EViews assignment – 2nd week

The problems that follow are from the course textbook. If it is referred to an EViews work file that
you yet have not created, you must import the data from the corresponding Excel file. This
procedure is described in the textbook.
Your solutions must be easy to follow and include figures and tables when prompted. The best
way to present the solutions is to use a Word document.
4.10 (slightly changed)
Re-open the Macro file and apply the same APT-type model to some of the other time-series of
stock returns contained in the CAPM-file (the APT-type model has already been applied for
Microsoft in chapter 3). Note: You need to add the stock series in the CAPM-file into the Macro-
file. So, import the capm.xls into the Macro file. Select Cancel for the series that already exist in
the Macro file.
a) Run the stepwise procedure in each case (Choose p-values = 0.2 as Stopping Criteria). Is
the same sub-set of variables selected for each stock? Can you explain the differences between
the series chosen?
b) Examine the sizes and signs of the parameters in the regressions in each case – do
these make sense?

5.13
Re-open the ‘macro.wf1’ and apply the stepwise procedure (on ermsoft) including all of the
explanatory variables as listed above, i.e., ersandp dprod dcredit dinflation dmoney dspread
rterm with a strict 5% threshold criterion for inclusion in the model. Then examine the resulting
model both financially and statistically by investigating the signs, sizes and significances of the
parameter estimates and by conducting all of the diagnostic tests for model adequacy.
QUESTION 4.10

STEPWISE REGRESSION RESULT FOR GENERAL ELECTRIC (GE)

Dependent Variable: ERGE

Method: Stepwise Regression
Date: 11/19/20 Time: 14:17
Sample (adjusted): 2002M02 2013M04
Included observations: 135 after adjustments
Number of always included regressors: 1
Number of search regressors: 7
Selection method: Stepwise forwards
Stopping criterion: p-value forwards/backwards = 0.2/0.2

Variable Coefficient Std. Error t-Statistic Prob.*  

C -0.876244 0.549867 -1.593557 0.1134

ERSANDP 1.398409 0.109842 12.73109 0.0000
DSPREAD 6.414402 3.384175 1.895411 0.0602
DCREDIT 6.93E-05 3.79E-05 1.828566 0.0697

R-squared 0.568847    Mean dependent var -0.242533

Adjusted R-squared 0.558973    S.D. dependent var 8.409521
S.E. of regression 5.584750    Akaike info criterion 6.307138
Sum squared resid 4085.816    Schwarz criterion 6.393220
Log likelihood -421.7318    Hannan-Quinn criter. 6.342120
F-statistic 57.61215    Durbin-Watson stat 2.122923
Prob(F-statistic) 0.000000

Selection Summary

Added ERSANDP
Added DSPREAD
Added DCREDIT

*Note: p-values and subsequent tests do not account for stepwise selection.
Source: Computation in E-view 11 output

Estimated Stepwise Model for General Electric (GE) with p-value 0.2 as stopping criteria.
ERGEt =−0.876244 +1.3984 ERSANDPt + 6.4144 DSPREADt + 6.93 DSCREDIT t
Where:
ERGE = Excess Return on General Electric Stock
ERSANDP = Excess Return on S&P500 index value / the excess market return
DCREDIT = Changes in Consumer credit series / credit variables
DSPREAD = Changes in credit spread / default spread
STEPWISE REGRESSION RESULT FOR ORACLE
Dependent Variable: ERORACLE
Method: Stepwise Regression
Date: 11/19/20 Time: 14:24
Sample (adjusted): 2002M02 2013M04
Included observations: 135 after adjustments
Number of always included regressors: 1
Number of search regressors: 7
Selection method: Stepwise forwards
Stopping criterion: p-value forwards/backwards = 0.2/0.2

Variable Coefficient Std. Error t-Statistic Prob.*  

C 0.342924 0.629420 0.544825 0.5868

ERSANDP 1.107528 0.137202 8.072223 0.0000
DPROD -1.350414 0.877849 -1.538321 0.1264

R-squared 0.333773    Mean dependent var 0.365143

Adjusted R-squared 0.323678    S.D. dependent var 8.834003
S.E. of regression 7.264975    Akaike info criterion 6.825978
Sum squared resid 6966.942    Schwarz criterion 6.890540
Log likelihood -457.7535    Hannan-Quinn criter. 6.852214
F-statistic 33.06531    Durbin-Watson stat 1.889542
Prob(F-statistic) 0.000000

Selection Summary

Added ERSANDP
Added DPROD

*Note: p-values and subsequent tests do not account for stepwise selection.
Source: Computation in E-view 11 output

Estimated Stepwise Model for Oracle with p-value 0.2 as stopping criteria.
ERORACLEt =0.342924+ 1.107528 ERSANDP t−1.3504 DPRODt
Where:
ERORACLE = Excess Return on Oracle Stock
ERSANDP = Excess Return on S&P500 index value / the excess market return
DPROD = Changes in Production index / Production variable
STEPWISE REGRESSION RESULT FOR FORD
Dependent Variable: ERFORD
Method: Stepwise Regression
Date: 11/19/20 Time: 14:28
Sample (adjusted): 2002M02 2013M04
Included observations: 135 after adjustments
Number of always included regressors: 1
Number of search regressors: 7
Selection method: Stepwise forwards
Stopping criterion: p-value forwards/backwards = 0.2/0.2

Variable Coefficient Std. Error t-Statistic Prob.*  

C -0.028495 1.057343 -0.026950 0.9785

ERSANDP 1.922524 0.240627 7.989643 0.0000
DPROD -4.230805 1.495299 -2.829404 0.0054
DSPREAD -15.69687 7.705852 -2.037007 0.0437
DINFLATION -4.801009 2.588314 -1.854879 0.0659

R-squared 0.408788    Mean dependent var -0.078204

Adjusted R-squared 0.390597    S.D. dependent var 15.63184
S.E. of regression 12.20288    Akaike info criterion 7.877555
Sum squared resid 19358.34    Schwarz criterion 7.985158
Log likelihood -526.7350    Hannan-Quinn criter. 7.921282
F-statistic 22.47181    Durbin-Watson stat 2.579251
Prob(F-statistic) 0.000000

Selection Summary

Added ERSANDP
Added DPROD
Added DSPREAD
Added DINFLATION
*Note: p-values and subsequent tests do not account for stepwise selection.
Source: Computation in E-view 11 output

Estimated Stepwise Model for General Electric (GE) with p-value 0.2 as stopping criteria.

ERFORD t=−0.028495+ 1.9225 ERSANDP t −4.2308 DPROD−15.6969 DSPREAD t −4.801 DINFLATION

Where:
ERFORD = Excess Return on Ford Stock
ERSANDP = Excess Return on S&P500 index value / the excess market return
DPROD = Changes in Production index
DINFLATION = Changes in inflation rate/ unexpected inflation
DSPREAD = Changes in credit spread / default spread
Table 4: Summary of the stepwise procedure for each stock
General Electric Stock Oracle Stock Ford Stock
Variable Coefficien Coefficien
t Std. Error P-value t Std. Error P-value Coefficient Std. Error P-value
C -0.87624 0.54987 0.11340 0.34292 0.62942 0.58680 -0.02850 1.05734 0.97850
ERSANDP 1.39841 0.10984 0.00000 1.10753 0.13720 0.00000 1.92252 0.24063 0.00000
DSPREAD 6.41440 3.38418 0.06020 - - - -15.69687 7.70585 0.04370
DCREDIT 0.00007 0.00004 0.06970 - - - - - -
DPROD - - - -1.35041 0.87785 0.12640 -4.23081 1.49530 0.00540
DINFLATIO
N - - - - - - -4.80101 2.58831 0.06590
ROBUSTNESS/DIAGNOSTICS TESTS
R-squared 0.56885 0.33377 0.40879
Adj. R-squared 0.55897 0.32368 0.39060
Prob (F-stat.) 0.00000 0.00000 0.00000
Durbin-Watson 2.12292 1.88954 2.57925
Source: Student’s compilation from e-view 11 output

(a). Question: Is the same sub-set of variables selected for each stock?
Table 4 above summarizes the variables selected for each stock at a p-value of 0.2 as stopping
criteria. It can be seen from the table that different sub-set of variables were selected for
individual excess stock returns except for the excess return on the market index (ERSANDP),
which was commonly selected for all the stocks.
Three variables were selected in the estimated General Electric (GE) stock model: excess market
return (ERSANDP), credit variables (DCREDIT), and default spread (DSPREAD), while money
supply, production variable, unexpected inflation, and term structure were omitted. The
estimated Oracle stock model, excess market return, and production variables were selected,
while money supply, credit variables, unexpected inflation, money supply, default spread, and
term structure were omitted. In the estimated Ford stock model, four variables were selected,
i.e., excess market return, production variable, unexpected inflation rate, and default spread,
while money supply, credit variables, and term structure were omitted.
(b). Question: Examine the sizes and signs of the parameters in the regressions in each case – do
these make sense?
The estimated constant (alpha) for the GE stock excess return has a negative value of 0.87624
and a negative value of 0.02850 for the Ford stock excess return. In contrast, the estimated
constant (alpha) for the return on the Oracle stock model has a positive value of 0.34292. The
estimated constant (alpha) for the three stocks is not statistically significant at the five percent
significance level. Considering the three estimated models, excess return on Oracle stock has the
highest constant (alpha) value with only a positive sign. This result means that Oracle stock
returned more than its expected return (abnormal returns) while GE and Ford's stock could not
'beat the market' and returned less, as shown by the negative alpha. Whether the estimated
constant (alpha) makes sense or not, alpha may be positive or negative depending on the stock
performance.
The estimated market excess return (ERSANDP) parameter for all three stock returns (GE stock,
Oracle stock, and Ford stock) is positive, greater than one, and statistically significant at the five
percent significance level. This result indicates that the three stocks move more than the market
and are all risky since their values are higher than one. Ford stock is more volatile and riskier than
the other two stock since it has the highest beta value, i.e., 1.92252. Whether the estimated
ERSANDP parameter makes sense or not, the estimated parameter for all the three stocks makes
sense since it is positive in all three cases.
The default spread (DSPREAD) estimated parameter for Ford stock excess return is negative, high,
and statistically significant at the five percent significance level. This result means that the higher
the default spread for Ford, the lower the excess return on Ford's stock. In contrast, the
estimated parameter of the default spread for GE stock excess return is positive but not
statistically significant at the five percent significance level. This result indicates that the higher
the default spread for GE, the higher the excess return on GE's stock. We can infer from these
results that the effect that default risk may have on equity returns is not apparent, which answers
the question of whether the estimated DSPREAD parameter makes sense or not.
The estimated parameter of the production index (DPROD) is negative for both Oracle stock
excess return and Ford stock excess return. However, it was statistically significant at the five
percent significance level for only Ford excess return. This result means that the Index of
Industrial production impact both stock returns negatively. Whether the estimated DPROD
parameter makes sense or not, the estimated parameter for the two-stock return (Oracle stock
and Ford stock) does not make sense as we expect increased industrial production to positively
impact the stock market activities.
The estimated parameter of the unexpected inflation (DINFLATION) is negative for Ford stock
excess return though not statistically significant at the five percent significance level. Unexpected
inflation can either positively or negatively impact the stock return, depending on the economic
cycle's timing.
For the robustness of the three stepwise estimated models, the R 2 value for the GE excess returns
regression was 0.56885, indicating that the selected variables (excess market return, credit
variables, and default spread) only explain 56.88 percent of the variation in GE excess returns.
The fit is not so good for the Oracle stock excess return; the R 2 is around 33.3%, implying that the
two selected variables (excess market return and production variables) only explained 33.3
percent of the variation in Oracle excess returns. Ford excess return stepwise estimated model
has an R2 value of 40.88 percent. This result implies that the four selected variables (excess
market return, production variable, unexpected inflation rate, and default spread) only explained
40.88 percent of Ford's variation in excess return. The p-values of the three models' F-statistic are
all 0.0000, which is less than 0.05, suggesting that all the parameters estimated in the three
models are all jointly statistically significant.

Question 5.13: Examine the resulting model both financially and statistically by investigating the
signs, sizes and significances of the parameter estimates and by conducting all the diagnostic
tests for model adequacy.
STEPWISE REGRESSION RESULT FOR MICROSOFT
Dependent Variable: ERMSOFT
Method: Stepwise Regression
Date: 11/22/20 Time: 04:35
Sample (adjusted): 1986M04 2013M04
Included observations: 325 after adjustments
Number of always included regressors: 1
Number of search regressors: 7
Selection method: Stepwise forwards
Stopping criterion: p-value forwards/backwards = 0.05/0.05
Note: final equation sample is larger than stepwise sample (rejected
        regressors contain missing values)

Variable Coefficient Std. Error t-Statistic Prob.*  

C -0.613701 0.705782 -0.869533 0.3852

ERSANDP 1.325376 0.153887 8.612649 0.0000

R-squared 0.186762    Mean dependent var -0.263150

Adjusted R-squared 0.184244    S.D. dependent var 14.06400
S.E. of regression 12.70249    Akaike info criterion 7.927608
Sum squared resid 52117.10    Schwarz criterion 7.950893
Log likelihood -1286.236    Hannan-Quinn criter. 7.936901
F-statistic 74.17773    Durbin-Watson stat 2.156763
Prob(F-statistic) 0.000000

Selection Summary

Added ERSANDP

*Note: p-values and subsequent tests do not account for stepwise selection.
Source: Computation in E-view 11 output
Answer
The stepwise estimated model for Microsoft excess returns with a strict 5% threshold criterion
revealed that the excess market return is the only variable included. In contrast, all other
variables: the money supply, default spread, Production index, credit variables, unexpected
inflation, and term structure, were omitted.
The estimated constant (alpha) has a negative value of 0.613701 but not statistically significant at
the five percent significance level. This negative constant implies that Microsoft stock under
performing and could not 'beat the market’, as a result, returned less. The estimated market
excess return (ERSANDP) parameter is positive, greater than one, and statistically significant at
the five percent significance level since its p-value is less than 0.05. This result implies that
Microsoft stock move more than the market (volatile) and risky since the estimated beta of
1.325376 is higher than one.
For the robustness of the Microsoft stepwise estimated models with a p-value of 0.05 as stopping
criteria, the R2 was 0.186762, indicating that the selected variable (excess market return) only
explains 18.67 percent of the variation in Microsoft stock excess returns; the fit is not so good.
The p-values of the stepwise regression models' F-statistic is 0.0000, which is less than 0.05,
suggesting that all the parameters estimated in the model are jointly statistically significant.

DIAGNOSTIC TESTS FOR MODEL ADEQUACY

Testing for Heteroscedasticity
 80

40

40 -40
20
-80
0
-20
-40
-60
-80
86 88 90 92 94 96 98 00 02 04 06 08 10 12

Residual Actual Fitted

From the above diagram, it is hard to see any clear pattern of the residuals of the regression to confirm if
there is any systematically changing variability over the sample (heteroscedasticity), so we need to run the
formal statistical test (White).
Heteroskedasticity Test: White
Null hypothesis: Homoskedasticity

F-statistic 0.207410    Prob. F(1,323) 0.6491

Obs*R-squared 0.208560    Prob. Chi-Square(1) 0.6479
Scaled explained SS 1.247780    Prob. Chi-Square(1) 0.2640

Test Equation:
Dependent Variable: RESID^2
Method: Least Squares
Date: 11/22/20 Time: 05:08
Sample: 1986M04 2013M04
Included observations: 325

Variable Coefficient Std. Error t-Statistic Prob.  

C 166.5625 33.90159 4.913117 0.0000

ERSANDP^2 -0.294854 0.647430 -0.455422 0.6491

R-squared 0.000642    Mean dependent var 160.3603

Adjusted R-squared -0.002452    S.D. dependent var 559.0048
S.E. of regression 559.6898    Akaike info criterion 15.49878
Sum squared resid 1.01E+08    Schwarz criterion 15.52206
Log likelihood -2516.551    Hannan-Quinn criter. 15.50807
F-statistic 0.207410    Durbin-Watson stat 2.005488
Prob(F-statistic) 0.649111

Source: Computation in E-view 11 output

From the White Heteroskedasticity Test above, the p-value of F-statistic is 0.6491 and p-value of the
χ2 (‘LM’) is 0.6479 which are both greater than 0.05. Both the F- and χ2 (‘LM’) versions of the test
statistic give the same conclusion that there is no evidence for the presence of heteroscedasticity
since the p-values are considerably more than 0.05.
In addition, scaled explained SS’ - a normalized version of the explained sum of squares from the
auxiliary regression, provide additional evidence of no heteroscedasticity (with the test result
significant at the 5%).
Testing for Autocorrelation
Breusch-Godfrey Serial Correlation LM Test:
Null hypothesis: No serial correlation at up to 10 lags

F-statistic 1.584605    Prob. F(10,313) 0.1100

Obs*R-squared 15.66072    Prob. Chi-Square(10) 0.1098

Test Equation:
Dependent Variable: RESID
Method: Least Squares
Date: 11/22/20 Time: 05:20
Sample: 1986M04 2013M04
Included observations: 325
Presample missing value lagged residuals set to zero.

Variable Coefficient Std. Error t-Statistic Prob.  

C -0.006437 0.699527 -0.009201 0.9927

ERSANDP -0.014537 0.154186 -0.094283 0.9249
RESID(-1) -0.112014 0.056574 -1.979937 0.0486
RESID(-2) -0.045388 0.056450 -0.804032 0.4220
RESID(-3) -0.008164 0.056395 -0.144765 0.8850
RESID(-4) -0.116305 0.056400 -2.062151 0.0400
RESID(-5) -0.101347 0.056497 -1.793845 0.0738
RESID(-6) -0.091067 0.056626 -1.608204 0.1088
RESID(-7) -0.041638 0.056490 -0.737076 0.4616
RESID(-8) -0.066444 0.056581 -1.174324 0.2412
RESID(-9) -0.128296 0.056616 -2.266057 0.0241
RESID(-10) -0.041189 0.056678 -0.726720 0.4679

R-squared 0.048187    Mean dependent var -5.79E-16

Adjusted R-squared 0.014737    S.D. dependent var 12.68287
S.E. of regression 12.58907    Akaike info criterion 7.939760
Sum squared resid 49605.74    Schwarz criterion 8.079470
Log likelihood -1278.211    Hannan-Quinn criter. 7.995519
F-statistic 1.440550    Durbin-Watson stat 1.996546
Prob(F-statistic) 0.153490

Source: Computation in E-view 11 output

From the Breusch-Godfrey Serial Correlation LM Test above, the p-value of the F-
statistic is 0.1100 and p-value of the χ2 (‘LM’) is 0.1098 which are both greater than
0.05. The conclusion from both versions of the test in this case is that the null
hypothesis of no autocorrelation should not be rejected since the p-values are above
0.05. This implies that there is presence of autocorrelation in the estimated model.
Testing for non-normality

70
Seri es: Residual s
60 Sample 1986M04 2013M04
Observations 325
50
Mea n -5.79e-16
40
Medi an 0.975692
30 Maxi mum 30.43601
Mi ni mum -66.52280
20 Std. Dev. 12.68287
Skewness -2.428564
10 Kurtosis 13.11432

0 Jarque-Bera 1704.775
-62.5 -50.0 -37.5 -25.0 -12.5 0.0 12.5 25.0
Probabi li ty 0.000000

Source: Computation in E-view 11 output

From the Bera-Jarque statistic Test above, the histogram is not bell-shaped. The
residuals are very negatively skewed and are leptokurtic. The null hypothesis for
residual normality is therefore rejected very strongly since the p-value for the Bera-
Jarque test is zero.

Multicollinearity test
We could not conduct a test for multicollinearity since we only have one selected
explanatory variable (ERSANDP) in the model.
RESET tests.

Ramsey RESET Test

Equation: MSOFTSTEPWISE
Omitted Variables: Squares of fitted values
Specification: ERMSOFT C ERSANDP

Value df Probability
t-statistic  1.612010  322  0.1079
F-statistic  2.598578 (1, 322)  0.1079
Likelihood ratio  2.612262  1  0.1060

F-test summary:
Mean
Sum of Sq. df Squares
Test SSR  417.2240  1  417.2240
Restricted SSR  52117.10  323  161.3532
Unrestricted SSR  51699.87  322  160.5586

LR test summary:
Value
Restricted LogL -1286.236
Unrestricted LogL -1284.930

Unrestricted Test Equation:

Dependent Variable: ERMSOFT
Method: Least Squares
Date: 11/22/20 Time: 05:31
Sample: 1986M04 2013M04
Included observations: 325

Variable Coefficient Std. Error t-Statistic Prob.  

C -1.198206 0.791928 -1.513025 0.1313

ERSANDP 1.464915 0.176232 8.312439 0.0000
FITTED^2 0.014842 0.009207 1.612010 0.1079

R-squared 0.193272    Mean dependent var -0.263150

Adjusted R-squared 0.188262    S.D. dependent var 14.06400
S.E. of regression 12.67117    Akaike info criterion 7.925724
Sum squared resid 51699.87    Schwarz criterion 7.960652
Log likelihood -1284.930    Hannan-Quinn criter. 7.939664
F-statistic 38.57171    Durbin-Watson stat 2.130219
Prob(F-statistic) 0.000000

Source: Computation in E-view 11 output

From the Ramsey RESET test above, both F−and χ2 versions of the test are presented, and
there is limited evidence for non-linearity in the regression equation (p-values above 10%)
and so it would be concluded that the linear model for the Microsoft returns is
appropriate.
Stability tests in EViews

Chow Breakpoint Test: 1996M01 

Null Hypothesis: No breaks at specified breakpoints
Varying regressors: C ERSANDP
Equation Sample: 1986M04 2013M04

F-statistic 0.841410 Prob. F(2,321) 0.4321

Log likelihood ratio 1.699339 Prob. Chi-Square(2) 0.4276
Wald Statistic  1.682820 Prob. Chi-Square(2) 0.4311

Source: Computation in E-view 11 output

From the Chow Breakpoint test above, F-statistic is 0.841410 with p-value of 0.4321,
Log likelihood ratio is 1.699339 with p-value of 0.4276 and Wald statistic is 1.682820
with p-value of 0.4311. This result implies that all the three test statistics are smaller
than their critical values and so the null hypothesis that the parameters are constant
across the two sub-samples is not rejected.