Table of Contents

Table of Contents.............................................................................................................
1. Introduction.................................................................................................................
2. Purpose......................................................................................................................
3. Data Collection...........................................................................................................
4. Model Specification.....................................................................................................
4.1. Estimating the simple linear regression model (mt = α + βxt + ϵt)......................
4.1.1. Estimated results in standard form.............................................................
4.1.2. Interpretation..............................................................................................
4.2. Log transformation...........................................................................................
4.2.1. Estimated results of the regression model (lmt = α + βlxt + ϵt) .....................
4.2.2. Interpretation..............................................................................................
4.3. Introduction of Lag...........................................................................................
4.3.1. Estimated results of the regression model (lmt = α + β1lxt + β2lmt-1 + ϵt).......
4.3.2. Interpretation..............................................................................................
4.4. Verification of CLRM assumptions...................................................................
4.4.1. 1st Assumption: E(ϵt) =0.............................................................................
4.4.2. 2nd Assumption: No Heteroskedasticity.......................................................
4.4.3. 3rd Assumption: No Autocorrelation............................................................
4.4.4. 4th Assumption: Multicollinearity................................................................10
4.4.5. 5th Assumption: Normality.........................................................................10
5. Conclusion................................................................................................................1
6. References............................................................................................................... 12

1
1. Introduction:
The field of econometrics plays an essential role in analyzing and understanding the
complex relationship between economic theory and real-world data. By using statistical
methods, econometricians try to quantify and measure the relationships between
economic variables, thereby enabling policymakers, and researchers to make informed
decisions based on empirical evidence. This assignment provides an overview of the
fundamental concepts and techniques in econometrics.

2. Purpose:
The purpose of this assignment is to apply the fundamental knowledge of econometrics
to analyze a real-world problem. By utilizing the concepts, techniques, and
methodologies learned in this course, the assignment aims to demonstrate the ability to
critically analyze economic data, interpret econometric results, and draw meaningful
conclusions.

3. Data Collection:
This study utilizes data on exports (x) and imports (m) of goods, obtained from the State
Bank of Pakistan. The dataset incorporates a period of nine years, i.e., the time series
with monthly frequency from July, 2005 to April 2023, allowing for a comprehensive
analysis of the trade dynamics over time.
To ensure the reliability and accuracy of the data, several steps were taken during the
data collection process. First, the selection criteria for the variables were carefully
defined, focusing specifically on exports and imports of goods. This precision allows for
a focused analysis of trade flows.

Note: The data was collected through online databases provided by the State Bank of
Pakistan.

4. Model Specification:

In terms of methodology, this study applies a simple linear regression model (mt = α +
βxt + ϵt), which was the core of this course, and employs monthly time series analysis.
The choice of a simple linear regression model enables us to assess the relationship
between the dependent variable and an independent variable. This approach allows for
the identification of any significant trends, patterns, or associations that may exist
between the variables.
The monthly time series analysis further enhances our understanding of the trade
patterns that may be present in the data. It enables us to identify any fluctuations, which

2
can provide valuable insights into the factors influencing exports and imports of goods
and services over time.

4.1. Estimating the simple linear regression model (mt = α + βxt + ϵt)
After using the eviews the following results were extracted;

4.1.1. Estimated Results of the simple regression model in standard form:

Where, m = import is our dependent variable,

x = export is our independent variable
t = monthly time series

mt = -918.53 + 2.31xt => eq01

R2 = 0.65
S.E. of regression = 619.99

3
DW stat = 0.53
4.1.2. Interpretation:

From the above results, we may interpret the following:

a. Equation (eq01):
The equation represents the simple linear regression model relating the dependent
variable to the independent variable. The equation is given as: m t = -918.53 + 2.31 xt.
This equation implies that for every unit increase in the export of goods, the import of
goods is expected to increase by 2.31 units. The negative intercept (-918.53) suggests
that even when xt is zero, there is still a negative value for mt.

b. R-squared (R2):
The R2 value measures the goodness-of-fit of the regression model. In this case, the R 2
value is 0.65, which is not close to 1 but even than it refers a moderate correlation,
whereas a higher R-squared value indicates a better fit of the model to the data.

c. Standard Error of Regression:

The S.E. of regression provides an estimate of the variability of the actual values of the
dependent variable around the regression line. In this case, the S.E. of regression is
619.99. It indicates that, on average, the actual values of the imports may deviate from
the predicted values by approximately 619.99 units.

d. Durbin-Watson statistic (DW stat):

The Durbin-Watson statistic is used to test for the presence of autocorrelation in the
residuals of the regression model. In this case, the DW stat is 0.53. The DW stat ranges
from 0 to 4, with a value close to 2 indicating no autocorrelation. A value less than 2
suggests positive autocorrelation, while a value greater than 2 suggests negative
autocorrelation. In this case, the DW stat of 0.53 suggests the presence of positive
autocorrelation in the residuals.

4.2. Log Transformation

Now we will be using log transformation of all the series used in the assignment to
smooth data and minimize the variance, we are now estimating simple regression in log
form, i.e., lmt = α + βlxt + ϵt.

4
4.2.1. Estimated Results of the regression model:

lmt = -1.36 + 1.26lxt => eq02

R2 = 0.67
S.E. of regression = 0.17
DW stat = 0.60

4.2.2. Interpretation:

From the above results, we may interpret the following:

a. Equation (eq02):
The equation in log form is lmt = -1.36 + 1.26lxt. This equation suggests that a one-unit
increase in the log-transformed independent variable (lxt) is associated with a 1.26 unit
increase in the log-transformed dependent variable (lm t). The negative intercept (-1.36)
indicates that even when lxt is zero, there is still a negative value for lm t in logarithmic
terms.

5
 b. R-squared (R2):
The R-squared value is 0.67. This suggests a reasonably high correlation good fit of the
model to the data.

c. Standard Error of Regression:

The S.E. of regression is 0.17, indicates a lower standard error which is better precision
of the model's predictions.

d. Durbin-Watson statistic (DW stat):

The DW stat of 0.60 suggests that there is still the presence of positive autocorrelation
in the residuals.

4.3. Introduction of Lag dependent variable as independent variable

Introducing a lagged dependent variable as an independent variable is a common

technique used in econometrics to address autocorrelation, which refers to the
correlation between the error terms of a regression model.

6
4.3.1. Estimated Results of the regression model (lmt = α + β1lxt + β2lmt-1 + ϵt):

lmt = -0.98 + 0.49lxt + 0.67lmt-1 => eq03

R2 = 0.87
S.E. of regression = 0.10
DW stat = 2.19

4.3.2. Interpretation:
From the above results, we can interpret the following:

a. Equation (eq03):
This equation indicates that both the log-transformed independent variable (lx t) and the
lagged log-transformed dependent variable (lmt-1) have significant effects on the current
log-transformed dependent variable (lmt). A one-unit increase in the log-transformed
independent variable (lxt) is associated with a 0.49 unit increase in the log-transformed
dependent variable (lmt), while a one-unit increase in the lagged log-transformed
dependent variable (lmt-1) is associated with a 0.67 unit increase in the current log-
transformed dependent variable (lmt).

b. R-squared (R2):
The R-squared value is 0.87, that suggests a significant correlation of the model to the
data.

c. Standard Error of Regression:

The S.E. of regression is 0.10, which is representing a lower standard error that
indicates a better precision of the model's predictions compared to the previous results.

d. Durbin-Watson statistic (DW stat):

The DW stat of 2.19 suggests no significant autocorrelation in the residuals. A value of
2.19 is close to the ideal value of 2, indicating that the residuals are not significantly
correlated over time. This suggests that the addition of the lagged dependent variable
(lmt-1) as an independent variable has effectively accounted for autocorrelation.

7
4.4. Verification of Classical Linear Regression Model (CLRM) Assumptions:

In econometrics, the CLRM assumptions are a set of assumptions that form the basis
for conducting regression analysis and obtaining reliable and efficient estimates.

4.4.1. 1st Assumption: E(ϵt) =0

Based on the result obtained from equation (eq03) indicates that the mean value of the
error term (ϵt) is approximately -1.28e-15, which is very close to zero, indicating that, on
average, the errors do not exhibit a systematic bias. This suggests that the first
assumption of the CLRM is verified.

4.4.2. 2nd Assumption: No Heteroskedasticity

Applying the white test of Heteroskedasticity, our H 0 of white test means there should be
no Heteroskedasticity, and it should be < 0.05 to reject our H0.

8
H0: No Heteroskedasticity

From the above result, White test of Heteroskedasticity and the obtained p-value from
equation (eq03), which is greater than 0.05, we can conclude that we accept our null
hypothesis (H0) that there is no heteroskedasticity. This indicates that the assumption of
no heteroskedasticity is verified for the regression model represented by equation
(eq03) and associates with the second assumption of the CLRM.

4.4.3. 3rd Assumption: No Autocorrelation

9
Based on the Durbin-Watson statistic (DW stat) result obtained from equation (eq03),
which indicates a value greater than 2, it can be concluded that there is no significant
autocorrelation present in the residuals. Therefore, for equation (eq03), the verification
of the third assumption of no autocorrelation in the CLRM is confirmed.

4.4.4. 4th Assumption: Multicollinearity

Since we are using simple regression model and we have only two variables so let us
assume, if the third variable was lms then;

Rlmlx = 0.82, Rlmlms = 0.80, Rlxlms = 0.77

R2lmlx = 0.67 R2lmlms = 0.64 R2lxlms = 0.59

Here R2lxlms = 0.59 which is < 0.65, so we may conclude that there is an evidence of
multicollinearity in the multiple regression model. Since we have option to drop any of
the two independent variables lx and lms, so we have to drop lms (but the correlation
between lx and lm is greater than the correlation between lms and lm as R 2lmlx > R2lxlms:
0.67 > 0.59). If we can find the replacement of lms then replace it otherwise we may
stick with this model as multicollinearity problem is not severe among the other
problems of econometrics.

4.4.5. 5th Assumption: Normality

Using the Jarque-Bera test our H0 means variables follow a normal distribution, and its
p-value should be < 0.05 to reject our H0.

10
H0: Variables follow a normal distribution

From the above result of the Jarque-Bera test for the residuals (e_eq03) obtained from
equation (eq03), the p-value is 0.59. Since this p-value is greater than the significance
level of 0.05, we accept our null hypothesis that the variables follow a normal
distribution. Therefore, we can conclude that the variables in equation (eq03) follow a
normal distribution. This confirms the fulfillment of the fifth assumption of the CLRM.

5. Conclusion:
Based on the analysis and interpretation of the regression results, as well as the
verification of the Classical Linear Regression Model assumptions, the following key
points can be highlighted:

a. The initial regression model (eq01) showed a significant relationship between the
dependent variable (mt) and the independent variable (xt), with an R-squared
value of 0.65. However, the presence of autocorrelation and heteroskedasticity
indicated the need for further analysis.

11
 b. The log transformation of variables was applied to address the issues of
autocorrelation and heteroskedasticity. The updated regression model (eq02) in
log form showed a stronger relationship between the log-transformed variables
(lmt), with an improved R-squared value of 0.67.

c. To address autocorrelation, a lagged dependent variable was introduced as an

independent variable in the regression model (eq03). The results indicated that
both the log-transformed independent variable (lx t) and the lagged log-
transformed dependent variable (lmt-1) had significant effects on the current log-
transformed dependent variable (lmt). The R-squared value increased to 0.87,
indicating a strong fit of the model.

d. The CLRM assumptions were verified for eq03. The error term had an expected
value close to zero, there was no heteroskedasticity, no significant
autocorrelation in the residuals, and the variables followed a normal distribution.
However, the presence of multicollinearity was detected when considering an
additional independent variable (lms). Based on the correlation and R-squared
values, it was recommended to drop lms and retain lx in the model.

6. References:

1. State Bank of Pakistan

12