Regression Arima Garch Var Model

download Regression Arima Garch Var Model

of 19

Transcript of Regression Arima Garch Var Model

  • 7/28/2019 Regression Arima Garch Var Model

    1/19

    Forecasting using Regression,ARIMA and VAR Models

    Submitted to Prof. Sajal Ghosh for partial fulfillment of Part TimePGPM Oct 2011 batch

    Modeling and Forecasting in Energy and Financial

    Markets

    Submitted by:

    Manish Jindal (11PT2-36)

  • 7/28/2019 Regression Arima Garch Var Model

    2/19

    1. RegressionInput Data Used (from E-Views Database):

    Table 7.3

    Real Gross Product, Labor Days and Real Capital Input in theAgricultural Sector, Taiwan, 1958-1972

    YEAR = YearY = Real Gross Product, Millions of NT $X2 = Labor Days, Millions of DaysX3 = Real Capital Input, Millions of NT $

    YEAR Y X2 X31958 16607.7 275.5 17803.71959 17511.3 274.4 18096.81960 20171.2 269.7 18271.81961 20932.9 267.0 19167.3

    1962 20406.0 267.8 19647.61963 20831.6 275.0 20803.51964 24806.3 283.0 22076.61965 26465.8 300.7 23445.21966 27403.0 307.5 24939.01967 28628.7 303.7 26713.71968 29904.5 304.7 29957.81969 27508.2 298.6 31585.91970 29035.5 295.5 33474.51971 29281.5 299.0 34821.81972 31535.8 288.1 41794.3

    E-views Output for multiple linear regression model:

    ls y c x2 x3

    Dependent Variable: Y

    Method: Least Squares

    Date: 04/24/13 Time: 17:39

    Sample: 1958 1972

    Included observations: 15

    Variable Coefficient Std. Error t-Statistic Prob.

    C -28067.17 9432.066 -2.975718 0.0116

    X2 147.9362 36.44344 4.059338 0.0016

    X3 0.403563 0.073561 5.486131 0.0001

    R-squared 0.909559 Mean dependent var 24735.33

    Adjusted R-squared 0.894485 S.D. dependent var 4874.173

    S.E. of regression 1583.279 Akaike info criterion 17.74924

    Sum squared resid 30081287 Schwarz criterion 17.89085

    Management Development Institute Page 2 October, 2011 PT-PGPM

  • 7/28/2019 Regression Arima Garch Var Model

    3/19

    Log likelihood -130.1193 F-statistic 60.34143

    Durbin-Watson stat 1.039019 Prob(F-statistic) 0.000001

    Analysis:

    High R-squared and Adjusted R-squared indicate a good explanatory power of the

    model.

    t-test (Probability values < 0.05) indicates that both Labour Days (X2) and Real Capital

    Input (X3) are significant in explaining the variability of the explained variable, Real

    GDP.

    Also, F-test (probability value < 0.05) indicates that the model as a whole is significant.

    However, the Durbin-Watson stat value of 1.039 is in the doubtful range and therefore,

    serial correlation LM test is performed to make sure no auto correlation exists in the

    model. Lag order of 1 is used for LM test. The outcome is given below:

    Serial Correlation LM Test:

    Breusch-Godfrey Serial Correlation LM Test:

    F-statistic 1.237281 Probability 0.289714

    Obs*R-squared 1.516612 Probability 0.218133

    Test Equation:

    Dependent Variable: RESID

    Method: Least SquaresDate: 04/24/13 Time: 18:00

    Presample missing value lagged residuals set to zero.

    Variable Coefficient Std. Error t-Statistic Prob.

    C 3243.336 9784.721 0.331469 0.7465

    X2 -12.76956 37.87033 -0.337192 0.7423

    X3 0.016801 0.074393 0.225839 0.8255

    RESID(-1) 0.333779 0.300071 1.112331 0.2897

    R-squared 0.101107 Mean dependent var -1.36E-11

    Adjusted R-squared -0.144045 S.D. dependent var 1465.832

    S.E. of regression 1567.854 Akaike info criterion 17.77598

    Sum squared resid 27039844 Schwarz criterion 17.96480

    Log likelihood -129.3199 F-statistic 0.412427

    Durbin-Watson stat 1.437784 Prob(F-statistic) 0.747397

    Management Development Institute Page 3 October, 2011 PT-PGPM

  • 7/28/2019 Regression Arima Garch Var Model

    4/19

    Here, the Null hypothesis is that no auto correlation exists in the model. A probability of

    0.21 and 0.28 are greater than the threshold of 0.05, therefore, Null Hypothesis of No

    Auto Correlation existing is accepted.

    Check for Multicollinearity:

    Correlation Matrix:

    X2 X3

    X2 1.000000 0.620523

    X3 0.620523 1.000000

    Correlation between X2 and X3 is found to be 0.620523 which is not high enough to

    indicate a multicollinearity problem existing in the model. Therefore, the model can now

    be used for forecasting.

    The final regression model used for forecasting is as follows:Y = -28067.16664 + 147.9362123*X2 + 0.4035625699*X3

    Forecasting using the model:

    12000

    16000

    20000

    24000

    28000

    32000

    36000

    58 59 60 61 62 63 64 65 66 67 68 69 70 71 72

    YF

    Forecast: YF

    Actual: Y

    Forecast sample: 1958 1972

    Included observations: 15

    Root Mean Squared Error 1416.128

    Mean Absolute Error 1100.348

    Mean Abs. Percent Error 5.169292

    Theil Inequality Coefficient 0.028143

    Bias Proportion 0.000000Variance Proportion 0.023694

    Covariance Proportion 0.976306

    Theil Inequality Coefficient of 0.023 is quite close to zero, which indicates a near perfect

    fit of the model for forecasting purpose.

    Management Development Institute Page 4 October, 2011 PT-PGPM

  • 7/28/2019 Regression Arima Garch Var Model

    5/19

    2. ARIMAInput Data Used peak demand.xls:

    peak demand.xls

    E-views Output for correlogram:

    Date: 06/23/13 Time: 14:59

    Sample: 2000M04 2008M07

    Included observations: 100

    Autocorrelation Partial Correlation AC PAC Q-Stat Prob

    . |*******| . |*******| 1 0.970 0.970 96.941 0.000

    . |*******| . | . | 2 0.940 -0.015 188.91 0.000

    . |*******| . | . | 3 0.910 -0.015 275.99 0.000

    . |*******| . | . | 4 0.880 -0.015 358.29 0.000

    . |*******| . | . | 5 0.850 -0.015 435.90 0.000

    . |****** | . | . | 6 0.820 -0.015 508.94 0.000

    . |****** | . | . | 7 0.791 -0.015 577.50 0.000

    . |****** | . | . | 8 0.761 -0.015 641.71 0.000

    . |****** | . | . | 9 0.731 -0.015 701.67 0.000

    . |***** | . | . | 10 0.702 -0.015 757.52 0.000

    . |***** | . | . | 11 0.673 -0.015 809.38 0.000

    . |***** | . | . | 12 0.643 -0.016 857.38 0.000

    . |***** | . | . | 13 0.614 -0.015 901.64 0.000

    . |**** | . | . | 14 0.586 -0.016 942.30 0.000

    . |**** | . | . | 15 0.557 -0.015 979.50 0.000

    . |**** | . | . | 16 0.528 -0.016 1013.4 0.000

    . |**** | . | . | 17 0.500 -0.015 1044.1 0.000

    . |**** | . | . | 18 0.472 -0.016 1071.8 0.000

    . |*** | . | . | 19 0.444 -0.015 1096.6 0.000

    . |*** | . | . | 20 0.416 -0.015 1118.6 0.000

    . |*** | . | . | 21 0.389 -0.016 1138.1 0.000

    . |*** | . | . | 22 0.361 -0.015 1155.2 0.000

    . |*** | . | . | 23 0.334 -0.015 1170.0 0.000

    . |** | . | . | 24 0.308 -0.016 1182.7 0.000

    . |** | . | . | 25 0.281 -0.015 1193.5 0.000

    . |** | . | . | 26 0.255 -0.016 1202.4 0.000

    . |** | . | . | 27 0.229 -0.015 1209.8 0.000

    . |** | . | . | 28 0.204 -0.016 1215.7 0.000

    . |*. | . | . | 29 0.179 -0.015 1220.3 0.000

    . |*. | . | . | 30 0.154 -0.016 1223.7 0.000

    Management Development Institute Page 5 October, 2011 PT-PGPM

  • 7/28/2019 Regression Arima Garch Var Model

    6/19

    . |*. | . | . | 31 0.130 -0.015 1226.2 0.000

    . |*. | . | . | 32 0.106 -0.015 1227.9 0.000

    . |*. | . | . | 33 0.082 -0.016 1228.9 0.000

    . | . | . | . | 34 0.059 -0.015 1229.4 0.000

    . | . | . | . | 35 0.036 -0.015 1229.6 0.000

    . | . | . | . | 36 0.013 -0.016 1229.7 0.000

    The correlogram shows signature of AR(1) process, however, it could be an illusion untilstationarity of the series is confirmed.

    Performing the ADF Unit Root test for check of

    stationarity:

    ADF unit root test output at level:

    Null Hypothesis: DEMAND has a unit root

    Exogenous: Constant, Linear Trend

    Lag Length: 0 (Automatic based on SIC, MAXLAG=11)

    t-Statistic Prob.*

    Augmented Dickey-Fuller test statistic -3.370152 0.0626

    Test critical values: 1% level -4.073859

    5% level -3.465548

    10% level -3.159372

    *MacKinnon (1996) one-sided p-values.

    Augmented Dickey-Fuller Test Equation

    Dependent Variable: D(DEMAND)

    Method: Least Squares

    Date: 06/23/13 Time: 15:03

    Sample (adjusted): 2000M05 2007M02

    Included observations: 82 after adjustments

    Variable Coefficient Std. Error t-Statistic Prob.

    DEMAND(-1) -0.249099 0.073913 -3.370152 0.0012

    C 4861.437 1432.497 3.393681 0.0011

    @TREND(2000M04) 29.69826 9.602382 3.092802 0.0027

    R-squared 0.125876 Mean dependent var 104.3415

    Adjusted R-squared 0.103747 S.D. dependent var 946.6032

    S.E. of regression 896.1555 Akaike info criterion 16.47000

    Sum squared resid 63444484 Schwarz criterion 16.55806

    Log likelihood -672.2702 F-statistic 5.688110

    Durbin-Watson stat 1.724296 Prob(F-statistic) 0.004922

    Management Development Institute Page 6 October, 2011 PT-PGPM

  • 7/28/2019 Regression Arima Garch Var Model

    7/19

    The probability value of >5% indicates that H0 of unit root present is accepted at 5%

    critical level and therefore, ADF Unit root test is done again on first difference with the

    following outcome:

    Null Hypothesis: D(DEMAND) has a unit root

    Exogenous: Constant

    Lag Length: 9 (Automatic based on SIC, MAXLAG=11)

    t-Statistic Prob.*

    Augmented Dickey-Fuller test statistic -7.571975 0.0000

    Test critical values: 1% level -3.524233

    5% level -2.902358

    10% level -2.588587

    *MacKinnon (1996) one-sided p-values.

    Augmented Dickey-Fuller Test Equation

    Dependent Variable: D(DEMAND,2)

    Method: Least Squares

    Date: 06/23/13 Time: 15:08

    Sample (adjusted): 2001M03 2007M02

    Included observations: 72 after adjustments

    Variable Coefficient Std. Error t-Statistic Prob.

    D(DEMAND(-1)) -4.811954 0.635495 -7.571975 0.0000

    D(DEMAND(-1),2) 3.506694 0.574176 6.107346 0.0000

    D(DEMAND(-2),2) 3.219195 0.511100 6.298564 0.0000

    D(DEMAND(-3),2) 2.809151 0.456585 6.152531 0.0000

    D(DEMAND(-4),2) 2.374129 0.408596 5.810456 0.0000

    D(DEMAND(-5),2) 2.039335 0.353954 5.761583 0.0000

    D(DEMAND(-6),2) 1.665176 0.295530 5.634549 0.0000

    D(DEMAND(-7),2) 1.228536 0.244074 5.033460 0.0000

    D(DEMAND(-8),2) 0.815078 0.190846 4.270875 0.0001

    D(DEMAND(-9),2) 0.463414 0.121058 3.828023 0.0003

    C 579.9569 122.9400 4.717396 0.0000

    R-squared 0.692708 Mean dependent var -1.263889

    Adjusted R-squared 0.642332 S.D. dependent var 1357.426

    S.E. of regression 811.8135 Akaike info criterion 16.37618

    Sum squared resid 40201512 Schwarz criterion 16.72401

    Log likelihood -578.5425 F-statistic 13.75081

    Durbin-Watson stat 1.980756 Prob(F-statistic) 0.000000

    Management Development Institute Page 7 October, 2011 PT-PGPM

  • 7/28/2019 Regression Arima Garch Var Model

    8/19

    The H0 of unit root is now rejected and therefore, the series is considered I(1)

    Checking the correlogram with first difference series for

    trend and seasonality:

    Trend component is added in the first difference series series dd=d(demand, 1,0) and

    correlogram checked with following outcome:

    Date: 06/23/13 Time: 15:14

    Sample: 2000M04 2008M07

    Included observations: 82

    Autocorrelation Partial Correlation AC PAC Q-Stat Prob

    . | . | . | . | 1 0.030 0.030 0.0747 0.785

    . | . | . | . | 2 0.016 0.015 0.0967 0.953

    .*| . | .*| . | 3 -0.111 -0.112 1.1721 0.760

    **| . | **| . | 4 -0.199 -0.195 4.6597 0.324

    .*| . | .*| . | 5 -0.112 -0.106 5.7834 0.328

    .*| . | .*| . | 6 -0.134 -0.147 7.4030 0.285

    .*| . | **| . | 7 -0.165 -0.229 9.9096 0.194

    .*| . | **| . | 8 -0.131 -0.243 11.508 0.175

    . | . | **| . | 9 -0.042 -0.200 11.675 0.232

    .*| . | ***| . | 10 -0.094 -0.353 12.528 0.251

    . |*** | . |*. | 11 0.352 0.085 24.557 0.011

    . |** | . |*. | 12 0.252 0.079 30.822 0.002

    . |** | . |*. | 13 0.260 0.167 37.577 0.000

    . | . | . | . | 14 0.003 -0.016 37.578 0.001

    .*| . | . | . | 15 -0.123 -0.046 39.137 0.001

    .*| . | .*| . | 16 -0.151 -0.090 41.512 0.000

    .*| . | . | . | 17 -0.099 -0.020 42.558 0.001

    .*| . | . | . | 18 -0.090 -0.005 43.427 0.001

    . | . | . |*. | 19 0.006 0.169 43.431 0.001

    .*| . | . | . | 20 -0.129 -0.036 45.292 0.001

    **| . | .*| . | 21 -0.195 -0.146 49.592 0.000

    . |*. | . | . | 22 0.103 -0.018 50.814 0.000

    . |*. | . | . | 23 0.098 -0.053 51.932 0.001

    . |** | . | . | 24 0.308 0.057 63.234 0.000

    . |*. | . | . | 25 0.156 0.035 66.175 0.000

    . | . | . | . | 26 -0.022 -0.049 66.234 0.000

    .*| . | . | . | 27 -0.084 -0.038 67.111 0.000

    . | . | . | . | 28 -0.044 0.062 67.359 0.000

    .*| . | . |*. | 29 -0.074 0.075 68.075 0.000

    .*| . | .*| . | 30 -0.083 -0.096 68.979 0.000

    Management Development Institute Page 8 October, 2011 PT-PGPM

  • 7/28/2019 Regression Arima Garch Var Model

    9/19

    . | . | . | . | 31 -0.050 -0.055 69.315 0.000

    .*| . | . | . | 32 -0.110 0.009 70.994 0.000

    .*| . | . | . | 33 -0.077 -0.039 71.830 0.000

    . |*. | . |*. | 34 0.079 0.173 72.732 0.000

    . | . | .*| . | 35 0.061 -0.061 73.273 0.000

    . |** | . | . | 36 0.221 0.034 80.603 0.000

    Nothing can be derived from the correlogram regarding AR and MA components so

    trying with the seasonal component (with 12 as seasonality): Series dd=d(demand,0,12)

    Date: 06/23/13 Time: 15:15

    Sample: 2000M04 2008M07

    Included observations: 71

    Autocorrelation Partial Correlation AC PAC Q-Stat Prob

    . |**** | . |**** | 1 0.486 0.486 17.479 0.000

    . |**** | . |** | 2 0.475 0.313 34.414 0.000

    . |*** | . |*. | 3 0.380 0.101 45.423 0.000

    . |*** | . | . | 4 0.340 0.063 54.342 0.000

    . |** | . | . | 5 0.303 0.050 61.572 0.000

    . |** | . | . | 6 0.222 -0.041 65.492 0.000

    . |*. | .*| . | 7 0.088 -0.166 66.118 0.000

    . |*. | . | . | 8 0.132 0.054 67.557 0.000

    . |** | . |** | 9 0.246 0.266 72.615 0.000

    . |*. | .*| . | 10 0.072 -0.152 73.050 0.000

    . |*. | . |*. | 11 0.174 0.085 75.652 0.000.*| . | ***| . | 12 -0.142 -0.398 77.429 0.000

    . | . | . |** | 13 0.062 0.201 77.767 0.000

    . | . | . | . | 14 0.043 0.058 77.939 0.000

    . | . | . |*. | 15 0.051 0.093 78.178 0.000

    . | . | . |*. | 16 0.031 0.072 78.271 0.000

    . |*. | . |*. | 17 0.096 0.069 79.151 0.000

    . |*. | . |*. | 18 0.141 0.072 81.100 0.000

    . |*. | . | . | 19 0.195 -0.022 84.903 0.000

    . |*. | . | . | 20 0.177 -0.053 88.092 0.000

    . | . | . | . | 21 0.008 -0.047 88.098 0.000

    . |*. | . | . | 22 0.145 0.001 90.326 0.000

    . |*. | . |*. | 23 0.071 0.124 90.864 0.000

    . |*. | .*| . | 24 0.125 -0.146 92.584 0.000

    . | . | . |*. | 25 0.063 0.095 93.038 0.000

    . |*. | . | . | 26 0.092 0.020 94.015 0.000

    . | . | .*| . | 27 0.058 -0.062 94.406 0.000

    . |*. | . | . | 28 0.106 -0.028 95.760 0.000

    . | . | .*| . | 29 0.021 -0.084 95.812 0.000

    .*| . | .*| . | 30 -0.123 -0.158 97.721 0.000

    Management Development Institute Page 9 October, 2011 PT-PGPM

  • 7/28/2019 Regression Arima Garch Var Model

    10/19

    .*| . | .*| . | 31 -0.151 -0.118 100.68 0.000

    **| . | .*| . | 32 -0.237 -0.112 108.16 0.000

    This correlogram shows an indication of AR(2), SMA(12) process. Confirming the

    significance of ar(1), ar(2) and sma(12) using regression

    Confirming AR and MA components from the

    correlogram:

    ls d(demand,0,12) c ar(1) ar(2) sma(12)

    Dependent Variable: D(DEMAND,0,12)

    Method: Least Squares

    Date: 06/23/13 Time: 15:18

    Sample (adjusted): 2001M06 2007M02

    Included observations: 69 after adjustmentsConvergence achieved after 12 iterations

    Backcast: 2000M06 2001M05

    Variable Coefficient Std. Error t-Statistic Prob.

    C 1701.522 214.0576 7.948896 0.0000

    AR(1) 0.471909 0.118750 3.973966 0.0002

    AR(2) 0.287270 0.117415 2.446625 0.0171

    MA(12) -0.854022 0.041146 -20.75597 0.0000

    R-squared 0.605499 Mean dependent var 1484.377

    Adjusted R-squared 0.587291 S.D. dependent var 1103.399

    S.E. of regression 708.8503 Akaike info criterion 16.02139

    Sum squared resid 32660470 Schwarz criterion 16.15090

    Log likelihood -548.7379 F-statistic 33.25499

    Durbin-Watson stat 2.055818 Prob(F-statistic) 0.000000

    Inverted AR Roots .82 -.35

    Inverted MA Roots .99 .85+.49i .85-.49i .49+.85i

    .49-.85i -.00-.99i -.00+.99i -.49-.85i

    -.49+.85i -.85+.49i -.85-.49i -.99

    All 3 processes are showing significance at 5% confidence level. Confirming whether the

    Residual is white Noise of not.

    Date: 06/23/13 Time: 15:20

    Sample: 2001M06 2007M02

    Included observations: 69

    Management Development Institute Page 10 October, 2011 PT-PGPM

  • 7/28/2019 Regression Arima Garch Var Model

    11/19

    Q-statisticprobabilities

    adjusted for 3 ARMAterm(s)

    Autocorrelation Partial Correlation AC PAC Q-Stat Prob

    . | . | . | . | 1 -0.043 -0.043 0.1322

    .*| . | .*| . | 2 -0.075 -0.077 0.5457

    . | . | . | . | 3 0.052 0.046 0.7499

    . | . | . | . | 4 -0.022 -0.024 0.7863 0.375

    . | . | . | . | 5 0.006 0.012 0.7892 0.674

    . | . | . | . | 6 -0.029 -0.035 0.8565 0.836

    **| . | **| . | 7 -0.203 -0.205 4.1276 0.389

    . | . | . | . | 8 -0.011 -0.038 4.1382 0.530

    . |*. | . |*. | 9 0.138 0.114 5.6902 0.459

    .*| . | .*| . | 10 -0.097 -0.075 6.4719 0.486

    . |*. | . |** | 11 0.182 0.199 9.2745 0.320

    .*| . | .*| . | 12 -0.119 -0.145 10.488 0.312

    . | . | . |*. | 13 0.043 0.070 10.650 0.385

    . | . | . | . | 14 0.037 -0.051 10.772 0.463

    . | . | . |*. | 15 0.035 0.073 10.885 0.539

    . | . | . | . | 16 -0.047 -0.016 11.092 0.603

    .*| . | .*| . | 17 -0.066 -0.078 11.508 0.646

    . | . | . | . | 18 -0.029 0.004 11.587 0.710

    . |*. | . |*. | 19 0.133 0.125 13.319 0.649

    . | . | . | . | 20 0.000 -0.044 13.319 0.715

    **| . | .*| . | 21 -0.227 -0.150 18.556 0.420

    . | . | .*| . | 22 0.016 -0.075 18.583 0.484

    .*| . | .*| . | 23 -0.109 -0.108 19.848 0.467

    . | . | .*| . | 24 -0.007 -0.079 19.854 0.530

    . | . | . |*. | 25 0.047 0.076 20.096 0.577

    . | . | . |*. | 26 0.025 0.080 20.169 0.632

    . | . | . | . | 27 -0.017 -0.006 20.202 0.685

    . |*. | . | . | 28 0.136 0.050 22.401 0.612

    Confirmed that the residuals components are white noise; so proceeding with forecasting:

    Forecasting using the model:

    Outcome of the static forecasting:

    Management Development Institute Page 11 October, 2011 PT-PGPM

  • 7/28/2019 Regression Arima Garch Var Model

    12/19

    18000

    20000

    22000

    24000

    26000

    28000

    30000

    32000

    34000

    2001 2002 2003 2004 2005 2006

    DEMANDF

    Forecast: DEMANDF

    Actual: DEMAND

    Forecast sample: 2000M04 2008M07

    Adjusted sample: 2001M06 2007M03

    Included observations: 69

    Root Mean Squared Error 687.9972

    Mean Absolute Error 588.4692

    Mean Abs. Percent Error 2.420978

    Theil Inequality Coefficient 0.013758

    Bias Proportion 0.013140

    Variance Proportion 0.020256

    Covariance Proportion 0.966604

    Outcome of dynamic forecasting:

    26000

    28000

    30000

    32000

    34000

    36000

    07M01 07M04 07M07 07M10 08M01 08M04 08M07

    DEMANDF

    Forecast: DEMANDF

    Actual: DEMAND

    Forecast sample: 2007M01 2008M07

    Included observations: 2

    Root Mean Squared Error 358.5803

    Mean Absolute Error 358.5267

    Mean Abs. Percent Error 1.239447

    Theil Inequality Coefficient 0.006235

    Bias Proportion 0.999701

    Variance Proportion 0.000299

    Covariance Proportion 0.000000

    MAPE is between 1% and 3%, which is good enough so model can be used for forecasting.

    Management Development Institute Page 12 October, 2011 PT-PGPM

  • 7/28/2019 Regression Arima Garch Var Model

    13/19

    3. VARInput Data Used sensex_ex_oil_VAR.xls:

    sensex_ex_oil_VAR.xls

    Stationarity test for all the three series:

    ADF Unit Root Test for Sensex:

    Null Hypothesis: SEN has a unit root

    Exogenous: Constant, Linear Trend

    Lag Length: 0 (Automatic based on SIC, MAXLAG=19)

    t-Statistic Prob.*

    Augmented Dickey-Fuller test statistic -1.535520 0.8165

    Test critical values: 1% level -3.972503

    5% level -3.416877

    10% level -3.130796

    *MacKinnon (1996) one-sided p-values.

    Augmented Dickey-Fuller Test Equation

    Dependent Variable: D(SEN)Method: Least Squares

    Date: 06/23/13 Time: 15:54

    Sample (adjusted): 2 639

    Included observations: 638 after adjustments

    Variable Coefficient Std. Error t-Statistic Prob.

    SEN(-1) -0.008067 0.005254 -1.535520 0.1252

    C 134.9426 66.52498 2.028450 0.0429

    @TREND(1) 0.037049 0.087721 0.422351 0.6729

    R-squared 0.007247 Mean dependent var 12.91376

    Adjusted R-squared 0.004121 S.D. dependent var 229.7398

    S.E. of regression 229.2660 Akaike info criterion 13.71233

    Sum squared resid 33377444 Schwarz criterion 13.73330

    Log likelihood -4371.235 F-statistic 2.317821

    Durbin-Watson stat 1.896014 Prob(F-statistic) 0.099321

    Management Development Institute Page 13 October, 2011 PT-PGPM

  • 7/28/2019 Regression Arima Garch Var Model

    14/19

  • 7/28/2019 Regression Arima Garch Var Model

    15/19

    Outcome with lag 2

    Vector Autoregression Estimates

    Date: 06/23/13 Time: 15:59

    Sample (adjusted): 4 639

    Included observations: 636 after adjustments

    Standard errors in ( ) & t-statistics in [ ]

    DLOG(SEN) DLOG(OIL) DLOG(EX)

    DLOG(SEN(-1)) 0.062916 0.041426 -0.007926

    (0.03938) (0.05878) (0.01270)

    [ 1.59786] [ 0.70476] [-0.62406]

    DLOG(SEN(-2)) -0.047235 0.069754 -0.032615

    (0.03926) (0.05861) (0.01266)

    [-1.20309] [ 1.19016] [-2.57540]

    DLOG(OIL(-1)) 0.027833 -0.046923 0.000759

    (0.02660) (0.03970) (0.00858)

    [ 1.04652] [-1.18187] [ 0.08852]

    DLOG(OIL(-2)) -0.046258 -0.019528 -0.012440

    (0.02550) (0.03807) (0.00823)

    [-1.81390] [-0.51295] [-1.51229]

    DLOG(EX(-1)) -0.043623 -0.018486 -0.032198

    (0.12249) (0.18286) (0.03951)

    [-0.35613] [-0.10110] [-0.81489]

    DLOG(EX(-2)) -0.461301 0.051077 -0.061351

    (0.12216) (0.18236) (0.03940)

    [-3.77614] [ 0.28008] [-1.55695]

    C 0.000873 0.001302 -4.56E-05

    (0.00063) (0.00094) (0.00020)

    [ 1.39264] [ 1.39148] [-0.22567]

    R-squared 0.033549 0.005954 0.018689

    Adj. R-squared 0.024330 -0.003529 0.009328

    Sum sq. resids 0.154639 0.344605 0.016089

    S.E. equation 0.015680 0.023406 0.005058

    F-statistic 3.639180 0.627866 1.996506

    Log likelihood 1743.907 1489.092 2463.518

    Akaike AIC -5.461973 -4.660667 -7.724899

    Schwarz SC -5.412938 -4.611632 -7.675864

    Mean dependent 0.000889 0.001300 -9.69E-05

    Management Development Institute Page 15 October, 2011 PT-PGPM

  • 7/28/2019 Regression Arima Garch Var Model

    16/19

    S.D. dependent 0.015874 0.023365 0.005081

    Determinant resid covariance (dof adj.) 3.43E-12

    Determinant resid covariance 3.32E-12

    Log likelihood 5698.165

    Akaike information criterion -17.85272Schwarz criterion -17.70561

    The lag length criteria output is as follows:

    VAR Lag Order Selection CriteriaEndogenous variables: DLOG(SEN) DLOG(OIL)DLOG(EX)

    Exogenous variables: C

    Date: 06/23/13 Time: 16:01

    Sample: 1 639Included observations: 630

    Lag LogL LR FPE AIC SC HQ

    0 5655.428 NA 3.23e-12 -17.94421 -17.92304* -17.93599*

    1 5658.995 7.088918 3.29e-12 -17.92697 -17.84229 -17.89408

    2 5675.220 32.09070 3.21e-12 -17.94991 -17.80172 -17.89234

    3 5697.734 44.31307* 3.08e-12* -17.99281* -17.78111 -17.91058

    4 5703.470 11.23404 3.11e-12 -17.98244 -17.70723 -17.87554

    5 5711.659 15.96324 3.12e-12 -17.97987 -17.64115 -17.84830

    6 5715.384 7.225424 3.17e-12 -17.96312 -17.56089 -17.80689

    7 5721.169 11.16626 3.20e-12 -17.95292 -17.48718 -17.77201

    8 5727.525 12.20586 3.23e-12 -17.94452 -17.41527 -17.73895

    * indicates lag order selected by the criterion

    LR: sequential modified LR test statistic (each test at 5% level)

    FPE: Final prediction error

    AIC: Akaike information criterion

    SC: Schwarz information criterion

    HQ: Hannan-Quinn information criterion

    The lag length 3 is selected by the criterion and therefore, it will be used for furtheranalysis.

    Cointegration Test: The Johansen and Jeselius cointegration test is performed to validate the existence of

    any cointegration between variables with the following outcome:

    Management Development Institute Page 16 October, 2011 PT-PGPM

  • 7/28/2019 Regression Arima Garch Var Model

    17/19

    Date: 06/23/13 Time: 16:06

    Sample (adjusted): 6 639

    Included observations: 634 after adjustments

    Trend assumption: Linear deterministic trend

    Series: DLOG(SEN) DLOG(OIL) DLOG(EX)

    Lags interval (in first differences): 1 to 3

    Unrestricted Cointegration Rank Test (Trace)

    Hypothesized Trace 0.05

    No. of CE(s) Eigenvalue Statistic Critical Value Prob.**

    None * 0.272322 472.1336 29.79707 0.0001

    At most 1 * 0.234106 270.5869 15.49471 0.0001

    At most 2 * 0.147925 101.4914 3.841466 0.0000

    Trace test indicates 3 cointegrating eqn(s) at the 0.05 level

    * denotes rejection of the hypothesis at the 0.05 level

    **MacKinnon-Haug-Michelis (1999) p-values

    Unrestricted Cointegration Rank Test (Maximum Eigenvalue)

    Hypothesized Max-Eigen 0.05

    No. of CE(s) Eigenvalue Statistic Critical Value Prob.**

    None * 0.272322 201.5467 21.13162 0.0001

    At most 1 * 0.234106 169.0955 14.26460 0.0001

    At most 2 * 0.147925 101.4914 3.841466 0.0000

    Max-eigenvalue test indicates 3 cointegrating eqn(s) at the 0.05 level

    * denotes rejection of the hypothesis at the 0.05 level

    **MacKinnon-Haug-Michelis (1999) p-values

    Unrestricted Cointegrating Coefficients (normalized by b'*S11*b=I):

    DLOG(SEN) DLOG(OIL) DLOG(EX)

    -106.7013 16.58454 -290.3165

    12.35526 88.88217 94.37331

    69.51591 11.45247 -287.3705

    Unrestricted Adjustment Coefficients (alpha):

    D(DLOG(SEN)) 0.007827 -0.001086 -0.003048

    D(DLOG(OIL)) -0.003808 -0.012067 -0.001340

    D(DLOG(EX)) 0.001309 -0.000759 0.001783

    Management Development Institute Page 17 October, 2011 PT-PGPM

  • 7/28/2019 Regression Arima Garch Var Model

    18/19

    1 Cointegrating Equation(s): Log likelihood 5590.689

    Normalized cointegrating coefficients (standard error in parentheses)

    DLOG(SEN) DLOG(OIL) DLOG(EX)

    1.000000 -0.155430 2.720834

    (0.05594) (0.25062)

    Adjustment coefficients (standard error in parentheses)

    D(DLOG(SEN)) -0.835138

    (0.06505)

    D(DLOG(OIL)) 0.406332

    (0.11089)

    D(DLOG(EX)) -0.139619

    (0.02283)

    2 Cointegrating Equation(s): Log likelihood 5675.236

    Normalized cointegrating coefficients (standard error in parentheses)

    DLOG(SEN) DLOG(OIL) DLOG(EX)

    1.000000 0.000000 2.824833

    (0.25049)

    0.000000 1.000000 0.669108

    (0.32764)

    Adjustment coefficients (standard error in parentheses)

    D(DLOG(SEN)) -0.848561 0.033247

    (0.06532) (0.05498)

    D(DLOG(OIL)) 0.257236 -1.135738(0.09882) (0.08318)

    D(DLOG(EX)) -0.149002 -0.045799

    (0.02275) (0.01915)

    The H0 of Cointegration is rejected for all 3 possible pairs for existence of cointegration.

    The unrestricted VAR can, therefore, be applied now.

    Granger Causality Test:

    VAR Granger Causality/Block Exogeneity Wald Tests

    Date: 06/23/13 Time: 16:12

    Sample: 1 639

    Included observations: 635

    Dependent variable: DLOG(SEN)

    Management Development Institute Page 18 October, 2011 PT-PGPM

  • 7/28/2019 Regression Arima Garch Var Model

    19/19

    Excluded Chi-sq df Prob.

    DLOG(OIL) 5.287301 3 0.1519

    DLOG(EX) 49.19361 3 0.0000

    All 54.27170 6 0.0000

    Dependent variable: DLOG(OIL)

    Excluded Chi-sq df Prob.

    DLOG(SEN) 1.856538 3 0.6027

    DLOG(EX) 0.518338 3 0.9148

    All 2.386966 6 0.8809

    Dependent variable: DLOG(EX)

    Excluded Chi-sq df Prob.

    DLOG(SEN) 9.790470 3 0.0204

    DLOG(OIL) 4.567472 3 0.2064

    All 13.33178 6 0.0381

    The granger causality test with H0 of non-causality running from the independent

    variable DLOG(EX) to the dependent variable DLOG(SEN) has been rejected and

    therefore, it is confirmed that Exchange Rate movement Granger causes Sensex

    movement.

    Similarly, the H0 of non-causality running from the independent variable DLOG(SEN) to

    the dependent variable DLOG(EX) has been rejected and therefore, it is confirmed that

    Sensex movement Granger causes Exchange rate movement.

    In other words, there is a bidirectional causality that exists between the sensex and the

    exchange rate movement.

    Forecasting can be done now using the above VAR model with lag structure of 3 between

    Sensex and Exchange Rate.

    Management Development Institute Page 19 October 2011 PT PGPM