Professional Documents
Culture Documents
MAY 2016
ROUGH ANSWERS
Coventry University
212ECN
Introduction to Econometrics
Instructions to candidates
You may take this exam paper away at the end of the examination:
please keep it in a safe place for future reference
Continued…
212ECN
Question 1
Table I
Model 1: Logit, using observations 1-100
Dependent variable: PC
Standard errors based on Hessian
Predicted
0 1
Actual
0 49 8
1 20 23
Variables are:
PC = 1 if student owns personal computer, = 0 otherwise,
SEX = 1 if male, = 0 otherwise,
CAMPUS = 1 if live on campus, = 0 otherwise,
WORK = 1 if work at least 20 hours a week, = 0 otherwise,
LECTURES – average number of lectures missed per week.
Page 1 Continued…
212ECN
Table II
Model 2: Logit, using observations 1-100
Dependent variable: PC
Standard errors based on Hessian
Coefficient Std. Error z Slope*
const −0.281851 0.201989 -1.3954
Predicted
0 1
Actual
0 57 0
1 43 0
Page 2 Continued…
212ECN
for LECTURES = 1:
0.170395 + 0.199162 * 0 + 1.15939 * 0 – 0.984154 * 1 – 0.483189 * 1 =
= -1.296948
D2 = Prob( PC = 1 ) = 1 / ( 1 + e-(-1.296948) ) = 0.214679
Absolute change:
D2 - D1 = 0.214679 - 0.307090 = -0.092411 = -9.2411 percentage point
(not %)
Relative change:
(D2 - D1) / D1 = (0.214679 - 0.307090) / 0.307090 = -0.092411 / 0.307090
= -0.300925 = -30.0925%
This requires you to: state the null and alternative hypotheses used,
determine a critical value at the 1% significance level; calculate z
statistic and present your decision. Please refer to Table I
[25 marks]
𝐻0 : 𝛽𝐿𝐸𝐶𝑇𝑈𝑅𝐸𝑆 = 0
𝐻1 : 𝛽𝐿𝐸𝐶𝑇𝑈𝑅𝐸𝑆 ≠ 0
Calculated value:
Page 3 Continued…
212ECN
Because -2.58 < -2.1355 so we can’t reject H0. This means that coefficient on
LECTURES in Model 1 is not statistically significant at 1% significance level
(99% confidence level) as the coefficient is not statistically different from
zero.
This requires you to: explain what the McFadden R-squared and Count
R2 are, show how they are calculated and make a comment on the
values. Please refer to Table I and II.
[10 marks]
2
McFadden R = 1 – (logLUnRest / logLRest) = 1 – (-62.09863 / -68.33149) =
0.091215 = 9.1215% (rather poor fit)
iv. Test whether the coefficients in Model 1 are jointly significant at the 1%
significance level.
This requires you to: state the null and alternative hypotheses used,
determine a critical value at the 1% significance level; show the
calculated likelihood ratio statistic and present your decision. Please
refer to Table I.
[25 marks]
𝐻0 : 𝛽𝑆𝐸𝑋 = 0 ∧ 𝛽𝐶𝐴𝑀𝑃𝑈𝑆 = 0 ∧ 𝛽𝑊𝑂𝑅𝐾 = 0 ∧ 𝛽𝐿𝐸𝐶𝑇𝑈𝑅𝐸𝑆 = 0
(all coefficients are simultaneously equal zero)
𝐻1 : 𝛽𝑆𝐸𝑋 ≠ 0 ∨ 𝛽𝐶𝐴𝑀𝑃𝑈𝑆 ≠ 0 ∨ 𝛽𝑊𝑂𝑅𝐾 ≠ 0 ∨ 𝛽𝐿𝐸𝐶𝑇𝑈𝑅𝐸𝑆 ≠ 0
(not all coefficients are simultaneously equal zero)
Calculated value:
LR = 2 * (logLUnRest - logLRest) = 2 * (-62.09863 – (-68.33149)) =
= 12.46572
Because 12.46572 < 13.28 so we can’t reject H0. This means that Model 1 is
not statistically significant at 1% significance level (99% confidence level) as
all the coefficients are simultaneously equal zero.
Page 4 Continued…
212ECN
̶ non-normality of residuals
̶ linear relationship → Pi = P(yi = 1|x) increases linearly with x,
that is, the marginal effect of x remains constant throughout
Page 5 Continued…
212ECN
Question 2
The data are taken from Whittington, Alm, and Peters (1990) and represent
time series with annual observations from 1917 to 1984. The variables are:
FERTILITY – number of children born per 1000 women in US,
EXEMPTION – average real dollar value of personal tax exemption on
dependents in US (in US dollars),
PILL – availability of birth control pill in US, 1 if year 1963-1984, 0
otherwise,
WW2 – US involvement in World War II, 1 if 1941-1945, 0 otherwise.
Table III
Model 3: OLS, using observations 1917-1984 (T = 68)
Dependent variable: l_FERTILITY
Table IV
Model 4: OLS, using observations 1917-1984 (T = 68)
Dependent variable: l_FERTILITY
Page 6 Continued…
212ECN
This requires you to: state the null and alternative hypotheses used,
determine a critical value at the 1% significance level; calculate t
statistic and present your decision.
[25 marks]
𝐻0 : 𝛽𝑙_𝐸𝑋𝐸𝑀𝑃𝑇𝐼𝑂𝑁 = 0
𝐻1 : 𝛽𝑙_𝐸𝑋𝐸𝑀𝑃𝑇𝐼𝑂𝑁 ≠ 0
Calculated value:
Because 2.58 < 3.400 so we reject H0. This means that coefficient on
l_EXEMPTION in Model 3 is statistically significant at 1% significance level
(99% confidence level) as the coefficient is statistically different from zero.
This requires you to: state the null and alternative hypotheses used,
determine a critical value at the 5% significance level; calculate F
statistic and present your decision.
[25 marks]
𝐻0 : 𝛽𝑙_𝐸𝑋𝐸𝑀𝑃𝑇𝐼𝑂𝑁 = 0 ∧ 𝛽𝑃𝐼𝐿𝐿 = 0
(all coefficients are simultaneously equal zero)
𝐻1 : 𝛽𝑙_𝐸𝑋𝐸𝑀𝑃𝑇𝐼𝑂𝑁 ≠ 0 ∨ 𝛽𝑃𝐼𝐿𝐿 ≠ 0
(not all coefficients are simultaneously equal zero)
Calculated value:
𝑅 2 /𝑘 0.427471/2
𝐹= = = 24.26568
(1 − 𝑅 2 )/(𝑛 − 𝑘 − 1) (1 − 0.427471)/65
Because 3.15 < 24.26568 so we reject H0. This means that Model 4 is
statistically significant at 5% significance level (95% confidence level) as all
the coefficients are not simultaneously equal zero.
iii. State how you would test for the omission of the WW2 in Model 3 in
Table III using an F-statistic. Please refer to Table III and IV.
This requires you to: state the null and alternative hypotheses used,
Page 7 Continued…
212ECN
Calculated value:
Because 4.00 < 8.9891 so we reject H0. This means that test is statistically
significant at 5% significance level (95% confidence level) as the coefficient
is statistically different from zero. It suggests that WW2 can’t be omitted.
Page 8 Continued…
212ECN
Question 3
Table V presents the Ramsey’s RESET test of the Model 3 in Table III in
Question 2.
Table VI presents the White test of the Model 3 in Table III in Question 2.
Table V
Table VI
Page 9 Continued…
212ECN
Explain in detail how we can test for the problem and how a regression
can be corrected if multicollinearity is found.
[50 marks]
Nature
The word collinear describes a linear correlation between two
independent variables, and multicollinearity indicates that more than
two independent variables are involved.
Perfect multicollinearity - an extreme case, two variables might be a
linear combination of each other: x1 = 3x2
Imperfect multicollinearity occurs when two or more RHS variables
are imperfectly linearly related: x1i = 3x2i + ui
Impact
Perfect multicollinearity is rare and normally the result of an error on
the user’s part
β1 & β2 - indeterminable
SE(β1) & SE(β2) = infinite
Detection
Page 10 Continued…
212ECN
Page 11 Continued…
212ECN
(2)
Where Ri2 is the unadjusted R2 from step one
There are no critical tables for VIF but just a rule of thumb.
If VIFi > 10 then it suggests multicollinearity problem as Ri2 >
0.9
VIF is a sufficient, not necessary, test for multicollinearity
because
Thus, whether variance of estimated coefficient is large or
small will depend on three factors
And not only VIF
So you might have large variances of estimated coefficients
in equation that has no large VIFs
Causes of multicollinearity
There are many reasons why we might find multicollinearity
Definitions of variables
e.g. nominal series reflect common trend due to changes in
price index
An identity relationship included
Too small a sample with too little variability
Too many dummies (dummy variable trap)
Entry errors
Solutions
1. Do nothing
A remedy should be considered only if the consequences
cause insignificant t ratios or unreliable estimated
coefficients!
The deletion of a multicollinear variable that belongs in a
equation will cause specification bias (omitting a relevant
variable).
Page 12 Continued…
212ECN
This requires you to: state the null and alternative hypotheses used,
determine a critical value at the 5% significance level; calculate F
statistic and present your decision. Please refer to Table V.
[25 marks]
𝐻0 : 𝛽𝑦ℎ𝑎𝑡^2 = 0 ∧ 𝛽𝑦ℎ𝑎𝑡^3 = 0 (no misspecification of Model 3)
𝐻1 : 𝛽𝑦ℎ𝑎𝑡^2 ≠ 0 ∨ 𝛽𝑦ℎ𝑎𝑡^3 ≠ 0 (misspecification of Model 3)
Calculated value:
(𝑅𝑆𝑆𝑅𝑒𝑠𝑡 − 𝑅𝑆𝑆𝑈𝑛𝑅𝑒𝑠𝑡 )/𝑟 (1.460845 − 1.025807)/2
𝐹= = = 13.1469
𝑅𝑆𝑆𝑈𝑛𝑅𝑒𝑠𝑡 /(𝑛 − 𝑘𝑈𝑛𝑅𝑒𝑠𝑡 − 1) 1.025807/62
Because 3.15 < 13.1469 so we reject H0. This means that the specification of
Page 13 Continued…
212ECN
This requires you to: state the null and alternative hypotheses used,
determine a critical value at the 5% significance level; calculate the
White statistic and present your decision. Please refer to Table VI.
[25 marks]
𝐻0 : 𝛽𝑙_𝐸𝑋𝐸𝑀𝑃𝑇𝐼𝑂𝑁 = 0 ∧ 𝛽𝑃𝐼𝐿𝐿 = 0 ∧ 𝛽𝑊𝑊2 = 0 ∧ 𝛽𝑠𝑞_𝑙_𝐸𝑋𝐸𝑀𝑃𝑇𝐼𝑂𝑁 = 0 ∧
𝛽𝑋2_𝑋3 = 0 ∧ 𝛽𝑋2_𝑋4 = 0 (homoskedasticity)
𝐻0 : 𝛽𝑙_𝐸𝑋𝐸𝑀𝑃𝑇𝐼𝑂𝑁 ≠ 0 ∨ 𝛽𝑃𝐼𝐿𝐿 ≠ 0 ∨ 𝛽𝑊𝑊2 ≠ 0 ∨ 𝛽𝑠𝑞_𝑙_𝐸𝑋𝐸𝑀𝑃𝑇𝐼𝑂𝑁 ≠ 0 ∨
𝛽𝑋2_𝑋3 ≠ 0 ∨ 𝛽𝑋2_𝑋4 ≠ 0 (heteroskedasticity)
Calculated value:
𝑊ℎ𝑖𝑡𝑒 ′ 𝑡𝑒𝑠𝑡 𝑠𝑡𝑎𝑡𝑖𝑠𝑡𝑖𝑐 = 𝑇𝑅 2 = 68 ∗ 0.342045 = 23.25906
Because 12.59 < 23.25906 so we reject H0. This means that residuals from
Model 3 are heteroskedastic at 5% significance level (95% confidence level)
as not all the coefficients are simultaneously equal zero.
Page 14 Continued…
212ECN
Question 4
Table VII presents the summary statistics of the residuals from the Model 3 in
Table III in Question 2.
Table VIII presents the Chow test of the Model 3 in Table III in Question 2.
Table VII
Table VIII
Augmented regression for Chow test
OLS, using observations 1917-1984 (T = 68)
Dependent variable: l_FERTILITY
Omitted due to exact collinearity: sd_PILL
Page 15 Continued…
212ECN
Table IX
Page 16 Continued…
212ECN
This requires you to: state the null and alternative hypotheses used,
determine a critical value at the 1% significance level; calculate Jarque-
Bera statistic and present your decision. Please refer to Table VII.
[25 marks]
H0 : skewnessuhat3 = 0 ∧ excess kurtosisuhat3 = 0
(residuals are normally distributed)
H1 : skewnessuhat3 ≠ 0 ∨ excess kurtosisuhat3 ≠ 0
(residuals are not normally distributed)
Calculated value:
𝑛 2
(𝑘 − 3)2 68 2
(−0.667137)2
𝐽 − 𝐵 = (𝑠𝑘 + )= (0.130942 + )
6 4 6 4
= 1.455356
Because 1.455356 < 9.21 so we can’t reject H0. This means that residuals
are normally distributed at 1% significance level (99% confidence level)
as skewness and excess kurtosis are jointly equal 0.
This requires you to: state the null and alternative hypotheses used,
determine a critical value at the 5% significance level; calculate Chow
statistic and present your decision. Please refer to Table VIII.
[25 marks]
𝐻0 : 𝛽𝑠𝑝𝑙𝑖𝑡𝑑𝑢𝑚 = 0 ∧ 𝛽𝑠𝑑_𝑙_𝐸𝑋𝐸𝑀𝑃𝑇𝐼𝑂𝑁 = 0
(no structural break in Model 3)
𝐻1 : 𝛽𝑠𝑝𝑙𝑖𝑑𝑢𝑚 ≠ 0 ∨ 𝛽𝑠𝑑_𝑙_𝐸𝑋𝐸𝑀𝑃𝑇𝐼𝑂𝑁 ≠ 0
(structural break in Model 3)
Calculated value:
(𝑅𝑆𝑆𝑅𝑒𝑠𝑡 − 𝑅𝑆𝑆𝑈𝑛𝑅𝑒𝑠𝑡 )/𝑟 (1.460845 − 1.370414)/2
𝐹= =
𝑅𝑆𝑆𝑈𝑛𝑅𝑒𝑠𝑡 /(𝑛 − 𝑘𝑈𝑛𝑅𝑒𝑠𝑡 − 1) 1.370414/62
= 2.045631
Because 2.045631 < 3.15 so we can’t reject H0. This means that there is no
structural break at observation 1980 in Model 3 at 5% significance level (95%
confidence level) as all of the coefficients are simultaneously equal zero.
iii. Explain briefly how to compare R2 between linear and log-log model.
[15 marks]
Let assume that we have two models:
Page 17 Continued…
212ECN
This requires you to: state the null and alternative hypotheses used,
determine a critical value at the 5% significance level; calculate
Breusch-Godfrey statistic and present your decision. Please refer to
Table IX.
[25 marks]
𝐻0 : 𝛽𝑢ℎ𝑎𝑡3_1 = 0 ∧ 𝛽𝑢ℎ𝑎𝑡3_2 = 0 (no autocorrelation up to order 2)
𝐻0 : 𝛽𝑢ℎ𝑎𝑡3_1 ≠ 0 ∨ 𝛽𝑢ℎ𝑎𝑡3_2 ≠ 0 (autocorrelation up to order 2)
Calculated value:
𝐵𝑟𝑒𝑢𝑠𝑐ℎ − 𝐺𝑜𝑑𝑓𝑟𝑒𝑦 𝑡𝑒𝑠𝑡 𝑠𝑡𝑎𝑡𝑖𝑠𝑡𝑖𝑐 = 𝑇𝑅 2 = 68 ∗ 0.774634 = 52.67511
or
𝐵𝑟𝑒𝑢𝑠𝑐ℎ − 𝐺𝑜𝑑𝑓𝑟𝑒𝑦 𝑡𝑒𝑠𝑡 𝑠𝑡𝑎𝑡𝑖𝑠𝑡𝑖𝑐 = (𝑇 − 2) ∗ 𝑅 2 = 66 ∗ 0.774634
= 51.12584
Because 5.99 < 52.67511 so we reject H0. This means that there is
autocorrelation up to order 2 in Model 3 at 5% significance level (95%
confidence level) as all the coefficients on lagged residuals are not
simultaneously equal zero.
Page 18 Continued…
212ECN
Statistical tables
Student t F
Z Durbin-Watson
Χ2 QLR
Student t
Page 19 Continued…
212ECN
Page 20 Continued…
212ECN
Page 21 Continued…
212ECN
Page 22 Continued…
212ECN
Page 23 Continued…
212ECN
Significance level
Number of restrictions (r)
10% 5% 1%
1 7.12 8.68 12.16
2 5.00 5.86 7.78
3 4.09 4.71 6.02
4 3.59 4.09 5.12
5 3.26 3.66 4.53
6 3.02 3.37 4.12
7 2.84 3.15 3.82
8 2.69 2.98 3.57
9 2.58 2.84 3.38
10 2.48 2.71 3.23
11 2.40 2.62 3.09
12 2.33 2.54 2.97
13 2.27 2.46 2.87
14 2.21 2.40 2.78
15 2.16 2.34 2.71
16 2.12 2.29 2.64
17 2.08 2.25 2.58
18 2.05 2.20 2.53
19 2.01 2.17 2.48
20 1.99 2.13 2.43
Source: Andrews (2003)
Page 24 End