Excluding the United States can be justified if the U.S.

is an outlier with a much

higher number of billionaires compared to other countries. Including it might distort
the estimates of the effects of variables such as GDP and population because the
U.S. could disproportionately influence the regression results. However, excluding it
also means losing valuable data that could help understand the global distribution of
wealth. The decision depends on the specific research question and whether the
U.S. is considered representative of the other countries.
PROBLEM 9.4
PROBLEM 10.4
Ý cuối là C11.6
Câu 21 :

Chap 4
Chap 5
23. The following equation was estimated by STATA for the first semester of
undergraduate students based on their semester results in a university in
Sydney. The standard errors reported below are not robust to
heteroskedasticity:
 a) Do the variables sthours, atar, female, and mode have the expected
estimated effects? Which of these variables are statistically significant at the
5% level? Does it matter which standard errors are used? Explain your
answer.
• sthour: The coefficient of 0.516 suggests that an increase in study hours is
expected to increase the average mark. This effect is significant at the 1%
level under both standard errors, as the p-value is 0.000 in both cases.
• atar: The coefficient of -0.411 indicates that an increase in the ATAR score is
expected to decrease the average mark. This is statistically significant at the
1% level under both standard errors (p-value = 0.000). The negative sign,
although counterintuitive, might reflect an issue with the data or model.
• female: The coefficient of 0.551 suggests that being female is associated
with higher average marks compared to males. This effect is significant at the
5% level under both standard errors.
• mode: The coefficient of 0.311 suggests that on-campus students tend to
have higher average marks. However, this variable is only significant at the
5% level under non-robust standard errors (p-value = 0.044), but not
significant when robust standard errors are used (p-value = 0.228)
b) Test the hypothesis H0 : βatar = 0 against the two-sided alternative at the
5% level, using both standard errors. Describe your conclusions.
●​ Null Hypothesis (H0): βatar = 0 (ATAR score does not affect the
average mark).
●​ Alternative Hypothesis (H1): βatar ̸= 0 (ATAR score affects the
average mark).
Using both standard errors, the p-value for the atar variable is 0.000, which is less
than the 5% significance level. Thus, we reject the null hypothesis in both cases.
This implies that the ATAR score has a statistically significant effect on the average
mark.
c) Test whether there is a positive impact of a student’s mode of learning as
on-campus. Does the significance level at which the null can be rejected depend on
the standard error used? Explain your answer.
●​ Null Hypothesis (H0): βmode ≤ 0 (Being an on-campus student does not
positively affect the average mark).
●​ Alternative Hypothesis (H1): βmode > 0 (Being an on-campus student
positively affects the average mark).
• Non-Robust Standard Errors: The coefficient is significant with a p-value of
0.044, allowing us to reject the null hypothesis at the 5% level.
• Robust Standard Errors: The coefficient is not significant with a p-value of 0.228,
meaning we cannot reject the null hypothesis at the 5% level using robust standard
errors.

This difference indicates that the significance of the mode variable depends on the
standard error used. The robust standard errors, which correct for heteroskedasticity,
suggest that the mode of learning does not have a statistically significant impact on
the average mark at the 5% level.

24.

a) Interpret the estimated coefficients of the model.

• lpopn: The coefficient 1.4237 represents the elasticity of the number of new
dwellings with respect to the total population. A 1% increase in the population is
expected to increase the number of new dwellings by approximately 1.42%, holding
other factors constant. This suggests a strong positive relationship between
population growth and new dwelling approvals.
• irate: The coefficient -0.0991 indicates that an increase in the commercial banks’
standard variable interest rate by 1% is expected to decrease the number of new
dwellings by approximately 0.0991%. This negative relationship suggests that higher
interest rates discourage new housing developments.
• cons: The constant term 2.9299 represents the baseline level of new dwellings
when the log of the population is zero and the interest rate is zero. Although this is a
hypothetical scenario, it serves as a baseline for the model.

b) Examine their individual and joint significance (the test result is reported in the
table above) and interpret the results.
• Individual Significance:
– lpopn: The t-statistic for lpopn is 9.59 with a p-value of 0.000, indicating that lpopn
is statistically significant at the 1% level. This means that population size significantly
affects the number of new dwellings.
– irate: The t-statistic for irate is -7.45 with a p-value of 0.000, also indicating
statistical significance at the 1% level. Thus, the interest rate significantly influences
the number of new dwellings.
– cons: The t-statistic for the constant is 2.32 with a p-value of 0.021, which is
significant at the 5% level. The constant term is meaningful in the context of this
model.
• Joint Significance:
– The F-test reported at the bottom of the output shows an F-statistic of 100.88 with
a p-value of 0.000, indicating that the model as a whole is statistically significant at
the 1% level. This implies that the independent variables jointly explain a significant
portion of the variation in the dependent variable ldwelling.
c) Suppose you find that the number of new dwellings in Victoria has seasonalities
and trends. Specify a model for dwelling or housing starts in Victoria that accounts
for possible trends and seasonality in the variables. Justify your answer.

To account for trends and seasonality in the number of new dwellings in Victoria, the
model should include both time trend variables and seasonal dummy variables. The
following model can be specified:

Where:
• 𝑡𝑟𝑒𝑛𝑑𝑡: A continuous time variable that captures the underlying trend in the data
over time.
• Qi: Seasonal dummy variables that capture the effects of different quarters (Q1,
Q2, Q3) on the number of new dwellings, with Q4 being the reference category.

Justification:
Including a trend variable accounts for long-term growth or decline in new dwellings
that might be driven by factors such as economic growth, demographic shifts, or
policy changes. Seasonal dummy variables capture the effects of seasonality,
recognising that housing starts might peak or dip during certain times of the year due
to factors such as weather conditions or construction cycles.
22.

a) Is there any evidence against the EMH? Explain your answer

The Efficient Market Hypothesis (EMH) suggests that past returns should not have
predictive power over future returns, implying that β1 = 0. The coefficient for return
𝑡−1
is 0.059, but the p-value is 0.122, which is greater than the standard
significance levels (e.g., 0.05). This indicates that we cannot reject the null
hypothesis that β1 = 0. Thus, there is no statistically significant evidence against the
EMH in this model.
b) Using the Breusch-Pagan test for heteroskedasticity, what is your conclusion?
Explain your answer.
The Breusch-Pagan test results in a chi-squared value of 95.22 with a p-value of
0.0000, indicating strong evidence against the null hypothesis of homoscedasticity.
Therefore, we conclude that there is heteroskedasticity in the model, meaning that
the variance of the error terms is not constant.
c) We compute the heteroskedasticity-robust standard errors as below:
What does using the heteroskedasticity-robust t statistic do to the significance of
return 𝑡−1 ? Explain your answer.

Using the heteroskedasticity-robust standard errors increases the standard error for
return 𝑡−1 from 0.038 to 0.059, which reduces the t-statistic to 0.995 and results in a
higher p-value of 0.395. This p-value is far greater than the standard significance
levels (e.g., 0.05), indicating that return 𝑡−1 is not statistically significant even when
accounting for heteroskedasticity. The lack of significance strengthens the
conclusion that there is no evidence against the EMH.
21.
a) Based on the test statistics Z(t) and the p-value for Z(t), is there any evidence for
a unit root in ln(DowJones) at 1% and 5% significance levels? Explain your answer.
The ADF test checks for the presence of a unit root in a time series, which would
imply non-stationarity. The Z(t) statistic here is -0.830, which is not as negative as
the critical values at both the 1% and 5% levels (i.e., -3.430 and -2.860,
respectively). Additionally, the p-value of 0.8103 is much higher than both 1% and
5%, suggesting we do not have enough evidence to reject the null hypothesis of a
unit root. Therefore, the test indicates that the ln(DowJones) series likely has a unit
root and is non-stationary

b) What is a convenient equation for carrying out the unit root test to test H0 : q = 0,
against H1 : q < 0 so that under H0 : yt−1 is I(1)? What problem could there be?
How to solve that problem?
The ADF test equation typically used to test for a unit root, as discussed by
Wooldridge, is:

where ∆yt = yt − yt−1 and ρ = q − 1. Under the null hypothesis H0, yt−1 is I(1),
meaning the series has a unit root.
One common issue is that the test might have low power if there’s a deterministic
trend that is not accounted for. To address this, Wooldridge suggests including a time
trend in the regression equation:

Including the trend term δt helps account for deterministic trends that could
otherwise bias the test results.
c) When we include a time trend in the regression, the critical values and the p-value
for Z(t) of the test change as follows:

Is there any evidence for a unit root in ln(DowJones) at 1% and 5% significance

levels, using both Z-test statistics and p-value approaches?
With the inclusion of a time trend, the Z(t) statistic is -0.973, which again is not as
negative as the critical values at both 1% (-3.960) and 5% (-3.410) significance
levels. Additionally, the p-value of 0.9477 is still much higher than both 1% and 5%,
which aligns with Wooldridge’s emphasis on the p-value approach. Thus, we still fail
to reject the null hypothesis of a unit root. This suggests that even after accounting
for a time trend, the ln(DowJones) series appears to have a unit root and remains
non-stationary at both 1% and 5% significance levels.