PRINCIPLES OF ECONOMETRICS

5TH EDITION

ANSWERS TO ODD-NUMBERED
EXERCISES IN CHAPTER 12

1
 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 2

EXERCISE 12.1
(a) Now, 1  c1 z 1  c2 z   1  c1 z  c2 z  c1c2 z 2  1   c1  c2  z  c1c2 z 2 . For this expression to be
equal to 1  1 z  2 z 2 , we require c1  c2  1 and c1c2  2 .

(b) We need to show that 1  2  1  0 implies c1  1 or c2  1 ; and that c1  1 or c2  1 implies

1  2  1  0 . From part (a),

1  2  1  c1  c2  c1c2  1
 c1 1  c2   (1  c2 )
 1  c2  c1  1

Thus, 1  2  1  0 only if c1  1 or c2  1 .

Now suppose c1  1 . Then, c2  1  c1  1  1 and c2  2 c1  2 .

Thus, 2  1  1 , and so 1  2  1  0 .

By symmetry the same result is obtained if c2  1 .

(c) The AR(2) model will be stationary if 1 c1  1 and 1 c2  1 . That is, stationarity implies c1  1
and c2  1 . Since 1  2  1  1  c2  c1  1 , if c1  1 and c2  1 , then 1  2  1  0 .

(d) Substituting for  and a1 , the equation yt     yt 1  a1yt 1  vt can be written as

yt  yt 1     1  2  1 yt 1  2  yt 1  yt 2   vt

yt  yt 1    1 yt 1  2 yt 1  yt 1  2 yt 1  2 yt 2  vt

Cancelling terms leads to the original model yt    1 yt 1  2 yt 2  vt .

The equivalence of the two equations and the results in parts (b) and (c) mean that a test for a
unit root (stationarity) can be based on the null and alternative hypotheses H 0 :   0 and
H0 :   0 .
(e) If   1   2     p  1  0 , then z  1 is a solution to the equation

1  1 z  2 z 2     p z p  0

Having z  1 implies the AR(p) model has a unit root.

(f) We wish to show that
p 1
yt     yt 1   as yt  s  vt
s 1

and
yt    1 yt 1  2 yt 2     p yt  p  vt

are equivalent representations of an AR(p) process if   1   2     p  1 and a j    r  j r 1.

p 1

Making these substitutions into the first equation, we have

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 3

yt  yt 1     1  2     p  1  yt 1    2  3     p   yt 1  yt 2 

   3     p   yt 2  yt 3        p  yt  p 1  yt  p   vt

Collecting coefficients of yt 1 , yt 2 , , yt  p leads to the AR(p) model

yt    1 yt 1  2 yt 2     p yt  p  vt

EXERCISE 12.3
(a) Both W and Y fluctuate around a nonzero mean with no obvious trend upwards or downwards,
and so Dickey-Fuller test equations with intercepts and no trend terms were used for these
variables. Trend terms were included for X and Z since they are both trending upwards. A lack
of serial correlation in the errors and insignificance of coefficient estimates for the lags of
Wt , Yt , X t and Zt would have led to the omission of augmentation terms.
(b) Using a 5% level of significance, we obtain the following Dickey-Fuller test results.
For W: since   3.23 is less than the 5% critical value of 2.86, the null hypothesis of
nonstationarity is rejected, and we infer that W is stationary.
For Y: since   1.98 is greater than the 5% critical value of 2.86, the null hypothesis of
nonstationarity is not rejected, and we infer that Y is not stationary.
For X: since   3.13 is greater than the 5% critical value of 3.41, the null hypothesis of
nonstationarity is not rejected, and we infer that X is not stationary.
For Z: since   1.87 is greater than the 5% critical value of 3.41, the null hypothesis of
nonstationarity is not rejected, and we infer that Z is not stationary.
(c) Using a 5% significance level, we find   2.83 is greater than the critical value of 3.42 . Thus,
we fail to reject a null hypothesis that the errors are nonstationary, and we conclude X and Z are
not cointegrated.
(d) The value   13.76 is less than the 5% critical value 2.86 . Thus, we reject the null hypothesis
that Z is nonstationary. Because Z is nonstationary, but Z is stationary, we conclude that Z
is integrated of order 1.

EXERCISE 12.5
(a) For the random walk yt  yt 1  vt , we have yˆT 1  10 and yˆT  2  10 .

(b) For the random walk with drift yt  5  yt 1  vt , we have yˆT 1  15 and yˆT 2  20 .

(c) For the random walk ln  yt   ln  yt 1   vt , we have yˆT 1  73.89 and yˆT  2  546 .

(d) For the deterministic trend model yt  10  0.1t  vt , we have yˆT 1  13 and yˆT 2  13.1 .

(e) For the ARDL model yt  6  0.6 yt 1  0.3xt  0.1xt 1  vt , we have yˆT 1  14 and yˆT 2  16.4 .

(f) For the error correction model yt  0.4  yt 1  15  xt 1   0.3xt  vt , we have yˆT 1  14 and
yˆT 2  16.4 . When x  5 , the long run equilibrium value for y is y  20 .

(g) In the first difference model yt  0.6yt 1  0.3xt  0.1xt 1  vt , we have yˆT 1  8.7 and
yˆT 2  7.92

Copyright © 2018 Wiley

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

EXERCISE 12.7
Unemployment
The plot and unit root test results for the unemployment series follow. There is no obvious trend in the
series and so a unit root test equation with a constant and no trend is used. Using the Schwarz criterion
to choose the number of augmentation terms leads to the inclusion of U t 1 in the test equation which
becomes
U t    U t 1  a1U t 1  vt
The null and alternative hypotheses are
H0 :   0 H1 :   0

Because the p-value = 0.0003, at a 5% significance level we reject H 0 :   0 in favour of H1 :   0 .

We conclude that U is stationary and of order I(0).
U
11

10

2
50 55 60 65 70 75 80 85 90 95 00 05 10 15

Copyright © 2018 Wiley

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

GDP growth rate

The plot and unit root test results for the GDP growth rate follow. There is no obvious trend in the series
and so a unit root test equation with a constant and no trend is used. Using the Schwarz criterion results
in no augmentation terms. The test equation is
Gt    Gt 1  vt

The null and alternative hypotheses are

H0 :   0 H1 :   0

Because the p-value = 0.0000, at a 5% significance level we reject H 0 :   0 in favour of H1 :   0 .

We conclude that G is stationary and of order I(0).
G
8

-2

-4
50 55 60 65 70 75 80 85 90 95 00 05 10 15

Inflation
The plot and unit root test results for the inflation series follow. There is no obvious trend in the series
and so a unit root test equation with a constant and no trend is used. Using the Schwarz criterion to
choose the number of augmentation terms leads to the inclusion of INFt 1 and INFt 2 in the test
equation which becomes
INFt    INFt 1  a1INFt 1  a2 INFt 2  vt

Copyright © 2018 Wiley

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

The null and alternative hypotheses are

H0 :   0 H1 :   0

Because the p-value = 0.0017, at a 5% significance level we reject H 0 :   0 in favour of H1 :   0 .

We conclude that INF is stationary and of order I(0).
INF
5

-1

-2

-3
50 55 60 65 70 75 80 85 90 95 00 05 10 15

EXERCISE 12.9
Unemployment
The plot and unit root test results for the unemployment series follow. There is an obvious downward
trend in the series from 1992 to 2008, but a less obvious trend over the complete sample period. Because
1992-2008 makes up a large part of the sample period, we include both a constant and a trend in the
unit root test equation. Using the Schwarz criterion to choose the number of augmentation terms, the
test equation becomes
U t    t  U t 1  a1U t 1  a2 U t 2  vt

The null and alternative hypotheses are H 0 :   0 and H1 :   0 . The null hypothesis is not rejected at
a 5% significance level, the p-value being 0.1009. Because H 0 :   0 cannot be rejected, we treat U as
nonstationary.

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 7

U
12

11

10

4
88 90 92 94 96 98 00 02 04 06 08 10 12 14

Concluding U is nonstationary means a unit root test on its first difference is needed to assess its order
of integration. The following output from this unit root test reveals a p-value of 0.0087, implying the
first difference is stationary and hence U is of order I(1).

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 8

Inflation
The plot and unit root test results for the inflation series follow. Although there is sharp drop in inflation
around 1990, there is no obvious trend in the series and so a unit root test equation with a constant and
no trend is used. The Schwarz criterion does not suggest inclusion of augmentation terms, and so the
test equation becomes
INFt    INFt 1  vt
The null and alternative hypotheses are
H0 :   0 H1 :   0

Because the p-value = 0.0000, at a 5% significance level we reject H 0 :   0 in favour of H1 :   0 .

We conclude that INF is stationary and of order I(0).
INF
3.0

2.5

2.0

1.5

1.0

0.5

0.0

-0.5
88 90 92 94 96 98 00 02 04 06 08 10 12 14

EXERCISE 12.11
(a) The results from estimating the equation XCHINAt  1  2 BEKLYt  et follow. They suggest a
very strong relationship between house prices in Beckley, West Virginia, and the value of
Australian exports to China. However, since there is no reason why these two variables should
be related, there is a need to ask whether the relationship is spurious.

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 9

(b) The plots of the three series given below show an increasing trend, with BEKLY and
ln( XCHINA) increasing approximately linearly, and XCHINA increasing at an increasing rate.
They suggest that the strong relationship estimated in part (a) is attributable to the upward
trends of the two variables and the omission of a trend from the equation.

BEKLY
160

140

120

100

80

60
88 90 92 94 96 98 00 02 04 06 08 10 12 14

XCHINA
10,000

8,000

6,000

4,000

2,000

0
88 90 92 94 96 98 00 02 04 06 08 10 12 14

LN_XCHINA
10

4
88 90 92 94 96 98 00 02 04 06 08 10 12 14

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 10

(c) Estimates for the equation ln  XCHINAt   1   t  2 BEKLYt  et follow. Since ln( XCHINA)
increases linearly, and XCHINA increases at an increasing rate, the linear trend on the right-hand
side of the equation is likely to be better at capturing the trend in ln( XCHINA) than that in
XCHINA.
Despite controlling for the trends by including a trend in the equation, house prices in Beckley
still seem to have a significant impact on Australian exports to China. We need to investigate
further why the equation might be spurious.

(d) The results from unit root tests on ln( XCHINA) and BEKLY follow. Given the trending behaviour
of both variables, a constant and a trend term have been included in the test equations. Two
augmentation terms were selected by the Schwarz criterion for the ln( XCHINA) equation and
nine were selected for the BEKLY equation. The test p-value of 0.0012 for ln( XCHINA) suggests
this series is trend stationary. However, for BEKLY, the test p-value is 0.7332. A null hypothesis
of a unit root cannot be rejected. Thus, we can treat BEKLY as nonstationary which means the
relationship seemingly uncovered in part (c) is likely to be spurious.

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 11

EXERCISE 12.13
(a) The plots of LCONS and INC displayed below exhibit increasing trends as well as a strong
seasonal component.
LCONS
5.4

5.2

5.0

4.8

4.6

4.4

4.2

4.0
1975 1980 1985 1990 1995 2000 2005 2010

INC
360

320

280

240

200

160

120

80
1975 1980 1985 1990 1995 2000 2005 2010

Copyright © 2018 Wiley

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

(b) Estimates for the trend equations LCONSt  a1  a2 t  u1t and INCt  c1  c2 t  u2t follow. The
residuals which represent the detrended series were stored as LCONS_T and INC_T.

(c) The plots of LCONS_T and INC_T are displayed below. The strong seasonal component
continues to be evident. Also, there is a predominance of negative values in the middle of the
sample, and positive values at the ends of the sample, particularly for INC_T. This outcome
suggests a linear trend may have been inadequate. Perhaps you would like to check out the
implications of trying a quadratic trend.
LCONS_T
.12

.08

.04

.00

‐.04

‐.08

‐.12
1975 1980 1985 1990 1995 2000 2005 2010

INC_T
60
50
40
30
20
10
0
‐10
‐20
‐30
1975 1980 1985 1990 1995 2000 2005 2010

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 13

(d) Estimates for the equations LCONSt  0 t  1D1t  2 D2t  3 D3t  4 D4t  ut and
INCt  0 t  1D1t  2 D2t  3 D3t  4 D4t  ut follow. The residuals which represent the
“seasonally-adjusted detrended” series were stored as LCONS_STAR and INC_STAR.

(e) Plots of LCONS  and INC  continue to display a predominance of negative values in the middle
of the sample, and positive values at the ends of the sample. The seasonal component has been
removed, but there is strong evidence of autocorrelation.
LCONS_STAR
.08

.06

.04

.02

.00

‐.02

‐.04

‐.06
1975 1980 1985 1990 1995 2000 2005 2010

INC_STAR
40

30

20

10

‐10

‐20

‐30
1975 1980 1985 1990 1995 2000 2005 2010

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 14

(f) The results for testing whether LCONS  is stationary or first-difference stationary follow.
Because LCONS  has been computed as least squares residuals, it will have a zero mean. Also,
the trend has been removed. Thus, a test equation without a trend or constant is appropriate. The
Schwarz criterion suggests no augmentation terms are necessary. Using the 5% critical value of
–3.41, found in the third row of Table 12.2, we conclude that LCONS  , with test value
  2.614 , is not stationary. However, its first difference, with test value   14.39 , is
stationary. The critical value –3.41 is relevant for an equation with a trend, and hence takes into
account the detrending used to create LCONS  . However, it does not allow for removal of the
seasonal component and hence will not be entirely suitable.

For the same reasons, test equations without trend or constant are chosen for INC  . Five
augmentation terms are chosen by the Schwarz criterion. Using the 5% critical value of –3.41,
found in the third row of Table 12.2, we conclude that INC  , with test value   0.454 , is not
stationary. However, its first difference, with test value   5.954 , is stationary. Again, the
critical value –3.41 will not be entirely suitable because it does not reflect removal of the seasonal
component.

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 15

(g) As can be seen from the output below, the estimate for  is the same from both equations,
LCONSt   t  1D1t  2 D2t  3 D3t  4 D4t   INCt  et , and LCONSt   INCt  et . The
equivalence follows from the Frisch-Waugh-Lovell theorem studied in Chapter 5. The standard
errors are different because the degrees of freedom used to estimate the error variance are
different. The R 2 ’s are different because of different variation in the different dependent
variables.

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 16

(h) The same set of residuals is obtained from both equations in part (g). Testing these residuals
EHAT_H for a unit root, we get the output below. Comparing the test value   5.088 with the
5% critical value –3.42, found in Table 12.4 of POE5, we conclude that the residuals are
stationary and hence that LCONSt and INCt are cointegrated.

(i) Estimating the error correction model LCONSt  eˆt  0 INCt  vt yields

The correlogram of residuals from this model reveals a significant autocorrelation at lag 4 which
is removed by estimating the model LCONSt  eˆt  0 INCt  4 LCONSt4  vt . The
results from estimating this model follow.

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 17

EXERCISE 12.15
(a) Because INFN is trending neither upwards nor downwards, to test whether it is stationary, we
use a test equation with a constant and no trend. A plot of INFN and the unit root test outcome
are given below. The Schwarz criterion chooses 15 augmentation terms. The test p-value is
0.0378. At a 5% level of significance, we reject a null hypothesis of a unit root and conclude that
INFN is stationary. At a 1% level of significance, we conclude INFN is nonstationary.

INFN
16

12

-4
55 60 65 70 75 80 85 90 95 00 05 10 15

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 18

(b) If, in part (a), we concluded that INFN is stationary, then we can say immediately that INFN is
of order I(0). If we used a 1% significance level in part (a), and concluded that INFN is
nonstationary, then we need to test whether the first differences of INFN are stationary. The p-
value for this purpose, given in the output below, is 0.0000. Thus, in this case we conclude that
the first differences are stationary and that INFN is of order I(1).

(c) The model chosen for forecasting is likely to depend on whether INFN was judged to be
stationary or nonstationary. Estimates for models for both cases are given on the next page.

INFN stationary
16
The autoregressive model INFN t     r INFNt r  et was estimated. Sixteen lags were
r 1

included in line with the Schwarz criterion. The forecast for 2017M1 from this model is
16

2017 M 1      r INFN 2017 M 1 r  1.9079
INFN ˆ ˆ
r 1

INFN nonstationary
15
The autoregressive model INFN t     r INFN t r  et was estimated. Fifteen lags were
r 1

included in line with the Schwarz criterion. The forecast for 2017M1 from this model is
15
2017 M 1  INFN 2016 M 12      r  INFN 2017 M 1 r  INFN 2017 M 1 r 1   1.8760

INFN ˆ ˆ
r 1

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 19

EXERCISE 12.17
(a) The estimated model is

Eˆ YIELDt | RAIN   1.4200  0.03024t  1.0713RAINt  0.12727 RAINt2

(se) (0.7829) (0.00332) (0.3825) (0.04559)
( p-values) (0.0765) (0.0000) (0.0075) (0.0077)

All coefficient estimates are significantly different from zero at a 5% level of significance, except
for the intercept which is significant at a 10% level. Yield is trending upwards. The response to
rainfall is positive initially, but with diminishing returns, and eventually becomes negative when
rainfall exceeds 4.2dm.
(b) The residuals are plotted in Figure 12.17(a) and the correlogram is displayed in Figure 12.17(b).
The residuals are positively correlated. In their plot, positive residuals tend to follow positive
residuals and negative residuals tend to follow negative ones. In the correlogram the first-order
autocorrelation is significantly different from zero.

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 20

.8

.6

.4

.2

.0

-.2

-.4

-.6

-.8
1950 1955 1960 1965 1970 1975 1980 1985 1990 1995

Figure 12.17(a) Residuals from linear trend model

Figure 12.17(b) Correlogram from linear trend model

(c) The estimated model is

Eˆ YIELDt | RAIN   0.2669  0.01902t  0.001005 t 2  0.7225RAINt  0.08821RAINt2

(se) (0.6990) (0.01120) (0.000221) (0.3273) (0.03888)
 p-value  (0.7045) (0.0967) (0.0000) (0.0327) (0.0284)

The estimated coefficients for the squared trend term and both RAIN terms are significant at a
5% level. The coefficient estimate for the linear trend term is significant at a 10% level. The
intercept estimate is not significant, but there is no reason to discard it. The quadratic trend is
such that yield declines until year 9 (1958) after which it increases at an increasing rate. Rainfall
has a positive impact on yield, at a diminishing rate. At rainfalls greater than 4.1dm extra rain
has a negative impact.
(d) The residuals are plotted in Figure 12.17(c) and the correlogram is displayed in Figure 12.17(d).
.6

.4

.2

.0

-.2

-.4

-.6
1950 1955 1960 1965 1970 1975 1980 1985 1990 1995

Figure 12.17(c) Residuals from quadratic trend model

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 21

Figure 12.17(b) Correlogram from quadratic trend model

(e) To test whether these variables are trend stationary after subtracting out a quadratic trend, we fit
regressions of the form
yt  1  2t  3t 2  ut

and then test the OLS residuals uˆt  yt  a1  a2t  a3t 2 for stationarity. In all 3 cases (YIELD,
RAIN and RAIN 2 ), the Dickey-Fuller test equation was of the form
uˆt  uˆt 1  vt
Intercept and trend terms, and augmentation terms, were not necessary. The  values were
For YIELD:   5.386
For RAIN:   5.931
For RAIN 2 :   5.719
Since we have eliminated a quadratic trend, not a linear trend, the critical values in the last row
of Table 12.2 are not necessarily accurate. Nevertheless, the calculated  values of less than 5
are sufficiently less than the linear trend value of 3.41 for us to safely conclude that all 3
variables are trend stationary after subtracting out the quadratic trend.

EXERCISE 12.19
Output for the error correction model, estimated with the lagged residuals from the cointegrating
relationship, is given below, followed by a table that compares the two sets of estimates of like
parameters.

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 22

Current Estimates
Parameter
estimates (12.32)
 -0.046369 0.046381
1 0.272374 0.272365
2 -0.242099 -0.242108
0 0.341750 0.341783
1 -0.105296 -0.105320
2 0.099082 0.099059
3 -0.065954 -0.065975
4 0.056073 0.056041

The two sets of estimates are virtually identical, except for a change in sign for ̂ . The change in sign
occurs because, in (12.32), the equation was written with a negative sign in front of  .

EXERCISE 12.21
(a) Results from the three Dickey-Fuller tests are:
(1) Dickey Fuller test (no intercept and no trend term)

CSI t  0.001CSI t 1
(tau ) ( 0.299)
Since the tau value (0.299) is greater than the 5% critical value of 1.94, the null
hypothesis of nonstationarity is not rejected. The variable CSI is not stationary.
(2) Dickey Fuller test (intercept but no trend term)

CSI t  0.051CSI t 1  4.500
(tau ) (3.001)
Since the tau value (3.001) is less than the 5% critical value of 2.86, the null hypothesis
of nonstationarity is rejected. The variable CSI is stationary.
(3) Dickey Fuller test (intercept and trend term)

CSI t  0.068CSI t 1  5.309  0.004t
(tau ) (3.483)
Since the tau value (3.483) is less than the 5% critical value of 3.41, the null hypothesis
of nonstationarity is rejected. The variable CSI is stationary.
The result of the Dickey-Fuller test without an intercept term is not consistent with the other two
results. This is because it assumes that when the alternative hypothesis of stationarity is true, the
series has a zero mean. This assumption is not correct (see graph in part (b)).

(b) The graph in Figure xr12.21 suggests that we should use the Dickey-Fuller test with an intercept
term.

(c) Since CSI is stationary, it suggests that the effect of news is temporary; hence consumers
“remember” and “retain” news information for only a short time.

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 23

Figure xr12.21 Plot of time series CSI

EXERCISE 12.23
Plots from the data in inter2.dat are shown below. The graphs show the level Yt , the first difference
DY  Yt  Yt  Yt 1 , and the second difference D2Y   2Yt  Yt  Yt 1 .

Figure xr12.23 Plots of Y and its first and second differences

The Dickey-Fuller unit root tests are shown below.


Y t  0.001Yt 1  0.001  0.00006t  0.991Yt 1
(tau ) (3.371)


 ( Yt )  0.011Yt 1  0.001 
 ( 2Yt )  0.987  2Yt 1
(tau ) (1.088) (tau ) (16.940)

Since Yt clearly has a trend, its Dickey-Fuller test equation includes an intercept and a trend. The tau
value (.371) is greater than the 5% critical value of 3.41; thus, the null hypothesis of nonstationarity
is not rejected. The variable Yt is not stationary.

Since DYt is fluctuating around a constant, its Dickey-Fuller test equation includes an intercept.
Because the tau value (.088) is greater than the 5% critical value of .86, the null hypothesis of
nonstationarity is not rejected. The variable DYt is not stationary.

Since D2Yt is fluctuating around zero, its Dickey-Fuller test equation does not include an intercept.
Because the tau value (.940) is less than the 5% critical value of .94, the null hypothesis of
nonstationarity is rejected. The variable D2Yt is stationary.

Thus, Yt needs to be differenced twice to achieve stationarity; thus Yt is integrated of order 2.

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 24

EXERCISE 12.25
(a) The plots of the two series appear in the following graph with GDP growth measured on the left
axis and sunspot activity measured on the right axis. There is some evidence that troughs in the
business cycle occur when sunspot activity is greatest, but it is difficult to conclude that the
business cycle tends to follow sunspot activity.
280
240
200
160
16 120
80
12 40
8 0

4
0
‐4
50 55 60 65 70 75 80 85 90 95 00 05 10

bcycles sunspots

(b) Output for unit root tests for each of the series follows. In both cases the p-values for the tests are
0.0000, indicating that both series are stationary.

(c) An ARDL model to test whether sunspots can be used to predict business cycles is of the form
p q
BCYCLESt     s BCYCLESt s   r SUNSPOTSt r  et
s 1 r 1

To use this model to test whether sunspots can be used to predict business cycles, we need to
select suitable values for p and q and then test the hypothesis
H 0 : 1  2    q  0

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 25

Schwarz criterion values for p  0,1, 2, ,8 and q  1, 2, ,8 are given in the following table.
The minimizing values are p  0 and q  1 . A value p  0 implies lagged values of BCYCLES
do not appear in the equation.

q 1 q2 q3 q4 q5 q6 q7 q 8

p0 1.724003 1.742675 1.778997 1.817294 1.874013 1.903025 1.963897 1.878685
p 1 1.729063 1.800063 1.848571 1.903059 1.988842 2.054519 2.100675 2.170251
p2 1.765240 1.827557 1.889745 1.928191 2.010593 2.074025 2.126620 2.209258
p3 1.816913 1.853814 1.918089 1.950025 2.034572 2.097677 2.153267 2.236409
p4 1.853126 1.891508 1.939690 2.002594 2.087666 2.156872 2.214355 2.290579
p5 1.919790 1.940930 2.000019 2.064166 2.114933 2.189382 2.238978 2.302102
p6 1.934487 1.974330 2.030969 2.093376 2.155330 2.224441 2.280530 2.356183
p7 1.974592 1.978672 2.043435 2.111826 2.178843 2.243109 2.298280 2.366466
p 8 1.915929 1.944536 2.013079 2.084007 2.154209 2.224637 2.269379 2.333242

The F-values for testing H 0 : 1  2    q  0 for the 3 smallest values of the Schwartz
criterion are given in the following table. In all cases we cannot reject the null hypothesis at a 5%
level of significance. The results do not support Jevons’ theory.

Model F-value p-value

p  0, q  1 2.529 0.1169
p  0, q  2 2.756 0.0716
p  1, q  1 1.506 0.2245

EXERCISE 12.27
(a) The two series are plotted in the figure below with XRATE on the left axis and IRON on the right
axis. They have similar trends, but it is less clear whether the fluctuations around those trends
move together.
200

160

1.1 120

1.0 80

0.9 40

0.8 0

0.7

0.6
2010 2011 2012 2013 2014 2015 2016

XRATE IRON

(b) In the equation to test for stationarity of the exchange rate, it is not clear whether a trend should
be included. Although the overall trend of the exchange rate is downwards, it trends upwards at
the beginning of the sample period, and there is no theory-based reason for it to always have a
downward trend. We therefore exclude a trend and obtain the following test results.

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 26

From the p-value of 0.8328, we cannot reject the null hypothesis of a unit root. We conclude that
the exchange rate is nonstationary. Accepting the null hypothesis of a unit root, and using the
model chosen by the Schwarz criterion, the model that best represents the relationship between
current and past exchange rates is the random walk
XRATEt  XRATEt 1  vt
If a trend term is included in the test equation, we obtain a p-value of 0.3329, which again leads
us to conclude that the exchange rate is nonstationary. In this case, the model that best represents
the relationship would be a random walk with drift
XRATEt    XRATEt 1  vt

(c) Output for testing whether the iron ore price is stationary follows. In the equation without a trend,
we obtain a p-value of 0.6672, indicating that the price is nonstationary. [Including a trend gives
a p-value of 0.0605; at the 5% level of significance we conclude the iron ore price is
nonstationary.]

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 27

(d) To investigate whether the exchange rate and the iron ore price move together, we estimate the
potential long-run cointegrating relationship XRATEt  1  2 IRONt  et and test whether the
residuals from this relationship are stationary. Estimates of the parameters of the equation
XRATEt  1  2 IRONt  et are given below, followed by the unit root test results on the
residuals. Comparing the value of the unit root test statistic   2.724 with the 5% critical value
of –3.37, found in Table 12.4 on page 583 of POE5, we cannot reject the null hypothesis of a unit
root in the residuals. We conclude therefore that the exchange rate and the iron ore price are not
cointegrated; we cannot establish that there is a long run relationship between them.

(e) Given that XRATE and IRON are both nonstationary and not cointegrated, to investigate whether
the iron price can be used to forecast the exchange rate, we estimate an ARDL model in first
differences
p q
XRATEt     s XRATEt s   r IRON t r  et
s 1 r 1

A check of the correlogram for XRATEt reveals little if any serial correlation, suggesting its
lags could be omitted from the equation, and we can focus on whether changes in the iron ore
price have a significant impact on exchange rate changes. The significance bounds are drawn at
1.96 83  0.215 .

.4

.2
correlation

.0

‐.2

‐.4

2 4 6 8 10 12 14 16 18 20 22 24
Lag

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 28

Estimating XRATEt     r 1  r IRON t r  et for q  1, 2, 3 and 4, and testing the null

q

hypothesis H 0 : 1  2    r  0 in each case gives the F-test values and corresponding p-

values that are displayed in the following table.

q F-value p-value
1 0.315 0.5763
2 0.170 0.8441
3 0.143 0.9336
4 0.123 0.9737

In all cases the null hypothesis H 0 : 1  2    r  0 cannot be rejected. We conclude that the
iron ore price cannot be used to help forecast the exchange rate.

EXERCISE 12.29
(a) The plot in Figure 12.29(a) suggests that EAF has an upward trend which is approximately linear.
Thus, it does not appear to be stationary, nor does the logistic growth curve appear to be a suitable
model for capturing changes in EAF.
.7

.6

.5
eaf

.4

.3

.2

.1
0 4 8 12 16 20 24 28 32 36 40 44 48

Figure 12.29(a) Share of U.S. steel produced by electric arc furnace technology

(b) Using a Dickey-Fuller test equation with a trend term and no augmentation terms gives a tau
value of   2.27 . Since this value is greater than the 5% critical value of 3.41 , we cannot
reject a null hypothesis that EAF has a unit root.
(c) Nonlinear least squares estimates are given in the following table.
Coefficient Std. Error t-value p-value

̂ 0.814375 0.051050 15.95247 0.0000

̂ -1.377672 0.056356 -24.44599 0.0000
̂ 0.057223 0.004304 13.29554 0.0000

The residuals are plotted in Figure 12.29(b). They are clearly autocorrelated, but it is difficult to
assess from the plot whether they are stationary. Carrying out a unit root test with no intercept
and no augmentation term, we have   2.31 and 5% critical value 1.94 . Including an
intercept term gives   2.28 , and 5% critical value 2.86 . The test outcome depends on which
test is chosen. Since we are testing the residuals from a nonlinear estimation, those residuals will

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 29

not necessarily have a zero mean. Also, although the critical values can be used as guide, they
are not necessarily the correct ones to use after a nonlinear estimation. Recognising all these
facts, we conclude that it has not been possible to establish that the residuals are stationary.
.05

.04

.03

.02

.01

.00

-.01

-.02

-.03

-.04
1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

Figure 12.29(b) Residuals from nonlinear least squares estimation

(d) Using a Dickey-Fuller test equation with an intercept and no augmentation terms gives a tau
value of   7.02 . Since this value is less than the 5% critical value of 2.86 , we reject a null
hypothesis that EAF has a unit root, and conclude that this first difference is stationary.

(e) A first differenced version of the model is

 
yt  yt 1    vt
1  exp(  t ) 1  exp    (t  1) 

Estimates from this specification are given in the following table.

Coefficient Std. Error t-value p-value

̂ 1.410410 3.323493 0.424376 0.6735

̂ -0.411822 2.667195 -0.154403 0.8780
̂ 0.032208 0.078014 0.412848 0.6818

The plot of the residuals below suggests that they are stationary and free from serial correlation.
Carrying out a unit root test with no intercept and no augmentation term, we have   7.13 and
5% critical value 1.94 . Including an intercept term gives   7.04 , and 5% critical value
2.86 . Once again, the critical values do not allow for the fact that the series is obtained from a
nonlinear estimation, but, nevertheless, the large values for  do suggest the residuals are
stationary.
.03

.02

.01

.00

-.01

-.02

-.03
1970 1975 1980 1985 1990 1995 2000 2005 2010 2015

Copyright © 2018 Wiley

 Chapter 12, Exercise Answers, Principles of Econometrics, 5e 30

(f) The unit root tests and the vast difference in the estimates from parts (c) and (e) suggest that EAF
is not trend stationary. The relatively small standard errors for the estimates in part (c) give the
appearance that those estimates are reliable. However, all other evidence suggests otherwise. The
linear plot in part (a) did not resemble a logistic curve, unit root tests suggested that EAF is
nonstationary, and estimating a differenced version of the model changed the estimates
dramatically – the saturation parameter became greater than 1, and the inflection point  
halved from 24.1 to 12.8. The large standard errors in part (e) show that none of the parameters
have been reliably estimated. We can conclude that the estimates in part (c) are spurious.

Copyright © 2018 Wiley