Member Student ID
Võ Lương Khánh Linh 31221027138
Hồ Trần Tú Linh 31221021568
Phùng Thị Phương Thảo 31221024093
Nguyễn Thị Kim Thoa 31221026671
Trương Khánh Linh 31221024012
CHAPTER 17 (WOOLDRIDGE)
17.1.
(i) If the spread is zero, it means there’s no clear favorite, so the favored team should
have a 50/50 chance of winning the game.
(ii)
^
favwin=.577+.0194 spread
(.028) (.0023)
[.032] [.0019]
n = 553, R2=0.111
The usual standard errors are denoted in parentheses (), while the heteroskedasticity-
robust standard errors are shown in square brackets []. With the usual standard error,
the t statistic is 2.73, which results in rejecting H 0 at the 1% significance level for a
two-sided alternative. However, using the robust standard error lowers the
significance level, but still leads to a strong rejection of H 0 at the 2% level for a two-
sided alternative: t = 2.43.
(iii)
As expected, the variable spread is highly statistically significant under both types of
standard errors, with a t-statistic over 8. When spread is 10, the predicted probability
of the favored team winning is .771
(iv)
The P value for H 0 that the intercept equals 0 is 0.9 that is failed to reject the H 0
(v)
When spread is 10, the predicted probability from the estimated probit model is .820,
which is slightly higher than the estimate from the LPM.
(vi)
When favhome, fav25, and und25 are included in the probit model, the log-likelihood
value changes to –262.64. This results in a likelihood ratio statistic of 2[–262.64 – (–
263.56)] = 2(263.56 – 262.64) = 1.84. The p-value from the x 23 distribution is
about .61. This indicates that the variables favhome, fav25, and und25 are not jointly
significant. In other words, after accounting for spread, these variables do not
contribute any meaningful additional predictive power to the model.
17.2
- At hsGPA = 3.0, SAT = 1,200, study = 10:
z = -1.17 + 0.24 x 3 + 0.00058 x 1,200 + 0.073 x 10 = 0.976
- At hsGPA = 3.0, SAT = 1,200, study = 5:
z = -1.17 + 0.24 x 3 + 0.00058 x 1,200 + 0.073 x 5 = 0.611
^
P(grad = 1¿hsGPA = 3.0, SAT = 1,200, study = 10) = exp(0.976)/[1 + exp(0.976)] =
0.726
^
P(grad = 1¿hsGPA = 3.0, SAT = 1,200, study = 5) = exp(0.611)/[1 + exp(0.611)] =
0.648
△ ^P = 0.726 - 0.648 = 0.078
17.6.
i) When the sample includes only families whose head is over 25 years old, it is said
to be truncated. And age is exogenous, hence the truncation is based on the exogenous
explanatory variable. That means the sample selection is correlated with age and age
has been controlled in the model. Therefore, if we use OLS on such a sample, we still
get unbiased estimators.
ii) In this case, we cannot estimate all of the parameters in the saving equation.
Because our sample includes only married couples without children, therefore the
hhsize variable is always 2, indicating that there is no variation in this variable. As a
result, we cannot estimate 𝛽 2. Regarding the remaining variables, we can still estimate
its parameters in case there is a variation in those variables.
iii) If we exclude from our sample families that save more than $25,000 per year, the
OLS estimators are no longer unbiased and consistent. Because the sample is now
truncated in the dependent variable, making it a potentially endogenous sample
selection.
17.7
The criticism regarding the analysis of admitted students highlighting potential bias is
partially valid but does not invalidate the findings. This study focuses on a specific
group of interested students, allowing for valuable insights into their enrollment
decisions. However, generalizations beyond this group may be limited.
COMPUTER ASSIGNMENT
17C3.
(i)
Out of 616 workers, 172 workers (about 18%) have no pension benefits. For the 444
workers who do receive pensions, the amounts range from $7.28 to $2,880.27. This
shows that a good number of workers have zero pensions, while the rest have a wide
range of pension amounts. Because of this, the Tobit model is a good choice—it helps
us study both who gets a pension and how much they get.
(ii)
(iii)
White male’s pension benefit:
As we can see the result is $893,06
Nonwhite female’s pension benefit:
As we can see the result is $521,86
=> The difference between the white male and nonwhite female is 893,06 – 521,86=
$371.
(iv)
The estimate of the coefficient on union is 439, it’s statistically significant at 1% level
with t statistic 7,03.
(v)
When peratio is used as the dependent variable in the Tobit model, white and male are
individually and jointly insignificant. The p-value for the test of joint significance is
about 0,74. Therefore, neither whites nor males seem to have different tastes for
pension benefits as a fraction of earnings. White males have higher pension benefits
because they have, on average, higher earnings
17C8.
i) First we use tabulate command in Rstudio to find the information on variable train
=> As can be seen from the above figure, 185 men out of 445 participated in the job
training program.
Then we use summarize command as follow
=> We can see that the highest number of months a man actually participated in the
program is 24 months
ii) We regress train on unem75, age, educ, black, hisp, married and get the result as
follow:
As can be seen, F-statistic = 1.43 and its p-value = 0.1915 > 0.05, therefore the
explanatory variables are jointly insignificant at the 5% level.
iii)
The likelihood ratio statistic = 10.18 and its p-value = 0.1785 > 0.05, therefore the
explanatory variables are jointly insignificant.
iv) Based on the answers in part (ii) and (iii), participation in job training does not
significantly correlate with the variables. Therefore, we can treat it as exogenous in
the 1978 unemployment equation.
v)
The estimated slope is -0.1106, implying that participating in the job training program
decreases the probability of 1978 unemployment by 0.1106. Its p-value is 0.013 <
0.05, therefore it is statistically significant at the 5% level.
vi)
Calculating the average partial effect, we can use the margins command to compute
the average partial effect for the probit estimates, which is minus 0.112 and
statistically significant at 5% level. It’s almost the same as the linear prob. model
estimate.
vii)
Both model have the same conclusion.
viii)
ix)
-.1131733 means that the job training program is estimated to lower the
unemployment prob. by 0.1131733, holding other variables in the model.
The linear prob. model gives an average partial effect of -.1117828:
CHAPTER 8 (CHRIS BROOK)
8.8. In EViews, open the ‘currencies.wf1’ file that will be discussed in detail in the
following chapter. Determine whether the exchange rate series (in their raw levels
forms) are non-stationary. If that is the case, test for cointegration between them using
both the Engle–Granger and Johansen approaches. Would you have expected the
series to cointegrate? Why or why not?
- Determine whether the exchange rate series (in their raw levels forms) are non-
stationary.
- Engle–Granger Cointegration Test
-> All p-values are well above 0.10, so we fail to reject the null hypothesis of non-
stationarity. This means the raw exchange rate series are not stationary and thus may
contain unit roots. The p-values were all well above 0.05, and none of the tau- or z-
statistics were significant -> no cointegration was found between the series using the
Engle–Granger method.
- Johansen Cointegration Test
-> Trace test indicates no cointegration at the 5% level. Max-Eigenvalue test also
indicates no cointegration at the 5% level.
- While economic theory suggests that major exchange rates like EUR, GBP, and
JPY might be cointegrated due to shared global influences and long-run parity
conditions, practical factors such as floating exchange rates, differing monetary
policies, and short-term shocks often prevent a stable long-run relationship.
This explains why cointegration is not observed in the test results.
CHAPTER 9 (CHRIS BROOK)
9.6.
Following the given hint, we express the model as: y i=α+β x i+ui
where: ui=c i+d i x i with c i=a i−α and d i=b ii−β
We know that E(c i)=0, and that: E(d i x i) = Cov (d i x i)
since E(d i)=E(b i−β)=E(b i)−β=0, given that b i and x i are uncorrelated by
assumption. Hence, E(d i x i )=0, and this implies that E(ui)=0, ui has zero mean.
Next, to ensure consistency of OLS, we must show that ui is uncorrelated with
x i. This will be true if both c i and d i x i are uncorrelated with x i. By assumption, a i (and
hence c i) is uncorrelated with x i.
As for d i x i, consider: Cov(d i x i, x i)=E[d i xi2]=Cov(b i x i2)
Again, since b i is assumed to be uncorrelated with x i2, this covariance is zero. Thus,
we have shown that ui has a zero mean and is uncorrelated with x i. Therefore, OLS
estimation of α and β is consistent. However, it is not necessarily unbiased in this
setup. For OLS to be unbiased, the stronger condition must hold.
CHAPTER 12 (CHRIS BROOK)
12.5.
(a) (a)
The results from the Probit regression show that none of the country's dummy
variables are statistically significant at the 5% level when using the UK as the
reference country. This means that, after taking into account individual and year-
related factors, there's no strong evidence that the probability of failure is different
between these countries and the UK. The coefficients for the other countries are
different, but all their p-values are above 0.05, so we can't say there's a significant
difference.We do with Logit model
In the Logit regression model, country dummies labeled C1 to C9 are used to capture
how each country might affect the probability of failure, with the UK being used as
the reference country. By doing this, we can compare the likelihood of failure in each
country to that of the UK by looking at the coefficients for these dummy variables.
The results show that none of the countries are significantly different from the UK at
the 5% significance level. In other words, all the p-values for the country dummies are
above 0.05, so there’s no strong statistical evidence that failure rates in these countries
are different from the UK. One exception is dummy variable C7, which has a p-value
of 0.0757. While this is still not below 0.05, it is below 0.10, which means C7 might
have a somewhat stronger effect on failure probability compared to the UK, but the
result is only weakly significant.
(b)
To assess the possible non-linear relationship between age and the probability of MSc
failure, a squared age term was added to the probit model. The coefficient on age was
-0.12 (p < 0.05), while the coefficient on age squared was 0.0031 (p < 0.05),
indicating a statistically significant U-shaped relationship. This shows that the
probability of failure initially decreases with age but begins to increase again at older
ages, suggesting that there is an optimal age with the lowest failure probability. This
shows a nonlinear relationship between age and failure probability.
