407 00197
D(DP(-13)) 0083808 0025169 3329787 00009
D(DP(-14)) 0064577 0023271 2774993 00055
D(DP(-15)) 0040232 0021310 1887948 00591
D(DP(-16)) 0067737 0019191 3529544 00004
D(DP(-17)) 0085886 0016976 5059121 00000
D(DP(-18)) 0038172 0014478 2636591 00084
D(DP(-19)) 0046510 0011535 4032085 00001
D(DP(-20)) 0046458 0007894 5884990 00000
C -0080367 0119239 -0674003 05003
TREND(1031950) 267E-05 129E-05 2074122 00381
R-squared 0539061 Mean dependent var 0000204
Adjusted R-squared 0538429 SD dependent var 1109785
SE of regression 7539766 Akaike info criterion 6879690
Sum squared resid 9120137 Schwarz criterion 6890691
Log likelihood -5524155 Hannan-Quinn criter 6883328
F-statistic 8528214 Durbin-Watson stat 1999071
Prob(F-statistic) 0000000
Correlogram analysis (series dp)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
This hints at ARMA(pq) model and not AR(p) or MA(q) model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0063 -0063 -799058 64008 0
2 -0039 -0043 -545389 87923 0
3 -0007 -0012 -152201 8871 0
4 -0001 -0004 -050734 88726 0
5 -004 -0041 -520022 11425 0
6 -0006 -0012 -152201 11482 0
7 -0026 -0031 -393187 1255 0
8 0009 0004 0507338 12688 0
9 -0006 -0008 -101468 12738 0
10 0034 0032 4058705 14622 0
11 -002 -0017 -215619 15252 0
12 005 0048 6088058 19218 0
13 0019 0025 3170863 19781 0
14 -0023 -0017 -215619 20648 0
15 -0029 -0026 -32977 2198 0
16 0035 003 3805036 23985 0
17 0014 0022 279036 24304 0
18 -0052 -0048 -608806 28721 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0013 001 1268345 28988 0
20 0008 0003 0380504 29104 0
21 -0048 -0046 -583439 32879 0
22 0018 001 1268345 3343 0
23 0007 0004 0507338 33519 0
24 0003 0004 0507338 33536 0
25 -001 -0013 -164885 33703 0
26 -0009 -0014 -177568 33843 0
27 0033 0035 4439209 35586 0
28 0007 0009 1141511 35656 0
29 0019 0019 2409856 36249 0
30 0005 0013 1648849 36284 0
31 -0003 0006 0761007 36299 0
32 0001 -0002 -025367 363 0
33 -0016 -0014 -177568 36714 0
34 -0066 -0061 -773691 4372 0
35 0006 -0004 -050734 4377 0
36 0018 0009 1141511 44317 0
ARMA Estimation ndash Information Criteria
Information Criteria(SIC)
MA(0) MA(1) MA(2)
AR(0) AIC 6895772
6896728
6891609
6892565
6890118
6891551SIC
AR(1) AIC 6892031
6892986
6889474
6890907
6889569
6891480SIC
AR(2) AIC 6890391
6891824
6889632
6891544
6889077
6891466SIC
Observations
ARMA(11) appears to be the best fitted model
The suitability has been in line with both AIC and SIC
Sufficient data present in the model so SIC is the best criteria
Estimation of the model
Observations
As expected all the terms are significant in the ARMA(11) model
The Durbin Watson Statistic is asymp 2 This suggests that the autocorrelation among residuals may be small
This is in fact confirmed by the correlogramQ-statistics of the residuals
Dependent Variable DP
Method Least Squares
Date 121113 Time 0144
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 8 iterations
MA Backcast 1041950
Variable Coefficient Std Error t-Statistic Prob
C 0111277 0049883 2230768 00257
AR(1) 0607851 0056980 1066776 00000
MA(1) -0672801 0053089 -1267311 00000
R-squared 0006645 Mean dependent var 0111371
Adjusted R-squared 0006522 SD dependent var 7606298
SE of regression 7581454 Akaike info criterion 6889474
Sum squared resid 9244258 Schwarz criterion 6890907
Log likelihood -5540904 Hannan-Quinn criter 6889948
F-statistic 5379525 Durbin-Watson stat 2004706
Prob(F-statistic) 0000000
Inverted AR Roots 61
Inverted MA Roots 67
Correlogram Q statistic (of residuals)
Observations
All ACF values from 5th lag onwards significant
No definite threshold found such that PACF vanishes above that threshold
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0002 -0002 -025366 00897
2 -0001 -0001 -012683 01221
3 0015 0015 1902459 35484 006
4 0011 0011 1395137 55919 0061
5 -0031 -0031 -393175 20833 0
6 -0002 -0003 -038049 20928 0
7 -0022 -0022 -279027 28685 0
8 0011 0011 1395137 30563 0
9 -0002 -0001 -012683 30637 0
10 0036 0035 4439071 50953 0
11 -0014 -0014 -177563 54261 0
12 005 0049 6214699 95031 0
13 002 002 2536612 10169 0
14 -0021 -0021 -266344 1086 0
15 -0027 -0025 -317076 12008 0
16 0033 003 3804918 13763 0
17 0012 0017 215612 14007 0
18 -0051 -005 -634153 18163 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 001 0011 1395137 18319 0
20 0006 0001 0126831 18372 0
21 -0047 -0045 -570738 21954 0
22 0016 0014 1775628 22378 0
23 0007 0006 0760984 22456 0
24 0003 0004 0507322 22474 0
25 -0009 -0011 -139514 226 0
26 -0007 -001 -126831 22681 0
27 0033 0037 4692732 24466 0
28 0009 0009 1141475 24598 0
29 002 0017 215612 25252 0
30 0005 0009 1141475 25299 0
31 -0003 0002 0253661 25318 0
32 -0002 -0006 -076098 25322 0
33 -0019 -0018 -228295 25902 0
34 -0066 -006 -760984 32879 0
35 0003 0001 0126831 32889 0
36 0017 0013 1648798 33334 0
Correlogram Q statistic (of squared residuals)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
However square of the lag values are larger (in ACF)
This suggests ARCH type modelling is more appropriate
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0236 0236 2993202 89426
2 0387 035 4439071 32984
3 0243 0123 1560016 42517 0
4 0297 0139 1762945 56715 0
5 0323 0189 2397098 73461 0
6 0311 0137 1737579 88992 0
7 0303 0098 124294 10377 0
8 0295 0089 1128792 11777 0
9 0283 0067 849765 13068 0
10 0311 0094 1192208 14621 0
11 0307 0089 1128792 16138 0
12 0287 0045 5707377 17464 0
13 0235 -0027 -342443 18356 0
14 022 -0039 -494639 19138 0
15 0237 0001 0126831 20041 0
16 0271 0051 646836 21225 0
17 0269 0045 5707377 22387 0
18 0272 0047 5961038 23577 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0209 -0026 -32976 24280 0
20 0253 003 3804918 25307 0
21 0289 0105 1331721 26653 0
22 0223 -0013 -16488 27456 0
23 0242 -0004 -050732 28402 0
24 0191 -002 -253661 28991 0
25 0209 -0012 -152197 29696 0
26 0205 -0013 -16488 30372 0
27 0276 0079 1001962 31600 0
28 0233 0023 2917104 32474 0
29 0214 -002 -253661 33215 0
30 0185 -0022 -279027 33768 0
31 0194 -0006 -076098 34373 0
32 0251 0067 849765 35388 0
33 02 -0005 -063415 36032 0
34 0232 003 3804918 36902 0
35 0169 -0018 -228295 37360 0
36 0208 0013 1648798 38055 0
ARCH Heteroskedasticity test
Observations
We reaffirm the ARCH effect using the Heteroskedasticity ndash ARCH test
Clearly all residue lags seem significant (uptolag 9)
This hints that we may have to opt for a GARCH(pq) model
Hence we have not tested for pure ARCH models any further
Heteroskedasticity Test ARCH
F-statistic 6246957 Prob F(916067) 00000
ObsR-squared 4167458 Prob Chi-Square(9) 00000
Variable Coefficient Std Error t-Statistic Prob
C 1160923 2045328 5675975 00000
RESID^2(-1) 0013088 0007872 1662714 00964
RESID^2(-2) 0213112 0007842 2717408 00000
RESID^2(-3) 0016465 0007996 2059229 00395
RESID^2(-4) 0062821 0007948 7903630 00000
RESID^2(-5) 0147233 0007879 1868761 00000
RESID^2(-6) 0111116 0007948 1397978 00000
RESID^2(-7) 0080432 0007996 1005965 00000
RESID^2(-8) 0087330 0007842 1113558 00000
RESID^2(-9) 0066907 0007872 8499328 00000
R-squared 0259219 Mean dependent var 5749988
Adjusted R-squared 0258804 SD dependent var 2862004
SE of regression 2463978 Akaike info criterion 1385239
Sum squared resid 975E+08 Schwarz criterion 1385717
Log likelihood -1113425 Hannan-Quinn criter 1385397
F-statistic 6246957 Durbin-Watson stat 2012623
Prob(F-statistic) 0000000
GARCH(11)
Observations
P-Value of Coefficient of GARCH(-1) is more less than 005
Thus we conclude volatility is of GARCH kind
Thus we do not check for ARCH We check for the best form of GARCH that fits the data
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0305
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 32 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0018850 0003353 5622112 00000
AR(1) -0141277 0075434 -1872843 00611
MA(1) 0242258 0073980 3274633 00011
Variance Equation
C 0000213 329E-05 6465720 00000
RESID(-1)^2 0066426 0001506 4410476 00000
GARCH(-1) 0939441 0001222 7689159 00000
R-squared -0022203 Mean dependent var 0111371
Adjusted R-squared -0022330 SD dependent var 7606298
SE of regression 7690752 Akaike info criterion 3800477
Sum squared resid 9512719 Schwarz criterion 3803344
Log likelihood -3056124 Hannan-Quinn criter 3801425
Durbin-Watson stat 2322505
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(12)
Observations
the coefficients of RESID(-1)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0312
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 33 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0019100 0003404 5610811 00000
AR(1) -0137936 0072289 -1908117 00564
MA(1) 0242221 0070576 3432082 00006
Variance Equation
C 0000161 262E-05 6145381 00000
RESID(-1)^2 0107198 0004442 2413017 00000
RESID(-2)^2 -0050822 0005119 -9928459 00000
GARCH(-1) 0948322 0001527 6212227 00000
R-squared -0023299 Mean dependent var 0111371
Adjusted R-squared -0023427 SD dependent var 7606298
SE of regression 7694877 Akaike info criterion 3798666
Sum squared resid 9522926 Schwarz criterion 3802010
Log likelihood -3054567 Hannan-Quinn criter 3799772
Durbin-Watson stat 2328892
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(22)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0319
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1) + C(8)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
Data
Data analyzed ndash Unadjusted closing price of SampP 500
Vintage ndash 03011950 ndash 09122013
Source ndash Yahoo Finance
Number of Observations ndash 16088
Nature ndash Non Seasonal (Daily data)
Variables
Level data ndash series P
First difference ndash Series DP
Methodology
Stationarity Analysis
bull Raw data plot
bull Unit root test
bull Removal of non-stationarity
bull Confirmation of stationarity
Conditional mean
bull Box-Jenkins analysis
bull Estimation
bull Information criteria ndash finalizing specification
bull Residual diagnostics
Volatility analysis
bull ARCH test
bull ARCHGARCH model fitting
bull Information criteria ndash finalizing specification
bull Residual diagnostics
Conclusion
bull Final specification of the model
bull Recommendation if any
Plot of Raw Data
0
400
800
1200
1600
2000
55 60 65 70 75 80 85 90 95 00 05 10
P
Observations
bull Uneven trendbull Non-Stationary data
Checking for Trend
Observations
Both trend and intercept terms are significant
Therefore we proceed to test for unit root using ADF test with trend and intercept
Dependent Variable P
Method Least Squares
Date 121113 Time 1132
Sample 1031950 12092013
Included observations 16088
Variable Coefficient Std Error t-Statistic Prob
C -3072818 3871625 -7936765 00000
TREND 0092369 0000417 2215926 00000
R-squared 0753242 Mean dependent var 4356903
Adjusted R-squared 0753226 SD dependent var 4942939
SE of regression 2455470 Akaike info criterion 1384498
Sum squared resid 970E+08 Schwarz criterion 1384593
Log likelihood -1113670 Hannan-Quinn criter 1384529
F-statistic 4910329 Durbin-Watson stat 0000960
Prob(F-statistic) 0000000
ADF test (series p)
Observations
P-Valuegt005
Null hypothesis rejected Unit root present (as expected)
First difference calculated dp=p-p(-1)
Null Hypothesis P has a unit root
Exogenous Constant Linear Trend
Lag Length 21 (Automatic - based on SIC maxlag=42)
t-Statistic Prob
Augmented Dickey-Fuller test statistic -0874981 09572
Test critical values 1 level -3958605
5 level -3410082
10 level -3126769
MacKinnon (1996) one-sided p-values
Variable Coefficient Std Error t-Statistic Prob
P(-1) -0000213 0000243 -0874981 03816
D(P(-1)) -0065128 0007889 -8255719 00000
D(P(-2)) -0042691 0007907 -5399168 00000
D(P(-3)) -0017087 0007914 -2159080 00309
D(P(-4)) -0009753 0007906 -1233615 02174
D(P(-5)) -0038168 0007906 -4827940 00000
D(P(-6)) -0012548 0007909 -1586680 01126
D(P(-7)) -0030904 0007907 -3908456 00001
D(P(-8)) 0007349 0007909 0929223 03528
D(P(-9)) -0004196 0007908 -0530685 05956
D(P(-10)) 0032448 0007898 4108179 00000
D(P(-11)) -0011639 0007902 -1472910 01408
D(P(-12)) 0048758 0007900 6172055 00000
D(P(-13)) 0021209 0007909 2681615 00073
D(P(-14)) -0019073 0007911 -2410950 00159
D(P(-15)) -0024191 0007909 -3058820 00022
D(P(-16)) 0027656 0007911 3496115 00005
D(P(-17)) 0018294 0007908 2313352 00207
D(P(-18)) -0047570 0007910 -6014067 00000
D(P(-19)) 0008477 0007918 1070654 02843
D(P(-20)) 817E-05 0007911 0010329 09918
D(P(-21)) -0046335 0007896 -5868435 00000
C -0145949 0140840 -1036276 03001
TREND(1031950) 463E-05 259E-05 1790536 00734
R-squared 0020029 Mean dependent var 0111498
Adjusted R-squared 0018624 SD dependent var 7611029
SE of regression 7539821 Akaike info criterion 6879767
Sum squared resid 9119701 Schwarz criterion 6891246
Log likelihood -5524117 Hannan-Quinn criter 6883563
F-statistic 1425548 Durbin-Watson stat 1999061
Prob(F-statistic) 0000000
Plot of Series dp (after first differencing of p)
Observations
bull Apparently stationarybull Conditionally
heteroskedasticbull Unit Root test required
for confirmation of stationarity
-120
-80
-40
0
40
80
120
55 60 65 70 75 80 85 90 95 00 05 10
DP
ADF test (series dp)
Observations
P-Valueltlt005 =gt Null hypothesis accepted ndash no unit root in the series dp
It is established that series dp is stationary We try to model it as an ARMA(pq) process by Box Jenkinrsquosmethod
Null Hypothesis P has a unit root
Exogenous Constant Linear Trend
Lag Length 20 (Automatic - based on SIC maxlag=42)
t-Statistic Prob
Augmented Dickey-Fuller test statistic -2906799 00000
Test critical values 1 level -3958605
5 level -3410082
10 level -3126769
MacKinnon (1996) one-sided p-values
Variable Coefficient Std Error t-Statistic Prob
DP(-1) -1208127 0041562 -2906799 00000
D(DP(-1)) 0142840 0040571 3520791 00004
D(DP(-2)) 0099996 0039534 2529330 00114
D(DP(-3)) 0082756 0038448 2152420 00314
D(DP(-4)) 0072855 0037418 1947064 00515
D(DP(-5)) 0034538 0036323 0950846 03417
D(DP(-6)) 0021837 0035136 0621521 05343
D(DP(-7)) -0009215 0033957 -0271356 07861
D(DP(-8)) -0002010 0032778 -0061307 09511
D(DP(-9)) -0006353 0031506 -0201653 08402
D(DP(-10)) 0025942 0030062 0862927 03882
D(DP(-11)) 0014150 0028576 0495183 06205
D(DP(-12)) 0062754 0026917 2331407 00197
D(DP(-13)) 0083808 0025169 3329787 00009
D(DP(-14)) 0064577 0023271 2774993 00055
D(DP(-15)) 0040232 0021310 1887948 00591
D(DP(-16)) 0067737 0019191 3529544 00004
D(DP(-17)) 0085886 0016976 5059121 00000
D(DP(-18)) 0038172 0014478 2636591 00084
D(DP(-19)) 0046510 0011535 4032085 00001
D(DP(-20)) 0046458 0007894 5884990 00000
C -0080367 0119239 -0674003 05003
TREND(1031950) 267E-05 129E-05 2074122 00381
R-squared 0539061 Mean dependent var 0000204
Adjusted R-squared 0538429 SD dependent var 1109785
SE of regression 7539766 Akaike info criterion 6879690
Sum squared resid 9120137 Schwarz criterion 6890691
Log likelihood -5524155 Hannan-Quinn criter 6883328
F-statistic 8528214 Durbin-Watson stat 1999071
Prob(F-statistic) 0000000
Correlogram analysis (series dp)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
This hints at ARMA(pq) model and not AR(p) or MA(q) model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0063 -0063 -799058 64008 0
2 -0039 -0043 -545389 87923 0
3 -0007 -0012 -152201 8871 0
4 -0001 -0004 -050734 88726 0
5 -004 -0041 -520022 11425 0
6 -0006 -0012 -152201 11482 0
7 -0026 -0031 -393187 1255 0
8 0009 0004 0507338 12688 0
9 -0006 -0008 -101468 12738 0
10 0034 0032 4058705 14622 0
11 -002 -0017 -215619 15252 0
12 005 0048 6088058 19218 0
13 0019 0025 3170863 19781 0
14 -0023 -0017 -215619 20648 0
15 -0029 -0026 -32977 2198 0
16 0035 003 3805036 23985 0
17 0014 0022 279036 24304 0
18 -0052 -0048 -608806 28721 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0013 001 1268345 28988 0
20 0008 0003 0380504 29104 0
21 -0048 -0046 -583439 32879 0
22 0018 001 1268345 3343 0
23 0007 0004 0507338 33519 0
24 0003 0004 0507338 33536 0
25 -001 -0013 -164885 33703 0
26 -0009 -0014 -177568 33843 0
27 0033 0035 4439209 35586 0
28 0007 0009 1141511 35656 0
29 0019 0019 2409856 36249 0
30 0005 0013 1648849 36284 0
31 -0003 0006 0761007 36299 0
32 0001 -0002 -025367 363 0
33 -0016 -0014 -177568 36714 0
34 -0066 -0061 -773691 4372 0
35 0006 -0004 -050734 4377 0
36 0018 0009 1141511 44317 0
ARMA Estimation ndash Information Criteria
Information Criteria(SIC)
MA(0) MA(1) MA(2)
AR(0) AIC 6895772
6896728
6891609
6892565
6890118
6891551SIC
AR(1) AIC 6892031
6892986
6889474
6890907
6889569
6891480SIC
AR(2) AIC 6890391
6891824
6889632
6891544
6889077
6891466SIC
Observations
ARMA(11) appears to be the best fitted model
The suitability has been in line with both AIC and SIC
Sufficient data present in the model so SIC is the best criteria
Estimation of the model
Observations
As expected all the terms are significant in the ARMA(11) model
The Durbin Watson Statistic is asymp 2 This suggests that the autocorrelation among residuals may be small
This is in fact confirmed by the correlogramQ-statistics of the residuals
Dependent Variable DP
Method Least Squares
Date 121113 Time 0144
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 8 iterations
MA Backcast 1041950
Variable Coefficient Std Error t-Statistic Prob
C 0111277 0049883 2230768 00257
AR(1) 0607851 0056980 1066776 00000
MA(1) -0672801 0053089 -1267311 00000
R-squared 0006645 Mean dependent var 0111371
Adjusted R-squared 0006522 SD dependent var 7606298
SE of regression 7581454 Akaike info criterion 6889474
Sum squared resid 9244258 Schwarz criterion 6890907
Log likelihood -5540904 Hannan-Quinn criter 6889948
F-statistic 5379525 Durbin-Watson stat 2004706
Prob(F-statistic) 0000000
Inverted AR Roots 61
Inverted MA Roots 67
Correlogram Q statistic (of residuals)
Observations
All ACF values from 5th lag onwards significant
No definite threshold found such that PACF vanishes above that threshold
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0002 -0002 -025366 00897
2 -0001 -0001 -012683 01221
3 0015 0015 1902459 35484 006
4 0011 0011 1395137 55919 0061
5 -0031 -0031 -393175 20833 0
6 -0002 -0003 -038049 20928 0
7 -0022 -0022 -279027 28685 0
8 0011 0011 1395137 30563 0
9 -0002 -0001 -012683 30637 0
10 0036 0035 4439071 50953 0
11 -0014 -0014 -177563 54261 0
12 005 0049 6214699 95031 0
13 002 002 2536612 10169 0
14 -0021 -0021 -266344 1086 0
15 -0027 -0025 -317076 12008 0
16 0033 003 3804918 13763 0
17 0012 0017 215612 14007 0
18 -0051 -005 -634153 18163 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 001 0011 1395137 18319 0
20 0006 0001 0126831 18372 0
21 -0047 -0045 -570738 21954 0
22 0016 0014 1775628 22378 0
23 0007 0006 0760984 22456 0
24 0003 0004 0507322 22474 0
25 -0009 -0011 -139514 226 0
26 -0007 -001 -126831 22681 0
27 0033 0037 4692732 24466 0
28 0009 0009 1141475 24598 0
29 002 0017 215612 25252 0
30 0005 0009 1141475 25299 0
31 -0003 0002 0253661 25318 0
32 -0002 -0006 -076098 25322 0
33 -0019 -0018 -228295 25902 0
34 -0066 -006 -760984 32879 0
35 0003 0001 0126831 32889 0
36 0017 0013 1648798 33334 0
Correlogram Q statistic (of squared residuals)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
However square of the lag values are larger (in ACF)
This suggests ARCH type modelling is more appropriate
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0236 0236 2993202 89426
2 0387 035 4439071 32984
3 0243 0123 1560016 42517 0
4 0297 0139 1762945 56715 0
5 0323 0189 2397098 73461 0
6 0311 0137 1737579 88992 0
7 0303 0098 124294 10377 0
8 0295 0089 1128792 11777 0
9 0283 0067 849765 13068 0
10 0311 0094 1192208 14621 0
11 0307 0089 1128792 16138 0
12 0287 0045 5707377 17464 0
13 0235 -0027 -342443 18356 0
14 022 -0039 -494639 19138 0
15 0237 0001 0126831 20041 0
16 0271 0051 646836 21225 0
17 0269 0045 5707377 22387 0
18 0272 0047 5961038 23577 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0209 -0026 -32976 24280 0
20 0253 003 3804918 25307 0
21 0289 0105 1331721 26653 0
22 0223 -0013 -16488 27456 0
23 0242 -0004 -050732 28402 0
24 0191 -002 -253661 28991 0
25 0209 -0012 -152197 29696 0
26 0205 -0013 -16488 30372 0
27 0276 0079 1001962 31600 0
28 0233 0023 2917104 32474 0
29 0214 -002 -253661 33215 0
30 0185 -0022 -279027 33768 0
31 0194 -0006 -076098 34373 0
32 0251 0067 849765 35388 0
33 02 -0005 -063415 36032 0
34 0232 003 3804918 36902 0
35 0169 -0018 -228295 37360 0
36 0208 0013 1648798 38055 0
ARCH Heteroskedasticity test
Observations
We reaffirm the ARCH effect using the Heteroskedasticity ndash ARCH test
Clearly all residue lags seem significant (uptolag 9)
This hints that we may have to opt for a GARCH(pq) model
Hence we have not tested for pure ARCH models any further
Heteroskedasticity Test ARCH
F-statistic 6246957 Prob F(916067) 00000
ObsR-squared 4167458 Prob Chi-Square(9) 00000
Variable Coefficient Std Error t-Statistic Prob
C 1160923 2045328 5675975 00000
RESID^2(-1) 0013088 0007872 1662714 00964
RESID^2(-2) 0213112 0007842 2717408 00000
RESID^2(-3) 0016465 0007996 2059229 00395
RESID^2(-4) 0062821 0007948 7903630 00000
RESID^2(-5) 0147233 0007879 1868761 00000
RESID^2(-6) 0111116 0007948 1397978 00000
RESID^2(-7) 0080432 0007996 1005965 00000
RESID^2(-8) 0087330 0007842 1113558 00000
RESID^2(-9) 0066907 0007872 8499328 00000
R-squared 0259219 Mean dependent var 5749988
Adjusted R-squared 0258804 SD dependent var 2862004
SE of regression 2463978 Akaike info criterion 1385239
Sum squared resid 975E+08 Schwarz criterion 1385717
Log likelihood -1113425 Hannan-Quinn criter 1385397
F-statistic 6246957 Durbin-Watson stat 2012623
Prob(F-statistic) 0000000
GARCH(11)
Observations
P-Value of Coefficient of GARCH(-1) is more less than 005
Thus we conclude volatility is of GARCH kind
Thus we do not check for ARCH We check for the best form of GARCH that fits the data
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0305
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 32 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0018850 0003353 5622112 00000
AR(1) -0141277 0075434 -1872843 00611
MA(1) 0242258 0073980 3274633 00011
Variance Equation
C 0000213 329E-05 6465720 00000
RESID(-1)^2 0066426 0001506 4410476 00000
GARCH(-1) 0939441 0001222 7689159 00000
R-squared -0022203 Mean dependent var 0111371
Adjusted R-squared -0022330 SD dependent var 7606298
SE of regression 7690752 Akaike info criterion 3800477
Sum squared resid 9512719 Schwarz criterion 3803344
Log likelihood -3056124 Hannan-Quinn criter 3801425
Durbin-Watson stat 2322505
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(12)
Observations
the coefficients of RESID(-1)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0312
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 33 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0019100 0003404 5610811 00000
AR(1) -0137936 0072289 -1908117 00564
MA(1) 0242221 0070576 3432082 00006
Variance Equation
C 0000161 262E-05 6145381 00000
RESID(-1)^2 0107198 0004442 2413017 00000
RESID(-2)^2 -0050822 0005119 -9928459 00000
GARCH(-1) 0948322 0001527 6212227 00000
R-squared -0023299 Mean dependent var 0111371
Adjusted R-squared -0023427 SD dependent var 7606298
SE of regression 7694877 Akaike info criterion 3798666
Sum squared resid 9522926 Schwarz criterion 3802010
Log likelihood -3054567 Hannan-Quinn criter 3799772
Durbin-Watson stat 2328892
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(22)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0319
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1) + C(8)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
Methodology
Stationarity Analysis
bull Raw data plot
bull Unit root test
bull Removal of non-stationarity
bull Confirmation of stationarity
Conditional mean
bull Box-Jenkins analysis
bull Estimation
bull Information criteria ndash finalizing specification
bull Residual diagnostics
Volatility analysis
bull ARCH test
bull ARCHGARCH model fitting
bull Information criteria ndash finalizing specification
bull Residual diagnostics
Conclusion
bull Final specification of the model
bull Recommendation if any
Plot of Raw Data
0
400
800
1200
1600
2000
55 60 65 70 75 80 85 90 95 00 05 10
P
Observations
bull Uneven trendbull Non-Stationary data
Checking for Trend
Observations
Both trend and intercept terms are significant
Therefore we proceed to test for unit root using ADF test with trend and intercept
Dependent Variable P
Method Least Squares
Date 121113 Time 1132
Sample 1031950 12092013
Included observations 16088
Variable Coefficient Std Error t-Statistic Prob
C -3072818 3871625 -7936765 00000
TREND 0092369 0000417 2215926 00000
R-squared 0753242 Mean dependent var 4356903
Adjusted R-squared 0753226 SD dependent var 4942939
SE of regression 2455470 Akaike info criterion 1384498
Sum squared resid 970E+08 Schwarz criterion 1384593
Log likelihood -1113670 Hannan-Quinn criter 1384529
F-statistic 4910329 Durbin-Watson stat 0000960
Prob(F-statistic) 0000000
ADF test (series p)
Observations
P-Valuegt005
Null hypothesis rejected Unit root present (as expected)
First difference calculated dp=p-p(-1)
Null Hypothesis P has a unit root
Exogenous Constant Linear Trend
Lag Length 21 (Automatic - based on SIC maxlag=42)
t-Statistic Prob
Augmented Dickey-Fuller test statistic -0874981 09572
Test critical values 1 level -3958605
5 level -3410082
10 level -3126769
MacKinnon (1996) one-sided p-values
Variable Coefficient Std Error t-Statistic Prob
P(-1) -0000213 0000243 -0874981 03816
D(P(-1)) -0065128 0007889 -8255719 00000
D(P(-2)) -0042691 0007907 -5399168 00000
D(P(-3)) -0017087 0007914 -2159080 00309
D(P(-4)) -0009753 0007906 -1233615 02174
D(P(-5)) -0038168 0007906 -4827940 00000
D(P(-6)) -0012548 0007909 -1586680 01126
D(P(-7)) -0030904 0007907 -3908456 00001
D(P(-8)) 0007349 0007909 0929223 03528
D(P(-9)) -0004196 0007908 -0530685 05956
D(P(-10)) 0032448 0007898 4108179 00000
D(P(-11)) -0011639 0007902 -1472910 01408
D(P(-12)) 0048758 0007900 6172055 00000
D(P(-13)) 0021209 0007909 2681615 00073
D(P(-14)) -0019073 0007911 -2410950 00159
D(P(-15)) -0024191 0007909 -3058820 00022
D(P(-16)) 0027656 0007911 3496115 00005
D(P(-17)) 0018294 0007908 2313352 00207
D(P(-18)) -0047570 0007910 -6014067 00000
D(P(-19)) 0008477 0007918 1070654 02843
D(P(-20)) 817E-05 0007911 0010329 09918
D(P(-21)) -0046335 0007896 -5868435 00000
C -0145949 0140840 -1036276 03001
TREND(1031950) 463E-05 259E-05 1790536 00734
R-squared 0020029 Mean dependent var 0111498
Adjusted R-squared 0018624 SD dependent var 7611029
SE of regression 7539821 Akaike info criterion 6879767
Sum squared resid 9119701 Schwarz criterion 6891246
Log likelihood -5524117 Hannan-Quinn criter 6883563
F-statistic 1425548 Durbin-Watson stat 1999061
Prob(F-statistic) 0000000
Plot of Series dp (after first differencing of p)
Observations
bull Apparently stationarybull Conditionally
heteroskedasticbull Unit Root test required
for confirmation of stationarity
-120
-80
-40
0
40
80
120
55 60 65 70 75 80 85 90 95 00 05 10
DP
ADF test (series dp)
Observations
P-Valueltlt005 =gt Null hypothesis accepted ndash no unit root in the series dp
It is established that series dp is stationary We try to model it as an ARMA(pq) process by Box Jenkinrsquosmethod
Null Hypothesis P has a unit root
Exogenous Constant Linear Trend
Lag Length 20 (Automatic - based on SIC maxlag=42)
t-Statistic Prob
Augmented Dickey-Fuller test statistic -2906799 00000
Test critical values 1 level -3958605
5 level -3410082
10 level -3126769
MacKinnon (1996) one-sided p-values
Variable Coefficient Std Error t-Statistic Prob
DP(-1) -1208127 0041562 -2906799 00000
D(DP(-1)) 0142840 0040571 3520791 00004
D(DP(-2)) 0099996 0039534 2529330 00114
D(DP(-3)) 0082756 0038448 2152420 00314
D(DP(-4)) 0072855 0037418 1947064 00515
D(DP(-5)) 0034538 0036323 0950846 03417
D(DP(-6)) 0021837 0035136 0621521 05343
D(DP(-7)) -0009215 0033957 -0271356 07861
D(DP(-8)) -0002010 0032778 -0061307 09511
D(DP(-9)) -0006353 0031506 -0201653 08402
D(DP(-10)) 0025942 0030062 0862927 03882
D(DP(-11)) 0014150 0028576 0495183 06205
D(DP(-12)) 0062754 0026917 2331407 00197
D(DP(-13)) 0083808 0025169 3329787 00009
D(DP(-14)) 0064577 0023271 2774993 00055
D(DP(-15)) 0040232 0021310 1887948 00591
D(DP(-16)) 0067737 0019191 3529544 00004
D(DP(-17)) 0085886 0016976 5059121 00000
D(DP(-18)) 0038172 0014478 2636591 00084
D(DP(-19)) 0046510 0011535 4032085 00001
D(DP(-20)) 0046458 0007894 5884990 00000
C -0080367 0119239 -0674003 05003
TREND(1031950) 267E-05 129E-05 2074122 00381
R-squared 0539061 Mean dependent var 0000204
Adjusted R-squared 0538429 SD dependent var 1109785
SE of regression 7539766 Akaike info criterion 6879690
Sum squared resid 9120137 Schwarz criterion 6890691
Log likelihood -5524155 Hannan-Quinn criter 6883328
F-statistic 8528214 Durbin-Watson stat 1999071
Prob(F-statistic) 0000000
Correlogram analysis (series dp)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
This hints at ARMA(pq) model and not AR(p) or MA(q) model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0063 -0063 -799058 64008 0
2 -0039 -0043 -545389 87923 0
3 -0007 -0012 -152201 8871 0
4 -0001 -0004 -050734 88726 0
5 -004 -0041 -520022 11425 0
6 -0006 -0012 -152201 11482 0
7 -0026 -0031 -393187 1255 0
8 0009 0004 0507338 12688 0
9 -0006 -0008 -101468 12738 0
10 0034 0032 4058705 14622 0
11 -002 -0017 -215619 15252 0
12 005 0048 6088058 19218 0
13 0019 0025 3170863 19781 0
14 -0023 -0017 -215619 20648 0
15 -0029 -0026 -32977 2198 0
16 0035 003 3805036 23985 0
17 0014 0022 279036 24304 0
18 -0052 -0048 -608806 28721 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0013 001 1268345 28988 0
20 0008 0003 0380504 29104 0
21 -0048 -0046 -583439 32879 0
22 0018 001 1268345 3343 0
23 0007 0004 0507338 33519 0
24 0003 0004 0507338 33536 0
25 -001 -0013 -164885 33703 0
26 -0009 -0014 -177568 33843 0
27 0033 0035 4439209 35586 0
28 0007 0009 1141511 35656 0
29 0019 0019 2409856 36249 0
30 0005 0013 1648849 36284 0
31 -0003 0006 0761007 36299 0
32 0001 -0002 -025367 363 0
33 -0016 -0014 -177568 36714 0
34 -0066 -0061 -773691 4372 0
35 0006 -0004 -050734 4377 0
36 0018 0009 1141511 44317 0
ARMA Estimation ndash Information Criteria
Information Criteria(SIC)
MA(0) MA(1) MA(2)
AR(0) AIC 6895772
6896728
6891609
6892565
6890118
6891551SIC
AR(1) AIC 6892031
6892986
6889474
6890907
6889569
6891480SIC
AR(2) AIC 6890391
6891824
6889632
6891544
6889077
6891466SIC
Observations
ARMA(11) appears to be the best fitted model
The suitability has been in line with both AIC and SIC
Sufficient data present in the model so SIC is the best criteria
Estimation of the model
Observations
As expected all the terms are significant in the ARMA(11) model
The Durbin Watson Statistic is asymp 2 This suggests that the autocorrelation among residuals may be small
This is in fact confirmed by the correlogramQ-statistics of the residuals
Dependent Variable DP
Method Least Squares
Date 121113 Time 0144
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 8 iterations
MA Backcast 1041950
Variable Coefficient Std Error t-Statistic Prob
C 0111277 0049883 2230768 00257
AR(1) 0607851 0056980 1066776 00000
MA(1) -0672801 0053089 -1267311 00000
R-squared 0006645 Mean dependent var 0111371
Adjusted R-squared 0006522 SD dependent var 7606298
SE of regression 7581454 Akaike info criterion 6889474
Sum squared resid 9244258 Schwarz criterion 6890907
Log likelihood -5540904 Hannan-Quinn criter 6889948
F-statistic 5379525 Durbin-Watson stat 2004706
Prob(F-statistic) 0000000
Inverted AR Roots 61
Inverted MA Roots 67
Correlogram Q statistic (of residuals)
Observations
All ACF values from 5th lag onwards significant
No definite threshold found such that PACF vanishes above that threshold
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0002 -0002 -025366 00897
2 -0001 -0001 -012683 01221
3 0015 0015 1902459 35484 006
4 0011 0011 1395137 55919 0061
5 -0031 -0031 -393175 20833 0
6 -0002 -0003 -038049 20928 0
7 -0022 -0022 -279027 28685 0
8 0011 0011 1395137 30563 0
9 -0002 -0001 -012683 30637 0
10 0036 0035 4439071 50953 0
11 -0014 -0014 -177563 54261 0
12 005 0049 6214699 95031 0
13 002 002 2536612 10169 0
14 -0021 -0021 -266344 1086 0
15 -0027 -0025 -317076 12008 0
16 0033 003 3804918 13763 0
17 0012 0017 215612 14007 0
18 -0051 -005 -634153 18163 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 001 0011 1395137 18319 0
20 0006 0001 0126831 18372 0
21 -0047 -0045 -570738 21954 0
22 0016 0014 1775628 22378 0
23 0007 0006 0760984 22456 0
24 0003 0004 0507322 22474 0
25 -0009 -0011 -139514 226 0
26 -0007 -001 -126831 22681 0
27 0033 0037 4692732 24466 0
28 0009 0009 1141475 24598 0
29 002 0017 215612 25252 0
30 0005 0009 1141475 25299 0
31 -0003 0002 0253661 25318 0
32 -0002 -0006 -076098 25322 0
33 -0019 -0018 -228295 25902 0
34 -0066 -006 -760984 32879 0
35 0003 0001 0126831 32889 0
36 0017 0013 1648798 33334 0
Correlogram Q statistic (of squared residuals)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
However square of the lag values are larger (in ACF)
This suggests ARCH type modelling is more appropriate
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0236 0236 2993202 89426
2 0387 035 4439071 32984
3 0243 0123 1560016 42517 0
4 0297 0139 1762945 56715 0
5 0323 0189 2397098 73461 0
6 0311 0137 1737579 88992 0
7 0303 0098 124294 10377 0
8 0295 0089 1128792 11777 0
9 0283 0067 849765 13068 0
10 0311 0094 1192208 14621 0
11 0307 0089 1128792 16138 0
12 0287 0045 5707377 17464 0
13 0235 -0027 -342443 18356 0
14 022 -0039 -494639 19138 0
15 0237 0001 0126831 20041 0
16 0271 0051 646836 21225 0
17 0269 0045 5707377 22387 0
18 0272 0047 5961038 23577 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0209 -0026 -32976 24280 0
20 0253 003 3804918 25307 0
21 0289 0105 1331721 26653 0
22 0223 -0013 -16488 27456 0
23 0242 -0004 -050732 28402 0
24 0191 -002 -253661 28991 0
25 0209 -0012 -152197 29696 0
26 0205 -0013 -16488 30372 0
27 0276 0079 1001962 31600 0
28 0233 0023 2917104 32474 0
29 0214 -002 -253661 33215 0
30 0185 -0022 -279027 33768 0
31 0194 -0006 -076098 34373 0
32 0251 0067 849765 35388 0
33 02 -0005 -063415 36032 0
34 0232 003 3804918 36902 0
35 0169 -0018 -228295 37360 0
36 0208 0013 1648798 38055 0
ARCH Heteroskedasticity test
Observations
We reaffirm the ARCH effect using the Heteroskedasticity ndash ARCH test
Clearly all residue lags seem significant (uptolag 9)
This hints that we may have to opt for a GARCH(pq) model
Hence we have not tested for pure ARCH models any further
Heteroskedasticity Test ARCH
F-statistic 6246957 Prob F(916067) 00000
ObsR-squared 4167458 Prob Chi-Square(9) 00000
Variable Coefficient Std Error t-Statistic Prob
C 1160923 2045328 5675975 00000
RESID^2(-1) 0013088 0007872 1662714 00964
RESID^2(-2) 0213112 0007842 2717408 00000
RESID^2(-3) 0016465 0007996 2059229 00395
RESID^2(-4) 0062821 0007948 7903630 00000
RESID^2(-5) 0147233 0007879 1868761 00000
RESID^2(-6) 0111116 0007948 1397978 00000
RESID^2(-7) 0080432 0007996 1005965 00000
RESID^2(-8) 0087330 0007842 1113558 00000
RESID^2(-9) 0066907 0007872 8499328 00000
R-squared 0259219 Mean dependent var 5749988
Adjusted R-squared 0258804 SD dependent var 2862004
SE of regression 2463978 Akaike info criterion 1385239
Sum squared resid 975E+08 Schwarz criterion 1385717
Log likelihood -1113425 Hannan-Quinn criter 1385397
F-statistic 6246957 Durbin-Watson stat 2012623
Prob(F-statistic) 0000000
GARCH(11)
Observations
P-Value of Coefficient of GARCH(-1) is more less than 005
Thus we conclude volatility is of GARCH kind
Thus we do not check for ARCH We check for the best form of GARCH that fits the data
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0305
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 32 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0018850 0003353 5622112 00000
AR(1) -0141277 0075434 -1872843 00611
MA(1) 0242258 0073980 3274633 00011
Variance Equation
C 0000213 329E-05 6465720 00000
RESID(-1)^2 0066426 0001506 4410476 00000
GARCH(-1) 0939441 0001222 7689159 00000
R-squared -0022203 Mean dependent var 0111371
Adjusted R-squared -0022330 SD dependent var 7606298
SE of regression 7690752 Akaike info criterion 3800477
Sum squared resid 9512719 Schwarz criterion 3803344
Log likelihood -3056124 Hannan-Quinn criter 3801425
Durbin-Watson stat 2322505
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(12)
Observations
the coefficients of RESID(-1)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0312
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 33 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0019100 0003404 5610811 00000
AR(1) -0137936 0072289 -1908117 00564
MA(1) 0242221 0070576 3432082 00006
Variance Equation
C 0000161 262E-05 6145381 00000
RESID(-1)^2 0107198 0004442 2413017 00000
RESID(-2)^2 -0050822 0005119 -9928459 00000
GARCH(-1) 0948322 0001527 6212227 00000
R-squared -0023299 Mean dependent var 0111371
Adjusted R-squared -0023427 SD dependent var 7606298
SE of regression 7694877 Akaike info criterion 3798666
Sum squared resid 9522926 Schwarz criterion 3802010
Log likelihood -3054567 Hannan-Quinn criter 3799772
Durbin-Watson stat 2328892
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(22)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0319
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1) + C(8)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
Plot of Raw Data
0
400
800
1200
1600
2000
55 60 65 70 75 80 85 90 95 00 05 10
P
Observations
bull Uneven trendbull Non-Stationary data
Checking for Trend
Observations
Both trend and intercept terms are significant
Therefore we proceed to test for unit root using ADF test with trend and intercept
Dependent Variable P
Method Least Squares
Date 121113 Time 1132
Sample 1031950 12092013
Included observations 16088
Variable Coefficient Std Error t-Statistic Prob
C -3072818 3871625 -7936765 00000
TREND 0092369 0000417 2215926 00000
R-squared 0753242 Mean dependent var 4356903
Adjusted R-squared 0753226 SD dependent var 4942939
SE of regression 2455470 Akaike info criterion 1384498
Sum squared resid 970E+08 Schwarz criterion 1384593
Log likelihood -1113670 Hannan-Quinn criter 1384529
F-statistic 4910329 Durbin-Watson stat 0000960
Prob(F-statistic) 0000000
ADF test (series p)
Observations
P-Valuegt005
Null hypothesis rejected Unit root present (as expected)
First difference calculated dp=p-p(-1)
Null Hypothesis P has a unit root
Exogenous Constant Linear Trend
Lag Length 21 (Automatic - based on SIC maxlag=42)
t-Statistic Prob
Augmented Dickey-Fuller test statistic -0874981 09572
Test critical values 1 level -3958605
5 level -3410082
10 level -3126769
MacKinnon (1996) one-sided p-values
Variable Coefficient Std Error t-Statistic Prob
P(-1) -0000213 0000243 -0874981 03816
D(P(-1)) -0065128 0007889 -8255719 00000
D(P(-2)) -0042691 0007907 -5399168 00000
D(P(-3)) -0017087 0007914 -2159080 00309
D(P(-4)) -0009753 0007906 -1233615 02174
D(P(-5)) -0038168 0007906 -4827940 00000
D(P(-6)) -0012548 0007909 -1586680 01126
D(P(-7)) -0030904 0007907 -3908456 00001
D(P(-8)) 0007349 0007909 0929223 03528
D(P(-9)) -0004196 0007908 -0530685 05956
D(P(-10)) 0032448 0007898 4108179 00000
D(P(-11)) -0011639 0007902 -1472910 01408
D(P(-12)) 0048758 0007900 6172055 00000
D(P(-13)) 0021209 0007909 2681615 00073
D(P(-14)) -0019073 0007911 -2410950 00159
D(P(-15)) -0024191 0007909 -3058820 00022
D(P(-16)) 0027656 0007911 3496115 00005
D(P(-17)) 0018294 0007908 2313352 00207
D(P(-18)) -0047570 0007910 -6014067 00000
D(P(-19)) 0008477 0007918 1070654 02843
D(P(-20)) 817E-05 0007911 0010329 09918
D(P(-21)) -0046335 0007896 -5868435 00000
C -0145949 0140840 -1036276 03001
TREND(1031950) 463E-05 259E-05 1790536 00734
R-squared 0020029 Mean dependent var 0111498
Adjusted R-squared 0018624 SD dependent var 7611029
SE of regression 7539821 Akaike info criterion 6879767
Sum squared resid 9119701 Schwarz criterion 6891246
Log likelihood -5524117 Hannan-Quinn criter 6883563
F-statistic 1425548 Durbin-Watson stat 1999061
Prob(F-statistic) 0000000
Plot of Series dp (after first differencing of p)
Observations
bull Apparently stationarybull Conditionally
heteroskedasticbull Unit Root test required
for confirmation of stationarity
-120
-80
-40
0
40
80
120
55 60 65 70 75 80 85 90 95 00 05 10
DP
ADF test (series dp)
Observations
P-Valueltlt005 =gt Null hypothesis accepted ndash no unit root in the series dp
It is established that series dp is stationary We try to model it as an ARMA(pq) process by Box Jenkinrsquosmethod
Null Hypothesis P has a unit root
Exogenous Constant Linear Trend
Lag Length 20 (Automatic - based on SIC maxlag=42)
t-Statistic Prob
Augmented Dickey-Fuller test statistic -2906799 00000
Test critical values 1 level -3958605
5 level -3410082
10 level -3126769
MacKinnon (1996) one-sided p-values
Variable Coefficient Std Error t-Statistic Prob
DP(-1) -1208127 0041562 -2906799 00000
D(DP(-1)) 0142840 0040571 3520791 00004
D(DP(-2)) 0099996 0039534 2529330 00114
D(DP(-3)) 0082756 0038448 2152420 00314
D(DP(-4)) 0072855 0037418 1947064 00515
D(DP(-5)) 0034538 0036323 0950846 03417
D(DP(-6)) 0021837 0035136 0621521 05343
D(DP(-7)) -0009215 0033957 -0271356 07861
D(DP(-8)) -0002010 0032778 -0061307 09511
D(DP(-9)) -0006353 0031506 -0201653 08402
D(DP(-10)) 0025942 0030062 0862927 03882
D(DP(-11)) 0014150 0028576 0495183 06205
D(DP(-12)) 0062754 0026917 2331407 00197
D(DP(-13)) 0083808 0025169 3329787 00009
D(DP(-14)) 0064577 0023271 2774993 00055
D(DP(-15)) 0040232 0021310 1887948 00591
D(DP(-16)) 0067737 0019191 3529544 00004
D(DP(-17)) 0085886 0016976 5059121 00000
D(DP(-18)) 0038172 0014478 2636591 00084
D(DP(-19)) 0046510 0011535 4032085 00001
D(DP(-20)) 0046458 0007894 5884990 00000
C -0080367 0119239 -0674003 05003
TREND(1031950) 267E-05 129E-05 2074122 00381
R-squared 0539061 Mean dependent var 0000204
Adjusted R-squared 0538429 SD dependent var 1109785
SE of regression 7539766 Akaike info criterion 6879690
Sum squared resid 9120137 Schwarz criterion 6890691
Log likelihood -5524155 Hannan-Quinn criter 6883328
F-statistic 8528214 Durbin-Watson stat 1999071
Prob(F-statistic) 0000000
Correlogram analysis (series dp)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
This hints at ARMA(pq) model and not AR(p) or MA(q) model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0063 -0063 -799058 64008 0
2 -0039 -0043 -545389 87923 0
3 -0007 -0012 -152201 8871 0
4 -0001 -0004 -050734 88726 0
5 -004 -0041 -520022 11425 0
6 -0006 -0012 -152201 11482 0
7 -0026 -0031 -393187 1255 0
8 0009 0004 0507338 12688 0
9 -0006 -0008 -101468 12738 0
10 0034 0032 4058705 14622 0
11 -002 -0017 -215619 15252 0
12 005 0048 6088058 19218 0
13 0019 0025 3170863 19781 0
14 -0023 -0017 -215619 20648 0
15 -0029 -0026 -32977 2198 0
16 0035 003 3805036 23985 0
17 0014 0022 279036 24304 0
18 -0052 -0048 -608806 28721 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0013 001 1268345 28988 0
20 0008 0003 0380504 29104 0
21 -0048 -0046 -583439 32879 0
22 0018 001 1268345 3343 0
23 0007 0004 0507338 33519 0
24 0003 0004 0507338 33536 0
25 -001 -0013 -164885 33703 0
26 -0009 -0014 -177568 33843 0
27 0033 0035 4439209 35586 0
28 0007 0009 1141511 35656 0
29 0019 0019 2409856 36249 0
30 0005 0013 1648849 36284 0
31 -0003 0006 0761007 36299 0
32 0001 -0002 -025367 363 0
33 -0016 -0014 -177568 36714 0
34 -0066 -0061 -773691 4372 0
35 0006 -0004 -050734 4377 0
36 0018 0009 1141511 44317 0
ARMA Estimation ndash Information Criteria
Information Criteria(SIC)
MA(0) MA(1) MA(2)
AR(0) AIC 6895772
6896728
6891609
6892565
6890118
6891551SIC
AR(1) AIC 6892031
6892986
6889474
6890907
6889569
6891480SIC
AR(2) AIC 6890391
6891824
6889632
6891544
6889077
6891466SIC
Observations
ARMA(11) appears to be the best fitted model
The suitability has been in line with both AIC and SIC
Sufficient data present in the model so SIC is the best criteria
Estimation of the model
Observations
As expected all the terms are significant in the ARMA(11) model
The Durbin Watson Statistic is asymp 2 This suggests that the autocorrelation among residuals may be small
This is in fact confirmed by the correlogramQ-statistics of the residuals
Dependent Variable DP
Method Least Squares
Date 121113 Time 0144
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 8 iterations
MA Backcast 1041950
Variable Coefficient Std Error t-Statistic Prob
C 0111277 0049883 2230768 00257
AR(1) 0607851 0056980 1066776 00000
MA(1) -0672801 0053089 -1267311 00000
R-squared 0006645 Mean dependent var 0111371
Adjusted R-squared 0006522 SD dependent var 7606298
SE of regression 7581454 Akaike info criterion 6889474
Sum squared resid 9244258 Schwarz criterion 6890907
Log likelihood -5540904 Hannan-Quinn criter 6889948
F-statistic 5379525 Durbin-Watson stat 2004706
Prob(F-statistic) 0000000
Inverted AR Roots 61
Inverted MA Roots 67
Correlogram Q statistic (of residuals)
Observations
All ACF values from 5th lag onwards significant
No definite threshold found such that PACF vanishes above that threshold
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0002 -0002 -025366 00897
2 -0001 -0001 -012683 01221
3 0015 0015 1902459 35484 006
4 0011 0011 1395137 55919 0061
5 -0031 -0031 -393175 20833 0
6 -0002 -0003 -038049 20928 0
7 -0022 -0022 -279027 28685 0
8 0011 0011 1395137 30563 0
9 -0002 -0001 -012683 30637 0
10 0036 0035 4439071 50953 0
11 -0014 -0014 -177563 54261 0
12 005 0049 6214699 95031 0
13 002 002 2536612 10169 0
14 -0021 -0021 -266344 1086 0
15 -0027 -0025 -317076 12008 0
16 0033 003 3804918 13763 0
17 0012 0017 215612 14007 0
18 -0051 -005 -634153 18163 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 001 0011 1395137 18319 0
20 0006 0001 0126831 18372 0
21 -0047 -0045 -570738 21954 0
22 0016 0014 1775628 22378 0
23 0007 0006 0760984 22456 0
24 0003 0004 0507322 22474 0
25 -0009 -0011 -139514 226 0
26 -0007 -001 -126831 22681 0
27 0033 0037 4692732 24466 0
28 0009 0009 1141475 24598 0
29 002 0017 215612 25252 0
30 0005 0009 1141475 25299 0
31 -0003 0002 0253661 25318 0
32 -0002 -0006 -076098 25322 0
33 -0019 -0018 -228295 25902 0
34 -0066 -006 -760984 32879 0
35 0003 0001 0126831 32889 0
36 0017 0013 1648798 33334 0
Correlogram Q statistic (of squared residuals)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
However square of the lag values are larger (in ACF)
This suggests ARCH type modelling is more appropriate
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0236 0236 2993202 89426
2 0387 035 4439071 32984
3 0243 0123 1560016 42517 0
4 0297 0139 1762945 56715 0
5 0323 0189 2397098 73461 0
6 0311 0137 1737579 88992 0
7 0303 0098 124294 10377 0
8 0295 0089 1128792 11777 0
9 0283 0067 849765 13068 0
10 0311 0094 1192208 14621 0
11 0307 0089 1128792 16138 0
12 0287 0045 5707377 17464 0
13 0235 -0027 -342443 18356 0
14 022 -0039 -494639 19138 0
15 0237 0001 0126831 20041 0
16 0271 0051 646836 21225 0
17 0269 0045 5707377 22387 0
18 0272 0047 5961038 23577 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0209 -0026 -32976 24280 0
20 0253 003 3804918 25307 0
21 0289 0105 1331721 26653 0
22 0223 -0013 -16488 27456 0
23 0242 -0004 -050732 28402 0
24 0191 -002 -253661 28991 0
25 0209 -0012 -152197 29696 0
26 0205 -0013 -16488 30372 0
27 0276 0079 1001962 31600 0
28 0233 0023 2917104 32474 0
29 0214 -002 -253661 33215 0
30 0185 -0022 -279027 33768 0
31 0194 -0006 -076098 34373 0
32 0251 0067 849765 35388 0
33 02 -0005 -063415 36032 0
34 0232 003 3804918 36902 0
35 0169 -0018 -228295 37360 0
36 0208 0013 1648798 38055 0
ARCH Heteroskedasticity test
Observations
We reaffirm the ARCH effect using the Heteroskedasticity ndash ARCH test
Clearly all residue lags seem significant (uptolag 9)
This hints that we may have to opt for a GARCH(pq) model
Hence we have not tested for pure ARCH models any further
Heteroskedasticity Test ARCH
F-statistic 6246957 Prob F(916067) 00000
ObsR-squared 4167458 Prob Chi-Square(9) 00000
Variable Coefficient Std Error t-Statistic Prob
C 1160923 2045328 5675975 00000
RESID^2(-1) 0013088 0007872 1662714 00964
RESID^2(-2) 0213112 0007842 2717408 00000
RESID^2(-3) 0016465 0007996 2059229 00395
RESID^2(-4) 0062821 0007948 7903630 00000
RESID^2(-5) 0147233 0007879 1868761 00000
RESID^2(-6) 0111116 0007948 1397978 00000
RESID^2(-7) 0080432 0007996 1005965 00000
RESID^2(-8) 0087330 0007842 1113558 00000
RESID^2(-9) 0066907 0007872 8499328 00000
R-squared 0259219 Mean dependent var 5749988
Adjusted R-squared 0258804 SD dependent var 2862004
SE of regression 2463978 Akaike info criterion 1385239
Sum squared resid 975E+08 Schwarz criterion 1385717
Log likelihood -1113425 Hannan-Quinn criter 1385397
F-statistic 6246957 Durbin-Watson stat 2012623
Prob(F-statistic) 0000000
GARCH(11)
Observations
P-Value of Coefficient of GARCH(-1) is more less than 005
Thus we conclude volatility is of GARCH kind
Thus we do not check for ARCH We check for the best form of GARCH that fits the data
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0305
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 32 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0018850 0003353 5622112 00000
AR(1) -0141277 0075434 -1872843 00611
MA(1) 0242258 0073980 3274633 00011
Variance Equation
C 0000213 329E-05 6465720 00000
RESID(-1)^2 0066426 0001506 4410476 00000
GARCH(-1) 0939441 0001222 7689159 00000
R-squared -0022203 Mean dependent var 0111371
Adjusted R-squared -0022330 SD dependent var 7606298
SE of regression 7690752 Akaike info criterion 3800477
Sum squared resid 9512719 Schwarz criterion 3803344
Log likelihood -3056124 Hannan-Quinn criter 3801425
Durbin-Watson stat 2322505
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(12)
Observations
the coefficients of RESID(-1)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0312
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 33 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0019100 0003404 5610811 00000
AR(1) -0137936 0072289 -1908117 00564
MA(1) 0242221 0070576 3432082 00006
Variance Equation
C 0000161 262E-05 6145381 00000
RESID(-1)^2 0107198 0004442 2413017 00000
RESID(-2)^2 -0050822 0005119 -9928459 00000
GARCH(-1) 0948322 0001527 6212227 00000
R-squared -0023299 Mean dependent var 0111371
Adjusted R-squared -0023427 SD dependent var 7606298
SE of regression 7694877 Akaike info criterion 3798666
Sum squared resid 9522926 Schwarz criterion 3802010
Log likelihood -3054567 Hannan-Quinn criter 3799772
Durbin-Watson stat 2328892
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(22)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0319
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1) + C(8)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
Checking for Trend
Observations
Both trend and intercept terms are significant
Therefore we proceed to test for unit root using ADF test with trend and intercept
Dependent Variable P
Method Least Squares
Date 121113 Time 1132
Sample 1031950 12092013
Included observations 16088
Variable Coefficient Std Error t-Statistic Prob
C -3072818 3871625 -7936765 00000
TREND 0092369 0000417 2215926 00000
R-squared 0753242 Mean dependent var 4356903
Adjusted R-squared 0753226 SD dependent var 4942939
SE of regression 2455470 Akaike info criterion 1384498
Sum squared resid 970E+08 Schwarz criterion 1384593
Log likelihood -1113670 Hannan-Quinn criter 1384529
F-statistic 4910329 Durbin-Watson stat 0000960
Prob(F-statistic) 0000000
ADF test (series p)
Observations
P-Valuegt005
Null hypothesis rejected Unit root present (as expected)
First difference calculated dp=p-p(-1)
Null Hypothesis P has a unit root
Exogenous Constant Linear Trend
Lag Length 21 (Automatic - based on SIC maxlag=42)
t-Statistic Prob
Augmented Dickey-Fuller test statistic -0874981 09572
Test critical values 1 level -3958605
5 level -3410082
10 level -3126769
MacKinnon (1996) one-sided p-values
Variable Coefficient Std Error t-Statistic Prob
P(-1) -0000213 0000243 -0874981 03816
D(P(-1)) -0065128 0007889 -8255719 00000
D(P(-2)) -0042691 0007907 -5399168 00000
D(P(-3)) -0017087 0007914 -2159080 00309
D(P(-4)) -0009753 0007906 -1233615 02174
D(P(-5)) -0038168 0007906 -4827940 00000
D(P(-6)) -0012548 0007909 -1586680 01126
D(P(-7)) -0030904 0007907 -3908456 00001
D(P(-8)) 0007349 0007909 0929223 03528
D(P(-9)) -0004196 0007908 -0530685 05956
D(P(-10)) 0032448 0007898 4108179 00000
D(P(-11)) -0011639 0007902 -1472910 01408
D(P(-12)) 0048758 0007900 6172055 00000
D(P(-13)) 0021209 0007909 2681615 00073
D(P(-14)) -0019073 0007911 -2410950 00159
D(P(-15)) -0024191 0007909 -3058820 00022
D(P(-16)) 0027656 0007911 3496115 00005
D(P(-17)) 0018294 0007908 2313352 00207
D(P(-18)) -0047570 0007910 -6014067 00000
D(P(-19)) 0008477 0007918 1070654 02843
D(P(-20)) 817E-05 0007911 0010329 09918
D(P(-21)) -0046335 0007896 -5868435 00000
C -0145949 0140840 -1036276 03001
TREND(1031950) 463E-05 259E-05 1790536 00734
R-squared 0020029 Mean dependent var 0111498
Adjusted R-squared 0018624 SD dependent var 7611029
SE of regression 7539821 Akaike info criterion 6879767
Sum squared resid 9119701 Schwarz criterion 6891246
Log likelihood -5524117 Hannan-Quinn criter 6883563
F-statistic 1425548 Durbin-Watson stat 1999061
Prob(F-statistic) 0000000
Plot of Series dp (after first differencing of p)
Observations
bull Apparently stationarybull Conditionally
heteroskedasticbull Unit Root test required
for confirmation of stationarity
-120
-80
-40
0
40
80
120
55 60 65 70 75 80 85 90 95 00 05 10
DP
ADF test (series dp)
Observations
P-Valueltlt005 =gt Null hypothesis accepted ndash no unit root in the series dp
It is established that series dp is stationary We try to model it as an ARMA(pq) process by Box Jenkinrsquosmethod
Null Hypothesis P has a unit root
Exogenous Constant Linear Trend
Lag Length 20 (Automatic - based on SIC maxlag=42)
t-Statistic Prob
Augmented Dickey-Fuller test statistic -2906799 00000
Test critical values 1 level -3958605
5 level -3410082
10 level -3126769
MacKinnon (1996) one-sided p-values
Variable Coefficient Std Error t-Statistic Prob
DP(-1) -1208127 0041562 -2906799 00000
D(DP(-1)) 0142840 0040571 3520791 00004
D(DP(-2)) 0099996 0039534 2529330 00114
D(DP(-3)) 0082756 0038448 2152420 00314
D(DP(-4)) 0072855 0037418 1947064 00515
D(DP(-5)) 0034538 0036323 0950846 03417
D(DP(-6)) 0021837 0035136 0621521 05343
D(DP(-7)) -0009215 0033957 -0271356 07861
D(DP(-8)) -0002010 0032778 -0061307 09511
D(DP(-9)) -0006353 0031506 -0201653 08402
D(DP(-10)) 0025942 0030062 0862927 03882
D(DP(-11)) 0014150 0028576 0495183 06205
D(DP(-12)) 0062754 0026917 2331407 00197
D(DP(-13)) 0083808 0025169 3329787 00009
D(DP(-14)) 0064577 0023271 2774993 00055
D(DP(-15)) 0040232 0021310 1887948 00591
D(DP(-16)) 0067737 0019191 3529544 00004
D(DP(-17)) 0085886 0016976 5059121 00000
D(DP(-18)) 0038172 0014478 2636591 00084
D(DP(-19)) 0046510 0011535 4032085 00001
D(DP(-20)) 0046458 0007894 5884990 00000
C -0080367 0119239 -0674003 05003
TREND(1031950) 267E-05 129E-05 2074122 00381
R-squared 0539061 Mean dependent var 0000204
Adjusted R-squared 0538429 SD dependent var 1109785
SE of regression 7539766 Akaike info criterion 6879690
Sum squared resid 9120137 Schwarz criterion 6890691
Log likelihood -5524155 Hannan-Quinn criter 6883328
F-statistic 8528214 Durbin-Watson stat 1999071
Prob(F-statistic) 0000000
Correlogram analysis (series dp)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
This hints at ARMA(pq) model and not AR(p) or MA(q) model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0063 -0063 -799058 64008 0
2 -0039 -0043 -545389 87923 0
3 -0007 -0012 -152201 8871 0
4 -0001 -0004 -050734 88726 0
5 -004 -0041 -520022 11425 0
6 -0006 -0012 -152201 11482 0
7 -0026 -0031 -393187 1255 0
8 0009 0004 0507338 12688 0
9 -0006 -0008 -101468 12738 0
10 0034 0032 4058705 14622 0
11 -002 -0017 -215619 15252 0
12 005 0048 6088058 19218 0
13 0019 0025 3170863 19781 0
14 -0023 -0017 -215619 20648 0
15 -0029 -0026 -32977 2198 0
16 0035 003 3805036 23985 0
17 0014 0022 279036 24304 0
18 -0052 -0048 -608806 28721 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0013 001 1268345 28988 0
20 0008 0003 0380504 29104 0
21 -0048 -0046 -583439 32879 0
22 0018 001 1268345 3343 0
23 0007 0004 0507338 33519 0
24 0003 0004 0507338 33536 0
25 -001 -0013 -164885 33703 0
26 -0009 -0014 -177568 33843 0
27 0033 0035 4439209 35586 0
28 0007 0009 1141511 35656 0
29 0019 0019 2409856 36249 0
30 0005 0013 1648849 36284 0
31 -0003 0006 0761007 36299 0
32 0001 -0002 -025367 363 0
33 -0016 -0014 -177568 36714 0
34 -0066 -0061 -773691 4372 0
35 0006 -0004 -050734 4377 0
36 0018 0009 1141511 44317 0
ARMA Estimation ndash Information Criteria
Information Criteria(SIC)
MA(0) MA(1) MA(2)
AR(0) AIC 6895772
6896728
6891609
6892565
6890118
6891551SIC
AR(1) AIC 6892031
6892986
6889474
6890907
6889569
6891480SIC
AR(2) AIC 6890391
6891824
6889632
6891544
6889077
6891466SIC
Observations
ARMA(11) appears to be the best fitted model
The suitability has been in line with both AIC and SIC
Sufficient data present in the model so SIC is the best criteria
Estimation of the model
Observations
As expected all the terms are significant in the ARMA(11) model
The Durbin Watson Statistic is asymp 2 This suggests that the autocorrelation among residuals may be small
This is in fact confirmed by the correlogramQ-statistics of the residuals
Dependent Variable DP
Method Least Squares
Date 121113 Time 0144
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 8 iterations
MA Backcast 1041950
Variable Coefficient Std Error t-Statistic Prob
C 0111277 0049883 2230768 00257
AR(1) 0607851 0056980 1066776 00000
MA(1) -0672801 0053089 -1267311 00000
R-squared 0006645 Mean dependent var 0111371
Adjusted R-squared 0006522 SD dependent var 7606298
SE of regression 7581454 Akaike info criterion 6889474
Sum squared resid 9244258 Schwarz criterion 6890907
Log likelihood -5540904 Hannan-Quinn criter 6889948
F-statistic 5379525 Durbin-Watson stat 2004706
Prob(F-statistic) 0000000
Inverted AR Roots 61
Inverted MA Roots 67
Correlogram Q statistic (of residuals)
Observations
All ACF values from 5th lag onwards significant
No definite threshold found such that PACF vanishes above that threshold
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0002 -0002 -025366 00897
2 -0001 -0001 -012683 01221
3 0015 0015 1902459 35484 006
4 0011 0011 1395137 55919 0061
5 -0031 -0031 -393175 20833 0
6 -0002 -0003 -038049 20928 0
7 -0022 -0022 -279027 28685 0
8 0011 0011 1395137 30563 0
9 -0002 -0001 -012683 30637 0
10 0036 0035 4439071 50953 0
11 -0014 -0014 -177563 54261 0
12 005 0049 6214699 95031 0
13 002 002 2536612 10169 0
14 -0021 -0021 -266344 1086 0
15 -0027 -0025 -317076 12008 0
16 0033 003 3804918 13763 0
17 0012 0017 215612 14007 0
18 -0051 -005 -634153 18163 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 001 0011 1395137 18319 0
20 0006 0001 0126831 18372 0
21 -0047 -0045 -570738 21954 0
22 0016 0014 1775628 22378 0
23 0007 0006 0760984 22456 0
24 0003 0004 0507322 22474 0
25 -0009 -0011 -139514 226 0
26 -0007 -001 -126831 22681 0
27 0033 0037 4692732 24466 0
28 0009 0009 1141475 24598 0
29 002 0017 215612 25252 0
30 0005 0009 1141475 25299 0
31 -0003 0002 0253661 25318 0
32 -0002 -0006 -076098 25322 0
33 -0019 -0018 -228295 25902 0
34 -0066 -006 -760984 32879 0
35 0003 0001 0126831 32889 0
36 0017 0013 1648798 33334 0
Correlogram Q statistic (of squared residuals)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
However square of the lag values are larger (in ACF)
This suggests ARCH type modelling is more appropriate
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0236 0236 2993202 89426
2 0387 035 4439071 32984
3 0243 0123 1560016 42517 0
4 0297 0139 1762945 56715 0
5 0323 0189 2397098 73461 0
6 0311 0137 1737579 88992 0
7 0303 0098 124294 10377 0
8 0295 0089 1128792 11777 0
9 0283 0067 849765 13068 0
10 0311 0094 1192208 14621 0
11 0307 0089 1128792 16138 0
12 0287 0045 5707377 17464 0
13 0235 -0027 -342443 18356 0
14 022 -0039 -494639 19138 0
15 0237 0001 0126831 20041 0
16 0271 0051 646836 21225 0
17 0269 0045 5707377 22387 0
18 0272 0047 5961038 23577 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0209 -0026 -32976 24280 0
20 0253 003 3804918 25307 0
21 0289 0105 1331721 26653 0
22 0223 -0013 -16488 27456 0
23 0242 -0004 -050732 28402 0
24 0191 -002 -253661 28991 0
25 0209 -0012 -152197 29696 0
26 0205 -0013 -16488 30372 0
27 0276 0079 1001962 31600 0
28 0233 0023 2917104 32474 0
29 0214 -002 -253661 33215 0
30 0185 -0022 -279027 33768 0
31 0194 -0006 -076098 34373 0
32 0251 0067 849765 35388 0
33 02 -0005 -063415 36032 0
34 0232 003 3804918 36902 0
35 0169 -0018 -228295 37360 0
36 0208 0013 1648798 38055 0
ARCH Heteroskedasticity test
Observations
We reaffirm the ARCH effect using the Heteroskedasticity ndash ARCH test
Clearly all residue lags seem significant (uptolag 9)
This hints that we may have to opt for a GARCH(pq) model
Hence we have not tested for pure ARCH models any further
Heteroskedasticity Test ARCH
F-statistic 6246957 Prob F(916067) 00000
ObsR-squared 4167458 Prob Chi-Square(9) 00000
Variable Coefficient Std Error t-Statistic Prob
C 1160923 2045328 5675975 00000
RESID^2(-1) 0013088 0007872 1662714 00964
RESID^2(-2) 0213112 0007842 2717408 00000
RESID^2(-3) 0016465 0007996 2059229 00395
RESID^2(-4) 0062821 0007948 7903630 00000
RESID^2(-5) 0147233 0007879 1868761 00000
RESID^2(-6) 0111116 0007948 1397978 00000
RESID^2(-7) 0080432 0007996 1005965 00000
RESID^2(-8) 0087330 0007842 1113558 00000
RESID^2(-9) 0066907 0007872 8499328 00000
R-squared 0259219 Mean dependent var 5749988
Adjusted R-squared 0258804 SD dependent var 2862004
SE of regression 2463978 Akaike info criterion 1385239
Sum squared resid 975E+08 Schwarz criterion 1385717
Log likelihood -1113425 Hannan-Quinn criter 1385397
F-statistic 6246957 Durbin-Watson stat 2012623
Prob(F-statistic) 0000000
GARCH(11)
Observations
P-Value of Coefficient of GARCH(-1) is more less than 005
Thus we conclude volatility is of GARCH kind
Thus we do not check for ARCH We check for the best form of GARCH that fits the data
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0305
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 32 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0018850 0003353 5622112 00000
AR(1) -0141277 0075434 -1872843 00611
MA(1) 0242258 0073980 3274633 00011
Variance Equation
C 0000213 329E-05 6465720 00000
RESID(-1)^2 0066426 0001506 4410476 00000
GARCH(-1) 0939441 0001222 7689159 00000
R-squared -0022203 Mean dependent var 0111371
Adjusted R-squared -0022330 SD dependent var 7606298
SE of regression 7690752 Akaike info criterion 3800477
Sum squared resid 9512719 Schwarz criterion 3803344
Log likelihood -3056124 Hannan-Quinn criter 3801425
Durbin-Watson stat 2322505
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(12)
Observations
the coefficients of RESID(-1)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0312
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 33 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0019100 0003404 5610811 00000
AR(1) -0137936 0072289 -1908117 00564
MA(1) 0242221 0070576 3432082 00006
Variance Equation
C 0000161 262E-05 6145381 00000
RESID(-1)^2 0107198 0004442 2413017 00000
RESID(-2)^2 -0050822 0005119 -9928459 00000
GARCH(-1) 0948322 0001527 6212227 00000
R-squared -0023299 Mean dependent var 0111371
Adjusted R-squared -0023427 SD dependent var 7606298
SE of regression 7694877 Akaike info criterion 3798666
Sum squared resid 9522926 Schwarz criterion 3802010
Log likelihood -3054567 Hannan-Quinn criter 3799772
Durbin-Watson stat 2328892
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(22)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0319
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1) + C(8)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
ADF test (series p)
Observations
P-Valuegt005
Null hypothesis rejected Unit root present (as expected)
First difference calculated dp=p-p(-1)
Null Hypothesis P has a unit root
Exogenous Constant Linear Trend
Lag Length 21 (Automatic - based on SIC maxlag=42)
t-Statistic Prob
Augmented Dickey-Fuller test statistic -0874981 09572
Test critical values 1 level -3958605
5 level -3410082
10 level -3126769
MacKinnon (1996) one-sided p-values
Variable Coefficient Std Error t-Statistic Prob
P(-1) -0000213 0000243 -0874981 03816
D(P(-1)) -0065128 0007889 -8255719 00000
D(P(-2)) -0042691 0007907 -5399168 00000
D(P(-3)) -0017087 0007914 -2159080 00309
D(P(-4)) -0009753 0007906 -1233615 02174
D(P(-5)) -0038168 0007906 -4827940 00000
D(P(-6)) -0012548 0007909 -1586680 01126
D(P(-7)) -0030904 0007907 -3908456 00001
D(P(-8)) 0007349 0007909 0929223 03528
D(P(-9)) -0004196 0007908 -0530685 05956
D(P(-10)) 0032448 0007898 4108179 00000
D(P(-11)) -0011639 0007902 -1472910 01408
D(P(-12)) 0048758 0007900 6172055 00000
D(P(-13)) 0021209 0007909 2681615 00073
D(P(-14)) -0019073 0007911 -2410950 00159
D(P(-15)) -0024191 0007909 -3058820 00022
D(P(-16)) 0027656 0007911 3496115 00005
D(P(-17)) 0018294 0007908 2313352 00207
D(P(-18)) -0047570 0007910 -6014067 00000
D(P(-19)) 0008477 0007918 1070654 02843
D(P(-20)) 817E-05 0007911 0010329 09918
D(P(-21)) -0046335 0007896 -5868435 00000
C -0145949 0140840 -1036276 03001
TREND(1031950) 463E-05 259E-05 1790536 00734
R-squared 0020029 Mean dependent var 0111498
Adjusted R-squared 0018624 SD dependent var 7611029
SE of regression 7539821 Akaike info criterion 6879767
Sum squared resid 9119701 Schwarz criterion 6891246
Log likelihood -5524117 Hannan-Quinn criter 6883563
F-statistic 1425548 Durbin-Watson stat 1999061
Prob(F-statistic) 0000000
Plot of Series dp (after first differencing of p)
Observations
bull Apparently stationarybull Conditionally
heteroskedasticbull Unit Root test required
for confirmation of stationarity
-120
-80
-40
0
40
80
120
55 60 65 70 75 80 85 90 95 00 05 10
DP
ADF test (series dp)
Observations
P-Valueltlt005 =gt Null hypothesis accepted ndash no unit root in the series dp
It is established that series dp is stationary We try to model it as an ARMA(pq) process by Box Jenkinrsquosmethod
Null Hypothesis P has a unit root
Exogenous Constant Linear Trend
Lag Length 20 (Automatic - based on SIC maxlag=42)
t-Statistic Prob
Augmented Dickey-Fuller test statistic -2906799 00000
Test critical values 1 level -3958605
5 level -3410082
10 level -3126769
MacKinnon (1996) one-sided p-values
Variable Coefficient Std Error t-Statistic Prob
DP(-1) -1208127 0041562 -2906799 00000
D(DP(-1)) 0142840 0040571 3520791 00004
D(DP(-2)) 0099996 0039534 2529330 00114
D(DP(-3)) 0082756 0038448 2152420 00314
D(DP(-4)) 0072855 0037418 1947064 00515
D(DP(-5)) 0034538 0036323 0950846 03417
D(DP(-6)) 0021837 0035136 0621521 05343
D(DP(-7)) -0009215 0033957 -0271356 07861
D(DP(-8)) -0002010 0032778 -0061307 09511
D(DP(-9)) -0006353 0031506 -0201653 08402
D(DP(-10)) 0025942 0030062 0862927 03882
D(DP(-11)) 0014150 0028576 0495183 06205
D(DP(-12)) 0062754 0026917 2331407 00197
D(DP(-13)) 0083808 0025169 3329787 00009
D(DP(-14)) 0064577 0023271 2774993 00055
D(DP(-15)) 0040232 0021310 1887948 00591
D(DP(-16)) 0067737 0019191 3529544 00004
D(DP(-17)) 0085886 0016976 5059121 00000
D(DP(-18)) 0038172 0014478 2636591 00084
D(DP(-19)) 0046510 0011535 4032085 00001
D(DP(-20)) 0046458 0007894 5884990 00000
C -0080367 0119239 -0674003 05003
TREND(1031950) 267E-05 129E-05 2074122 00381
R-squared 0539061 Mean dependent var 0000204
Adjusted R-squared 0538429 SD dependent var 1109785
SE of regression 7539766 Akaike info criterion 6879690
Sum squared resid 9120137 Schwarz criterion 6890691
Log likelihood -5524155 Hannan-Quinn criter 6883328
F-statistic 8528214 Durbin-Watson stat 1999071
Prob(F-statistic) 0000000
Correlogram analysis (series dp)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
This hints at ARMA(pq) model and not AR(p) or MA(q) model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0063 -0063 -799058 64008 0
2 -0039 -0043 -545389 87923 0
3 -0007 -0012 -152201 8871 0
4 -0001 -0004 -050734 88726 0
5 -004 -0041 -520022 11425 0
6 -0006 -0012 -152201 11482 0
7 -0026 -0031 -393187 1255 0
8 0009 0004 0507338 12688 0
9 -0006 -0008 -101468 12738 0
10 0034 0032 4058705 14622 0
11 -002 -0017 -215619 15252 0
12 005 0048 6088058 19218 0
13 0019 0025 3170863 19781 0
14 -0023 -0017 -215619 20648 0
15 -0029 -0026 -32977 2198 0
16 0035 003 3805036 23985 0
17 0014 0022 279036 24304 0
18 -0052 -0048 -608806 28721 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0013 001 1268345 28988 0
20 0008 0003 0380504 29104 0
21 -0048 -0046 -583439 32879 0
22 0018 001 1268345 3343 0
23 0007 0004 0507338 33519 0
24 0003 0004 0507338 33536 0
25 -001 -0013 -164885 33703 0
26 -0009 -0014 -177568 33843 0
27 0033 0035 4439209 35586 0
28 0007 0009 1141511 35656 0
29 0019 0019 2409856 36249 0
30 0005 0013 1648849 36284 0
31 -0003 0006 0761007 36299 0
32 0001 -0002 -025367 363 0
33 -0016 -0014 -177568 36714 0
34 -0066 -0061 -773691 4372 0
35 0006 -0004 -050734 4377 0
36 0018 0009 1141511 44317 0
ARMA Estimation ndash Information Criteria
Information Criteria(SIC)
MA(0) MA(1) MA(2)
AR(0) AIC 6895772
6896728
6891609
6892565
6890118
6891551SIC
AR(1) AIC 6892031
6892986
6889474
6890907
6889569
6891480SIC
AR(2) AIC 6890391
6891824
6889632
6891544
6889077
6891466SIC
Observations
ARMA(11) appears to be the best fitted model
The suitability has been in line with both AIC and SIC
Sufficient data present in the model so SIC is the best criteria
Estimation of the model
Observations
As expected all the terms are significant in the ARMA(11) model
The Durbin Watson Statistic is asymp 2 This suggests that the autocorrelation among residuals may be small
This is in fact confirmed by the correlogramQ-statistics of the residuals
Dependent Variable DP
Method Least Squares
Date 121113 Time 0144
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 8 iterations
MA Backcast 1041950
Variable Coefficient Std Error t-Statistic Prob
C 0111277 0049883 2230768 00257
AR(1) 0607851 0056980 1066776 00000
MA(1) -0672801 0053089 -1267311 00000
R-squared 0006645 Mean dependent var 0111371
Adjusted R-squared 0006522 SD dependent var 7606298
SE of regression 7581454 Akaike info criterion 6889474
Sum squared resid 9244258 Schwarz criterion 6890907
Log likelihood -5540904 Hannan-Quinn criter 6889948
F-statistic 5379525 Durbin-Watson stat 2004706
Prob(F-statistic) 0000000
Inverted AR Roots 61
Inverted MA Roots 67
Correlogram Q statistic (of residuals)
Observations
All ACF values from 5th lag onwards significant
No definite threshold found such that PACF vanishes above that threshold
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0002 -0002 -025366 00897
2 -0001 -0001 -012683 01221
3 0015 0015 1902459 35484 006
4 0011 0011 1395137 55919 0061
5 -0031 -0031 -393175 20833 0
6 -0002 -0003 -038049 20928 0
7 -0022 -0022 -279027 28685 0
8 0011 0011 1395137 30563 0
9 -0002 -0001 -012683 30637 0
10 0036 0035 4439071 50953 0
11 -0014 -0014 -177563 54261 0
12 005 0049 6214699 95031 0
13 002 002 2536612 10169 0
14 -0021 -0021 -266344 1086 0
15 -0027 -0025 -317076 12008 0
16 0033 003 3804918 13763 0
17 0012 0017 215612 14007 0
18 -0051 -005 -634153 18163 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 001 0011 1395137 18319 0
20 0006 0001 0126831 18372 0
21 -0047 -0045 -570738 21954 0
22 0016 0014 1775628 22378 0
23 0007 0006 0760984 22456 0
24 0003 0004 0507322 22474 0
25 -0009 -0011 -139514 226 0
26 -0007 -001 -126831 22681 0
27 0033 0037 4692732 24466 0
28 0009 0009 1141475 24598 0
29 002 0017 215612 25252 0
30 0005 0009 1141475 25299 0
31 -0003 0002 0253661 25318 0
32 -0002 -0006 -076098 25322 0
33 -0019 -0018 -228295 25902 0
34 -0066 -006 -760984 32879 0
35 0003 0001 0126831 32889 0
36 0017 0013 1648798 33334 0
Correlogram Q statistic (of squared residuals)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
However square of the lag values are larger (in ACF)
This suggests ARCH type modelling is more appropriate
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0236 0236 2993202 89426
2 0387 035 4439071 32984
3 0243 0123 1560016 42517 0
4 0297 0139 1762945 56715 0
5 0323 0189 2397098 73461 0
6 0311 0137 1737579 88992 0
7 0303 0098 124294 10377 0
8 0295 0089 1128792 11777 0
9 0283 0067 849765 13068 0
10 0311 0094 1192208 14621 0
11 0307 0089 1128792 16138 0
12 0287 0045 5707377 17464 0
13 0235 -0027 -342443 18356 0
14 022 -0039 -494639 19138 0
15 0237 0001 0126831 20041 0
16 0271 0051 646836 21225 0
17 0269 0045 5707377 22387 0
18 0272 0047 5961038 23577 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0209 -0026 -32976 24280 0
20 0253 003 3804918 25307 0
21 0289 0105 1331721 26653 0
22 0223 -0013 -16488 27456 0
23 0242 -0004 -050732 28402 0
24 0191 -002 -253661 28991 0
25 0209 -0012 -152197 29696 0
26 0205 -0013 -16488 30372 0
27 0276 0079 1001962 31600 0
28 0233 0023 2917104 32474 0
29 0214 -002 -253661 33215 0
30 0185 -0022 -279027 33768 0
31 0194 -0006 -076098 34373 0
32 0251 0067 849765 35388 0
33 02 -0005 -063415 36032 0
34 0232 003 3804918 36902 0
35 0169 -0018 -228295 37360 0
36 0208 0013 1648798 38055 0
ARCH Heteroskedasticity test
Observations
We reaffirm the ARCH effect using the Heteroskedasticity ndash ARCH test
Clearly all residue lags seem significant (uptolag 9)
This hints that we may have to opt for a GARCH(pq) model
Hence we have not tested for pure ARCH models any further
Heteroskedasticity Test ARCH
F-statistic 6246957 Prob F(916067) 00000
ObsR-squared 4167458 Prob Chi-Square(9) 00000
Variable Coefficient Std Error t-Statistic Prob
C 1160923 2045328 5675975 00000
RESID^2(-1) 0013088 0007872 1662714 00964
RESID^2(-2) 0213112 0007842 2717408 00000
RESID^2(-3) 0016465 0007996 2059229 00395
RESID^2(-4) 0062821 0007948 7903630 00000
RESID^2(-5) 0147233 0007879 1868761 00000
RESID^2(-6) 0111116 0007948 1397978 00000
RESID^2(-7) 0080432 0007996 1005965 00000
RESID^2(-8) 0087330 0007842 1113558 00000
RESID^2(-9) 0066907 0007872 8499328 00000
R-squared 0259219 Mean dependent var 5749988
Adjusted R-squared 0258804 SD dependent var 2862004
SE of regression 2463978 Akaike info criterion 1385239
Sum squared resid 975E+08 Schwarz criterion 1385717
Log likelihood -1113425 Hannan-Quinn criter 1385397
F-statistic 6246957 Durbin-Watson stat 2012623
Prob(F-statistic) 0000000
GARCH(11)
Observations
P-Value of Coefficient of GARCH(-1) is more less than 005
Thus we conclude volatility is of GARCH kind
Thus we do not check for ARCH We check for the best form of GARCH that fits the data
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0305
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 32 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0018850 0003353 5622112 00000
AR(1) -0141277 0075434 -1872843 00611
MA(1) 0242258 0073980 3274633 00011
Variance Equation
C 0000213 329E-05 6465720 00000
RESID(-1)^2 0066426 0001506 4410476 00000
GARCH(-1) 0939441 0001222 7689159 00000
R-squared -0022203 Mean dependent var 0111371
Adjusted R-squared -0022330 SD dependent var 7606298
SE of regression 7690752 Akaike info criterion 3800477
Sum squared resid 9512719 Schwarz criterion 3803344
Log likelihood -3056124 Hannan-Quinn criter 3801425
Durbin-Watson stat 2322505
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(12)
Observations
the coefficients of RESID(-1)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0312
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 33 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0019100 0003404 5610811 00000
AR(1) -0137936 0072289 -1908117 00564
MA(1) 0242221 0070576 3432082 00006
Variance Equation
C 0000161 262E-05 6145381 00000
RESID(-1)^2 0107198 0004442 2413017 00000
RESID(-2)^2 -0050822 0005119 -9928459 00000
GARCH(-1) 0948322 0001527 6212227 00000
R-squared -0023299 Mean dependent var 0111371
Adjusted R-squared -0023427 SD dependent var 7606298
SE of regression 7694877 Akaike info criterion 3798666
Sum squared resid 9522926 Schwarz criterion 3802010
Log likelihood -3054567 Hannan-Quinn criter 3799772
Durbin-Watson stat 2328892
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(22)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0319
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1) + C(8)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
Plot of Series dp (after first differencing of p)
Observations
bull Apparently stationarybull Conditionally
heteroskedasticbull Unit Root test required
for confirmation of stationarity
-120
-80
-40
0
40
80
120
55 60 65 70 75 80 85 90 95 00 05 10
DP
ADF test (series dp)
Observations
P-Valueltlt005 =gt Null hypothesis accepted ndash no unit root in the series dp
It is established that series dp is stationary We try to model it as an ARMA(pq) process by Box Jenkinrsquosmethod
Null Hypothesis P has a unit root
Exogenous Constant Linear Trend
Lag Length 20 (Automatic - based on SIC maxlag=42)
t-Statistic Prob
Augmented Dickey-Fuller test statistic -2906799 00000
Test critical values 1 level -3958605
5 level -3410082
10 level -3126769
MacKinnon (1996) one-sided p-values
Variable Coefficient Std Error t-Statistic Prob
DP(-1) -1208127 0041562 -2906799 00000
D(DP(-1)) 0142840 0040571 3520791 00004
D(DP(-2)) 0099996 0039534 2529330 00114
D(DP(-3)) 0082756 0038448 2152420 00314
D(DP(-4)) 0072855 0037418 1947064 00515
D(DP(-5)) 0034538 0036323 0950846 03417
D(DP(-6)) 0021837 0035136 0621521 05343
D(DP(-7)) -0009215 0033957 -0271356 07861
D(DP(-8)) -0002010 0032778 -0061307 09511
D(DP(-9)) -0006353 0031506 -0201653 08402
D(DP(-10)) 0025942 0030062 0862927 03882
D(DP(-11)) 0014150 0028576 0495183 06205
D(DP(-12)) 0062754 0026917 2331407 00197
D(DP(-13)) 0083808 0025169 3329787 00009
D(DP(-14)) 0064577 0023271 2774993 00055
D(DP(-15)) 0040232 0021310 1887948 00591
D(DP(-16)) 0067737 0019191 3529544 00004
D(DP(-17)) 0085886 0016976 5059121 00000
D(DP(-18)) 0038172 0014478 2636591 00084
D(DP(-19)) 0046510 0011535 4032085 00001
D(DP(-20)) 0046458 0007894 5884990 00000
C -0080367 0119239 -0674003 05003
TREND(1031950) 267E-05 129E-05 2074122 00381
R-squared 0539061 Mean dependent var 0000204
Adjusted R-squared 0538429 SD dependent var 1109785
SE of regression 7539766 Akaike info criterion 6879690
Sum squared resid 9120137 Schwarz criterion 6890691
Log likelihood -5524155 Hannan-Quinn criter 6883328
F-statistic 8528214 Durbin-Watson stat 1999071
Prob(F-statistic) 0000000
Correlogram analysis (series dp)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
This hints at ARMA(pq) model and not AR(p) or MA(q) model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0063 -0063 -799058 64008 0
2 -0039 -0043 -545389 87923 0
3 -0007 -0012 -152201 8871 0
4 -0001 -0004 -050734 88726 0
5 -004 -0041 -520022 11425 0
6 -0006 -0012 -152201 11482 0
7 -0026 -0031 -393187 1255 0
8 0009 0004 0507338 12688 0
9 -0006 -0008 -101468 12738 0
10 0034 0032 4058705 14622 0
11 -002 -0017 -215619 15252 0
12 005 0048 6088058 19218 0
13 0019 0025 3170863 19781 0
14 -0023 -0017 -215619 20648 0
15 -0029 -0026 -32977 2198 0
16 0035 003 3805036 23985 0
17 0014 0022 279036 24304 0
18 -0052 -0048 -608806 28721 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0013 001 1268345 28988 0
20 0008 0003 0380504 29104 0
21 -0048 -0046 -583439 32879 0
22 0018 001 1268345 3343 0
23 0007 0004 0507338 33519 0
24 0003 0004 0507338 33536 0
25 -001 -0013 -164885 33703 0
26 -0009 -0014 -177568 33843 0
27 0033 0035 4439209 35586 0
28 0007 0009 1141511 35656 0
29 0019 0019 2409856 36249 0
30 0005 0013 1648849 36284 0
31 -0003 0006 0761007 36299 0
32 0001 -0002 -025367 363 0
33 -0016 -0014 -177568 36714 0
34 -0066 -0061 -773691 4372 0
35 0006 -0004 -050734 4377 0
36 0018 0009 1141511 44317 0
ARMA Estimation ndash Information Criteria
Information Criteria(SIC)
MA(0) MA(1) MA(2)
AR(0) AIC 6895772
6896728
6891609
6892565
6890118
6891551SIC
AR(1) AIC 6892031
6892986
6889474
6890907
6889569
6891480SIC
AR(2) AIC 6890391
6891824
6889632
6891544
6889077
6891466SIC
Observations
ARMA(11) appears to be the best fitted model
The suitability has been in line with both AIC and SIC
Sufficient data present in the model so SIC is the best criteria
Estimation of the model
Observations
As expected all the terms are significant in the ARMA(11) model
The Durbin Watson Statistic is asymp 2 This suggests that the autocorrelation among residuals may be small
This is in fact confirmed by the correlogramQ-statistics of the residuals
Dependent Variable DP
Method Least Squares
Date 121113 Time 0144
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 8 iterations
MA Backcast 1041950
Variable Coefficient Std Error t-Statistic Prob
C 0111277 0049883 2230768 00257
AR(1) 0607851 0056980 1066776 00000
MA(1) -0672801 0053089 -1267311 00000
R-squared 0006645 Mean dependent var 0111371
Adjusted R-squared 0006522 SD dependent var 7606298
SE of regression 7581454 Akaike info criterion 6889474
Sum squared resid 9244258 Schwarz criterion 6890907
Log likelihood -5540904 Hannan-Quinn criter 6889948
F-statistic 5379525 Durbin-Watson stat 2004706
Prob(F-statistic) 0000000
Inverted AR Roots 61
Inverted MA Roots 67
Correlogram Q statistic (of residuals)
Observations
All ACF values from 5th lag onwards significant
No definite threshold found such that PACF vanishes above that threshold
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0002 -0002 -025366 00897
2 -0001 -0001 -012683 01221
3 0015 0015 1902459 35484 006
4 0011 0011 1395137 55919 0061
5 -0031 -0031 -393175 20833 0
6 -0002 -0003 -038049 20928 0
7 -0022 -0022 -279027 28685 0
8 0011 0011 1395137 30563 0
9 -0002 -0001 -012683 30637 0
10 0036 0035 4439071 50953 0
11 -0014 -0014 -177563 54261 0
12 005 0049 6214699 95031 0
13 002 002 2536612 10169 0
14 -0021 -0021 -266344 1086 0
15 -0027 -0025 -317076 12008 0
16 0033 003 3804918 13763 0
17 0012 0017 215612 14007 0
18 -0051 -005 -634153 18163 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 001 0011 1395137 18319 0
20 0006 0001 0126831 18372 0
21 -0047 -0045 -570738 21954 0
22 0016 0014 1775628 22378 0
23 0007 0006 0760984 22456 0
24 0003 0004 0507322 22474 0
25 -0009 -0011 -139514 226 0
26 -0007 -001 -126831 22681 0
27 0033 0037 4692732 24466 0
28 0009 0009 1141475 24598 0
29 002 0017 215612 25252 0
30 0005 0009 1141475 25299 0
31 -0003 0002 0253661 25318 0
32 -0002 -0006 -076098 25322 0
33 -0019 -0018 -228295 25902 0
34 -0066 -006 -760984 32879 0
35 0003 0001 0126831 32889 0
36 0017 0013 1648798 33334 0
Correlogram Q statistic (of squared residuals)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
However square of the lag values are larger (in ACF)
This suggests ARCH type modelling is more appropriate
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0236 0236 2993202 89426
2 0387 035 4439071 32984
3 0243 0123 1560016 42517 0
4 0297 0139 1762945 56715 0
5 0323 0189 2397098 73461 0
6 0311 0137 1737579 88992 0
7 0303 0098 124294 10377 0
8 0295 0089 1128792 11777 0
9 0283 0067 849765 13068 0
10 0311 0094 1192208 14621 0
11 0307 0089 1128792 16138 0
12 0287 0045 5707377 17464 0
13 0235 -0027 -342443 18356 0
14 022 -0039 -494639 19138 0
15 0237 0001 0126831 20041 0
16 0271 0051 646836 21225 0
17 0269 0045 5707377 22387 0
18 0272 0047 5961038 23577 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0209 -0026 -32976 24280 0
20 0253 003 3804918 25307 0
21 0289 0105 1331721 26653 0
22 0223 -0013 -16488 27456 0
23 0242 -0004 -050732 28402 0
24 0191 -002 -253661 28991 0
25 0209 -0012 -152197 29696 0
26 0205 -0013 -16488 30372 0
27 0276 0079 1001962 31600 0
28 0233 0023 2917104 32474 0
29 0214 -002 -253661 33215 0
30 0185 -0022 -279027 33768 0
31 0194 -0006 -076098 34373 0
32 0251 0067 849765 35388 0
33 02 -0005 -063415 36032 0
34 0232 003 3804918 36902 0
35 0169 -0018 -228295 37360 0
36 0208 0013 1648798 38055 0
ARCH Heteroskedasticity test
Observations
We reaffirm the ARCH effect using the Heteroskedasticity ndash ARCH test
Clearly all residue lags seem significant (uptolag 9)
This hints that we may have to opt for a GARCH(pq) model
Hence we have not tested for pure ARCH models any further
Heteroskedasticity Test ARCH
F-statistic 6246957 Prob F(916067) 00000
ObsR-squared 4167458 Prob Chi-Square(9) 00000
Variable Coefficient Std Error t-Statistic Prob
C 1160923 2045328 5675975 00000
RESID^2(-1) 0013088 0007872 1662714 00964
RESID^2(-2) 0213112 0007842 2717408 00000
RESID^2(-3) 0016465 0007996 2059229 00395
RESID^2(-4) 0062821 0007948 7903630 00000
RESID^2(-5) 0147233 0007879 1868761 00000
RESID^2(-6) 0111116 0007948 1397978 00000
RESID^2(-7) 0080432 0007996 1005965 00000
RESID^2(-8) 0087330 0007842 1113558 00000
RESID^2(-9) 0066907 0007872 8499328 00000
R-squared 0259219 Mean dependent var 5749988
Adjusted R-squared 0258804 SD dependent var 2862004
SE of regression 2463978 Akaike info criterion 1385239
Sum squared resid 975E+08 Schwarz criterion 1385717
Log likelihood -1113425 Hannan-Quinn criter 1385397
F-statistic 6246957 Durbin-Watson stat 2012623
Prob(F-statistic) 0000000
GARCH(11)
Observations
P-Value of Coefficient of GARCH(-1) is more less than 005
Thus we conclude volatility is of GARCH kind
Thus we do not check for ARCH We check for the best form of GARCH that fits the data
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0305
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 32 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0018850 0003353 5622112 00000
AR(1) -0141277 0075434 -1872843 00611
MA(1) 0242258 0073980 3274633 00011
Variance Equation
C 0000213 329E-05 6465720 00000
RESID(-1)^2 0066426 0001506 4410476 00000
GARCH(-1) 0939441 0001222 7689159 00000
R-squared -0022203 Mean dependent var 0111371
Adjusted R-squared -0022330 SD dependent var 7606298
SE of regression 7690752 Akaike info criterion 3800477
Sum squared resid 9512719 Schwarz criterion 3803344
Log likelihood -3056124 Hannan-Quinn criter 3801425
Durbin-Watson stat 2322505
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(12)
Observations
the coefficients of RESID(-1)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0312
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 33 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0019100 0003404 5610811 00000
AR(1) -0137936 0072289 -1908117 00564
MA(1) 0242221 0070576 3432082 00006
Variance Equation
C 0000161 262E-05 6145381 00000
RESID(-1)^2 0107198 0004442 2413017 00000
RESID(-2)^2 -0050822 0005119 -9928459 00000
GARCH(-1) 0948322 0001527 6212227 00000
R-squared -0023299 Mean dependent var 0111371
Adjusted R-squared -0023427 SD dependent var 7606298
SE of regression 7694877 Akaike info criterion 3798666
Sum squared resid 9522926 Schwarz criterion 3802010
Log likelihood -3054567 Hannan-Quinn criter 3799772
Durbin-Watson stat 2328892
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(22)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0319
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1) + C(8)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
ADF test (series dp)
Observations
P-Valueltlt005 =gt Null hypothesis accepted ndash no unit root in the series dp
It is established that series dp is stationary We try to model it as an ARMA(pq) process by Box Jenkinrsquosmethod
Null Hypothesis P has a unit root
Exogenous Constant Linear Trend
Lag Length 20 (Automatic - based on SIC maxlag=42)
t-Statistic Prob
Augmented Dickey-Fuller test statistic -2906799 00000
Test critical values 1 level -3958605
5 level -3410082
10 level -3126769
MacKinnon (1996) one-sided p-values
Variable Coefficient Std Error t-Statistic Prob
DP(-1) -1208127 0041562 -2906799 00000
D(DP(-1)) 0142840 0040571 3520791 00004
D(DP(-2)) 0099996 0039534 2529330 00114
D(DP(-3)) 0082756 0038448 2152420 00314
D(DP(-4)) 0072855 0037418 1947064 00515
D(DP(-5)) 0034538 0036323 0950846 03417
D(DP(-6)) 0021837 0035136 0621521 05343
D(DP(-7)) -0009215 0033957 -0271356 07861
D(DP(-8)) -0002010 0032778 -0061307 09511
D(DP(-9)) -0006353 0031506 -0201653 08402
D(DP(-10)) 0025942 0030062 0862927 03882
D(DP(-11)) 0014150 0028576 0495183 06205
D(DP(-12)) 0062754 0026917 2331407 00197
D(DP(-13)) 0083808 0025169 3329787 00009
D(DP(-14)) 0064577 0023271 2774993 00055
D(DP(-15)) 0040232 0021310 1887948 00591
D(DP(-16)) 0067737 0019191 3529544 00004
D(DP(-17)) 0085886 0016976 5059121 00000
D(DP(-18)) 0038172 0014478 2636591 00084
D(DP(-19)) 0046510 0011535 4032085 00001
D(DP(-20)) 0046458 0007894 5884990 00000
C -0080367 0119239 -0674003 05003
TREND(1031950) 267E-05 129E-05 2074122 00381
R-squared 0539061 Mean dependent var 0000204
Adjusted R-squared 0538429 SD dependent var 1109785
SE of regression 7539766 Akaike info criterion 6879690
Sum squared resid 9120137 Schwarz criterion 6890691
Log likelihood -5524155 Hannan-Quinn criter 6883328
F-statistic 8528214 Durbin-Watson stat 1999071
Prob(F-statistic) 0000000
Correlogram analysis (series dp)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
This hints at ARMA(pq) model and not AR(p) or MA(q) model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0063 -0063 -799058 64008 0
2 -0039 -0043 -545389 87923 0
3 -0007 -0012 -152201 8871 0
4 -0001 -0004 -050734 88726 0
5 -004 -0041 -520022 11425 0
6 -0006 -0012 -152201 11482 0
7 -0026 -0031 -393187 1255 0
8 0009 0004 0507338 12688 0
9 -0006 -0008 -101468 12738 0
10 0034 0032 4058705 14622 0
11 -002 -0017 -215619 15252 0
12 005 0048 6088058 19218 0
13 0019 0025 3170863 19781 0
14 -0023 -0017 -215619 20648 0
15 -0029 -0026 -32977 2198 0
16 0035 003 3805036 23985 0
17 0014 0022 279036 24304 0
18 -0052 -0048 -608806 28721 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0013 001 1268345 28988 0
20 0008 0003 0380504 29104 0
21 -0048 -0046 -583439 32879 0
22 0018 001 1268345 3343 0
23 0007 0004 0507338 33519 0
24 0003 0004 0507338 33536 0
25 -001 -0013 -164885 33703 0
26 -0009 -0014 -177568 33843 0
27 0033 0035 4439209 35586 0
28 0007 0009 1141511 35656 0
29 0019 0019 2409856 36249 0
30 0005 0013 1648849 36284 0
31 -0003 0006 0761007 36299 0
32 0001 -0002 -025367 363 0
33 -0016 -0014 -177568 36714 0
34 -0066 -0061 -773691 4372 0
35 0006 -0004 -050734 4377 0
36 0018 0009 1141511 44317 0
ARMA Estimation ndash Information Criteria
Information Criteria(SIC)
MA(0) MA(1) MA(2)
AR(0) AIC 6895772
6896728
6891609
6892565
6890118
6891551SIC
AR(1) AIC 6892031
6892986
6889474
6890907
6889569
6891480SIC
AR(2) AIC 6890391
6891824
6889632
6891544
6889077
6891466SIC
Observations
ARMA(11) appears to be the best fitted model
The suitability has been in line with both AIC and SIC
Sufficient data present in the model so SIC is the best criteria
Estimation of the model
Observations
As expected all the terms are significant in the ARMA(11) model
The Durbin Watson Statistic is asymp 2 This suggests that the autocorrelation among residuals may be small
This is in fact confirmed by the correlogramQ-statistics of the residuals
Dependent Variable DP
Method Least Squares
Date 121113 Time 0144
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 8 iterations
MA Backcast 1041950
Variable Coefficient Std Error t-Statistic Prob
C 0111277 0049883 2230768 00257
AR(1) 0607851 0056980 1066776 00000
MA(1) -0672801 0053089 -1267311 00000
R-squared 0006645 Mean dependent var 0111371
Adjusted R-squared 0006522 SD dependent var 7606298
SE of regression 7581454 Akaike info criterion 6889474
Sum squared resid 9244258 Schwarz criterion 6890907
Log likelihood -5540904 Hannan-Quinn criter 6889948
F-statistic 5379525 Durbin-Watson stat 2004706
Prob(F-statistic) 0000000
Inverted AR Roots 61
Inverted MA Roots 67
Correlogram Q statistic (of residuals)
Observations
All ACF values from 5th lag onwards significant
No definite threshold found such that PACF vanishes above that threshold
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0002 -0002 -025366 00897
2 -0001 -0001 -012683 01221
3 0015 0015 1902459 35484 006
4 0011 0011 1395137 55919 0061
5 -0031 -0031 -393175 20833 0
6 -0002 -0003 -038049 20928 0
7 -0022 -0022 -279027 28685 0
8 0011 0011 1395137 30563 0
9 -0002 -0001 -012683 30637 0
10 0036 0035 4439071 50953 0
11 -0014 -0014 -177563 54261 0
12 005 0049 6214699 95031 0
13 002 002 2536612 10169 0
14 -0021 -0021 -266344 1086 0
15 -0027 -0025 -317076 12008 0
16 0033 003 3804918 13763 0
17 0012 0017 215612 14007 0
18 -0051 -005 -634153 18163 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 001 0011 1395137 18319 0
20 0006 0001 0126831 18372 0
21 -0047 -0045 -570738 21954 0
22 0016 0014 1775628 22378 0
23 0007 0006 0760984 22456 0
24 0003 0004 0507322 22474 0
25 -0009 -0011 -139514 226 0
26 -0007 -001 -126831 22681 0
27 0033 0037 4692732 24466 0
28 0009 0009 1141475 24598 0
29 002 0017 215612 25252 0
30 0005 0009 1141475 25299 0
31 -0003 0002 0253661 25318 0
32 -0002 -0006 -076098 25322 0
33 -0019 -0018 -228295 25902 0
34 -0066 -006 -760984 32879 0
35 0003 0001 0126831 32889 0
36 0017 0013 1648798 33334 0
Correlogram Q statistic (of squared residuals)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
However square of the lag values are larger (in ACF)
This suggests ARCH type modelling is more appropriate
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0236 0236 2993202 89426
2 0387 035 4439071 32984
3 0243 0123 1560016 42517 0
4 0297 0139 1762945 56715 0
5 0323 0189 2397098 73461 0
6 0311 0137 1737579 88992 0
7 0303 0098 124294 10377 0
8 0295 0089 1128792 11777 0
9 0283 0067 849765 13068 0
10 0311 0094 1192208 14621 0
11 0307 0089 1128792 16138 0
12 0287 0045 5707377 17464 0
13 0235 -0027 -342443 18356 0
14 022 -0039 -494639 19138 0
15 0237 0001 0126831 20041 0
16 0271 0051 646836 21225 0
17 0269 0045 5707377 22387 0
18 0272 0047 5961038 23577 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0209 -0026 -32976 24280 0
20 0253 003 3804918 25307 0
21 0289 0105 1331721 26653 0
22 0223 -0013 -16488 27456 0
23 0242 -0004 -050732 28402 0
24 0191 -002 -253661 28991 0
25 0209 -0012 -152197 29696 0
26 0205 -0013 -16488 30372 0
27 0276 0079 1001962 31600 0
28 0233 0023 2917104 32474 0
29 0214 -002 -253661 33215 0
30 0185 -0022 -279027 33768 0
31 0194 -0006 -076098 34373 0
32 0251 0067 849765 35388 0
33 02 -0005 -063415 36032 0
34 0232 003 3804918 36902 0
35 0169 -0018 -228295 37360 0
36 0208 0013 1648798 38055 0
ARCH Heteroskedasticity test
Observations
We reaffirm the ARCH effect using the Heteroskedasticity ndash ARCH test
Clearly all residue lags seem significant (uptolag 9)
This hints that we may have to opt for a GARCH(pq) model
Hence we have not tested for pure ARCH models any further
Heteroskedasticity Test ARCH
F-statistic 6246957 Prob F(916067) 00000
ObsR-squared 4167458 Prob Chi-Square(9) 00000
Variable Coefficient Std Error t-Statistic Prob
C 1160923 2045328 5675975 00000
RESID^2(-1) 0013088 0007872 1662714 00964
RESID^2(-2) 0213112 0007842 2717408 00000
RESID^2(-3) 0016465 0007996 2059229 00395
RESID^2(-4) 0062821 0007948 7903630 00000
RESID^2(-5) 0147233 0007879 1868761 00000
RESID^2(-6) 0111116 0007948 1397978 00000
RESID^2(-7) 0080432 0007996 1005965 00000
RESID^2(-8) 0087330 0007842 1113558 00000
RESID^2(-9) 0066907 0007872 8499328 00000
R-squared 0259219 Mean dependent var 5749988
Adjusted R-squared 0258804 SD dependent var 2862004
SE of regression 2463978 Akaike info criterion 1385239
Sum squared resid 975E+08 Schwarz criterion 1385717
Log likelihood -1113425 Hannan-Quinn criter 1385397
F-statistic 6246957 Durbin-Watson stat 2012623
Prob(F-statistic) 0000000
GARCH(11)
Observations
P-Value of Coefficient of GARCH(-1) is more less than 005
Thus we conclude volatility is of GARCH kind
Thus we do not check for ARCH We check for the best form of GARCH that fits the data
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0305
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 32 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0018850 0003353 5622112 00000
AR(1) -0141277 0075434 -1872843 00611
MA(1) 0242258 0073980 3274633 00011
Variance Equation
C 0000213 329E-05 6465720 00000
RESID(-1)^2 0066426 0001506 4410476 00000
GARCH(-1) 0939441 0001222 7689159 00000
R-squared -0022203 Mean dependent var 0111371
Adjusted R-squared -0022330 SD dependent var 7606298
SE of regression 7690752 Akaike info criterion 3800477
Sum squared resid 9512719 Schwarz criterion 3803344
Log likelihood -3056124 Hannan-Quinn criter 3801425
Durbin-Watson stat 2322505
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(12)
Observations
the coefficients of RESID(-1)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0312
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 33 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0019100 0003404 5610811 00000
AR(1) -0137936 0072289 -1908117 00564
MA(1) 0242221 0070576 3432082 00006
Variance Equation
C 0000161 262E-05 6145381 00000
RESID(-1)^2 0107198 0004442 2413017 00000
RESID(-2)^2 -0050822 0005119 -9928459 00000
GARCH(-1) 0948322 0001527 6212227 00000
R-squared -0023299 Mean dependent var 0111371
Adjusted R-squared -0023427 SD dependent var 7606298
SE of regression 7694877 Akaike info criterion 3798666
Sum squared resid 9522926 Schwarz criterion 3802010
Log likelihood -3054567 Hannan-Quinn criter 3799772
Durbin-Watson stat 2328892
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(22)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0319
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1) + C(8)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
Correlogram analysis (series dp)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
This hints at ARMA(pq) model and not AR(p) or MA(q) model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0063 -0063 -799058 64008 0
2 -0039 -0043 -545389 87923 0
3 -0007 -0012 -152201 8871 0
4 -0001 -0004 -050734 88726 0
5 -004 -0041 -520022 11425 0
6 -0006 -0012 -152201 11482 0
7 -0026 -0031 -393187 1255 0
8 0009 0004 0507338 12688 0
9 -0006 -0008 -101468 12738 0
10 0034 0032 4058705 14622 0
11 -002 -0017 -215619 15252 0
12 005 0048 6088058 19218 0
13 0019 0025 3170863 19781 0
14 -0023 -0017 -215619 20648 0
15 -0029 -0026 -32977 2198 0
16 0035 003 3805036 23985 0
17 0014 0022 279036 24304 0
18 -0052 -0048 -608806 28721 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0013 001 1268345 28988 0
20 0008 0003 0380504 29104 0
21 -0048 -0046 -583439 32879 0
22 0018 001 1268345 3343 0
23 0007 0004 0507338 33519 0
24 0003 0004 0507338 33536 0
25 -001 -0013 -164885 33703 0
26 -0009 -0014 -177568 33843 0
27 0033 0035 4439209 35586 0
28 0007 0009 1141511 35656 0
29 0019 0019 2409856 36249 0
30 0005 0013 1648849 36284 0
31 -0003 0006 0761007 36299 0
32 0001 -0002 -025367 363 0
33 -0016 -0014 -177568 36714 0
34 -0066 -0061 -773691 4372 0
35 0006 -0004 -050734 4377 0
36 0018 0009 1141511 44317 0
ARMA Estimation ndash Information Criteria
Information Criteria(SIC)
MA(0) MA(1) MA(2)
AR(0) AIC 6895772
6896728
6891609
6892565
6890118
6891551SIC
AR(1) AIC 6892031
6892986
6889474
6890907
6889569
6891480SIC
AR(2) AIC 6890391
6891824
6889632
6891544
6889077
6891466SIC
Observations
ARMA(11) appears to be the best fitted model
The suitability has been in line with both AIC and SIC
Sufficient data present in the model so SIC is the best criteria
Estimation of the model
Observations
As expected all the terms are significant in the ARMA(11) model
The Durbin Watson Statistic is asymp 2 This suggests that the autocorrelation among residuals may be small
This is in fact confirmed by the correlogramQ-statistics of the residuals
Dependent Variable DP
Method Least Squares
Date 121113 Time 0144
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 8 iterations
MA Backcast 1041950
Variable Coefficient Std Error t-Statistic Prob
C 0111277 0049883 2230768 00257
AR(1) 0607851 0056980 1066776 00000
MA(1) -0672801 0053089 -1267311 00000
R-squared 0006645 Mean dependent var 0111371
Adjusted R-squared 0006522 SD dependent var 7606298
SE of regression 7581454 Akaike info criterion 6889474
Sum squared resid 9244258 Schwarz criterion 6890907
Log likelihood -5540904 Hannan-Quinn criter 6889948
F-statistic 5379525 Durbin-Watson stat 2004706
Prob(F-statistic) 0000000
Inverted AR Roots 61
Inverted MA Roots 67
Correlogram Q statistic (of residuals)
Observations
All ACF values from 5th lag onwards significant
No definite threshold found such that PACF vanishes above that threshold
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0002 -0002 -025366 00897
2 -0001 -0001 -012683 01221
3 0015 0015 1902459 35484 006
4 0011 0011 1395137 55919 0061
5 -0031 -0031 -393175 20833 0
6 -0002 -0003 -038049 20928 0
7 -0022 -0022 -279027 28685 0
8 0011 0011 1395137 30563 0
9 -0002 -0001 -012683 30637 0
10 0036 0035 4439071 50953 0
11 -0014 -0014 -177563 54261 0
12 005 0049 6214699 95031 0
13 002 002 2536612 10169 0
14 -0021 -0021 -266344 1086 0
15 -0027 -0025 -317076 12008 0
16 0033 003 3804918 13763 0
17 0012 0017 215612 14007 0
18 -0051 -005 -634153 18163 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 001 0011 1395137 18319 0
20 0006 0001 0126831 18372 0
21 -0047 -0045 -570738 21954 0
22 0016 0014 1775628 22378 0
23 0007 0006 0760984 22456 0
24 0003 0004 0507322 22474 0
25 -0009 -0011 -139514 226 0
26 -0007 -001 -126831 22681 0
27 0033 0037 4692732 24466 0
28 0009 0009 1141475 24598 0
29 002 0017 215612 25252 0
30 0005 0009 1141475 25299 0
31 -0003 0002 0253661 25318 0
32 -0002 -0006 -076098 25322 0
33 -0019 -0018 -228295 25902 0
34 -0066 -006 -760984 32879 0
35 0003 0001 0126831 32889 0
36 0017 0013 1648798 33334 0
Correlogram Q statistic (of squared residuals)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
However square of the lag values are larger (in ACF)
This suggests ARCH type modelling is more appropriate
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0236 0236 2993202 89426
2 0387 035 4439071 32984
3 0243 0123 1560016 42517 0
4 0297 0139 1762945 56715 0
5 0323 0189 2397098 73461 0
6 0311 0137 1737579 88992 0
7 0303 0098 124294 10377 0
8 0295 0089 1128792 11777 0
9 0283 0067 849765 13068 0
10 0311 0094 1192208 14621 0
11 0307 0089 1128792 16138 0
12 0287 0045 5707377 17464 0
13 0235 -0027 -342443 18356 0
14 022 -0039 -494639 19138 0
15 0237 0001 0126831 20041 0
16 0271 0051 646836 21225 0
17 0269 0045 5707377 22387 0
18 0272 0047 5961038 23577 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0209 -0026 -32976 24280 0
20 0253 003 3804918 25307 0
21 0289 0105 1331721 26653 0
22 0223 -0013 -16488 27456 0
23 0242 -0004 -050732 28402 0
24 0191 -002 -253661 28991 0
25 0209 -0012 -152197 29696 0
26 0205 -0013 -16488 30372 0
27 0276 0079 1001962 31600 0
28 0233 0023 2917104 32474 0
29 0214 -002 -253661 33215 0
30 0185 -0022 -279027 33768 0
31 0194 -0006 -076098 34373 0
32 0251 0067 849765 35388 0
33 02 -0005 -063415 36032 0
34 0232 003 3804918 36902 0
35 0169 -0018 -228295 37360 0
36 0208 0013 1648798 38055 0
ARCH Heteroskedasticity test
Observations
We reaffirm the ARCH effect using the Heteroskedasticity ndash ARCH test
Clearly all residue lags seem significant (uptolag 9)
This hints that we may have to opt for a GARCH(pq) model
Hence we have not tested for pure ARCH models any further
Heteroskedasticity Test ARCH
F-statistic 6246957 Prob F(916067) 00000
ObsR-squared 4167458 Prob Chi-Square(9) 00000
Variable Coefficient Std Error t-Statistic Prob
C 1160923 2045328 5675975 00000
RESID^2(-1) 0013088 0007872 1662714 00964
RESID^2(-2) 0213112 0007842 2717408 00000
RESID^2(-3) 0016465 0007996 2059229 00395
RESID^2(-4) 0062821 0007948 7903630 00000
RESID^2(-5) 0147233 0007879 1868761 00000
RESID^2(-6) 0111116 0007948 1397978 00000
RESID^2(-7) 0080432 0007996 1005965 00000
RESID^2(-8) 0087330 0007842 1113558 00000
RESID^2(-9) 0066907 0007872 8499328 00000
R-squared 0259219 Mean dependent var 5749988
Adjusted R-squared 0258804 SD dependent var 2862004
SE of regression 2463978 Akaike info criterion 1385239
Sum squared resid 975E+08 Schwarz criterion 1385717
Log likelihood -1113425 Hannan-Quinn criter 1385397
F-statistic 6246957 Durbin-Watson stat 2012623
Prob(F-statistic) 0000000
GARCH(11)
Observations
P-Value of Coefficient of GARCH(-1) is more less than 005
Thus we conclude volatility is of GARCH kind
Thus we do not check for ARCH We check for the best form of GARCH that fits the data
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0305
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 32 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0018850 0003353 5622112 00000
AR(1) -0141277 0075434 -1872843 00611
MA(1) 0242258 0073980 3274633 00011
Variance Equation
C 0000213 329E-05 6465720 00000
RESID(-1)^2 0066426 0001506 4410476 00000
GARCH(-1) 0939441 0001222 7689159 00000
R-squared -0022203 Mean dependent var 0111371
Adjusted R-squared -0022330 SD dependent var 7606298
SE of regression 7690752 Akaike info criterion 3800477
Sum squared resid 9512719 Schwarz criterion 3803344
Log likelihood -3056124 Hannan-Quinn criter 3801425
Durbin-Watson stat 2322505
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(12)
Observations
the coefficients of RESID(-1)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0312
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 33 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0019100 0003404 5610811 00000
AR(1) -0137936 0072289 -1908117 00564
MA(1) 0242221 0070576 3432082 00006
Variance Equation
C 0000161 262E-05 6145381 00000
RESID(-1)^2 0107198 0004442 2413017 00000
RESID(-2)^2 -0050822 0005119 -9928459 00000
GARCH(-1) 0948322 0001527 6212227 00000
R-squared -0023299 Mean dependent var 0111371
Adjusted R-squared -0023427 SD dependent var 7606298
SE of regression 7694877 Akaike info criterion 3798666
Sum squared resid 9522926 Schwarz criterion 3802010
Log likelihood -3054567 Hannan-Quinn criter 3799772
Durbin-Watson stat 2328892
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(22)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0319
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1) + C(8)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
ARMA Estimation ndash Information Criteria
Information Criteria(SIC)
MA(0) MA(1) MA(2)
AR(0) AIC 6895772
6896728
6891609
6892565
6890118
6891551SIC
AR(1) AIC 6892031
6892986
6889474
6890907
6889569
6891480SIC
AR(2) AIC 6890391
6891824
6889632
6891544
6889077
6891466SIC
Observations
ARMA(11) appears to be the best fitted model
The suitability has been in line with both AIC and SIC
Sufficient data present in the model so SIC is the best criteria
Estimation of the model
Observations
As expected all the terms are significant in the ARMA(11) model
The Durbin Watson Statistic is asymp 2 This suggests that the autocorrelation among residuals may be small
This is in fact confirmed by the correlogramQ-statistics of the residuals
Dependent Variable DP
Method Least Squares
Date 121113 Time 0144
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 8 iterations
MA Backcast 1041950
Variable Coefficient Std Error t-Statistic Prob
C 0111277 0049883 2230768 00257
AR(1) 0607851 0056980 1066776 00000
MA(1) -0672801 0053089 -1267311 00000
R-squared 0006645 Mean dependent var 0111371
Adjusted R-squared 0006522 SD dependent var 7606298
SE of regression 7581454 Akaike info criterion 6889474
Sum squared resid 9244258 Schwarz criterion 6890907
Log likelihood -5540904 Hannan-Quinn criter 6889948
F-statistic 5379525 Durbin-Watson stat 2004706
Prob(F-statistic) 0000000
Inverted AR Roots 61
Inverted MA Roots 67
Correlogram Q statistic (of residuals)
Observations
All ACF values from 5th lag onwards significant
No definite threshold found such that PACF vanishes above that threshold
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0002 -0002 -025366 00897
2 -0001 -0001 -012683 01221
3 0015 0015 1902459 35484 006
4 0011 0011 1395137 55919 0061
5 -0031 -0031 -393175 20833 0
6 -0002 -0003 -038049 20928 0
7 -0022 -0022 -279027 28685 0
8 0011 0011 1395137 30563 0
9 -0002 -0001 -012683 30637 0
10 0036 0035 4439071 50953 0
11 -0014 -0014 -177563 54261 0
12 005 0049 6214699 95031 0
13 002 002 2536612 10169 0
14 -0021 -0021 -266344 1086 0
15 -0027 -0025 -317076 12008 0
16 0033 003 3804918 13763 0
17 0012 0017 215612 14007 0
18 -0051 -005 -634153 18163 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 001 0011 1395137 18319 0
20 0006 0001 0126831 18372 0
21 -0047 -0045 -570738 21954 0
22 0016 0014 1775628 22378 0
23 0007 0006 0760984 22456 0
24 0003 0004 0507322 22474 0
25 -0009 -0011 -139514 226 0
26 -0007 -001 -126831 22681 0
27 0033 0037 4692732 24466 0
28 0009 0009 1141475 24598 0
29 002 0017 215612 25252 0
30 0005 0009 1141475 25299 0
31 -0003 0002 0253661 25318 0
32 -0002 -0006 -076098 25322 0
33 -0019 -0018 -228295 25902 0
34 -0066 -006 -760984 32879 0
35 0003 0001 0126831 32889 0
36 0017 0013 1648798 33334 0
Correlogram Q statistic (of squared residuals)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
However square of the lag values are larger (in ACF)
This suggests ARCH type modelling is more appropriate
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0236 0236 2993202 89426
2 0387 035 4439071 32984
3 0243 0123 1560016 42517 0
4 0297 0139 1762945 56715 0
5 0323 0189 2397098 73461 0
6 0311 0137 1737579 88992 0
7 0303 0098 124294 10377 0
8 0295 0089 1128792 11777 0
9 0283 0067 849765 13068 0
10 0311 0094 1192208 14621 0
11 0307 0089 1128792 16138 0
12 0287 0045 5707377 17464 0
13 0235 -0027 -342443 18356 0
14 022 -0039 -494639 19138 0
15 0237 0001 0126831 20041 0
16 0271 0051 646836 21225 0
17 0269 0045 5707377 22387 0
18 0272 0047 5961038 23577 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0209 -0026 -32976 24280 0
20 0253 003 3804918 25307 0
21 0289 0105 1331721 26653 0
22 0223 -0013 -16488 27456 0
23 0242 -0004 -050732 28402 0
24 0191 -002 -253661 28991 0
25 0209 -0012 -152197 29696 0
26 0205 -0013 -16488 30372 0
27 0276 0079 1001962 31600 0
28 0233 0023 2917104 32474 0
29 0214 -002 -253661 33215 0
30 0185 -0022 -279027 33768 0
31 0194 -0006 -076098 34373 0
32 0251 0067 849765 35388 0
33 02 -0005 -063415 36032 0
34 0232 003 3804918 36902 0
35 0169 -0018 -228295 37360 0
36 0208 0013 1648798 38055 0
ARCH Heteroskedasticity test
Observations
We reaffirm the ARCH effect using the Heteroskedasticity ndash ARCH test
Clearly all residue lags seem significant (uptolag 9)
This hints that we may have to opt for a GARCH(pq) model
Hence we have not tested for pure ARCH models any further
Heteroskedasticity Test ARCH
F-statistic 6246957 Prob F(916067) 00000
ObsR-squared 4167458 Prob Chi-Square(9) 00000
Variable Coefficient Std Error t-Statistic Prob
C 1160923 2045328 5675975 00000
RESID^2(-1) 0013088 0007872 1662714 00964
RESID^2(-2) 0213112 0007842 2717408 00000
RESID^2(-3) 0016465 0007996 2059229 00395
RESID^2(-4) 0062821 0007948 7903630 00000
RESID^2(-5) 0147233 0007879 1868761 00000
RESID^2(-6) 0111116 0007948 1397978 00000
RESID^2(-7) 0080432 0007996 1005965 00000
RESID^2(-8) 0087330 0007842 1113558 00000
RESID^2(-9) 0066907 0007872 8499328 00000
R-squared 0259219 Mean dependent var 5749988
Adjusted R-squared 0258804 SD dependent var 2862004
SE of regression 2463978 Akaike info criterion 1385239
Sum squared resid 975E+08 Schwarz criterion 1385717
Log likelihood -1113425 Hannan-Quinn criter 1385397
F-statistic 6246957 Durbin-Watson stat 2012623
Prob(F-statistic) 0000000
GARCH(11)
Observations
P-Value of Coefficient of GARCH(-1) is more less than 005
Thus we conclude volatility is of GARCH kind
Thus we do not check for ARCH We check for the best form of GARCH that fits the data
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0305
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 32 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0018850 0003353 5622112 00000
AR(1) -0141277 0075434 -1872843 00611
MA(1) 0242258 0073980 3274633 00011
Variance Equation
C 0000213 329E-05 6465720 00000
RESID(-1)^2 0066426 0001506 4410476 00000
GARCH(-1) 0939441 0001222 7689159 00000
R-squared -0022203 Mean dependent var 0111371
Adjusted R-squared -0022330 SD dependent var 7606298
SE of regression 7690752 Akaike info criterion 3800477
Sum squared resid 9512719 Schwarz criterion 3803344
Log likelihood -3056124 Hannan-Quinn criter 3801425
Durbin-Watson stat 2322505
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(12)
Observations
the coefficients of RESID(-1)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0312
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 33 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0019100 0003404 5610811 00000
AR(1) -0137936 0072289 -1908117 00564
MA(1) 0242221 0070576 3432082 00006
Variance Equation
C 0000161 262E-05 6145381 00000
RESID(-1)^2 0107198 0004442 2413017 00000
RESID(-2)^2 -0050822 0005119 -9928459 00000
GARCH(-1) 0948322 0001527 6212227 00000
R-squared -0023299 Mean dependent var 0111371
Adjusted R-squared -0023427 SD dependent var 7606298
SE of regression 7694877 Akaike info criterion 3798666
Sum squared resid 9522926 Schwarz criterion 3802010
Log likelihood -3054567 Hannan-Quinn criter 3799772
Durbin-Watson stat 2328892
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(22)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0319
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1) + C(8)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
Estimation of the model
Observations
As expected all the terms are significant in the ARMA(11) model
The Durbin Watson Statistic is asymp 2 This suggests that the autocorrelation among residuals may be small
This is in fact confirmed by the correlogramQ-statistics of the residuals
Dependent Variable DP
Method Least Squares
Date 121113 Time 0144
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 8 iterations
MA Backcast 1041950
Variable Coefficient Std Error t-Statistic Prob
C 0111277 0049883 2230768 00257
AR(1) 0607851 0056980 1066776 00000
MA(1) -0672801 0053089 -1267311 00000
R-squared 0006645 Mean dependent var 0111371
Adjusted R-squared 0006522 SD dependent var 7606298
SE of regression 7581454 Akaike info criterion 6889474
Sum squared resid 9244258 Schwarz criterion 6890907
Log likelihood -5540904 Hannan-Quinn criter 6889948
F-statistic 5379525 Durbin-Watson stat 2004706
Prob(F-statistic) 0000000
Inverted AR Roots 61
Inverted MA Roots 67
Correlogram Q statistic (of residuals)
Observations
All ACF values from 5th lag onwards significant
No definite threshold found such that PACF vanishes above that threshold
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0002 -0002 -025366 00897
2 -0001 -0001 -012683 01221
3 0015 0015 1902459 35484 006
4 0011 0011 1395137 55919 0061
5 -0031 -0031 -393175 20833 0
6 -0002 -0003 -038049 20928 0
7 -0022 -0022 -279027 28685 0
8 0011 0011 1395137 30563 0
9 -0002 -0001 -012683 30637 0
10 0036 0035 4439071 50953 0
11 -0014 -0014 -177563 54261 0
12 005 0049 6214699 95031 0
13 002 002 2536612 10169 0
14 -0021 -0021 -266344 1086 0
15 -0027 -0025 -317076 12008 0
16 0033 003 3804918 13763 0
17 0012 0017 215612 14007 0
18 -0051 -005 -634153 18163 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 001 0011 1395137 18319 0
20 0006 0001 0126831 18372 0
21 -0047 -0045 -570738 21954 0
22 0016 0014 1775628 22378 0
23 0007 0006 0760984 22456 0
24 0003 0004 0507322 22474 0
25 -0009 -0011 -139514 226 0
26 -0007 -001 -126831 22681 0
27 0033 0037 4692732 24466 0
28 0009 0009 1141475 24598 0
29 002 0017 215612 25252 0
30 0005 0009 1141475 25299 0
31 -0003 0002 0253661 25318 0
32 -0002 -0006 -076098 25322 0
33 -0019 -0018 -228295 25902 0
34 -0066 -006 -760984 32879 0
35 0003 0001 0126831 32889 0
36 0017 0013 1648798 33334 0
Correlogram Q statistic (of squared residuals)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
However square of the lag values are larger (in ACF)
This suggests ARCH type modelling is more appropriate
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0236 0236 2993202 89426
2 0387 035 4439071 32984
3 0243 0123 1560016 42517 0
4 0297 0139 1762945 56715 0
5 0323 0189 2397098 73461 0
6 0311 0137 1737579 88992 0
7 0303 0098 124294 10377 0
8 0295 0089 1128792 11777 0
9 0283 0067 849765 13068 0
10 0311 0094 1192208 14621 0
11 0307 0089 1128792 16138 0
12 0287 0045 5707377 17464 0
13 0235 -0027 -342443 18356 0
14 022 -0039 -494639 19138 0
15 0237 0001 0126831 20041 0
16 0271 0051 646836 21225 0
17 0269 0045 5707377 22387 0
18 0272 0047 5961038 23577 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0209 -0026 -32976 24280 0
20 0253 003 3804918 25307 0
21 0289 0105 1331721 26653 0
22 0223 -0013 -16488 27456 0
23 0242 -0004 -050732 28402 0
24 0191 -002 -253661 28991 0
25 0209 -0012 -152197 29696 0
26 0205 -0013 -16488 30372 0
27 0276 0079 1001962 31600 0
28 0233 0023 2917104 32474 0
29 0214 -002 -253661 33215 0
30 0185 -0022 -279027 33768 0
31 0194 -0006 -076098 34373 0
32 0251 0067 849765 35388 0
33 02 -0005 -063415 36032 0
34 0232 003 3804918 36902 0
35 0169 -0018 -228295 37360 0
36 0208 0013 1648798 38055 0
ARCH Heteroskedasticity test
Observations
We reaffirm the ARCH effect using the Heteroskedasticity ndash ARCH test
Clearly all residue lags seem significant (uptolag 9)
This hints that we may have to opt for a GARCH(pq) model
Hence we have not tested for pure ARCH models any further
Heteroskedasticity Test ARCH
F-statistic 6246957 Prob F(916067) 00000
ObsR-squared 4167458 Prob Chi-Square(9) 00000
Variable Coefficient Std Error t-Statistic Prob
C 1160923 2045328 5675975 00000
RESID^2(-1) 0013088 0007872 1662714 00964
RESID^2(-2) 0213112 0007842 2717408 00000
RESID^2(-3) 0016465 0007996 2059229 00395
RESID^2(-4) 0062821 0007948 7903630 00000
RESID^2(-5) 0147233 0007879 1868761 00000
RESID^2(-6) 0111116 0007948 1397978 00000
RESID^2(-7) 0080432 0007996 1005965 00000
RESID^2(-8) 0087330 0007842 1113558 00000
RESID^2(-9) 0066907 0007872 8499328 00000
R-squared 0259219 Mean dependent var 5749988
Adjusted R-squared 0258804 SD dependent var 2862004
SE of regression 2463978 Akaike info criterion 1385239
Sum squared resid 975E+08 Schwarz criterion 1385717
Log likelihood -1113425 Hannan-Quinn criter 1385397
F-statistic 6246957 Durbin-Watson stat 2012623
Prob(F-statistic) 0000000
GARCH(11)
Observations
P-Value of Coefficient of GARCH(-1) is more less than 005
Thus we conclude volatility is of GARCH kind
Thus we do not check for ARCH We check for the best form of GARCH that fits the data
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0305
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 32 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0018850 0003353 5622112 00000
AR(1) -0141277 0075434 -1872843 00611
MA(1) 0242258 0073980 3274633 00011
Variance Equation
C 0000213 329E-05 6465720 00000
RESID(-1)^2 0066426 0001506 4410476 00000
GARCH(-1) 0939441 0001222 7689159 00000
R-squared -0022203 Mean dependent var 0111371
Adjusted R-squared -0022330 SD dependent var 7606298
SE of regression 7690752 Akaike info criterion 3800477
Sum squared resid 9512719 Schwarz criterion 3803344
Log likelihood -3056124 Hannan-Quinn criter 3801425
Durbin-Watson stat 2322505
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(12)
Observations
the coefficients of RESID(-1)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0312
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 33 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0019100 0003404 5610811 00000
AR(1) -0137936 0072289 -1908117 00564
MA(1) 0242221 0070576 3432082 00006
Variance Equation
C 0000161 262E-05 6145381 00000
RESID(-1)^2 0107198 0004442 2413017 00000
RESID(-2)^2 -0050822 0005119 -9928459 00000
GARCH(-1) 0948322 0001527 6212227 00000
R-squared -0023299 Mean dependent var 0111371
Adjusted R-squared -0023427 SD dependent var 7606298
SE of regression 7694877 Akaike info criterion 3798666
Sum squared resid 9522926 Schwarz criterion 3802010
Log likelihood -3054567 Hannan-Quinn criter 3799772
Durbin-Watson stat 2328892
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(22)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0319
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1) + C(8)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
Correlogram Q statistic (of residuals)
Observations
All ACF values from 5th lag onwards significant
No definite threshold found such that PACF vanishes above that threshold
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 -0002 -0002 -025366 00897
2 -0001 -0001 -012683 01221
3 0015 0015 1902459 35484 006
4 0011 0011 1395137 55919 0061
5 -0031 -0031 -393175 20833 0
6 -0002 -0003 -038049 20928 0
7 -0022 -0022 -279027 28685 0
8 0011 0011 1395137 30563 0
9 -0002 -0001 -012683 30637 0
10 0036 0035 4439071 50953 0
11 -0014 -0014 -177563 54261 0
12 005 0049 6214699 95031 0
13 002 002 2536612 10169 0
14 -0021 -0021 -266344 1086 0
15 -0027 -0025 -317076 12008 0
16 0033 003 3804918 13763 0
17 0012 0017 215612 14007 0
18 -0051 -005 -634153 18163 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 001 0011 1395137 18319 0
20 0006 0001 0126831 18372 0
21 -0047 -0045 -570738 21954 0
22 0016 0014 1775628 22378 0
23 0007 0006 0760984 22456 0
24 0003 0004 0507322 22474 0
25 -0009 -0011 -139514 226 0
26 -0007 -001 -126831 22681 0
27 0033 0037 4692732 24466 0
28 0009 0009 1141475 24598 0
29 002 0017 215612 25252 0
30 0005 0009 1141475 25299 0
31 -0003 0002 0253661 25318 0
32 -0002 -0006 -076098 25322 0
33 -0019 -0018 -228295 25902 0
34 -0066 -006 -760984 32879 0
35 0003 0001 0126831 32889 0
36 0017 0013 1648798 33334 0
Correlogram Q statistic (of squared residuals)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
However square of the lag values are larger (in ACF)
This suggests ARCH type modelling is more appropriate
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0236 0236 2993202 89426
2 0387 035 4439071 32984
3 0243 0123 1560016 42517 0
4 0297 0139 1762945 56715 0
5 0323 0189 2397098 73461 0
6 0311 0137 1737579 88992 0
7 0303 0098 124294 10377 0
8 0295 0089 1128792 11777 0
9 0283 0067 849765 13068 0
10 0311 0094 1192208 14621 0
11 0307 0089 1128792 16138 0
12 0287 0045 5707377 17464 0
13 0235 -0027 -342443 18356 0
14 022 -0039 -494639 19138 0
15 0237 0001 0126831 20041 0
16 0271 0051 646836 21225 0
17 0269 0045 5707377 22387 0
18 0272 0047 5961038 23577 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0209 -0026 -32976 24280 0
20 0253 003 3804918 25307 0
21 0289 0105 1331721 26653 0
22 0223 -0013 -16488 27456 0
23 0242 -0004 -050732 28402 0
24 0191 -002 -253661 28991 0
25 0209 -0012 -152197 29696 0
26 0205 -0013 -16488 30372 0
27 0276 0079 1001962 31600 0
28 0233 0023 2917104 32474 0
29 0214 -002 -253661 33215 0
30 0185 -0022 -279027 33768 0
31 0194 -0006 -076098 34373 0
32 0251 0067 849765 35388 0
33 02 -0005 -063415 36032 0
34 0232 003 3804918 36902 0
35 0169 -0018 -228295 37360 0
36 0208 0013 1648798 38055 0
ARCH Heteroskedasticity test
Observations
We reaffirm the ARCH effect using the Heteroskedasticity ndash ARCH test
Clearly all residue lags seem significant (uptolag 9)
This hints that we may have to opt for a GARCH(pq) model
Hence we have not tested for pure ARCH models any further
Heteroskedasticity Test ARCH
F-statistic 6246957 Prob F(916067) 00000
ObsR-squared 4167458 Prob Chi-Square(9) 00000
Variable Coefficient Std Error t-Statistic Prob
C 1160923 2045328 5675975 00000
RESID^2(-1) 0013088 0007872 1662714 00964
RESID^2(-2) 0213112 0007842 2717408 00000
RESID^2(-3) 0016465 0007996 2059229 00395
RESID^2(-4) 0062821 0007948 7903630 00000
RESID^2(-5) 0147233 0007879 1868761 00000
RESID^2(-6) 0111116 0007948 1397978 00000
RESID^2(-7) 0080432 0007996 1005965 00000
RESID^2(-8) 0087330 0007842 1113558 00000
RESID^2(-9) 0066907 0007872 8499328 00000
R-squared 0259219 Mean dependent var 5749988
Adjusted R-squared 0258804 SD dependent var 2862004
SE of regression 2463978 Akaike info criterion 1385239
Sum squared resid 975E+08 Schwarz criterion 1385717
Log likelihood -1113425 Hannan-Quinn criter 1385397
F-statistic 6246957 Durbin-Watson stat 2012623
Prob(F-statistic) 0000000
GARCH(11)
Observations
P-Value of Coefficient of GARCH(-1) is more less than 005
Thus we conclude volatility is of GARCH kind
Thus we do not check for ARCH We check for the best form of GARCH that fits the data
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0305
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 32 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0018850 0003353 5622112 00000
AR(1) -0141277 0075434 -1872843 00611
MA(1) 0242258 0073980 3274633 00011
Variance Equation
C 0000213 329E-05 6465720 00000
RESID(-1)^2 0066426 0001506 4410476 00000
GARCH(-1) 0939441 0001222 7689159 00000
R-squared -0022203 Mean dependent var 0111371
Adjusted R-squared -0022330 SD dependent var 7606298
SE of regression 7690752 Akaike info criterion 3800477
Sum squared resid 9512719 Schwarz criterion 3803344
Log likelihood -3056124 Hannan-Quinn criter 3801425
Durbin-Watson stat 2322505
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(12)
Observations
the coefficients of RESID(-1)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0312
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 33 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0019100 0003404 5610811 00000
AR(1) -0137936 0072289 -1908117 00564
MA(1) 0242221 0070576 3432082 00006
Variance Equation
C 0000161 262E-05 6145381 00000
RESID(-1)^2 0107198 0004442 2413017 00000
RESID(-2)^2 -0050822 0005119 -9928459 00000
GARCH(-1) 0948322 0001527 6212227 00000
R-squared -0023299 Mean dependent var 0111371
Adjusted R-squared -0023427 SD dependent var 7606298
SE of regression 7694877 Akaike info criterion 3798666
Sum squared resid 9522926 Schwarz criterion 3802010
Log likelihood -3054567 Hannan-Quinn criter 3799772
Durbin-Watson stat 2328892
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(22)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0319
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1) + C(8)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
Correlogram Q statistic (of squared residuals)
Observations
All ACF values significant
No definite threshold found such that PACF vanishes above that threshold
However square of the lag values are larger (in ACF)
This suggests ARCH type modelling is more appropriate
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0236 0236 2993202 89426
2 0387 035 4439071 32984
3 0243 0123 1560016 42517 0
4 0297 0139 1762945 56715 0
5 0323 0189 2397098 73461 0
6 0311 0137 1737579 88992 0
7 0303 0098 124294 10377 0
8 0295 0089 1128792 11777 0
9 0283 0067 849765 13068 0
10 0311 0094 1192208 14621 0
11 0307 0089 1128792 16138 0
12 0287 0045 5707377 17464 0
13 0235 -0027 -342443 18356 0
14 022 -0039 -494639 19138 0
15 0237 0001 0126831 20041 0
16 0271 0051 646836 21225 0
17 0269 0045 5707377 22387 0
18 0272 0047 5961038 23577 0
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0209 -0026 -32976 24280 0
20 0253 003 3804918 25307 0
21 0289 0105 1331721 26653 0
22 0223 -0013 -16488 27456 0
23 0242 -0004 -050732 28402 0
24 0191 -002 -253661 28991 0
25 0209 -0012 -152197 29696 0
26 0205 -0013 -16488 30372 0
27 0276 0079 1001962 31600 0
28 0233 0023 2917104 32474 0
29 0214 -002 -253661 33215 0
30 0185 -0022 -279027 33768 0
31 0194 -0006 -076098 34373 0
32 0251 0067 849765 35388 0
33 02 -0005 -063415 36032 0
34 0232 003 3804918 36902 0
35 0169 -0018 -228295 37360 0
36 0208 0013 1648798 38055 0
ARCH Heteroskedasticity test
Observations
We reaffirm the ARCH effect using the Heteroskedasticity ndash ARCH test
Clearly all residue lags seem significant (uptolag 9)
This hints that we may have to opt for a GARCH(pq) model
Hence we have not tested for pure ARCH models any further
Heteroskedasticity Test ARCH
F-statistic 6246957 Prob F(916067) 00000
ObsR-squared 4167458 Prob Chi-Square(9) 00000
Variable Coefficient Std Error t-Statistic Prob
C 1160923 2045328 5675975 00000
RESID^2(-1) 0013088 0007872 1662714 00964
RESID^2(-2) 0213112 0007842 2717408 00000
RESID^2(-3) 0016465 0007996 2059229 00395
RESID^2(-4) 0062821 0007948 7903630 00000
RESID^2(-5) 0147233 0007879 1868761 00000
RESID^2(-6) 0111116 0007948 1397978 00000
RESID^2(-7) 0080432 0007996 1005965 00000
RESID^2(-8) 0087330 0007842 1113558 00000
RESID^2(-9) 0066907 0007872 8499328 00000
R-squared 0259219 Mean dependent var 5749988
Adjusted R-squared 0258804 SD dependent var 2862004
SE of regression 2463978 Akaike info criterion 1385239
Sum squared resid 975E+08 Schwarz criterion 1385717
Log likelihood -1113425 Hannan-Quinn criter 1385397
F-statistic 6246957 Durbin-Watson stat 2012623
Prob(F-statistic) 0000000
GARCH(11)
Observations
P-Value of Coefficient of GARCH(-1) is more less than 005
Thus we conclude volatility is of GARCH kind
Thus we do not check for ARCH We check for the best form of GARCH that fits the data
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0305
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 32 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0018850 0003353 5622112 00000
AR(1) -0141277 0075434 -1872843 00611
MA(1) 0242258 0073980 3274633 00011
Variance Equation
C 0000213 329E-05 6465720 00000
RESID(-1)^2 0066426 0001506 4410476 00000
GARCH(-1) 0939441 0001222 7689159 00000
R-squared -0022203 Mean dependent var 0111371
Adjusted R-squared -0022330 SD dependent var 7606298
SE of regression 7690752 Akaike info criterion 3800477
Sum squared resid 9512719 Schwarz criterion 3803344
Log likelihood -3056124 Hannan-Quinn criter 3801425
Durbin-Watson stat 2322505
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(12)
Observations
the coefficients of RESID(-1)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0312
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 33 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0019100 0003404 5610811 00000
AR(1) -0137936 0072289 -1908117 00564
MA(1) 0242221 0070576 3432082 00006
Variance Equation
C 0000161 262E-05 6145381 00000
RESID(-1)^2 0107198 0004442 2413017 00000
RESID(-2)^2 -0050822 0005119 -9928459 00000
GARCH(-1) 0948322 0001527 6212227 00000
R-squared -0023299 Mean dependent var 0111371
Adjusted R-squared -0023427 SD dependent var 7606298
SE of regression 7694877 Akaike info criterion 3798666
Sum squared resid 9522926 Schwarz criterion 3802010
Log likelihood -3054567 Hannan-Quinn criter 3799772
Durbin-Watson stat 2328892
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(22)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0319
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1) + C(8)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
ARCH Heteroskedasticity test
Observations
We reaffirm the ARCH effect using the Heteroskedasticity ndash ARCH test
Clearly all residue lags seem significant (uptolag 9)
This hints that we may have to opt for a GARCH(pq) model
Hence we have not tested for pure ARCH models any further
Heteroskedasticity Test ARCH
F-statistic 6246957 Prob F(916067) 00000
ObsR-squared 4167458 Prob Chi-Square(9) 00000
Variable Coefficient Std Error t-Statistic Prob
C 1160923 2045328 5675975 00000
RESID^2(-1) 0013088 0007872 1662714 00964
RESID^2(-2) 0213112 0007842 2717408 00000
RESID^2(-3) 0016465 0007996 2059229 00395
RESID^2(-4) 0062821 0007948 7903630 00000
RESID^2(-5) 0147233 0007879 1868761 00000
RESID^2(-6) 0111116 0007948 1397978 00000
RESID^2(-7) 0080432 0007996 1005965 00000
RESID^2(-8) 0087330 0007842 1113558 00000
RESID^2(-9) 0066907 0007872 8499328 00000
R-squared 0259219 Mean dependent var 5749988
Adjusted R-squared 0258804 SD dependent var 2862004
SE of regression 2463978 Akaike info criterion 1385239
Sum squared resid 975E+08 Schwarz criterion 1385717
Log likelihood -1113425 Hannan-Quinn criter 1385397
F-statistic 6246957 Durbin-Watson stat 2012623
Prob(F-statistic) 0000000
GARCH(11)
Observations
P-Value of Coefficient of GARCH(-1) is more less than 005
Thus we conclude volatility is of GARCH kind
Thus we do not check for ARCH We check for the best form of GARCH that fits the data
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0305
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 32 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0018850 0003353 5622112 00000
AR(1) -0141277 0075434 -1872843 00611
MA(1) 0242258 0073980 3274633 00011
Variance Equation
C 0000213 329E-05 6465720 00000
RESID(-1)^2 0066426 0001506 4410476 00000
GARCH(-1) 0939441 0001222 7689159 00000
R-squared -0022203 Mean dependent var 0111371
Adjusted R-squared -0022330 SD dependent var 7606298
SE of regression 7690752 Akaike info criterion 3800477
Sum squared resid 9512719 Schwarz criterion 3803344
Log likelihood -3056124 Hannan-Quinn criter 3801425
Durbin-Watson stat 2322505
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(12)
Observations
the coefficients of RESID(-1)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0312
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 33 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0019100 0003404 5610811 00000
AR(1) -0137936 0072289 -1908117 00564
MA(1) 0242221 0070576 3432082 00006
Variance Equation
C 0000161 262E-05 6145381 00000
RESID(-1)^2 0107198 0004442 2413017 00000
RESID(-2)^2 -0050822 0005119 -9928459 00000
GARCH(-1) 0948322 0001527 6212227 00000
R-squared -0023299 Mean dependent var 0111371
Adjusted R-squared -0023427 SD dependent var 7606298
SE of regression 7694877 Akaike info criterion 3798666
Sum squared resid 9522926 Schwarz criterion 3802010
Log likelihood -3054567 Hannan-Quinn criter 3799772
Durbin-Watson stat 2328892
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(22)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0319
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1) + C(8)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
GARCH(11)
Observations
P-Value of Coefficient of GARCH(-1) is more less than 005
Thus we conclude volatility is of GARCH kind
Thus we do not check for ARCH We check for the best form of GARCH that fits the data
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0305
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 32 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0018850 0003353 5622112 00000
AR(1) -0141277 0075434 -1872843 00611
MA(1) 0242258 0073980 3274633 00011
Variance Equation
C 0000213 329E-05 6465720 00000
RESID(-1)^2 0066426 0001506 4410476 00000
GARCH(-1) 0939441 0001222 7689159 00000
R-squared -0022203 Mean dependent var 0111371
Adjusted R-squared -0022330 SD dependent var 7606298
SE of regression 7690752 Akaike info criterion 3800477
Sum squared resid 9512719 Schwarz criterion 3803344
Log likelihood -3056124 Hannan-Quinn criter 3801425
Durbin-Watson stat 2322505
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(12)
Observations
the coefficients of RESID(-1)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0312
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 33 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0019100 0003404 5610811 00000
AR(1) -0137936 0072289 -1908117 00564
MA(1) 0242221 0070576 3432082 00006
Variance Equation
C 0000161 262E-05 6145381 00000
RESID(-1)^2 0107198 0004442 2413017 00000
RESID(-2)^2 -0050822 0005119 -9928459 00000
GARCH(-1) 0948322 0001527 6212227 00000
R-squared -0023299 Mean dependent var 0111371
Adjusted R-squared -0023427 SD dependent var 7606298
SE of regression 7694877 Akaike info criterion 3798666
Sum squared resid 9522926 Schwarz criterion 3802010
Log likelihood -3054567 Hannan-Quinn criter 3799772
Durbin-Watson stat 2328892
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(22)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0319
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1) + C(8)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
GARCH(12)
Observations
the coefficients of RESID(-1)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0312
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 33 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1)
Variable Coefficient Std Error z-Statistic Prob
C 0019100 0003404 5610811 00000
AR(1) -0137936 0072289 -1908117 00564
MA(1) 0242221 0070576 3432082 00006
Variance Equation
C 0000161 262E-05 6145381 00000
RESID(-1)^2 0107198 0004442 2413017 00000
RESID(-2)^2 -0050822 0005119 -9928459 00000
GARCH(-1) 0948322 0001527 6212227 00000
R-squared -0023299 Mean dependent var 0111371
Adjusted R-squared -0023427 SD dependent var 7606298
SE of regression 7694877 Akaike info criterion 3798666
Sum squared resid 9522926 Schwarz criterion 3802010
Log likelihood -3054567 Hannan-Quinn criter 3799772
Durbin-Watson stat 2328892
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(22)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0319
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1) + C(8)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
GARCH(22)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 0319
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)GARCH(-1) + C(8)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
Inverted AR Roots -14
Inverted MA Roots -24
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
GARCH(23)
Observations
the coefficients of RESID(-1)^2 RESID(-2)^2 RESID(-3)^2 GARCH(-1) and GARCH(-2) are all significant at 5 level of significance
we cannot draw any inference talking about the best fitted volatility model to the data
Likewise all models upto GARCH(33) have been estimated though they have not been pasted in this presentation for brevity
To see which of these models best fit the data we consider the minimum AIC and BIC values
Dependent Variable DP
Method ML - ARCH (Marquardt) - Normal distribution
Date 121113 Time 1223
Sample (adjusted) 1051950 12092013
Included observations 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast 1041950
Presample variance backcast (parameter = 07)
GARCH = C(4) + C(5)RESID(-1)^2 + C(6)RESID(-2)^2 + C(7)RESID(-3)^2+ + C(8)GARCH(-1) + C(9)GARCH(-2)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
Inverted AR Roots -14
Inverted MA Roots -25
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
Volatility Estimation ndash Information Criteria
Information Criteria(SIC)
AIC BIC
GARCH(11) 3800477 3803344
GARCH(12) 3798666 3802010
GARCH(13) 3798186 3802009
GARCH(21) 3800477 3802872
GARCH(22) 3794703 3798525
GARCH(23) 3794321 3798621
GARCH (31) 3798906 3802729
GARCH(32) 3799317 3803617
GARCH(33) 3799790 3804568
Observations
GARCH (22) and GARCH (23) appear to be the best fitted models
We proceed to estimate both models
The final model selection will depend upon the residual diagnostics
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0019824 0003325 5962788 00000
AR(1) -0142799 0071582 -1994902 00461
MA(1) 0248253 0069856 3553797 00004
Variance Equation
C 195E-06 427E-07 4569525 00000
RESID(-1)^2 0103399 0004377 2362315 00000
RESID(-2)^2 -0133792 0009751 -1372090 00000
RESID(-3)^2 0031082 0005770 5386980 00000
GARCH(-1) 1904230 0005893 3231577 00000
GARCH(-2) -0904878 0005823 -1554076 00000
R-squared -0023694 Mean dependent var 0111371
Adjusted R-squared -0023822 SD dependent var 7606298
SE of regression 7696361 Akaike info criterion 3794321
Sum squared resid 9526601 Schwarz criterion 3798621
Log likelihood -3050872 Hannan-Quinn criter 3795743
Durbin-Watson stat 2331385
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15477
2 0002 0002 0253661 15945
3 0 0 0 15952 0207
4 0015 0015 1902459 52146 0074
5 -0004 -0004 -050732 54646 0141
6 -0014 -0014 -177563 8698 0069
7 -0003 -0002 -025366 88176 0117
8 0008 0008 1014645 99152 0128
9 -0003 -0003 -038049 10031 0187
10 0018 0019 2409781 15317 0053
11 -0007 -0008 -101464 16206 0063
12 0013 0013 1648798 19079 0039
13 0005 0005 0634153 19468 0053
14 -0004 -0005 -063415 19747 0072
15 0 0 0 19748 0102
16 0011 0011 1395137 21735 0084
17 -001 -0011 -139514 23501 0074
18 -001 -0009 -114148 25015 007
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0003 0003 0380492 25131 0092
20 0005 0004 0507322 25486 0112
21 -0006 -0006 -076098 26126 0127
22 -001 -001 -126831 27786 0115
23 -0003 -0004 -050732 2797 0141
24 0013 0012 1521967 30601 0105
25 -0013 -0013 -16488 3345 0074
26 -0016 -0015 -190246 37473 0039
27 0006 0007 0887814 38039 0046
28 0005 0004 0507322 38424 0055
29 0008 0009 1141475 3955 0056
30 0004 0005 0634153 3977 0069
31 -0013 -0013 -16488 42344 0052
32 0005 0004 0507322 42673 0063
33 0011 0011 1395137 44582 0054
34 -0012 -0012 -152197 46866 0044
35 -0007 -0006 -076098 47736 0047
36 0001 0001 0126831 4777 0059
Estimation Correlogram Q statistics of Residuals
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
GARCH (23)
Variable Coefficient Std Error z-Statistic Prob
C 0020273 0003412 5940926 00000
AR(1) -0137986 0072567 -1901495 00572
MA(1) 0242984 0070989 3422855 00006
Variance Equation
C 430E-06 860E-07 5000378 00000
RESID(-1)^2 0085055 0002502 3400018 00000
RESID(-2)^2 -0083594 0002442 -3423010 00000
GARCH(-1) 1871332 0006706 2790638 00000
GARCH(-2) -0872703 0006578 -1326706 00000
R-squared -0023536 Mean dependent var 0111371
Adjusted R-squared -0023663 SD dependent var 7606298
SE of regression 7695767 Akaike info criterion 3794703
Sum squared resid 9525129 Schwarz criterion 3798525
Log likelihood -3051279 Hannan-Quinn criter 3795967
Durbin-Watson stat 2330303
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 001 001 1268306 15963
2 0001 0001 0126831 16214
3 0 0 0 16214 0203
4 0014 0014 1775628 49305 0085
5 -0004 -0004 -050732 51411 0162
6 -0014 -0014 -177563 85074 0075
7 -0003 -0003 -038049 86544 0124
8 0008 0008 1014645 96626 014
9 -0002 -0002 -025366 97411 0204
10 0018 0018 2282951 14693 0065
11 -0007 -0008 -101464 15589 0076
12 0013 0013 1648798 18335 005
13 0005 0005 0634153 18735 0066
14 -0004 -0005 -063415 19059 0087
15 0 0 0 19059 0121
16 0011 0011 1395137 21107 0099
17 -0011 -0011 -139514 22955 0085
18 -001 -0009 -114148 24507 0079
Lags AC PACSQRT(N)PAC Q-Stat Prob
19 0002 0003 0380492 24605 0104
20 0005 0004 0507322 24978 0126
21 -0006 -0006 -076098 25604 0142
22 -001 -001 -126831 27214 0129
23 -0003 -0004 -050732 27395 0158
24 0013 0013 1648798 30322 0111
25 -0014 -0014 -177563 33438 0074
26 -0016 -0016 -202929 37617 0038
27 0006 0007 0887814 38194 0044
28 0005 0004 0507322 38651 0053
29 0008 0008 1014645 39671 0055
30 0004 0005 0634153 39933 0067
31 -0013 -0013 -16488 42486 0051
32 0005 0004 0507322 42841 0061
33 0011 0011 1395137 44682 0053
34 -0012 -0012 -152197 46952 0043
35 -0007 -0006 -076098 47832 0046
36 0001 0001 0126831 4786 0058
Estimation Correlogram Q statistics of Residuals
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
The best fit model
Observations
The Residual diagnostics (previous slides) hint that GARCH(23) is a better fit compared to GARCH(22)
We check the squared residuals of GARCH(23) to affirm the goodness of fit
The correlogram affirms our result that GARCH (23) is indeed the best fit model
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
1 0011 0011 1395137 20998
2 0002 0002 0253661 21909
3 0001 0001 0126831 22065 0137
4 -0002 -0002 -025366 22434 0326
5 0 0 0 22444 0523
6 -001 -001 -126831 37929 0435
7 -0012 -0012 -152197 6194 0288
8 -0001 -0001 -012683 62132 04
9 001 001 1268306 77643 0354
10 0001 0 0 7769 0456
11 -0002 -0002 -025366 7816 0553
12 -0006 -0006 -076098 83879 0591
13 -0004 -0004 -050732 8619 0657
14 0002 0002 0253661 86817 073
15 -0005 -0005 -063415 91452 0762
16 -0004 -0004 -050732 94626 08
17 -0009 -0009 -114148 10793 0767
18 -0005 -0005 -063415 11272 0792
lags AC PAC
SQRT(N)
PAC Q-Stat Prob
19 0002 0002 0253661 11368 0837
20 -0002 -0002 -025366 11424 0876
21 -0002 -0001 -012683 1146 0907
22 -0005 -0005 -063415 11849 0921
23 0 0 0 11849 0944
24 0002 0001 0126831 11894 096
25 -0001 -0001 -012683 11921 0972
26 -0022 -0021 -266344 19371 0732
27 -0005 -0004 -050732 19765 0759
28 -0008 -0008 -101464 20862 0749
29 0001 0001 0126831 20871 0792
30 -0008 -0008 -101464 21971 0783
31 -0005 -0005 -063415 22344 0806
32 -0007 -0008 -101464 23234 0806
33 0009 0009 1141475 24675 0782
34 0004 0003 0380492 24903 081
35 0008 0008 1014645 25907 0805
36 0002 0002 0253661 25965 0837
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
Conclusion
The data has significant stochastic trend which got removed after first differencing
No presence of seasonality in the data
The best fitted conditional mean model is ARMA(11)
Heteroskedasticity exists in the data
The best fitted volatility model is GARCH(23)
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das
DATA SOURCE Yahoo FinanceSOFTWARE USEDEviews 7Microsoft Office 2013
Reference ARCHGARCH Models by Prof Samarjit Das