TABLE OF EVALUATION

No. Full name Level of completion

1. Le Ngoc Phương Khanh 100%
2. Mai Linh 100%
3. Doan My Huyen 100%
4. Nguyen Tri Khang 100%

CHRIS BROOKS
1.
(a) It is because the main goal of using OLS estimation is to fit a line to the data by
minimizing the total sum of the squares, also known as residuals. If we consider
horizontal distance, that means we are adjusting the value of x to determine y, which
is not realistic because x are variables that are assumed to be fixed rather than be
random like y.

(b) Since some points can lie above or below the line, minimizing the sum of squared
distances ensures a unique solution, unlike simply minimizing the sum of residuals,
which could cancel out. When the total distance of points above equals the total
below, many lines could fit the data, as their residuals would sum to zero. Squaring
the residuals prevents this cancellation, ensuring a unique line that provides the best fit
without needing a special solution for the coefficients.

(c) Applying squaring to residuals would make it simpler for calculations compared to
using the absolute value. Moreover, it would prevent a case in which positive and
negative residuals cancel each other out.

2.
Sample Regression Function (SRF): The Sample Regression Function (SRF)
represents the relationship derived from analyzing sample observations. It is
essentially the estimated form of the Population Regression Function (PRF),
expressed through its equation.

^y t =α^ + ^β x t
In this equation:

● α^ and ^β are the estimates of α and β , obtained from the sample data.

● ^y t is the predicted value of y based on the sample x .

Population Regression Function (PRF): The Population Regression Function (PRF)

describes the actual relationship between the independent variable xxx and dependent
variable y and for the entire population. It is typically represented by the equation:

y t =α + β x t +ut

Where:

● α is the intercept, representing the expected value of y when x = 0.

● β is the regression coefficient, indicating the alteration in y for a 1-unit rise in x

● ut is the random error term, accounting for factors not explained by x .

Key Differences:

● The PRF describes the true relationship in the entire population, but since the
population data is typically unobservable, the PRF is often unknown.

● is derived from sample data and serves as an approximation of the Population

Regression Function (PRF). It decomposes the observed value of y t into two

parts: y t = ^y t +u^ t

Where:
o ^y t : The predicted value of yy.
o u^ t : The residual (error term), representing the difference between the
observed value and the predicted value.

3.
An estimator is a formula used to determine parameter estimates, which describe the
relationship between two or more explanatory variables. There are many different
types of estimators, but the OLS estimator is one of the most commonly preferred.
OLS is considered "optimal" because it has the lowest variance among linear and
unbiased estimators, meaning no other linear and unbiased estimator has a smaller
sampling variance. While an estimator with a lower sampling variance than OLS can
be defined, it would either be non-linear, biased, or both. Therefore, choosing an
estimator always involves a trade-off between bias and variance.

The OLS estimator is considered optimal only when the assumptions of the classical
linear regression model are satisfied. These assumptions include linearity, no
endogeneity and homoscedasticity (the error terms have constant variance). Under
these conditions, the Gauss-Markov theorem states that OLS is the Best Linear
Unbiased Estimator (BLUE).

However, if these assumptions are violated—such as in cases of endogeneity or

heteroscedasticity—OLS can still produce unbiased estimates but may lose efficiency,
meaning it might no longer have the smallest variance. In such situations, alternative
estimators, such as Generalized Least Squares (GLS) or Instrumental Variables (IV),
may perform better by addressing these violations. Thus, while OLS is optimal when
the classical assumptions hold, it may not always be the best choice if those
assumptions are breached.

4.
Five assumptions are usually made about the unobservable error terms in the CLRM:
● E(ut) = 0 → The error terms are assumed to have an expected value of zero

● Var(ut) = 𝜎2 → The error term is constant and finite across all values of x t.

● Cov(ui, uj) = 0 for i ≠ j→ The errors are linearly independent of one

another

● Cov(ut, xt) = 0 → The error term is uncorrelated with corresponding x.

● ut ~ N(0,𝜎2) → The error term ut follows a normal distribution.

In Box 3.3 of the book, the assumptions of the disturbance terms of the classical linear
regression model are listed. To demonstrate that the ordinary least squares estimators
of α and β are the "best," or that among the class of linear unbiased estimators they
have the minimum variance, we must first establish the first four assumptions. The
Gauss-Markov theorem establishes that OLS estimators are BLUE (had fulfilled
assumptions). It's possible that OLS estimators lose their unbiased and "efficient" if
these assumptions are violated, a topic covered in Chapter 4. In other words, they
might be not accurate or subject to variations between samples. The fifth assumption,
stating that the disturbances follow a normal distribution, is necessary for drawing
statistical inferences about population parameters from the sample data. This
assumption allows hypothesis testing of the coefficients, ensuring that the test
statistics adhere to a t-distribution, assuming all other conditions are met.
conclusion: five assumptions of the CLRM's error terms ensure that OLS estimators
are BLUE, as established by the Gauss-Markov theorem. The first four assumptions
guarantee unbiasedness and efficiency, while the fifth assumption, requiring normally
distributed errors, enables valid hypothesis testing using the t-distribution. Violations
of these assumptions can compromise the reliability of OLS estimators.

5.

Evaluation:

1. Model (3.39):

y t =α + β x t +ut

This is linear in parameters (α , β ), so it can be estimated using OLS.

2. Model (3.40):
α β ut
y t =e x t e

This is nonlinear in parameters. However, by taking the natural log of both

sides:
ln( y t ¿=α + β ln (x)+ut
It becomes linear in parameters and can be estimated using OLS.
 3. Model (3.41):
y t =α + β y x t +ut

This is nonlinear in parameters ( β y ), so it cannot be estimated using OLS.

4. Model (3.42):
ln( y t ¿=α + β ln (x t )+u t
This is already linear in parameters (α , β ), so it can be estimated using OLS.
5. Model (3.43):
y t =α + β1 x t + β 2 z t +u t

This is linear in parameters (α , β 1, β 2), so it can be estimated using OLS.

Conclusion:

Models (3.39), (3.40) (after transformation), (3.42), and (3.43) can be estimated using
OLS. Model (3.41) cannot.

6.
It is said by the null hypothesis that the value of β = 1, so

H0: β = 1

H1: β > 1

^
−B∗¿ 1.147−1
Test statistic = B ¿= = 2.682
SE ¿ ¿ 0.0548

We have n is the sample size => T = 62, and there are 2 parameters (αi, βi). In this
case, for a one-tailed test at the 5% significance level with T – 2 = 60 degrees of
freedom, so according to the t-distribution table, the critical value is 1.671.
Since the calculated test statistic is 2.682 > 1.671, which falls in the 5% rejection
region, the null hypothesis must be rejected. In addition, we can say that the stock is
more risky than the market and the analyst’s claim is not empirically verified.

9.

We test hypotheses regarding the actual coefficients rather than the estimated
values.This is due to the fact that the objective is to test hypotheses regarding the
underlying population parameters in order to infer their likely values. We don't need to
test hypotheses regarding estimated values because we know exactly what our
estimates are because we calculated them.

Conclusion: In order to draw inferences about the population parameters, we test

hypotheses regarding the actual coefficients (β). Since the estimated values are already
known from the calculations, there is no need to test them.
 WOOLDRIDGE
4.
(i) Using the regression equation ^
bwght =119.77−0.514 × cigs:
- When cigs = 0, predicted birth weigh is 119.77 ounces.

- When cigs = 20, predicted birth weigh is: ^

bwght ¿ 119.77−0.514 × 20=109.49 ounces.

The difference is 10.28 ounces, which indicates that smoking 20 cigarettes per day
during pregnancy is associated with a reduction of approximately 8.6% in predicted
birth weight.

(ii) Not necessarily. The regression shows a statistical association between smoking
and birth weight, but it does not confirm causation. Other factors, such as the mother's
overall health, nutrition, access to prenatal care, or exposure to other harmful
substances, might influence birth weight and could also correlate with smoking.
Without controlling for these potential confounders, we cannot conclude that smoking
directly causes a decrease in birth weight.

(iii) Solving the equation:125=119.77−0.514 × cigs

119.77−125
cigs= ≈−10.18
−0.514

This result suggests that for a predicted birth weight of 125 ounces, the mother would
need to smoke a negative number of cigarettes, which is nonsensical. This highlights a
limitation of simple linear regression: predictions outside the observed data range
(extrapolation) may lead to unrealistic results. Additionally, the regression model
suggests the maximum predicted birth weight is 119.77 ounces, as this corresponds to
zero cigarette consumption.

(iv) The fact that about 85% of women in the sample do not smoke while pregnant
implies that most observed birth weights are clustered near the upper limit of the
regression prediction: predicted birth weight is 119.77 ounces. This suggests that the
data does not adequately represent heavier birth weights, limiting the model's ability
to explain such cases. Therefore, the sample composition reinforces why predicting a
birth weight of 125 ounces is not realistic within this framework.
5.
(i) According to the intercept, cons is expected to be - $124.84 when inc = 0.
Obviously, this is untrue and illustrates how consumption function may not be reliable
indicator of consumption at extremely low levels of income. On an annual basis,
however, $124.84 is not so far from zero.
(ii) Simply substitute 30,000: cons^= –124.84 + 0.853 * 30,000 = 25,465.16 dollars.
(iii) The following graph displays the MPC and the APC. Despite the negative
intercept, the smallest APC observed in the sample remains positive. An annual
income level of $1,000 (in 1970 dollars) is where the graph begins.

6.

(i) The coefficient on log(dist) is 0.312, which tells that a 1% increase in the distance
from the incinerator leads to a 0.312% increase in the price of the house, holding all
else constant. This is a positive relationship, indicating that the further houses are
from the incinerator, the higher they tend to have higher prices.

(ii) The simple regression may not provide an unbiased estimator of the ceteris paribus
elasticity of price with respect to distance from the incinerator. This is because the
placement of the incinerator could be correlated with other unobserved factors that
influence housing prices, such as socioeconomic characteristics, zoning laws, or city
planning decisions. Consequently, housing prices do not directly correlate with
distance from the incinerator in a proportional manner.

(iii) Many other factors could affect a house's price, including:

● Size of the house (square footage): Larger houses generally sell for more.
● Age of the house: Older houses might be less expensive due to wear and tear,
unless they are historic or have unique features.
● Neighborhood characteristics: Proximity to schools, parks, shopping areas,
and public transportation.
● House condition: Well-maintained homes are more expensive.
● Local amenities: Features like swimming pools, garages, and modern
appliances can add value.