PRINCIPLES OF ECONOMETRICS

5TH EDITION

ANSWERS TO ODD-NUMBERED
EXERCISES IN CHAPTER 4

1
 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 2

EXERCISE 4.1

(a) R 2 = 0.843
(b) σˆ 2 = 8.4249
(c) R 2 = 0.711

EXERCISE 4.3

(a) yˆ 0 = 4.4
(b) se( f ) = 1.5875
(c) (−0.6520, 9.4520)
(d) (−4.8722, 13.6722)
(e) (−1.8189, 5.8189 )

EXERCISE 4.5

(a) R 2 = 0.204

(b) ryyˆ = ryx = 0.4517 .

(c) h5 = 0.01131, h16 = 0.001351, h21 = 0.0063175

(d) DFBETAS2i = −0.05237256 .

(e) DFFITSi = −0.06474186

(f) eˆistu = −0.71691235

EXERCISE 4.7

(a) R 2 = 0. 113

(b)  = 60.12
ENTERT
(c) [ −52.95, 173.19] or [0, 173.19]
(d) [ −158.85, 519.57 ] or [ 0, 519.57 ]

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 3

EXERCISE 4.9

(a) The Jarque-Bera = 30.405483. The test statistic value is larger than the critical value and we
reject the null hypothesis.
(b) In this case JB = 1.9153333. Thus we fail to reject the null.
(c) In this case JB = 0.88941667. Thus, we fail to reject the null hypothesis.
(d) The log-log model fits the data the best.
(e) (i) The magnitude of the correlation between y and x is the same as the correlation between
y and a + bx.
(ii) The R 2 = 0.31315216.
(iii) YHAT and RPRICE3 have an exact linear relationship.
(iv) 0.43046721
(v) 0.45616516
(f) The log-log model fits the data best.

EXERCISE 4.11

(a) [5.7059548, 42.134045].

(b) [0.61731625, 56.582684].

EXERCISE 4.13

(a) α =
 xi yi
. Substituting we have α =
 ( xi − x )( yi − y ) = b
 xi
2
 ( xi − x )
2 2

(b) ei = yi − α xi = ( yi − y ) − b2 ( xi − x ) = yi − ( y − b2 x ) − b2 xi = eˆi

Copyright © 2018 Wiley

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

EXERCISE 4.15

(a) For all values of x the dependent variable will be positive. An x = 0 will create an undefined
value.

(b) dy dx = exp β1 + β 2 (1 x )  × ( −β 2 x −2 )

Assuming that x > 0 the slope will be positive if β2 < 0 .

(c) When β2 < 0 , as x approaches zero from above, we see that β1 − β2 (1 x ) → −∞ and
y = exp β1 − β 2 (1 x )  → 0 . If x approaches infinity β1 − β2 (1 x ) → β1 and
y = exp β1 − β 2 (1 x )  → exp ( β1 ) . In Figure XR 4-15(c) part 1 we chose β1 = 1 , so
y → e ≅ 2.7182818 .

See Figure XR 4-15(c) part 2. If β2 > 0 , as x approaches zero from above, we see that
β1 + β2 (1 x ) → ∞  1 + β 2 (1 x )  → ∞ .
and y = exp β If x approaches infinity
β1 + β2 (1 x ) → β1 and y = exp β1 + β 2 (1 x )  → exp ( β1 ) . In Figure XR 4-15(c) part 2 we
chose β1 = 1 , so y → e ≅ 2.7182818 .

(d) See Figure XR 4-15(d).

The slopes at the x-values 0.5, 1.0, 1.5, 2.0, 2.5, 3.0, 3.5, 4.0 are

Copyright © 2018 Wiley

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

x dy/dx
.5 .0053674
1 .07326256
1.5 .12352614
2 .13533528
2.5 .12921377
3 .11715428
3.5 .10413275
4 .09196986

We see that the slope increases but then begins to decrease. So, the answer is both.

 1 + β 2 (1 x )  × ( −β 2 x ) . The second derivative, uses

The first derivative is dy dx = exp β −2
(e)
Derivative Rule 6. To solve for the point where the second derivative equals zero we set the
second term to zero and solve for x:
−β2
x=
2
EXERCISE 4.17

(a) A possible match is the linear-log model, shown in Figure 4.5(f).


RATE _ UK = −278.834 + 50.729ln( SPEND _ UK ) R 2 = 0.9276
(b)
(se) (20.4216) (3.4378)

UK linear-log
XR 4.17(b)
35
30
25
20
15
10

300 350 400 450 500

per capita consumers' expenditures (1970 pounds)

total ownership rate, per 100 Fitted values

In this case the log-linear model fits the data well, except at the very low and very high ends.

(c) If SPEND_UK = 1, then ln(SPEND_UK) = 0, then E ( RATE _ UK | SPEND _ UK = 1) = β1

Copyright © 2018 Wiley

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

(d)

RATE _ UK = −33.0221 +12.296 ln( SPEND _ UK − 280) R 2 = 0.9840
(se) (1.7274) (0.3805)

UK linear-log 2
XR 4.17(d)

30
25
20
15
10

300 350 400 450 500

per capita consumers' expenditures (1970 pounds)

total ownership rate, per 100 Fitted values

The adjustment improves the fit in the lower and upper ends.

(e)

ln ( RATE _ UK ) = 5.72032 − 1002.266 (1 SPEND _ UK ) R 2 = 0.8642
(se) (0 .2574) ( 96.3413)

UK log-reciprocal
XR 4.17(e)
40
30
20
10

300 350 400 450 500

per capita consumers' expenditures (1970 pounds)

total ownership rate, per 100 ratelr_uk

For these data the log-reciprocal model is almost a straight line, missing the curvature of the
data.

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 7

(f) The problem is that in the reciprocal the value of the explanatory variable x, here
SPEND _ UK , becomes large, which makes 1 / SPEND _ UK very small.


ln ( RATE _ UK ) = 3.6684 − 48.3067 (1 [ SPEND _ UK − 280]) R 2 = 0.9818
(g)
(se) (0.0229) (1.5957)
UK log-reciprocal 2
XR 4.17(g)

40
30
20
10

300 350 400 450 500

per capita consumers' expenditures (1970 pounds)

total ownership rate, per 100 ratelr_uk2

ratelr_uk

The solid fitted line if from the modified model. For comparison we plot the dashed fitted
values from part (e).

(h)

(i) The data for Ireland is shown below.

Ireland data line plot
XR 4.17(h)-i
20
total ownership rate, per 100
5 10
0 15

250 300 350 400

per capita consumers' expenditures (1968 pounds)

(ii) The linear log model for Ireland is


RATE _ IR = −181.762 + 33.768ln( SPEND _ IR) R 2 = 0.9326
(se) (16.412) (2.871)

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 8

Ireland linear-log
XR 4.17(h)-ii

20
15
10
5
0
250 300 350 400
per capita consumers' expenditures (1968 pounds)

total ownership rate, per 100 Fitted values

(iii) The modified linear-log model is


RATE _ IR = −14.4554 +6.4279 ln( SPEND _ IR − 240) R 2 = 0.9845
(se) ( 1.0317) (0.2548)

Ireland linear-log 2
XR 4.17(h)-iii
20
15
10
5
0

250 300 350 400

per capita consumers' expenditures (1968 pounds)

total ownership rate, per 100 Fitted values

(iv) The log-reciprocal model for Ireland is


ln ( RATE _ IR ) = 5.9641 − 1096.784 (1 SPEND _ IR ) R 2 = 0.8190
(se) (0.5448 ) ( 163.0473)

Ireland log-reciprocal
XR 4.17(h)-iv
20
15
10
5

250 300 350 400

per capita consumers' expenditures (1968 pounds)

total ownership rate, per 100 ratelr_ir

(v) The modified log reciprocal model for Ireland is

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 9


ln ( RATE _ IR ) = 2.9895 − 29.0909 (1 [ SPEND _ IR − 240]) R 2 = 0.9824
(se) (0.0345) (1.2325)
Ireland log-reciprocal 2
XR 4.17(h)-v

20
0

250 300 350 400

per capita consumers' expenditures (1968 pounds)

total ownership rate, per 100 ratelr_ir2

ratelr_ir

EXERCISE 4.19

(a) RYIELD can be interpreted as the number of hectares needed to produce one tonne of wheat.

(b) In the figure RMULL is the RYIELD series for Mullewa and RNORTH is the RYIELD series
for Northampton. In both shires amount of land required to produce show a spike in 1963
and for Mullewa again in 1976-1977 and 1979.

There is an outlier in 1963 in the two shires, implying that a greater number of hectares
was needed to produce one tonne of wheat than in any other year. There were similar but
less extreme outliers in Mullewa in 1976, 1977 and 1979. Wheat production in Western
Australia is highly dependent on rainfall, and so one would suspect that rainfall was low in
the above years. A check of rainfall data at [Link] reveals
that rainfall was lower than usual in 1976 and 1977, but higher than normal in 1963. Thus,
it is difficult to assess why 1963 was a bad year; excess rainfall may have caused rust or
other disease problems during the growing season, or rain at harvest time may have led to
a deterioration in wheat quality.

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 10

(c) The estimated equations are given in Table XR4-19c

Table XR4-19c RYIELD estimations
(1) (2)
MULLEWA NORTHMAPTON
C 1.4552 1.3934
(11.1903) (12.8220)
TIME -0.0121 -0.0169
(-2.6287) (-4.3871)
N 48 48

t statistics in parentheses

(d) In each case the null hypothesis is rejected indicating that the required number of hectares
is decreasing over time.

(e) The threshold for leverage is 2(2 / N ) = 0.08333333 . For studentized residuals the threshold
is 2. For DFBETAS it is 2 N = 0.28867513 and for DFFITS it is 2 K / N = 0.40824829
. For Mullewa the values exceeding thresholds are given in Table XR 4-19(e). In the table
RESID is the OLS residual, and STU_RESID is the studentized residual.

Table XR 4-19 (e) MULLEWA MEASURES

YEAR RYIELD LEVER RESID STU_RESID DFBETAS DFFITS
1952 2.038321 .0710124 .6195211 1.467464 -.3410547 .4057231
1963 2.522068 .0328014 1.236875 3.08792 -.3434951 .5686624
1976 2.198769 .0215118 1.071473 2.589605 .0681897 .3839671
1977 2.351835 .0221631 1.236684 3.066973 .1131014 .4617343
1979 2.222222 .0241171 1.131364 2.762376 .1602395 .4342566

For Northampton only 1963 produced unusual results.

Table XR 4-19 (e) NORTHHAMPTON MEASURES

YEAR RYIELD LEVER RESID STU_RESID DFBETAS DFFITS
1963 3.306878 .0328014 2.150595 11.8371 -1.31674 2.179886

(f) The year 1963 had three measures above the usual thresholds. Other years did not have as
many, so we choose 1963 as the most unusual year.

(g) Omitting 1963

Mullewa

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 11

 = 1.3929 − 0.0107TIME
RYIELD
(t) (11.499) ( − 2.5033)
Northampton
 = 1.2850 − 0.0144TIME
RYIELD
(t) (23.389) ( −7.4383)

EXERCISE 4.21

(a) Malwai is located in southeast Africa and is bordered by Zambia to the northwest, Tanzania
to the northeast and Mozambique to the east, south and west. See, for example,
[Link] As of April 2017, the exchange rate was 1 Malawian
Kwacha to 0.0014 US$. The 2015 population is just over 17 million according to the world
bank, [Link] The main industry is agriculture.

(b) The fitted model is

 = 0.5179241 − 0.0726372ln (TOTEXP ) R 2 = 0.1180
PFOOD
(se) ( 0.0112454 )( 0.0057363)

We estimate β2 < 0 , so that as total expenditure rises the share devoted to food declines. A
95% interval estimate of β2 is

b2 ± t( 0.975, N − 2)se ( b2 ) = −0.0726372 ± 1.96 ( 0.0057363) = [ −0.0838916, −0.0613828]

The interval estimate is relatively narrow due to the small standard error of the coefficient
estimate. We have estimated β2 relatively precisely.

(c) The 5th percentile of TOTEXP is 1.5 and the 75th percentile is 10.
εˆ 5 = 0.8512972, se ( εˆ 5 ) = 0.0095097

εˆ 75 = 0.7928621, se ( εˆ 75 ) = 0.0184522

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 12

(d)

.6
8

.4
6

.2
Residuals
Percent
4

0 -.2
2

-.4
0

-.4 -.2 0 .2 .4 .6 -2 0 2 4 6
Residuals ltotexp

The histogram shows a not quite bell-shaped distribution. The residual plot shows no strong
“spray” or other pattern.

The values for the skewness and kurtosis are approximately 0.261 and 2.519, and the
calculated JB test statistic value is 25.192. Thus, we reject the null hypothesis that the
regression errors are normally distributed at the 1% level.

(e) The estimated log-log model is


ln ( FOOD ) = −0.647313 + 0.7498553ln (TOTEXP ) R 2 = 0.5785
(se) (0.036251) (.0184918)

The log-log model is a constant elasticity function.

(f)
15

1
0
10

Residuals
Percent

-1
5

-2
-3
0

-3 -2 -1 0 1 -2 0 2 4 6
Residuals ltotexp

The value of the Jarque-Bera statistic is 298.16 and far exceeds χ(20.99,2) = 9.210 . We reject
the normality of the regression errors.

(g) The estimated model is

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 13

 = −0.7754008 + 2.017512 ln (TOTEXP ) R 2 = 0.4426

FOOD
(se) (0.1282382) (0.0654148)

εˆ 50 = 0.7373919, se ( εˆ 50 ) = 0.0286551

εˆ 75 = 0.5213085 , se ( εˆ 75 ) = 0.0143217

(h)
15

30
20
10

Residuals
Percent

10
5

0
-10
0

-3 -2 -1 0 1 -2 0 2 4 6
Residuals ltotexp

The estimated skewness is 3.422 and the estimated kurtosis is 34.168. The value of the
Jarque-Bera statistic is 50913.77 which is greater than the critical value 9.210. We reject the
normality of the model random errors.

(i) The correlations with FOOD for the three models are 0.7153, 0.7162 and 0.6653,
respectively. The first of the models seems the best choice.

EXERCISE 4.23

(a) The estimates are given in Table XR 4.23

Table XR 4.23
i ii iii iv v vi
Telephone Clothes Fuel
C -0.0160 -3.2731 0.0490 -1.6186 0.0666 -2.8179
(0.0031) (0.2443) (0.0071) (0.0975) (0.0033) (0.0523)

LTOTEXP 0.0186 1.1042 0.0111 0.6918 -0.0063 0.8046

(0.0016) (0.0930) (0.0036) (0.0448) (0.0017) (0.0265)
N 1200 245 1200 556 1200 1159
R2 0.105 0.367 0.008 0.301 0.012 0.443
ε at p25 3.5634 1.1042 1.1762 0.6918 0.8928 0.8046
ε at p75 1.6945 1.1042 1.1487 0.6918 0.8792 0.8046
Standard errors in parentheses

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 14

In model (i), we estimate that a 1% increase in total expenditure leads to about a 0.000186
increase in the proportion of expenditures devoted to telephone services. In model (iii) we
estimate that a 1% increase in total expenditure leads to about a 0.000111 increase in the
proportion of expenditures devoted to clothing. And in model (v) we estimate that a 1%
increase in total expenditure leads to about a -0.000063 decrease in the proportion of
expenditures devoted to fuel. In models (ii), (iv) and (vi) the coefficients are elasticities
which are discussed in the next part of the solution.

(b) In the log-log models, (ii), (iv) and (vi), the estimated coefficients are constant elasticities;
they do not vary. For the telephone expenditures the elasticity is 1.1042, suggesting that it
is a luxury item. For the clothing expenditures the elasticity is 0.6918, suggesting that it is
a necessity item. For the fuel expenditures the elasticity is 0.8046, suggesting that it is a
necessity item.

(c) The variable TOTEXP is “Total household expenditure last month, in thousands of
Malawian Kwacha”. The median expenditure is 5.7 and the mean 8.5. The 25th percentile is
3.5 and the 75th percentile is 10. As we can see the elasticities vary quite a bit across total
expenditure levels. For telephone services the elasticity at the 25th percentile is 3.56, and at
the 75th percentile it is 1.69. In both cases we would classify telephone services as a luxury,
but the responsiveness of household’s budget share is quite different. For clothing and fuel
there are slight differences across the percentiles, but not very much. And the elasticities are
quite similar to the constant elasticity estimated by the log-log model, for fuel.

EXERCISE 4.25

(a)
Table XR4.25 Solutions
Model (1) Model (2) (3)
ln(PRICE) ln(PRICE) PRICE
C 4.3939 (0.0433) 2.0497 (0.1580) -115.4236 (13.0882)
SQFT 0.0360 (0.0015) 13.4029 (0.4492)
ln(SQFT) 1.0248 (0.0484)
N 500 500 500
R2 0.542 0.474 0.641
Rg2 0.662 0.645 0.641
JB 26.6790 14.2787 221.1115
JB p-value 0.0000 0.0008 0.0000
σ̂ 0.34001 0.36432 102.8
Standard errors in parentheses

(b) We estimate a 1% increase in SQFT to increase expected priced by 1.02%. At the mean
values, we estimate that an added 100 square feet of interior space increases expected price
by $9,400.

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 15

(c) The linear model results are in Table XR4.25, Model (3). The R 2 = 0.641 , or 64.1% of the
variation in PRICE about its mean is explained by the model. The generalized R 2 for the
log-linear model is largest, indicating that it fits the data better than the other two models,
at least based on this measure.

(d) In each case we reject normality of the regression errors at any usual level of significance.

(e) It is fair to say the logarithmic models have less clear evidence about the violation of the
homoskedasticity assumption.

(f) The predictions are

Linear model: $246,456

Log-linear model “Natural predictor”: $214,234

Log-linear model “Corrected predictor”: $226,982

Log-log model “Natural predictor”: $227,539

Log-log model “Corrected predictor”: $243,151

(g) The 95% prediction intervals are

Linear model: $44,277 to $448,634

Log-linear model: $109,769 to $418,117

Log-log model: $111,141 to $465,841

In each case the intervals are quite wide and not too informative. House prices depend on
much more than their size.

(h) There is little difference between the log-log and log-linear model results. The log-linear
model has a slightly higher generalized R 2 but the log-log model has a slightly smaller
Jarque –Bera Statistic. The slopes and elasticities are similar. At this point either model is
preferable to the linear relationship.

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 16

Figure XR 4.25

Linear Model Residuals Linear Model Residuals

20

400 200
15

Residuals
Percent
10

0 -200
5

-400
0

-400 -200 0 200 400 0 20 40 60 80 100

Residuals total interior square feet of house (100s sqft)

Log-Linear Model Residuals Log-Linear Model Residuals

15

1
.5
10

0
Residuals
Percent

-.5
5

-1
-1.5
0

-1.5 -1 -.5 0 .5 1 0 20 40 60 80 100

Residuals total interior square feet of house (100s sqft)

Log-Log Model Residuals Log-Log Model Residuals

15

1
.5
10

Residuals
Percent

-.5 0
5

-1
-1.5
0

-1.5 -1 -.5 0 .5 1 0 20 40 60 80 100

Residuals total interior square feet of house (100s sqft)

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 17

EXERCISE 4.27

(a) The summary statistics for each partition are shown below. We observe that the mean and
median wage of white males is highest, followed by white female, black male and black
female. Female wages have a higher standard deviation than males: the standard deviation
is lowest for black male, followed by black female, white male and white female.
White male Wage
N mean Std. Dev. CV min p50 max
4740 24.9481 15.2378 61.07801 2.5 21 221.1

White female Wage

N mean Std. Dev. CV min p50 max
3674 22.06724 17.6279 79.88264 2.56 18 466

Black male Wage

N mean Std. Dev. CV min p50 max
404 19.81181 11.22209 56.64343 5.49 16.83 72.13

Black female Wage

N mean Std. Dev. CV min p50 max
473 19.67019 13.51805 68.72353 2.75 16.15 173.13

(b) The coefficient of variation is given in the part (a) tables, CV. We see that the variation in
female wages, scaled by the mean, is higher than for males. The variation for white females
is higher than the variation for black females, and the variation for white males is higher
than the variation for black males.

(c) The estimated models are in Table XR 4.27.

Table XR 4.27
(1) (2) (3) (4)
White male White female Black male Black female
C 1.7274 1.3944 1.5160 1.3998
(0.0359) (0.0452) (0.1292) (0.1270)

EDUC 0.0951 0.1058 0.0983 0.1015

(0.0025) (0.0031) (0.0093) (0.0089)
N 4740 3674 404 473
R2 0.231 0.244 0.216 0.216
Standard errors in parentheses

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 18

(d) Interval estimates are given in Table XR 4.27(d).

Table XR 4.27(d)
(1) (2) (3) (4)
White male White female Black male Black female
C 1.727 1.394 1.516 1.400
[1.6570,1.7977] [1.3057,1.4830] [1.2620,1.7700] [1.1503,1.6493]

EDUC 0.0951 0.106 0.0983 0.101

[0.0902,0.1001] [0.0998,0.1118] [0.0799,0.1166] [0.0840,0.1190]
N 4740 3674 404 473
R2 0.231 0.244 0.216 0.216
95% confidence intervals in brackets

The interval estimates for the slope coefficients all overlap. The returns to education are
similar in the sense that the range of values are not distinct. If we were to test the hypothesis
that the slope parameter β2 = c , where c is a value in another interval, we would find values
that we would not reject.

(e) For the intercept parameter, the white male and white female intervals do not overlap. The
white male and black female intervals do not overlap.

(f) The models fit similarly well for all groups, but the fit is higher for whites than blacks, and
is higher for white females than white males.

EXERCISE 4.29

(a) The summary statistics are

variable N mean p50 min max Std. Dev.
FOOD 1200 114.4431 99.8 9.63 476.67 72.6575
INCOME 1200 72.14264 65.29 10 200 41.65228

The histograms are

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 19

Both distributions are positively skewed with mean’s greater than medians. They are not
bell shaped or symmetrical. For INCOME the Jarque-Bera statistic is 148.21 and for FOOD
expenditure it is 648.65. The critical value for a test at the 5% level is 5.99. We reject the
null hypothesis of normality for each variable.

(b) The estimation results are in Table XR 4.29. The linear relation estimates are in column (1).
Table XR 4.29
(1) (2) (3)
Linear relation Log-Log relation Linear-Log relation
C 88.5665 3.7789 23.5685
(4.1082) (0.1203) (13.3696)

INCOME 0.3587
(0.0493)

ln(INCOME) 0.1863 22.1874

(0.0290) (3.2253)
N 1200 1200 1200
R2 0.042 0.033 0.038
Standard errors in parentheses

The scatter plot and fitted line are in Figure XR 4.29(b).

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 20

The 95% interval estimate of the slope is [0.2619, 0.4555].

(c) The least squares residuals are plotted in Figure XR 4.29(c). The positive skew at each
income is clear. There is not a clear “spray” pattern except at high incomes. The residual
histogram shows the skewness. The Jarque-Bera statistic is 624.186, which is far greater
than the 5% critical value 5.99.

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 21

(d) The estimated elasticity and intervals are

INCOME b1 + b2 INCOME ε̂ se ( εˆ ) LB UB

19 95.3815 0.0715 0.009825 0.0522 0.09073

65 111.8811 0.2084 0.02865 0.1522 0.2646
160 145.9564 0.3932 0.0541 0.2871 0.4993

(e) The log-log model estimation results are in column (2) of Table XR 4.29. The data and fitted
relationship are in Figure XR 2.29(e)

The generalized R2 is 0.03965 which is slightly smaller than the R2 from the linear model.

(f) The 95% interval estimate of the elasticity is [0.1293, 0.2433].

(g) The residual scatter from the log-log model is shown in Figure XR 4.29(g).

The residual histogram is given in Figure XR 4.29(g).

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 22

There is a slight negative skew (Skewness = -0.3577) and Kurtosis is 3.0719. The Jarque-
Bera statistic is 25.85 which is greater than the 5% critical value 5.99. So, we reject the null
hypothesis that the log-log regression errors are normal.

(h) The estimated linear-log model is in column (3) of Table XR 4.29. The data and fitted line
are in Figure XR 4.29(h). The figure is much like that for the linear model, and not as well
defined as that for the log-log model. The R2 = 0.038, which is smaller than that of the linear
model, and smaller than the generalized R2 from the log-log model.

(i) At the given income values the estimates and intervals are:
INCOME αˆ 1 + αˆ 2 ln( INCOME ) ε̂ se ( εˆ ) LB UB

19 88.8979 0.2496 0.0363 0.1784 0.3208

65 116.1872 0.1910 0.0278 0.1365 0.2454
160 136.1733 0.1629 0.0237 0.1165 0.2094

(j) The residual diagrams for the linear-log model are in Figures XR 4.29(j). The residual
scatter shows positive skewness at each income level and overall. The Jarque-Bera statistic
is 628.07 which is far greater than the 5.99 critical value. We reject the normality of the
model errors. The data scatter suggests a slight “spray” pattern.

Copyright © 2018 Wiley

 Chapter 4, Exercise Answers, Principles of Econometrics, 5e 23

(k) The linear model is counter-intuitive with increasing income elasticity. The linear-log model
certainly satisfies economic reasoning, but the residual pattern is not an ideal random scatter.
The log-log model implies that the income elasticity is constant for all income levels, which
is not impossible to imagine, and the residual scatter is the most random, and the residuals
are the least non-normal, based on skewness and kurtosis. On these grounds the log-log
model seems like a good choice.

Copyright © 2018 Wiley