PRINCIPLES OF ECONOMETRICS

5TH EDITION

ANSWERS TO ODD-NUMBERED
EXERCISES IN CHAPTER 6

1
 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 2

EXERCISE 6.1
SSE 2132.65
(a) R2  1  1   0.5501
SST 4740.11595
( SST  SSE ) ( K  1) (4740.11595  2132.65) 2
(b) F    28.73
SSE ( N  K ) 2132.65 (50  3)

At   0.01 , the critical value is F(0.99, 2, 47)  5.09 . We reject H 0 . There is evidence from
the data to suggest that  2  0 and/or 3  0 .

(c) The F-value for testing this hypothesis is

( SSE R  SSEU ) J (2132.65  1072.88) 2
F    22.23
SSEU ( N  K ) 1072.88 (50  5)

The critical value for significance level   0 .0 1 is F(0.99,2,45)  5.11 . Since the calculated
F is greater than the critical F we reject H 0 and conclude that the model is misspecified.

SSE 401.179
(d) R2  1   1  0.9154
SST 4740.11595

F
 SSER  SSEU J (401.179  388.684) 2
  0.707
(e)
SSEU ( N  K ) 388.684 (50  6)

The critical value for significance level   0.05 is F(0.95,2,44)  3.209 . Since the
calculated F is less than the critical F we fail to reject H 0 . There is no evidence from
RESET to suggest the model is misspecified. Including zi2 appears to have remedied the
initial problem.

EXERCISE 6.3
(a) From the definition of wi , we have

b2   wi  yi  y    wi yi

with the second equality holding because  wi y  y  wi  0 . Then,

b   wi yi

2

  wi  1  2 xi  3 zi  ei 
 1  wi  2  wi xi  3  wi zi   wi ei
 2  wi  xi  x   3  wi zi   wi ei
 2  3  wi zi   wi ei

 wi  xi  x     xi  x    xi  x   1.
2 2
with the last equality holding because

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 3

(b) From part (a),

E  b2* | X   2  3  wi  zi  z    wi E  ei | X 

1
N
  xi  x  zi  z 
  2  3
1
  xi  x 
2

N
  x, z 
cov
  2  3
  x
var

(c) From part (a), the conditional variance of b2* is given by

var  b2* | X   var  2  3  wi zi   wi ei  | X 

  wi2 var  ei | X 

  xi  x  
2
2 2
 2 
  xi  x  N var(
 x)
2 2
   xi  x 2 
 
(d) From equation (5.13),
2
var  b2 | X  
1  r    x  x 
2
xz i
2

where rxz is the sample correlation between x and z. Then,

2 var  b2 | X 
var  b2 | X   
1  r  N var(
2
xz
 x) 1  r  2
xz

var  b2 | X 

Because 1  r 2
 1,  var  b2 | X 
xz
1  r 2
xz

 
and hence var b2* | X  var  b2 | X  . Although b2 suffers from omitted variable bias, it
does have a lower variance than b2 .

EXERCISE 6.5
The following assumptions are sufficient
E  EXPER | EDUC , AGE   E  EXPER | AGE   1  2 AGE  3 AGE 2

E  EXPER 2 | EDUC , AGE   E  EXPER 2 | AGE   1   2 AGE  3 AGE 2

E  e | EDUC , AGE   0

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 4

EXERCISE 6.7
 0  76.974 and
The point and interval predictions for SALES from Example 6.15 are SALES
(67.533, 86.415), respectively.
The point estimate for E ( SALES | PRICE  6, ADVERT  1.9) is

Eˆ ( SALES | PRICE  6, ADVERT  1.9)  76.974

The standard error is se E ( SALES )  0.84215  0.9177 , leading to the following interval estimate for
expected sales
76.974  1.9939  0.9177   75.144, 78.804 

This interval is much narrower than the one for the prediction interval in Example 6.15. Because it
is concerned with estimating average sales from all cities rather than predicting sales for a specific
city, the error variance for an observation from a specific city is not included.

EXERCISE 6.9
Consider, for example, the model
y  1   2 x  3 z  e

If we augment the model with the predictions ŷ the model becomes

y  1   2 x  3 z   yˆ  e
However, ŷ  b1  b2 x  b3 z is perfectly collinear with x and z . This perfect collinearity means
that least squares estimation of the augmented model will fail.

EXERCISE 6.11
There are several ways in which the restrictions can be substituted into the model, with each one
resulting in a different restricted model. We have chosen to substitute out 1 and 3 , leading to
the model

 SALES  ADVERT  78.1  2  PRICE  6  4  3.61  3.8 ADVERT  ADVERT 2   e

EXERCISE 6.13
(a) Let 2  the coefficient of GUNRATE in the firearm suicide equation. To answer the
question is there evidence that the gun buyback has reduced firearm suicides, we test
H 0 : 2  0 against the alternative H1 : 2  0 . The t-value for this test is
t  0.223 0.069  3.232 . The 1% critical value is t(0.01, 195)  2.346 . Because
3.232  2.346 , we reject H 0 : 2  0 at 10%, 5% and 1% significance levels. We
conclude that the gun buyback has been successful in reducing the number of firearm
suicides.
Let  2  the coefficient of GUNRATE in the non-firearm suicide equation. To answer the

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 5

question has there been substitution away from firearms to other means of suicide, we test
H 0 : 2  0 against the alternative H1 :  2  0 . The t-value for this test is
t  0.553 0.144  3.840 . The 1% critical value is t(0.99, 195)  2.346 . Because
3.840  2.346 , we reject H 0 : 2  0 at 10%, 5% and 1% significance levels. Although the
gun buyback has been successful in reducing the number of firearm suicides, it has led to
an increase in the number of non-firearm suicides. There has been substitution away from
firearms to other means of suicide.
Let  5  the coefficient of YEAR in the equation for the overall suicide rate. To answer the
question is there a separate trend in the suicide rate, we test H 0 :  5  0 against the
alternative H1 :  5  0 . The t-value for this test is t  0.056 0.393  0.142 . The 10%
critical values are t(0.95, 195)  1.653 . Because 1.653  0.142  1.653 , we fail to reject
H 0 :  5  0 at 10%, 5% and 1% significance levels. There is no evidence to suggest there
is a separate overall trend in the suicide rate.
Tests for the coefficients of YEAR in the other two equations suggest there has been a
decreasing trend in firearm suicides and an increasing trend in non-firearm suicides, not
related to the gun buyback, but there is no evidence of any overall trend in the suicide rate.

(b) Let  3  the coefficient of URATE in the equation for the overall suicide rate. To answer
the question is there evidence that greater unemployment increases the suicide rate, we test
H 0 :  3  0 against the alternative H1 :  3  0 . The t-value for this test is
t  1.147 1.206  0.951 . The 10% critical value is t(0.90, 195)  1.286 . Because
0.951  1.286 , we fail to reject H 0 :  3  0 at 10%, 5% and 1% significance levels. There
is no evidence to suggest that greater unemployment increases the suicide rate.
A similar conclusion is reached from the firearm equation. In the non-firearm equation,
testing H 0 : 3  0 against the alternative H1 : 3  0 , t  1.710 ; we reject H 0 : 3  0
at a 5% significance level, but not at a 1% significance level. There is some evidence that
greater unemployment increases the non-firearm suicide rate.

(c) For the firearm equation, we have H 0 : 3  0, 4  0 versus H1 : 3  0, and/or 4  0 .

The F-value is

F
 SSER  SSEU  J

 50641  29745  2
 68.494
SSEU ( N  K ) 29745 (200  5)

For the non-firearm equation, we have H 0 : 3  0, 4  0 versus

H1 : 3  0, and/or  4  0 . The F-value is

F
 SSER  SSEU  J

131097  129122  2
 1.491
SSEU ( N  K ) 129122 (200  5)

For the overall equation, we have H 0 :  3  0,  4  0 versus H1 :  3  0, and/or  4  0 . The

F-value is

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 6

F
 SSER  SSEU  J

175562  151890  2
 15.195
SSEU ( N  K ) 151890 (200  5)

The 1%, 5% and 10% critical F-values are, respectively, F(0.99, 2, 195)  4.716 ,
F(0.95, 2, 195)  3.042 , and F(0.90, 2, 195)  2.330 . All three significance levels lead to the same
conclusion. URATE and CITY contribute to the firearm equation and the overall equation
(15.195 and 68.494 are greater than 4.716), but not to the non-firearm equation (1.491 is
less than 2.330).

EXERCISE 6.15
(a) We test the null hypothesis H 0 : 3  0, 4  0 against the alternative that one of the
coefficients is not zero. The F-test statistic is
 SSER  SSEU  J 1513.6  1358.7  2
F   2.74
SSEU  N  K  1358.7  52  4 

The 10% critical value is 2.417. We reject the null hypothesis that both coefficients are zero
at the 10% level of significance and conclude that at least one of the competing brands prices
should be included in the equation.

(b) When PRB and PRC are omitted from the equation, the coefficient of PRA will change by
3.63944 . The intercept will change by 17.7507 .

(c) Let m  E (CANS | PRA  0.91, PRB  0.91, PRC  0.90) , then mˆ  3.4392

The interval estimate is mˆ  t 0.975,48se  mˆ   0.26239708, 6.6160029

(d) From (b), Eˆ (CANS | PRA  0.91)  1.17979 .

Using (c),
Eˆ (CANS | PRA  0.91)  22.96  47.08  0.91  9.30 Eˆ ( PRB | PRA  0.91)
 16.51Eˆ ( PRC | PRA  0.91)
The estimate from this equation will differ from that in part (c) because in part (c) we set
PRB  0.91 and PRC  0.90 . These values do not coincide with Eˆ ( PRB | PRA  0.91)
and Eˆ ( PRC | PRA  0.91) . The values for PRB and PRC for which the two estimates are
the same are Eˆ ( PRB | PRA  0.91)  0.84925 and Eˆ ( PRC | PRA  0.91)  0.79737 .

(e)   3.4392 . An estimate of the variance of the prediction error

The point prediction is CANS
  f   30.80265 .The prediction interval is
f is var

 0.975,48 se  f   3.4392  2.0106348  5.5500135    7.7198501, 14.59825

 t
CANS

We estimate that at the given prices the sales are between zero and 14,598 cans. This interval
is very much wider than the interval for the average number of cans sold.

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 7

(f)  2 is
The RESET statistic to test the significance of CANS

RESET 
1358.7  1198.9   6.2645759
1198.9 47

For degrees of freedom 1, 47  the 10%, 5% and 1% critical values are 2.8154381,
4.0470999, 7.2068389, respectively. Thus, we reject the null hypothesis of no specification
error at 10% and 5%, but not at 1%.

EXERCISE 6.17
Values for the various model selection criteria for each of the equations are reported in the table
below. All 4 criteria suggest that all the new models with BATHS and BEDROOMS included are
an improvement over the best model in Example 6.16. From the set of new models, no.11, with the
extra variables BATHS and BEDROOMS  SQFT , is favored by R 2 , AIC and SC, while the
RMSE for the predictions in the hold-out sample favors model no.10, with
BATHS and BEDROOMS .

Variables included in addition to those in

Model R2 AIC SC RMSE
(xr6.17)
2 None 0.7198 –2.609 –2.587 0.2714
9 BATHS 0.7377 –2.674 –2.647 0.2615
10 BATHS , BEDROOMS 0.7405 –2.684 –2.652 0.2590
11 BATHS , BEDROOMS  SQFT 0.7408 –2.685 –2.653 0.2633
12 BATHS , BEDROOMS  SQFT , BATHS  SQFT 0.7405 –2.682 –2.645 0.2648

EXERCISE 6.19
(a) The change in E (WFOOD | TOTEXP, NK ) from adding an extra child when
TOTEXP  90 is
1   2 ln(90)  3  NK  1   4  NK  1  ln(90) 

 1   2 ln(90)  3 NK   4  NK  ln(90) 

 3   4 ln(90)

(b) The change in E (WFOOD | TOTEXP, NK ) from an increase in total expenditure from
£80/week to £120/week when NK  2 is
1  2 ln(120)  23  4  2ln(120) 

 1  2 ln(80)  23  4  2 ln(80) 

 2 ln 1.5   2 4 ln 1.5 

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 8

(c) Direct substitution yields

E (WFOOD | TOTEXP  90, NK  2)  1  2 ln(90)  23  24 ln(90)

(d) For H 0(1) , the relationship between the F and  2 test values is F =  2 . The relationship
between the critical values is F(1 ,1,846)  (12  ,1) . The relationship between the t and F tests
for H 0(1) is F  t 2 for the test values and F(1 ,1,846)  t(12  2,846) for the critical values.

For H 0(2) , the relationship between the F and  2 test values is F = 2 2 . The relationship
between the critical values is F(1 , 2,846)  (12  , 2) 2 .
For H 0(3) , the relationship between the F and  2 test values is F =  2 3 . The relationship
between the critical values is F(1 ,3,846)  (12  ,3) 3 .

(e) The p-values are given in the following table

p-values for hypotheses

H 0(1) H 0(2) H 0(3)

Number of observations F 2 F 2 F 2

100 0.652 0.651 0.462 0.459 0.386 0.381

400 0.953 0.953 0.053 0.052 0.108 0.106
850 0.094 0.094 0.008 0.008 0.022 0.022

(f) With the exception of going from 100 to 400 observations for H 0(1) , increasing the number
of observations decreases the level of significance at which the null hypothesis is rejected.
This finding is in line with a general result that increasing the number of observations
increases the power of a test. Adding more hypotheses does not necessarily increase nor
decrease the level of significance at which a null hypothesis is rejected. With 400
observations and a 10% significance level we reject H 0(2) , but after adding an extra
condition for testing H 0(3) the null hypothesis is no longer rejected at this significance
level. There are no dramatic differences between the F-test outcomes and the  2 -test
outcomes.

EXERCISE 6.21
(a) (i) A 95% interval estimate for the saturation level  is given by

ˆ  t(0.975, 43) se  ˆ   0.81438  2.0167  0.05105  (0.7114, 0.9173)

(ii) A 95% interval estimate for the inflection point t I    is given by

 
ˆ ˆ  t(0.975, 43) se ˆ ˆ  1.37767 0.05722  2.0167  2.4777  (19.08, 29.07)

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 9

This interval corresponds to years 1988 to 1998.

(iii) A 95% interval estimate for the EAF share in 1969 when t  0 is given by

 
ˆ ˆ
 t(0.975, 43) se  
1  exp ˆ    
 1  exp ˆ 

0.81438
  2.0167  0.00671869  (0.1505, 0.1776)
1  exp(1.37767)

(iv) The predicted shares up to 2050 are given in Table 6.21. They are plotted, along with
their 95% bounds, in Figure 6.21. The predictions are approaching the saturation level
very slowly. From 2040 to 2050, they increase only a small amount, from 0.7624 to
0.7842, when the saturation level is 0.8144. The 95% bounds become wider the further
into the future we are forecasting, reflecting the extra uncertainty from doing so.

Table 6.21 Predicted share of electric arc furnace technology

Year Share Year Share Year Share
2016 0.6416 2028 0.7172 2040 0.7624
2017 0.6492 2029 0.7220 2041 0.7651
2018 0.6566 2030 0.7265 2042 0.7677
2019 0.6638 2031 0.7309 2043 0.7701
2020 0.6707 2032 0.7351 2044 0.7725
2021 0.6773 2033 0.7391 2045 0.7747
2022 0.6837 2034 0.7429 2046 0.7768
2023 0.6899 2035 0.7466 2047 0.7788
2024 0.6958 2036 0.7501 2048 0.7807
2025 0.7015 2037 0.7534 2049 0.7825
2026 0.7069 2038 0.7565 2050 0.7842
2027 0.7122 2039 0.7595

Figure 6.21 Predicted share of electric arc furnace technology and 95% interval bounds

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 10

(b) The saturation level is given by  and the point of inflection is given by t I    .
Having    25 is equivalent to 25    0 . Thus, we set up the hypotheses
H 0 :   0.85, 25    0
H1 : at least one of the equalities in H 0 does not hold
Values of the F(2, 43) and  (2)
2
test statistics for testing this hypothesis are F  4.460 (p-
value = 0.0174) and  2  8.92 (p-value = 0.0116). Since both p-values are less than 0.05,
both test statistics lead to the same conclusion: we reject H 0 .

Since   0.85 and tI  25 both lie within the respective 95% interval estimates for 
and t I , it is perhaps surprising that the joint hypothesis H 0 :   0.85, tI  25 is rejected.
However, the separate interval estimates do not allow for the correlation between the
estimates for  and t I . If these estimates are negatively correlated, then it is unlikely that
both  and t I could take on these values which occur in the upper regions of their
respective interval estimates.

(c) When the restrictions   0.85 and 25    0 are imposed, the model becomes
0.85
yt   et
1  exp(25  t )
The F-test value is

F
 SSER  SSEU  J

 0.020462  0.017272  2
 3.97
SSEU (T  K ) 0.017272 43
which has a corresponding p-value of 0.0262. The test outcome is the same. The null
hypothesis is rejected. However, the F-test value is different. The equivalence of the
approaches only holds for equations which are linear in the parameters.

EXERCISE 6.23
(a)

Coefficient Std. Error

̂  2.13711 0.43777
̂ 0.77897 0.10978
̂  0.27247 0.74451
̂ 0.63834 0.09435

(b)
Interval estimates
Parameter Estimate Standard Error
Lower Upper
 0.1180 0.05165 0.0153 0.2207
 0.7790 0.1098 0.5607 0.9973
 1.3745 1.4066  1.4226 4.1716
 0.6383 0.0944 0.4507 0.8260

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 11

(c) For testing H 0 :   1,   0 against H1 :   1 or   0 , we reject H 0 if

F  F(0.95,2,84)  3.105 . Using computer software, the computed value is F  2.113 with p-
 
value Pr F(2, 84)  2.113  0.1272 . There is insufficient evidence to suggest a constant-
returns-to-scale Cobb-Douglas function is inadequate. Alternatively, carrying out a 2 -test,
we have 2  4.226 , and Pr (2)
2

 4.226  0.1209 
EXERCISE 6.25
(a)

ln( PRICE )  4.20037  0.039603SQFT  0.018677 AGE  0.00020558 AGE 2
(se) (0.03009) (0.000974) (0.001436) (0.00002436)

(b) Graphing Eˆ [ln( PRICE ) | SQFT  22, AGE ] against AGE, we obtain
5.1

5.0

4.9
log_PRICE

4.8

4.7

4.6
0 10 20 30 40 50 60 70 80 90

AGE

Figure 6.25 Estimated log-price for houses of different ages

(c) Simplifying and combining into the one joint null hypothesis, we have
753  63754  0
H0 : 
82  253  8754  0
H 1 : at least one equality in H 0 does not hold

The value of the F-statistic for testing H 0 is F  1.710 with p-value = 0.1814. Also, the
5% critical value is F(0.95, 2, 896)  3.006 . Thus, we fail to reject H 0 . There is no evidence to
suggest the joint hypothesis about expected log prices is incorrect.
(d) Simplifying and combining into one joint null hypothesis, we have.
  1004  0
H0 :  3
1  222  503  25004  4.60517
H 1 : at least one equality in H 0 does not hold

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 12

The value of the F-statistic for testing H 0 is F  3.109 with p-value = 0.0451. Also, the
5% critical value is F(0.95, 2, 896)  3.006 . Thus, we reject H 0 . The data do not support the
joint hypothesis about expected log prices.

(e) Estimates with BATHS and SQFT  BEDROOMS added are

Variable Coefficient Std. Error t-value p-value

C 3.944206 0.042032 93.83751 0.0000

SQFT 0.041407 0.003037 13.63533 0.0000
AGE -0.016961 0.001404 -12.07745 0.0000
AGE^2 0.000195 2.35E-05 8.286746 0.0000
BATHS 0.175518 0.021777 8.059791 0.0000
BEDROOMS*SQFT -0.002007 0.000585 -3.433094 0.0006

(f) Adding BATHS and SQFT  BEDROOMS will have improved the predictive ability of
the model if 5  0 and/or 6  0 . Thus, we test H 0 : 5  0, 6  0 against
H1 :5  0 and/or 6  0 . The calculated F-value for this test is F  37.358 with p-value
= 0.0000. Also, the 5% critical value is F(0.95, 2, 894)  3.006 . Thus, we reject H 0 . Adding
BATHS and SQFT  BEDROOMS improves the predictive ability of the model.

(g) Using the “natural predictor”, before adding one extra bedroom and bathroom the value of
the house is estimated to be
 BEFORE  exp  b  23b  2b  (3  23)b   165.525
PRICE 1 2 5 6

After adding one extra bedroom and bathroom the value of the house is estimated to be
 AFTER  exp  b  25.6b  3b  (4  25.6)b   205.463
PRICE 1 2 5 6

The estimated increase in value is $205,463  $165,525 = $39,938.

(h) Assuming there is no inflation in house prices, and the same equation is relevant after 20
years, in 20 years’ time the estimated values without and with the extra bedroom and
bathroom are
 BEFORE  exp  b  23b  20b  202 b  2b  (3  23)b   127, 466
PRICE 1 2 3 4 5 6

 AFTER  exp  b  25.6b  20b  202 b  3b  (4  25.6)b   158, 221

PRICE 1 2 3 4 5 6

The estimated increase in value is $158,221  $127,466 = $30,755. The extra bedroom and
bathroom has less value in the older house.

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 13

EXERCISE 6.27
(a) The least-squares estimated equation using observations where EDUC  7 ( N  1178) is

ln(WAGE )  1.1391  0.11673EDUC  0.008321EXPER
(se) (0.0869) (0.00547) (0.001059)
The hypotheses for the claim about the marginal effects of education and experience are
H 0 : 2  0.112, 3  0.008 H1 : 2  0.112 or 3  0.008
We reject H 0 if the computed value for the F-statistic, Fcalc , is greater than or equal to
F(0.95, 2, 1175)  3.003 . We find Fcalc  0.389  3.003 , and hence do not reject H 0 . There is
no evidence to dispute the claim made about the marginal effects of experience and
education.

(b) Augmenting the equation with the squares of the predictions, the RESET F-value for testing
the adequacy of the model is

F
 SSER  SSEU  1  261.8566  260.4315  6.424
SSEU 1174 260.4315 1174

Since 6.424  F(0.95, 1, 1174)  3.849 (p-value = 0.0114), we conclude that the squares of the
predictions have significant explanatory power and hence the model is inadequate.
Augmenting the equation with the squares and cubes of the predictions, the RESET F-value
for testing the adequacy of the model is

F
 SSER  SSEU  2

 261.8566  259.8446  2
 4.541
SSEU 1173 259.8446 1173

Since 4.541  F(0.95, 2, 1173)  3.003 (p-value = 0.0108), we conclude that the squares and
cubes of the predictions have significant explanatory power and hence the model is
inadequate.

(c) The results from estimating the extended model are

Variable Coefficient Std. Error t-Value p-value

C 0.065994 0.424611 0.155421 0.8765

EDUC 0.226881 0.053758 4.220395 0.0000
EXPER 0.040795 0.007593 5.372761 0.0000
EDUC^2 -0.003287 0.001724 -1.906540 0.0568
EXPER^2 -0.000481 7.79E-05 -6.181598 0.0000
EDUC*EXPER -0.000673 0.000419 -1.605578 0.1086

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 14

The hypotheses and test results for each part of the question are given in the following table.

Part Null hypothesis Fcalc F(0.95, 2, 1172) p-value Test decision

2  20 4  5 6  0.112
(i) 16.039 3.003 0.0000 Reject H 0
3  10 5  10 6  0.008
2  20 4  5 6  0.112
(ii) 0.580 3.003 0.5599 Fail to reject H 0
3  10 5  10 6  0.008
2  20 4  5 6  0.112
(iii) 13.638 3.003 0.0000 Reject H 0
3  10 5  10 6  0.008

(d) With the squares of the predictions, the RESET F-value is

F
 SSER  SSEU  1  252.7137  252.6801  0.156
SSEU 1171 252.6801 1171

Since 0.156  F(0.95, 1, 1171)  3.849 (p-value = 0.6930), we conclude that the squares of the
predictions do not have significant explanatory power; the test does not suggest the model
is inadequate.
With the squares and cubes of the predictions, the RESET F-value is

F
 SSER  SSEU  2

 252.7137  251.8972  2
 1.896
SSEU 1170 251.8972 1170

Since 1.896  F(0.95, 2, 1170)  3.003 (p-value = 0.1506), we conclude that the squares and
cubes of the predictions do not have significant explanatory power; the test does not suggest
the model is inadequate.

(e) The claim is not true for all levels of education and experience. The medians for EDUC and
EXPER are 14 and 23, respectively, values that approximately correspond to those tested in
part (ii). Thus, the claim is reasonable for those at the median levels of education and
experience, but it does not hold for values of EDUC and EXPER which are much greater or
less than the medians.

EXERCISE 6.29
(a) The estimated equation is

ln( SHARE )  2.9569  0.015074 TAX
(se) (0.0405) (0.001038)
We find that an increase in the marginal tax rate for top earners by one percentage point is
associated with a decrease in the income share of the top earners by 1.5 percent. It is unlikely
we could interpret this estimate as causal. SHARE will depend not only on TAX, but also on
variables related to the general state of the economy. If these variables are correlated with
TAX, which is likely, 2 cannot be given a causal interpretation.

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 15

(b) The estimated model with the quadratic trend included is


ln( SHARE )  3.1325  0.0079849 TAX  0.027217 YEAR  0.00031104 YEAR 2
(se) (0.0411) (0.0015703) (0.004054) (0.00004250)
After taking out the effect of the quadratic trend, we find that an increase in the marginal
tax rate for top earners by one percentage point is associated with a decrease in the income
share of the top earners by 0.8 percent. Adding the trend has decreased the estimated effect
of the marginal tax rate by a considerable amount.
The correlations between TAX and YEAR and TAX and YEAR 2 are given in the following
matrix.

TAX YEAR YEAR_SQ

TAX 1.000000 0.629566 0.462882

YEAR 0.629566 1.000000 0.969008
YEAR_SQ 0.462882 0.969008 1.000000

The correlations between TAX and YEAR and TAX and YEAR 2 are 0.630 and 0.463,
respectively. These values are moderately high and sufficient for omission of the trend from
the equation to result in overestimation of the effect of changes in TAX on ln(SHARE). A
regression of TAX on YEAR and YEAR 2 gives R 2  0.751 , a large enough value to explain
the change in magnitude for ̂ 2 .

(c) SHARE will be smallest when ln(SHARE) is smallest. To find that year, we solve the
following equation for YEAR
 ln( SHARE ) 3
  3  2 4YEAR  0 giving YEARMIN  
YEAR 2 4

with point estimate

 MIN   ˆ 3   0.0272166  43.751
YEAR
2ˆ 4 2  0.000311042

 
 MIN  1.243246
Its standard error is given by se YEAR

A 95% interval estimate for YEARMIN is

 MIN  t
YEAR 

(0.975, 76) se YEAR MIN   41.27, 46.23

After rounding to the closest year, the value YEAR  43.75 corresponds to the year 1964
while YEAR  41.27 and YEAR  46.23 correspond to the years 1961 and 1966,
respectively. Thus, we estimate that SHARE will be smallest in 1964, but, based on a 95%
interval estimate, the year when it is smallest could be as early as 1961 or as late as 1966.
The smallest actual value for SHARE is 7.74 in 1973, a year which does not fall within the
interval estimate.

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 16

(d) The hypotheses are

H 0 : 1  50 2  803  802  4  ln(12)
H1 : 1  50 2  803  802  4  ln(12)

The calculated t-value for this hypothesis is

ˆ 1  50ˆ 2  80ˆ 3  802 ˆ 4  ln(12) 0.061677
t   1.299
se  ˆ 1  50ˆ 2  80ˆ 3  802 ˆ 4  0.047484

 
with p-value Pr t(76)  1.299  0.1979 . Thus, we fail to reject H 0 at all conventional
significance levels. The data do not contradict the conjecture about the expected log income
share of top earners in the year 2000 for a marginal tax rate of 50%.

(e) The hypotheses are

  50 2  80 3  80  4  ln(12) 

2
H0 :  1 
 1  50 2  5 3  5  4  ln(12) 
2

H1 : at least one equality in H 0 does not hold

The calculated F-value for this hypothesis is 1.270 with

p-value Pr  F(2,76)  1.270   0.2867 . Thus, we fail to reject H 0 at the 10%, 5% and 1%
levels of significance. We conclude that the data are consistent with the claim that a marginal
tax rate of 50% leads to the same expected log share of ln(12) for the top income earners in
both 1925 and 2000.

(f) Adding GWTH to the equation, we obtain


ln( SHARE )  3.1264  0.0081225 TAX  0.027073YEAR  0.00030983YEAR 2  0.0034719GWTH
(se) (0.0409) (0.0015585) (0.004018) (0.00004212) (0.0022362)
Comparing the estimates of the coefficients in this equation with the corresponding ones in
part (b), we find that adding GWTH has led to very little change in the estimates. They are
almost identical. Also, the same test outcomes are obtained when the tests in parts (d) and
(e) are repeated using the equation with GWTH included. For the test in part (d), we obtain
 
t  1.104 and Pr t(75)  1.104  0.2730 ; for the test in part (e), F  0.999 and

Pr  F(2,75)  0.999   0.3731 .

The lack of changes can be explained by the low correlations between GWTH and the other
right-hand side variables in the equation. These correlations are given in the following
matrix. They are all less than 0.1.

TAX YEAR YEAR_SQ

GWTH 0.063947 0.019622 0.009369

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 17

(g) To obtain an expected log income share of ln(12) in 2001, when GWTH 2001  3 , we require

ln(12)  1  81 3  812  4  3 5

TAX 
2
  60.54 .
A point estimate for this quantity is TAX

 
  7.514183 .
Using the delta rule to get its standard error, we obtain se TAX

A 95% interval estimate is given by

 t
TAX 

(0.975,75) se TAX  (45.57, 75.51)

We estimate that a marginal tax rate of 60.5% would be required. However, the interval
estimate that ranges from 45.6% to 75.5% shows there is a great deal of uncertainty about
this estimate.

EXERCISE 6.31
(a) Estimates for the model CSUMPTN  1  2 HOURS  3GOV  4 R  5 INC  e are
given in the following output.

The coefficient of 3 , with p-value 0.2559 and the intercept with a p-value of 0.2257 are
not significantly different from zero at a 5% significance level.

(b) The positive estimates for 2 and 3 suggest that both hours worked and government
consumption are complements for private consumption. Testing H 0 : 2  0, 3  0 against
the alternative that at least one of 2 or 3 is zero yields a test value of F  2.662 .
Comparing this value with the critical value F(0.95, 2, 32)  3.295 , or examining its p-value
of 0.0852, we find we cannot reject H 0 : 2  0, 3  0 at a 5% significance level. Thus, it
is possible that both hours worked and government consumption are neither complements
nor substitutes for private consumption.
(c) Re-estimating the equation CSUMPTN  1  2 HOURS  3GOV  4 R  5 INC  e
with GOV omitted yields the following results:

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 18

Omitting the variable GOV has led to a decrease in the magnitude of the coefficient estimate
for HOURS, making it no longer significantly different from zero at a 5% level of
significance, and an increase in the magnitude of the coefficient estimate for INC. The
coefficient estimate for R has changed very little.
(d) Estimates for the equation GOV  1  2 HOURS  3 R  4 INC  v are

The p-values in this output suggest that GOV is highly correlated with HOURS and INC, but
less so with R. These results explain why omission of GOV from the equation in part (c) has
led to relatively large changes in the coefficient estimates for HOURS and INC, and only a
small change in the coefficient estimate for R. Also, the signs of the coefficients of HOURS
and INC in part (d) explain the direction of the changes. In part (c) the effect on CSUMPTN
of a change in HOURS includes the negative indirect effect of HOURS on GOV. Similarly,
the effect on CSUMPTN of a change in INC includes the positive indirect effect of INC
on GOV. The changes can be reconciled explicitly as follows:
E  CSUMPTN | HOURS , R, INC   1  31   2  32  HOURS   4  33  R   5  34  INC
 1   2 HOURS  3 R   4 INC

Estimates for  1,  2 ,  3 ,  4  were given in part (c). We can derive those estimates from
those given in parts (a) and (d):
ˆ 1  b1  b3a1  0.003980  0.164079  0.013171  0.006141
ˆ 2  b2  b3a2  0.424608  0.164079  0.710646  0.308006
ˆ 3  b4  b3a3  0.243435  0.164079  0.100382  0.259906
ˆ 4  b5  b3a4  0.608275  0.164079  0.845469  0.746999

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 19

(e) Re-estimating the models in parts (a) and (c) with the year 2007 omitted yields:

Point and 95% interval forecasts for CSUMPTN in 2007 are:

  0.008779 and the 95% interval forecast is
From model (a): CSUMPTN 2007

(0.975, 31)se( f )   0.01381, 0.03137 

 t
CSUMPTN 2007

  0.006649 and the 95% interval forecast is

From model (c): CSUMPTN 2007

(0.975, 32) se( f )   0.01578, 0.02908 

 t
CSUMPTN 2007

(f) The 2007 realized value for consumption growth was CSUMPTN 2007  0.003606 . The
model in part (c) produced the more accurate forecast for 2007.

EXERCISE 6.33 (a)

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 20

Casual inspection suggests separate equations are not needed for the three different varieties.
Coefficient estimates for corresponding variables are all similar in magnitude, particularly
when considered relative to their standard errors.
(b) SSEU  5.924196  2.933984  4.006701  12.864881
(c) The total number of coefficients in the 3 equations is 12. If a single equation is estimated
for all 3 varieties, the number of parameters is 4. Thus, if we restrict corresponding
coefficients for all varieties to be equal, there are 12  4  8 restrictions.
(d) Using all the data to estimate one equation gives the following results:

Copyright © 2018 Wiley

 Chapter 6, Exercise Answers, Principles of Econometrics, 5e 21

(e) Let the coefficients of the cabernet, pinot and other varieties be denoted by
 C1 , C 2 , C 3 , C 4  ,  P1, P 2 , P3 , P 4  , and  O1 , O 2 , O3 , O 4  , respectively. The null and
alternative hypotheses are
H 0 : C1  P1  O1 , C 2  P 2  O 2 , C 3  P 3  O 3 , C 4  P 4  O 4
H1 : At least one of the inequalities in H 0 does not hold.
The value of the test statistic is

F
 SSER  SSEU  J

12.93496  12.864881 8
 0.688
SSEU ( N  K ) 12.864881 1010

This value is less than the 5% critical value F(0.95, 8, 1010)  1.948 . Thus, the null hypothesis
is not rejected. There is no evidence to suggest that there should be different equations for
different varieties.

Copyright © 2018 Wiley