CONCORDIA UNIVERSITY

DEPARTMENT OF ECONOMICS
ECON 324, Fall 2019
Midterm Exam Solution

TA’s instructions
Please give 25% of the marks to any student who attempted to solve a question and
failed to get the right answer (wrong steps+incorrect final answer). Also feel free to
give 75% of the marks to students who have the correct steps but wrong final answers.

1 Question 1 [20 marks]

Answer the following questions and explain briefly to get the full marks.

1. Assume you want to use the linear y = β1 +β2 x+e model to predict the y values for x0 = 100
and x1 = 150. Assume the average of x is x = 130. Which of these two predictions will you
expect to be the most accurate. Explain. [04 marks]
Solution: The prediction is more accurate when x is the closest to the average x. Here, the
prediction will be more accurate for x1 = 150.

2. In modelling time-series data, when do use the Almon Distributed Lag Model? How does
the model work? [04 marks]
Solution: The Almon Polynomial DL model is used when the standard OLS estimates fail
to deliver a shape of the lags effects that is consistent with economic theory. Indeed, we
expect the lag effects to be initially small, then increase with time lags and finally decrease
as we move too far in the past.

3. Is the autoregressive error (et = ρet−1 + ut ) correlated at the third order? Explain. [04
marks]
Solution: This process is known to be autocorrelated at all orders although its autocorrela-
tion exponentially declines with the lag. The autocorrelation is ρ3 at order 3.

4. Assume you have T =30 observations on a time-series variable yt and you estimated the first
order correlation to be 0.19. Is yt serially correlated? Explain [04 marks]
√ √
Solution: yt will be serially correlated if | T ρb|=| 30 ∗ 0.19|=1.040672859 is greater than
1.96. Here, we conclude that there is no autocorrelation present in yt .

5. Give one reason that will push you to use an ARDL model instead of a DL representation.
[04 marks]
Solution: We can prefer an ARDL model to a DL model if the DL model has an infinite
number of lags (geometric model for example), or if we want to purge the regression residuals
from serial correlation.

1
2 Question 2 [25 marks]
Assume the estimation of the Koyck transformation of the Geometric Lag DL model gives the
following results:

C
bt = 0.0875 + 0.65 Ct−1 + 0.25 Yt ,
(0.0442) (0.369) (0.0819)

where Yt denotes income and Ct is consumption. The standard errors of the estimators are reported
in parenthesis below each estimate.
1. Are the slope coefficients statistically significant in this regression? [05 marks]
Solution: The slope estimate is statistically significant if the tstat=estimate/standard er-
ror is greater than 1.96 in absolute value. The slope coefficient of Ct−1 has a t-stat=
0.65/0.369=1.761517615 thus this estimate is not statistically significant. The slope coef-
ficient of Yt has a t-stat= 0.25/0.0819=3.052503053, and we conclude that this estimate is
statistically significant.

2. Compute the i-delayed multiplier of a change in Yt on Ct for i = 1, 2, 3, 4. [08 marks]

Solution: Here, we can identify that φ = 0.65 and β = β0 =0.25.
We deduce that β1 = β × φ = 0.1625, β2 = β × φ2 = 0.105625,β3 = β × φ3 = 0.06865625,
β4 = β × φ4 = 0.044626563. The i−delayed multiplier is given by γi =β0 + β1 + . . . + βi . We
obtain γ1 = 0.4125, γ2 = 0.518125, γ3 = 0.58678125, γ4 = 0.631407813.

3. What is the long-run multiplier? [07 marks]

Solution: The long-run multiplier is given by γ0 = β/(1 − φ) = 0.714285714.

3 Question 3 [40 marks]

Assume we use OLS to estimate the following relationship between consumption (Ct ) and income
(Yt ) :

Ct = 0.09557 + 0.40990Yt + 0.27662Yt−1 + et .

(0.129) (0.175737) (0.153332)

Where T = 90, Adj. R2 =0.3538, (et − et−1 )2 = 0.950,

P P 2
et = 0.575, dL =1.12, dU = 1.78 at the
5% significance level. Standard errors are provided below each parameter estimate.
1. What is the economic motivation for including Yt−1 in this regression? [05 marks]
Solution: Yt−1 in included in this regression to capture the delayed effect of past income on
current consumption. The argument is that we can not expect the full influence on income
on consumption to be fully observed at time t. It takes time to the economic agents to make
their new optimal choice of consumption.

2. Test for first order serial correlation in the equation error terms at the 5% level using the
DW statistic. [10 marks]
Solution: The DW test statistic is given by (et −et−1 )2 / e2t = 0.950/0.575 = 1.652173913.
P P

Since dL < DW < dU , the DW test in inconclusive.

2
 3. What other test can be used to verify your results in (2)? [05 marks]
Solution: The Lagrange multiplier (LM) test can be also used.
Assume, the following model was estimated

ebt = −0.3 + 0.03Yt + 0.02Yt−1 + 0.25b

et−1 + ut ,

where ebt is the residual from the first model. The standard errors are provided in parenthesis.
This models has T = 89 observations and the R2 of the regression is 0.05477. The critical
value of the chi-square distribution with one degree of freedom is 3.8414 at the 5% level .
Test for the existence of serial correlation using the alternative test that you mentioned. [
05 marks].
Solution: The LM test statistic is given by T R2 = 89 ∗ 0.05477 = 4.87453 which is greater
than the critical value=3.8414. We thus reject the null of absence of correlation and conclude
that there is correlation is the errors terms.

It is suggested that you re-estimate the equation as a first-difference model, which you do
and get the following results:

Ct? = 0.065555 + 0.488269Yt? + 0.236173Yt−1

?
+ et .
(0.129415) (0.175737) (0.153332)

where Ct? = (Ct − ρCt−1 ), Yt? = (Yt − ρYt−1 ), and Yt−1

? = (Y
t−1 − ρYt−2 ), with T = 90, Adj.
2
R =0.06, ρb = 0.184695.

4. How is this remedy called? Explain the steps involved to get the results presented here. [08
marks]
Solution: The remedy is called the generalized least squares or GLS. Here, one can imple-
ment the GLS using the Cochrane-Orcutt methodology which consists in first estimating the
ρ using the first order autocorrelation of ebt , which is the residual from the first regression.
Using this ρ, we estimate the transformed equation by OLS and save the slope and the in-
tercepts coefficients. The two steps are repeated until we observe some sort of convergence
in the OLS estimates.

5. What alternative correction for serial correlation might have been used? How does it work?
[07 marks]
Solution: One can alternatively use the Newey-West standard errors to correct the OLS
standard errors that are biased under serial correlation. The Newey-West estimates of the
slope and intercept coefficients are identical the the initial OLS estimates, only the standard
errors will be inflated.

4 Question 4 [15 marks]

Consider the following distributed lag model relating the percentage growth in private investment
(INVGWTH) to the federal funds rate of interest (FFRATE):

INVGWTHt = 4% − 0.4FFRATEt − 0.8FFRATEt−1 − 0.6FFRATEt−2 − 0.2FFRATEt−3 .

3
1. Is the shape of the lags weights consistent with economic theory? Explain. [03 marks]
Solution We have |β0 = 0.4|, |β1 = 0.8|, |β2 = 0.6|, |β3 = 0.2|. It clearly appears that the
lags effects β’s initially increase and then start decreasing after the first lag. This behaviour
is consistent with the economy theory.

2. Suppose FFRATE=1% for t = 1, 2, 3, 4. Use the above equation to forecast INVGWTH for
t = 4. [04 marks]
Solution: Replace t by 4 in the above equation gives: INVGWTH4 = 4% − 0.4FFRATE4 −
0.8FFRATE3 − 0.6FFRATE2 − 0.2FFRATE1 = (4 − 0.4 − 0.8 − 0.6 − 0.2)% = 2%.

3. Suppose FFRATE is raised 1.5% in period t = 5 and then returned to its original level of
1% for t = 6, 7, 8, 9. Use the equation to forecast INVGWTH for periods t = 5, 6, 7, 8, 9. [04
marks]
Solution: Here it is assumed that FFRATE1 = 1%,FFRATE2 = 1%,FFRATE3 = 1%,FFRATE4 =
1%, FFRATE5 = 1.5%, FFRATE6 = 1%, FFRATE7 = 1%, FFRATE8 = 1%, FFRATE9 =
1%.
To answer this question, simply replace t by 5 (6,7,8,9) in the initial equation. We obtain
INVGWTH5 = 1.8%, INVGWTH6 = 1.6%, INVGWTH7 = 1.7%, INVGWTH8 = 1.9%
,INVGWTH9 = 2%.

4. Suppose FFRATE is raised to 1.5% for periods t = 5, 6, 7, 8, 9. Use the equation to forecast
INVGWTH for periods t = 5, 6, 7, 8, 9. [04 marks]
Solution: Here it is assumed that FFRATE1 = 1%,FFRATE2 = 1%,FFRATE3 = 1%,FFRATE4 =
1%, FFRATE5 = 1.5%, FFRATE6 = 1.5%, FFRATE7 = 1.5%, FFRATE8 = 1.5%, FFRATE9 =
1.5%.
To answer this question, simply replace t by 5 (6,7,8,9) in the initial equation. We obtain
INVGWTH5 = 1.8%, INVGWTH6 = 1.4%, INVGWTH7 = 1.1%, INVGWTH8 = 1.0%
,INVGWTH9 = 1.0%.