Assignment Fisher 1
-
Upload
ranjit-aulak -
Category
Documents
-
view
89 -
download
1
Transcript of Assignment Fisher 1
Testing the Existence
and Forecasting the
Fisher Effect in Malaysia
and Canada
BSc Banking and Finance
073604149
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
1.0 Introduction
For my project, I wish to test, forecast and compare to see if there is a fisher
effect in both Canada and Malaysia between the years of 1995 to 2009 with
quarterly seasonally adjusted data which I think will give me a good span of data to
build a model and so I can have a good comparison of the two countries. It is not as
much as I wanted, but I think it is enough for this project. My project is structured as
follows;
Section 2.0 is the economic and econometrics background theory of the
fisher effect.
Section 3.0 I will talk about previous published papers on the fisher effect and
its existence.
In section 4.0 I will explain my data, my variables and how I got them and
what I did to them.
Section 5.0 will be all my results from the tests I used from econometric
models and graphs.
Section 6.0 I will then finally evaluate my results and then compare my
results in the conclusion.
The reason I choose the two countries, that I thought I might get results that
differ due to continental reasons, which may affect the factors that affect the fisher
effect, and also I think that Malaysia is not such a well developed country in terms of
its economy as Canada is.
2.0 Economics Theory
The fisher effect was introduced in 1930, and it shows the relationship
between interest rates and inflation, and it shows that the nominal interest rate at a
given time is equal to the sum of the real interest and the expected rate of inflation.
There are three variables in the fisher equation, they are nominal inflation rate
2
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
which is denoted by ‘p’, which is the change in price level, the nominal interest rate
’i’ which is the actual interest rate without any adjustments in the economy and the
real interest rates ‘r’ which is the interest rate that has been adjusted to remove the
effects of inflation. This relationship can be denoted by:
r=i- π
The purchasing power can be found by finding the difference between the nominal
interest rate and the rate of inflation, and the above can be rearranged to give:
i=r+ π
the more accurate fisher effect equation is written as:
it=rt+ π et
where π et is the expected inflation rate. Once the nominal interest rate has been set,
the inflation rate is known, therefore only adjust to the expected inflation.
There are two types of Real interest rates, ex ante and ex post. Ex ante is when for
example the interest rate that the borrower and the lender expect to receive when
the take out a loan, the actual real interest rate realised is the ex post real interest
rate.
The fisher effect hypothesis says that the efficient capital markets
compensate for changes in the purchasing power of money in the long run will not
have an effect on a countries relative prices. For example, if the rate of expected
inflation increased by 2%, this should change the nominal interest rate and increase
that by 2%. This movement together of the expected inflation and nominal interest
rate is what the fisher effect is.
2.1 Econometric Theory
3
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
The nominal interest rate can be divided into two parts, the ex ante real
interest rate ret, and the expected inflation π e
t. This was proposed by Fisher (1930).
Using this, the fisher effect can be written as it= ret + πt
e . It is possible that the
expected inflation may be different from actual inflation. The errors are said to be
stationary, both types of interest rates have similar properties when they differ by a
stationary factor, therefore:
it= πte+rt+e where rt=it- πt
e
Changes in expected rate of inflation are shown by nominal interest rate which can
be shown by It= πte
This means that it+(it- πte)+( πt
e- πt)= πt+et, where E(et)=0 and is stationary.
The international fisher effect says that the difference in two countries
interest rate should be equal to the difference in its expected inflation. This means
that if a country has a high nominal interest rate then it will have a high expected
inflation.
3.0 Literature review
In the paper ‘Is there really the Fisher effect’ written by Paul A. Johnson,
Johnson writes about how the cointegration of the inflation and nominal interest
rates is constant with any theory implying a stationary interest rate and therefore is
not a sufficient condition for the fisher effect to hold.
Mishkin also wrote a paper ‘Is the Fisher effect for real’ where he found
support for the long-run fisher relationship where inflation and interest rates are
cointegrated. In his later paper he found that both interest rate and inflation
contained unit roots and the rest indicated that there was evidence of long run fisher
effect but none for short run fisher effect. His paper solves the problem of why
strong fisher effect occurs only in some periods and not others by re-examining the
4
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
relationship between inflation and interest rates with modern techniques. Mishkin
was one of the first to use the Engle-Granger concept of Cointegration successfully.
In another paper, ‘the fisher effect: new evidence and implications’ written
by Fahmy and M.Kandil, the results that they achieved did not support the short run
fisher effect because short-term interest rates are associated with small changes in
expected inflation. The results also did not favour the long run fisher effect and the
correlation between nominal interest rates and inflation rates until they move
together in relation. They use the Johansen test for cointegration.
Perez, S.J and M.V Siegler (2003) performed similar studies where they
explore the field of potential non linearity’s data sets.
A paper published by Arusha Cooray ‘THE FISHER EFFECT: A REVIEW OF THE
LITERATURE’ showed that by using the Johansen cointegration approach, there was a
presence of a relationship between nominal interest rates and inflation for some
developing countries, which were Malaysia, Sri Lanka and Pakistan, while no
evdidence of the fisher effect for other developing countries like Fiji, India and
Thailand.
4.0 Description of my data.
Below is a list of my data that I have used, and the abbreviated name for that
variable that has been used in eviews. The abbreviation is in italics.
Cpi_mly: CPI Malaysia
Nominal_mly: nominal interest rate of Malaysia
Logcpi_mly: log CPI Malaysia (inflation rate Malaysia)
Lognominal_mly: log nominal Malaysia
5
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
dlogcpi_mly: first differential inflation Malaysia
dlognominal_mly: first differential nominal interest rate Malaysia
resid_mly: Residual Malaysia
Cpi_can: CPI Canada
Nominal_can: nominal interest rate of Canada
Logcpi_can: log CPI Canada (inflation rate Canada)
Lognominal_can: log nominal Canada
dlogcpi_can: first differential inflation Canada
dlognominal_can: first differential nominal interest rate Canada
resid_can: Residual Canada
I collected my data from two different sources; I got data from the ESDS
website and from the Bank of Canada website. I have quarterly data from 1999 to
2009, which is not as much as I would have liked but I still think the number of
observations I have will give enough evidence to see if there is an existence of the
fisher effect in Canada and Malaysia. It was difficult to get data for a longer time
because there was data missing and the earliest data I did find was starting from
1999. To get the inflation rate of the two countries I had to get the CPI data then log
it to get the inflation rate.
5.0 Analysis
First I will show a comparison of my regular data and my logged data, then I
will do a test for normality using histograms, then I will test my data for stationarity,
to do this I will have to use the augmented dickey fuller test to test the presence of a
unit root. The Dickey fuller test is one of the most used tests, it tests whether a unit
root is present in an autoregressive model. Then I will test my variables for
cointegration using engle granger test, I will test for heteroskedasticity, then with
this I will run the regression of the model and test for stationarity of my residuals.
Then I will compare my results of the two countries. I will perform a unit root test to
6
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
test if a time series variable is stationary or non stationary using an autoregressive
model. I will also do a forecasting test, by using the Chow forecast test.
5.1 Comparison of original data and transformed data.
In the appendix the plotted variables are against their logged values. The
reason for logging the values is to convert the data into a smaller range which makes
the data to work with and to create a model with. The graphs show what I expected,
showing a smaller version of the normal variable for the respected variables. The
graphs show that there isn’t a structural break, and it follows a trend, this is good for
the tests that I want to implement.
5.2 Histograms, Test for normality.
There are many tests that you could do to check if the data is normally
distributed, one test is to plot your data into a histogram and to see if is normally
distributed with this graphical method. The results are summerised in the table
below and the histograms are in the appendix.
Skewness Kurtosis Jarque-Bera Probability
Log Cpi Canada -2.366404 9.155652 150.7288 0.000000
Dlog Cpi Canada -2.802828 12.97633 321.9201 0.000000
Log Cpi Malaysia -0.028557 2.23713 1.438698 0.487069
Dlog Cpi Malaysia 1.350618 9.752628 127.8287 0.000000
Log nominal Canada 0.041422 1.62081 4.772571 0.091971
Dlog nominal
Canada -0.479133 3.348232 2.555537 0.278658
Log nominal
Malaysia 0.66748 2.178807 6.038844 0.048829
Dlog nominal
Malaysia -1.255697 10.05003 135.3577 0.000000
7
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
As you can see my histograms in the appendix for all of my log variables they are not
normally distributed, therefore the variables are not normal. But when I differenced
the log of the variables, there is a slightly more consistency, but it is not normally
distributed, just there is a bit more of a left skweness. Dlogcpi Canada is left skewed,
I did this by eye by looking at the histogram itself, but as you can see on the right of
every histogram is a table of numbers, and at the skewness there is a number, for it
to be normally distributed, the number should be zero, the number should be as
close to zero as possible for it to be normally distributed. The range at which the
skewness is is from -1 which is left skewed to 1 which is right skewed. Also if you look
at the Kurtosis, this number should ranged between -2 and 2, if it is between this it is
normally distributed, and again as you can see most of the data is outside this range.
So as you can see my data has a non-normal distribution. Also if you look at the
Jarque-Bera statistic and the corresponding probability, if that probability is less than
0.05 then again the data is not normal.
5.3 Stationary properties of variables
To test whether my data for stationarity, I used the ADF test at both level
and first difference to test for a unit root of all my logged variables and I achieved
the following results:
Variable DF Used Level (H0)
1st
Difference
(H1)
t-statistic
Critical
value t-statistic
Critical
value
lognominal_mly
Trend and
intercept -1.97244 -3.495295 -3.66315 -3.495295
lognominal_can
Trend and
intercept -2.73596 -3.489228 -5.25478 -3.493692
logcpi_mly
Trend and
intercept -2.61554 -3.490662 -5.85026 -3.492149
8
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
logcpi_can
Trend and
intercept -1.75387 -3.489228 -4.34684 -3.489228
And as you can see in 1st difference the t-statistic is smaller than the critical value of
5% which shows that the data is stationary. Stationary is when the statistical
properties of a time series’ data, such as mean and variance, are all constant over
time. We are given an idea for finding appropriate forecasting models whilst making
data stationary. Having stationary data makes extrapolation more valid when using
statistical findings. I could have also used the Phillips Perron test, which would have
done the same thing.
5.4 Autocorrelation function of the variables.
The autocorrelation function (ACF) of a variable expresses the correlation
between the two variables at different points in time and this is displayed graphically
in the appendix. When the autocorrelation function moves in a steady pattern it
means that the variables are non stationary.
In the appendix are the auto correlation and partial correlation function of the
inflation and interest rates for both Canada and Malaysia. There are two sets of
graphs per variable of log and dlog, which is the level and 1st difference part, and this
is done because at level the variables my not be stationary, so doing it at 1st
difference makes the variables stationary. The reason for using difference log on my
logged variables is because for my logged variables there is negative inflation values,
so I had to difference them.
5.5 Cointegration
Cointegration is and econometric property for time series variables. It is a
way of testing the hypothesis of the relationship between variables that have a unit
root. This is used to help describe the movement of data that is measured over a
period of time. If two or more series are individually integrated but some linear
9
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
combinations of them have a lower order of integration then the series are said to
be cointegrated.
Engle and Granger introduced the notion of Cointegration. They showed that
a linear combination of integrated variables can possibly be stationary. This process
tries to determine whether or not the residuals of the equilibrium relationship are
stationary. Engle and Granger (1997) found that if Cointegration exists between two
variables in the long-run, then there must exist either unidirectional or bi-directional
Granger causality between the two variables. They also found that the cointegrated
variables must have an error correction model representation.
What I expect to find is that there is a relationship between the inflation and
nominal interest rates for Canada. However not such an obvious relationship in
Malaysia. To test for the existence of Cointegration, I have to use Engel Granger. I
have to run a regression and test it for unit root using ADF test. If at level it is
stationary and at 1st difference it is stationary, then I will have to run an error
correction model. But before this I will first run my regression, then I will check for
heteroskedasticity, using the white test, it basically means that a series of random
variances that are different, there is also series of data which have the same
variances which is called homoskedasticity. And the white test tests to see whether
the variance of residuals is constant. My results are summerised below and the full
output is in my appendix:
White test
Obs*R-
squared Prob Chi2 value
Canada 21.2838 0.0000
11.0704976
9
Malaysia 8.064592 0.0000
11.0704976
9
So if the R2 value is less than or equal to the Chi squared then you can not reject the
null hypothesis, therefore it is heteroskedasticity.
10
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
Now I will do the engel granger method for Cointegration. The output for my
regression for Canada gives the regression:
πt=22.67175-4.678007it+et
The output for my regression for Malaysia gives:
πt=4.775677-0.181368it+et
The two regression tables shown in the appendix shows the regressions of
linear relationships between interest rates and inflation. Also if you look at the prob
(F-statistic) value on the two regressions, the regression for Canada and the
regression for Malaysia are 0.000001 and 0.000000 respectively. The two regression
outputs show the regressions of linear relationships between interest rates and
inflation.
The coefficients are estimated by OLS and to check if there are any unit
roots present in the residuals I will test my them for unit roots, with the
Augmented dickey fuller test, my results are summerised as followed and the
tables are output is in the appendix:
Residual Level
1st
Difference
t-statistic
Critical
value t-statistic
Critical
value
Residual Canada -1.969410 -3.489228 -4.350184 -3.490662
Residual Malaysia -2.166312 -3.490662 -6.192455 -3.490662
So as you can see, for my critical value which is 5% is larger than the t-statistic figure,
therefore the unit root test confirms that the residuals are stationary at 1st
difference, and are non stationary at level, therefore my variables are not
cointegrated, so there is no need to do an error correction model. If I was to do an
error correction model, then I would use two step engel granger procedures, to
correct my model then I would have to redo the ADF test again and then the
amended residuals would be stationary at 1st difference.
5.6 Chow Forecasting
11
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
The Chow test was invented by economist Gregory Chow. In econometrics,
the Chow test is most commonly used in time series analysis. The chow forecast test
estimates the model for a subsample. The estimated model is then used to predict
the values of the dependent variable. A large difference between the actual and
predicted values casts doubt on the stability of the estimated relation over the two
subsamples. The Chow forecast test can be used with least squares and two-stage
least squares regressions.
My chow forecasting output is summerised below:
F-
statistic
F-
statistics(Chow) Prob.F
Canada
from
2005 Q1
39.21799
12.85481 0.0000
Malaysia
from
2005Q1
87.97009
9.423081 0.0000
So if you look at the ‘prob f’ number, for both the output for the chow test says its
0.0000 for both regression then this means that there is a high margin of error, so in
other words the regression will be very inaccurate in forecasting future figures. The
figure closest to one means it will be a better forecaster. I have tried various
different years, and I get the same results, the errors could have occurred from a
number of things, this could be because in inaccurate data, or because I tried to
forecast before the financial crisis and the data I have for that period may have been
impaired.
6 Conclusion
12
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
After analysing both countries by carrying out a number of tests I found that
both countries generated very similar results for unit root and Cointegration, I have
found that because the residuals at level are stationary and at 1st difference are non
stationary shows that they are not cointegrated, which shows that fisher effect holds
for both countries. For Canada I did expect the fisher effect to hold, because being a
well developed country and for having a good economy. However for Malaysia I had
my doubts because it is still a developing country in some sense. There is a negative
relationship between nominal interest rates, and inflation. And because they are not
cointegrated at this point, then I do not have to run an error correction model.
Another test for Cointegration I could have used is Johansen test, but I could have
only used this test if I had more variables which would make it multivariate, but it
was not, it was univariate. However I have seen papers and journals that used the
Johansen test for the fisher effect and with univariable data. And an alternative unit
root test I could have done is the Phillips Perron. And I would have done another
forecasting test to just verify the forecasting accuracy of my regression, and to
compare the two forecasters. I would do this because for the chow forecasting I saw
that both of the regressions are very poor forecasters to predict future values.
If I were to make changes to this project, I would pick more variables to make
my model multivariate and try the Johnsen test, because I have heard that it is more
sophisticated to use than the Engle Granger, and gives more of an accurate result.
And I would have chosen two very different countries, for example a developing
country, for example India and a well developed country like the United Kingdom to
see and compare them. Also what would be interesting to see in these very different
countries, that in quarterly data, how the holidays for the countries may affect the
results. For example in the UK it is a highly Christian populated country so in the final
quarter when Christmas and in the second quarter when Easter takes place, to see if
this has any significance on the results, and India is a high number of Hindus, and to
see whether the results change in corresponding to their holidays and festivals.
I would have also changed the CPI data, and find another source to show
inflation, because I read that even though I can find out inflation from calculating the
log of CPI, I could get slightly different results from another source for example if I
13
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
found the inflation from the countries GDP data then I might have got different
results, for example my data may not have been stationary at any point, or may
result into the regression being a better forecaster with a low probability of errors.
Also I think if I had more practice and experience in using eviews or any other
econometrics software, I could execute and interpret more tests and to get better
results and more accurate results.
Appendix of raw data
TimeNominal MLY CPI MLY Nominal Can CPI Can
1995q01 5.03 77.8749
86.93333333 7.99
1995q02 5.41067 78.3058 87.6 7.343331995q03 5.557 78.8379
87.76666667 6.47333
1995q04 6.01733 79.5222
87.86666667 5.76333
1996q01 6.276 80.4851 88.2 5.106671996q02 6.32333 81.1694
88.83333333 4.68667
1996q03 6.535 81.6762 89 4.141996q04 6.51367 82.183
89.66666667 2.89333
1997q01 6.29133 83.0446 90.1 2.963331997q02 6.433 83.1714
90.23333333 3
1997q03 6.20267 83.5515
90.53333333 3.18333
1997q04 6.70133 84.4131
90.56666667 3.88667
1998q01 5.94867 86.6178 91.1 4.439671998q02 8.879 87.9356 91.1 4.819
14
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
1998q03 6.93067 88.2904 91.2 4.912671998q04 5.698 88.9493 91.6 4.7521999q01 5.427 90.0643
91.86666667 4.74733
1999q02 3.10433 90.267
92.56666667 4.54767
1999q03 2.837 90.343
93.16666667 4.719
1999q04 2.73467 90.7739
93.76666667 4.865
2000q01 2.74733 91.5118
94.36666667 5.13467
2000q02 2.75733 91.5424 94.8 5.583672000q03 2.989 91.6648
95.66666667 5.60367
2000q04 2.951 92.2765 96.7 5.637332001q01 2.85 92.9188
96.93333333 4.85
2001q02 2.80733 92.9799 98.2 4.3512001q03 2.79 92.9188
98.23333333 3.62433
2001q04 2.72133 93.3776
97.73333333 2.24767
2002q01 2.72967 94.2645
98.43333333 2.12633
2002q02 2.72467 94.7845 99.5 2.592002q03 2.725 94.8457 100.6 2.890332002q04 2.749 95.0292
101.4666667 2.736
2003q01 2.79633 95.488
102.7666667 2.93867
2003q02 2.78133 95.6409
102.2333333 3.18833
2003q03 2.80333 95.7938
102.7666667 2.70167
2003q04 2.77233 95.7632 103.2 2.651672004q0 2.53 96.3749 103.7 2.12767
15
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
12004q02 2.54667 96.742 104.5 2.001332004q03 2.457 97.2007
104.7666667 2.212
2004q04 2.0495 98.1795
105.5666667 2.53867
2005q01 2.3475 98.6913
105.9666667 2.48233
2005q02 2.32 99.5249
106.4333333 2.46167
2005q03 2.42067 100.458
107.4666667 2.71733
2005q04 2.84883 101.325 108 3.241332006q01 2.94667 102.392
108.6666667 3.66333
2006q02 3.102 103.626
109.1333333 4.14967
2006q03 3.409 104.026 109.2 4.150332006q04 3.45133 104.393 109.5 4.1692007q01 3.41733 105.093
110.6666667 4.171
2007q02 3.38967 105.16
111.4333333 4.29067
2007q03 3.48967 105.893
111.4666667 4.22533
2007q04 3.438 106.693
112.2333333 3.916
2008q01 3.374 107.794
112.7333333 2.91
2008q02 3.439 110.261 114 2.706672008q03 3.433 114.795 115.3 2.273332008q04 3.314 112.995
114.3666667 1.67
2009q01 2.377 111.795
114.0666667 0.716667
2009q02 1.898 111.695 114.1 0.232009q03 1.96267 112.161
114.2666667 0.233333
16
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
Appendix of outputsComparison of original data and transformed data.
70
80
90
100
110
120
1996 1998 2000 2002 2004 2006 2008
CPI MLYLOGCPI_MLY
-2
0
2
4
6
8
10
12
1996 1998 2000 2002 2004 2006 2008
CPI CanLOGCPI_CAN
17
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
0
20
40
60
80
100
120
140
1996 1998 2000 2002 2004 2006 2008
Nominal CanLOGNOMINAL_CAN
0
2
4
6
8
10
12
1996 1998 2000 2002 2004 2006 2008
Nominal MLYLOGNOMINAL_MLY
Histograms, test for normality.
Log CPI Canada
0
2
4
6
8
10
12
14
16
-1 0 1 2
Series: LOGCPI_CANSample 1995Q1 2009Q4Observations 60
Mean 1.124237Median 1.293021Maximum 2.078191Minimum -1.514128Std. Dev. 0.723547Skewness -2.366404Kurtosis 9.155652
Jarque-Bera 150.7288Probability 0.000000
Dlogcpi Canada
18
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
0
4
8
12
16
20
-1.2 -1.0 -0.8 -0.6 -0.4 -0.2 -0.0 0.2
Series: DLOGCPI_CANSample 1995Q1 2009Q4Observations 59
Mean -0.060887Median -0.008358Maximum 0.199625Minimum -1.136532Std. Dev. 0.221683Skewness -2.802828Kurtosis 12.97633
Jarque-Bera 321.9201Probability 0.000000
Log CPI malaysia
0
2
4
6
8
10
4.4 4.5 4.6 4.7
Series: LOGCPI_MLYSample 1995Q1 2009Q4Observations 59
Mean 4.546965Median 4.551606Maximum 4.743148Minimum 4.355104Std. Dev. 0.103172Skewness -0.028557Kurtosis 2.237130
Jarque-Bera 1.438698Probability 0.487069
Dlogcpi malaysia
0
2
4
6
8
10
12
14
-0.01 0.00 0.01 0.02 0.03 0.04
Series: DLOGCPI_MLYSample 1995Q1 2009Q4Observations 58
Mean 0.006290Median 0.005852Maximum 0.040298Minimum -0.015804Std. Dev. 0.007680Skewness 1.350618Kurtosis 9.752628
Jarque-Bera 127.8287Probability 0.000000
19
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
Log nominal can
0
2
4
6
8
10
4.45 4.50 4.55 4.60 4.65 4.70 4.75
Series: LOGNOMINAL_CANSample 1995Q1 2009Q4Observations 60
Mean 4.606131Median 4.605655Maximum 4.747537Minimum 4.465142Std. Dev. 0.090665Skewness 0.041422Kurtosis 1.620810
Jarque-Bera 4.772571Probability 0.091971
Dlog nominal Canada
0
2
4
6
8
10
-0.005 0.000 0.005 0.010
Series: DLOGNOMINAL_CANSample 1995Q1 2009Q4Observations 59
Mean 0.004786Median 0.004821Maximum 0.012983Minimum -0.008128Std. Dev. 0.004417Skewness -0.479133Kurtosis 3.348232
Jarque-Bera 2.555537Probability 0.278658
log nominal malaysia
20
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
0
4
8
12
16
20
24
0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2
Series: LOGNOMINAL_MLYSample 1995Q1 2009Q4Observations 59
Mean 1.261041Median 1.094939Maximum 2.183689Minimum 0.640801Std. Dev. 0.389120Skewness 0.667480Kurtosis 2.178807
Jarque-Bera 6.038844Probability 0.048829
Dlognominal malaysia
0
4
8
12
16
20
24
-0.6 -0.4 -0.2 -0.0 0.2 0.4
Series: DLOGNOMINAL_MLYSample 1995Q1 2009Q4Observations 58
Mean -0.016226Median -0.001790Maximum 0.400521Minimum -0.558589Std. Dev. 0.124558Skewness -1.255697Kurtosis 10.05003
Jarque-Bera 135.3577Probability 0.000000
Autocorrelation function of the variables.
Log and dlog of CPI of Canada at level
log and dlog of CPI Canada at 1st difference
21
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
log and dlog of CPI malaysia level
log and dlog of CPI malaysia at 1st difference
log and dlog of nominal interest in Canada level
22
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
log and dlog of nominal interest rates in Canada 1st difference
log and dlog of nominal interest rates in malaysia level
log and dlog of nominal interest rates in malaysia 1st difference
23
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
CointegrationWhite TestWhite test for Canada:
Heteroskedasticity Test: White
F-statistic 15.66755 Prob. F(2,57) 0.0000Obs*R-squared 21.28380 Prob. Chi-Square(2) 0.0000Scaled explained SS 61.36587 Prob. Chi-Square(2) 0.0000
Test Equation:Dependent Variable: RESID^2Method: Least SquaresDate: 04/17/10 Time: 23:30Sample: 1995Q1 2009Q4Included observations: 60
VariableCoefficien
t Std. Error t-Statistic Prob.
C 1091.274 298.0955 3.660821 0.0006LOGNOMINAL_CAN -477.8290 129.4198 -3.692087 0.0005LOGNOMINAL_CAN
^2 52.29830 14.04255 3.724274 0.0005
R-squared 0.354730 Mean dependent var 0.337905Adjusted R-squared 0.332089 S.D. dependent var 0.846490S.E. of regression 0.691801 Akaike info criterion 2.149670Sum squared resid 27.27957 Schwarz criterion 2.254388Log likelihood -61.49011 Hannan-Quinn criter. 2.190631F-statistic 15.66755 Durbin-Watson stat 0.423025Prob(F-statistic) 0.000004
White Test for Malaysia
24
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
RegressionRegression for Canada:
Dependent Variable: LOGCPI_CANMethod: Least SquaresDate: 04/17/10 Time: 23:10Sample: 1995Q1 2009Q4Included observations: 60
VariableCoefficien
t Std. Error t-Statistic Prob.
C 22.67175 3.911218 5.796597 0.0000LOGNOMINAL_CAN -4.678007 0.848971 -5.510207 0.0000
R-squared 0.343612 Mean dependent var 1.124237Adjusted R-squared 0.332295 S.D. dependent var 0.723547S.E. of regression 0.591234 Akaike info criterion 1.819555Sum squared resid 20.27433 Schwarz criterion 1.889366Log likelihood -52.58664 Hannan-Quinn criter. 1.846862F-statistic 30.36239 Durbin-Watson stat 0.154905Prob(F-statistic) 0.000001
Regression for malaysia:
Dependent Variable: LOGCPI_MLYMethod: Least SquaresDate: 04/17/10 Time: 23:14Sample (adjusted): 1995Q1 2009Q3Included observations: 59 after adjustments
VariableCoefficien
t Std. Error t-Statistic Prob.
C 4.775677 0.033783 141.3643 0.0000LOGNOMINAL_ML
Y -0.181368 0.025617 -7.079841 0.0000
R-squared 0.467907 Mean dependent var 4.546965Adjusted R-squared 0.458572 S.D. dependent var 0.103172S.E. of regression 0.075916 Akaike info criterion -
25
Heteroskedasticity Test: White
F-statistic 4.433234 Prob. F(2,56) 0.0163Obs*R-squared 8.064592 Prob. Chi-Square(2) 0.0177Scaled explained SS 5.959772 Prob. Chi-Square(2) 0.0508
Test Equation:Dependent Variable: RESID^2Method: Least SquaresDate: 04/17/10 Time: 23:32Sample: 1995Q1 2009Q3Included observations: 59
VariableCoefficien
t Std. Error t-Statistic Prob.
C -0.030538 0.012416 -2.459677 0.0170LOGNOMINAL_ML
Y 0.056650 0.019105 2.965162 0.0044LOGNOMINAL_ML
Y^2 -0.020316 0.006825 -2.976605 0.0043
R-squared 0.136688 Mean dependent var 0.005568Adjusted R-squared 0.105855 S.D. dependent var 0.007067
S.E. of regression 0.006682 Akaike info criterion-
7.129215
Sum squared resid 0.002501 Schwarz criterion-
7.023578
Log likelihood 213.3118 Hannan-Quinn criter.-
7.087979F-statistic 4.433234 Durbin-Watson stat 0.497435Prob(F-statistic) 0.016319
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
2.285065
Sum squared resid 0.328506 Schwarz criterion-
2.214640
Log likelihood 69.40941 Hannan-Quinn criter.-
2.257574F-statistic 50.12415 Durbin-Watson stat 0.114388Prob(F-statistic) 0.000000
Augmented dickey fuller tests on residualsFor Canada at level:Null Hypothesis: RESIDCAN has a unit rootExogenous: Constant, Linear TrendLag Length: 1 (Automatic based on AIC, MAXLAG=10)
t-Statistic Prob.*
Augmented Dickey-Fuller test statistic -1.969410 0.6053Test critical values: 1% level -4.124265
5% level -3.48922810% level -3.173114
*MacKinnon (1996) one-sided p-values.
Augmented Dickey-Fuller Test EquationDependent Variable: D(RESIDCAN)Method: Least SquaresDate: 04/17/10 Time: 23:47Sample (adjusted): 1995Q3 2009Q4Included observations: 58 after adjustments
Canada at 1st difference:Null Hypothesis: D(RESIDCAN) has a unit rootExogenous: Constant, Linear TrendLag Length: 1 (Automatic based on AIC, MAXLAG=10)
t-Statistic Prob.*
Augmented Dickey-Fuller test statistic -4.350184 0.0054Test critical values: 1% level -4.127338
5% level -3.49066210% level -3.173943
*MacKinnon (1996) one-sided p-values.
Augmented Dickey-Fuller Test Equation
26
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
Dependent Variable: D(RESIDCAN,2)Method: Least SquaresDate: 04/17/10 Time: 23:47Sample (adjusted): 1995Q4 2009Q4Included observations: 57 after adjustments
And my results for Malaysia at level is:
Null Hypothesis: RESID01 has a unit rootExogenous: Constant, Linear TrendLag Length: 1 (Automatic based on AIC, MAXLAG=10)
t-Statistic Prob.*
Augmented Dickey-Fuller test statistic -2.166312 0.4986Test critical values: 1% level -4.127338
5% level -3.49066210% level -3.173943
*MacKinnon (1996) one-sided p-values.
Augmented Dickey-Fuller Test EquationDependent Variable: D(RESID01)Method: Least SquaresDate: 04/17/10 Time: 23:41Sample (adjusted): 1995Q3 2009Q3Included observations: 57 after adjustments
And at 1st difference:Null Hypothesis: D(RESID01) has a unit rootExogenous: Constant, Linear TrendLag Length: 0 (Automatic based on AIC, MAXLAG=10)
t-Statistic Prob.*
Augmented Dickey-Fuller test statistic -6.192455 0.0000Test critical values: 1% level -4.127338
5% level -3.49066210% level -3.173943
*MacKinnon (1996) one-sided p-values.
Augmented Dickey-Fuller Test EquationDependent Variable: D(RESID01,2)Method: Least SquaresDate: 04/17/10 Time: 23:41
27
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
Sample (adjusted): 1995Q3 2009Q3Included observations: 57 after adjustments
Chow Forecasting
Chow forecaseting from 2005Q1 for Canada:
Chow Forecast Test: Forecast from 2005Q1 to 2009Q4
F-statistic 12.85481 Prob. F(20,38) 0.0000Log likelihood ratio 122.9829 Prob. Chi-Square(20) 0.0000
Test Equation:Dependent Variable: LOGCPI_CANMethod: Least SquaresDate: 04/30/10 Time: 14:28Sample: 1995Q1 2004Q4Included observations: 40
Variable Coefficient Std. Error t-Statistic Prob.
C 21.29822 3.187329 6.682154 0.0000LOGNOMINAL_CAN -4.383027 0.699893 -6.262427 0.0000
R-squared 0.507887 Mean dependent var 1.339495Adjusted R-squared 0.494936 S.D. dependent var 0.368823S.E. of regression 0.262115 Akaike info criterion 0.208638Sum squared resid 2.610757 Schwarz criterion 0.293082Log likelihood -2.172758 Hannan-Quinn criter. 0.239170F-statistic 39.21799 Durbin-Watson stat 0.301117Prob(F-statistic) 0.000000
Chow forecaseting from 2005Q1 for Malaysia:
Chow Forecast Test: Forecast from 2005Q1 to 2009Q3
F-statistic 9.423081 Prob. F(19,38) 0.0000Log likelihood ratio 102.8068 Prob. Chi-Square(19) 0.0000
Test Equation:Dependent Variable: LOGCPI_MLYMethod: Least SquaresDate: 04/30/10 Time: 14:26Sample: 1995Q1 2004Q4Included observations: 40
Variable Coefficient Std. Error t-Statistic Prob.
C 4.677912 0.020858 224.2786 0.0000LOGNOMINAL_MLY -0.138210 0.014736 -9.379237 0.0000
R-squared 0.698341 Mean dependent var 4.490985
28
073604149Testing the Existence and Forecasting the Fisher Effect in Malaysia and Canada
Adjusted R-squared 0.690403 S.D. dependent var 0.069920S.E. of regression 0.038905 Akaike info criterion -3.606692Sum squared resid 0.057516 Schwarz criterion -3.522248Log likelihood 74.13384 Hannan-Quinn criter. -3.576160F-statistic 87.97009 Durbin-Watson stat 0.255183Prob(F-statistic) 0.000000
Bibliography
Johnson, P.A. (2006), ‘Is there really the Fisher effect?’, Applied Economics Letters 13, pp. 201-203.
Perez, S.J. and M.V. Siegler (2003), ‘Inflationary expectations and the Fisher effect prior to World War I’, Journal of Money, Credit and Banking 35, pp. 947-965.
Fahmy, Y.A.F. and M. Kandil (2003), ‘The Fisher effect: new evidence and implications’, International Review of Economics and Finance 12, pp. 451-465.
Gujarati D.N. (1995), Basic Econometrics, 3rd edition, McGraw-Hill International Editions.
Mankiw,N.G. (2000), Macroeconomics, 4th edition, Worth Publishers, chapters 7 and 18.
Arusha Cooray (2002), ‘THE FISHER EFFECT: A REVIEW OF THE LITERATURE’ . (http://www.econ.mq.edu.au/research/2002/6-2002Cooray.PDF)
29