QUESTION 1

a.

The dependent variable is GDP, PPP (2017 constant international $). The independent
variables are as follows:

FDI brings capital, technology, and managerial expertise to the host country, which can
enhance productivity and economic growth, it fosters capital accumulation and technological
advancement (Borensztein et al. 1998).

Government Expenditure on Education (ExpenEdu) is critical for human capital

development. Increased spending on education improves labor productivity and innovation
capacity, which are essential drivers of economic growth (Barro 1991). A well-educated
workforce can adapt to new technologies and processes, enhancing overall economic
performance.

Trade Trade openness allows countries to specialize in industries where they have a
comparative advantage, leading to more efficient resource allocation and economic growth.
Trade also provides access to larger markets and advanced technologies. (Frankel and Romer
1999).

Population (Pop) Population size affects the labor force and market size. A larger population
can contribute to economic growth by providing a bigger workforce and higher consumer
demand.
Industry Value Added (Industry): The industrial sector, including construction, is a key driver
of economic growth, especially in developing economies. Industrialization leads to increased
production capabilities, job creation, and technological advancement (Szirmai 2012).

Multicollinearity
VIF value:

billion_FDI ExpenEdu Trade Pop billion_Industry

1.085917 1.008738 1.048844 2.031282 2.031096

As all of the values is less than 10, this indicate that there is no serious multicollinearity
problems in the model (Wooldridge 2019)

Non-linear, level, and logarithmic form

For the Population variable, logarithmic transformation is used as the population size affects
economic output in a proportional manner, and the absolute number of population is
increasing proportionally , potentially influencing GDP PPP significantly.

Scatter Plot
Regression Output
Model 1 equation:

b. Interpretation on coefficient

billion_FDI: Coefficient is -9.455, indicating for one billion increase in billion_FDI (Foreign
Direct Investment), billion_GDPppp is estimated to decrease by 9.455 billion, ceteris paribus

ExpenEdu: Coefficient is 1.145, indicating a 1% increase in ExpenEdu (Government

Expenditure on Education, % of total GDP), GDPppp is estimated to increase by 11.45
billion, which is a considerable effect, ceteris paribus

Trade: Coefficient is -0.1757, indicating for 1% increase in Trade (% of total GDP),

billion_GDPppp is estimated to decrease by 0.1757 billion, ceteris paribus

Log(pop): Coefficient is 101.9, indicating for a one-unit increase in log(Pop) (Population),

billion_GDPppp is estimated to increase by 101.9 units, ceteris paribus

billion_Industry: Coefficient is 4.253, indicating one billion increase in Industry (Industry

Value Added), GDPppp is estimated to increase by 4.253 billion, ceteris paribus
All independent variables, except for Log(pop), have a simple level coefficient, indicating the
direct impact of these variables on GDP,PPP without considering any proportional
relationship. Regarding Log(pop), this is a semi-elasticity coefficient, as it measures the
percentage change in billion_GDPppp for a 1% change in Population (assuming a log-linear
relationship).

c. Statistical significance

The significance level to determine whether a variable is statistically significant in this case is
set at 5%

d. Breusch-Pagan test
The Breusch-Pagan test, is a statistical test used in econometrics to assess the presence of
heteroscedasticity in a regression model (Bobbit 2020).

The Breusch-Pagan has a null hypothesis: H0 = The model has heteroskedasticity

Result:
BP = 18.096, df = 5, p-value = 0.002828
As the p-value is 0.002828 <0.05, thus there is enough evidence to reject the H0, hence the
model has no heteroskedasticity problem.

e. HAC Robust Errors

With the HAC Robust Error there are some significant changes in the SE of all explanatory
variables, and usually present larger values. This is due to they accommodate the increased
variability caused by heteroscedasticity. Unlike classical standard errors, which assume
constant variance of residuals across all levels of the independent variables, robust standard
errors adjust for the varying levels of variance. This adjustment is crucial for downweighting
high-variance observations, ensuring they have less influence on the estimation process,
provide a more conservative estimate of uncertainty, which is essential for valid inference in
the presence of heteroscedasticity. (Wooldrige 2019)

f.
Reverse causality: occurs when the assumed cause and effect in a relationship may actually
be reversed. In the context of GDP per capita (GDPpc) and Trade (% of GDP), reverse
causality suggests that not only does trade affect GDPpc, but GDPpc may also influence
trade. For example, a higher GDPpc may lead to increased trade activities as countries with
higher income levels tend to engage more in international trade. However, higher trade levels
can also contribute to economic growth by boosting exports and imports, thus increasing
GDPpc.

To address reverse causality, instrumental variable (IV) regression can be used. IV regression
utilizes an instrument, a variable that is correlated with the endogenous explanatory variable
(Trade (% of GDP)) but not directly related to the dependent variable (GDPpc), to estimate
causal effects.

A potential instrumental variable for Trade (% of GDP) could be "Geographic Distance to

Major Markets". This variable captures the geographical proximity of a country to major
trading partners and is likely correlated with Trade (% of GDP) but not directly related to
GDPpc.

The criteria for a good instrumental variable

 ● Relevance: The instrument should be correlated with the endogenous variable (Trade
(% of GDP)).
● Exogeneity: The instrument should be unrelated to the error term in the regression
model.
● Exclusion Restriction: The instrument should affect GDPpc only through its effect
on Trade (% of GDP) and not through any other channels.

g.
From model 1 result, FDI, population, and industrial output emerge as statistically significant
determinants of GDP. There are several policy suggestion to increase GDP

Promoting Industrial Growth : To harness this potential, policymakers should focus on

creating an environment conducive to industrial expansion and innovation. Investing in
infrastructure, such as transportation and energy, is essential to support industrial activities
(UNIDO, 2020). Additionally, fostering technological innovation through research and
development (R&D) incentives can enhance productivity and competitiveness. Policies that
support small and medium-sized enterprises (SMEs) are also vital, as SMEs often drive job
creation and economic diversification (OECD, 2017).

Strategic Management of Foreign Direct Investment (FDI) The negative coefficient for FDI in
our model highlights the complexity of its impact on GDP, suggesting that not all FDI is
equally beneficial. To maximize the positive effects of FDI, policymakers should adopt a
strategic approach that targets high-quality investments. This includes attracting FDI that
promotes technology transfer, creates skilled jobs, chain (Javorcik 2004). Establishing clear
guidelines and incentives for FDI can ensure alignment with national development goals. For
instance, offering tax breaks or subsidies for foreign companies that invest in high-tech
industries or R&D can drive economic growth (OECD 2015). Additionally, improving the
business environment by enhancing legal frameworks and ensuring political stability can
make a country more attractive to investors.
h. R code
install.packages("car")
install.packages("ggplot2")
install.packages("lmtest")
install.packages("VIM")
install.packages("stargazer")
install.packages("sandwich")
library(VIM)
library(lmtest)
library(car)
library(ggplot2)
library(stargazer)

Data <- dt2019

View(Data)

#Data re-create

Data$GDPppp <- Data$`GDP, PPP (constant 2017 international $)`

Data$FDI <-Data$`Foreign direct investment, net (BoP, current US$)`
Data$ExpenEdu <- Data$`Government expenditure on education, total (% of GDP) `
Data$Trade <- Data$`Trade (% of GDP)`
Data$Pop <- Data$`Population, total`
Data$Industry <-Data$`Industry (including construction), value added (current US$)`

#Replacing missing values using K-nearest Neighbors imputation

cleaned_data <- Data[complete.cases(Data[c("GDPppp", "FDI",
"ExpenEdu","Trade","Pop","Industry")]), ]
imputed_data <- kNN(Data, k = 5)
View(imputed_data)
# Select final dataset for analysis
Data_final <- imputed_data[, c("GDPppp", "FDI", "ExpenEdu","Trade","Pop","Industry")]
View(Data_final)

#Data rescaling

Data_final$billion_GDPppp <-Data$GDPppp/1000000000
Data_final$billion_FDI <-Data$FDI/1000000000
Data_final$billion_Industry <-Data$Industry/1000000000

########Part a)
#Scatter Plot
# Linear regression for FDI
FDI_reg <- lm(billion_GDPppp ~ billion_FDI, data = Data_final)
summary(FDI_reg)
plot(Data_final$billion_FDI, Data_final$billion_GDPppp, main = "FDI vs GDPppp", xlab =
"GDPppp", ylab = "FDI")
abline(FDI_reg, col = "red")

# Linear regression for ExpenEdu

ExpenEdu_reg <- lm(billion_GDPppp ~ ExpenEdu, data = Data_final)
summary(ExpenEdu_reg)
plot(Data_final$ExpenEdu, Data_final$billion_GDPppp, main = "ExpenEdu vs GDPppp",
xlab = "ExpenEdu", ylab = "GDPppp")
abline(ExpenEdu_reg, col = "red")

# Linear regression for Trade

Trade_reg <- lm(billion_GDPppp ~ Trade, data = Data_final)
summary(Trade_reg)
plot(Data_final$Trade, Data_final$billion_GDPppp, main = "Trade vs GDPppp", xlab =
"Trade", ylab = "GDPppp")
abline(Trade_reg, col = "red")

# Linear regression for Pop

Pop_reg <- lm(billion_GDPppp ~ Pop, data = Data_final)
summary(Pop_reg)
plot(Data_final$Pop, Data_final$billion_GDPppp, main = "Pop vs GDPppp", xlab =
"Population", ylab = "GDPppp")
abline(Pop_reg, col = "red")

# Linear regression for billion_Industry

Industry_reg <- lm(billion_GDPppp ~ billion_Industry, data = Data_final)
summary(Industry_reg)
plot(Data_final$billion_Industry, Data_final$billion_GDPppp, main = "Industry vs
GDPppp", xlab = "Industry (billion)", ylab = "GDPppp")
abline(Industry_reg, col = "red")

##Multiple linear regression

model1 <- lm(billion_GDPppp~billion_FDI + ExpenEdu + Trade + log(Pop) +
billion_Industry, data = Data_final)
model1_results <- summary(model1)
print(model1_results)
stargazer(model1,type="text")

##Check Multi-collinearity
correlation_matrix <- cor(Data_final[, c("GDPppp", "FDI",
"ExpenEdu","Trade","Pop","Industry")])
rounded_correlation_matrix <- round(correlation_matrix, 3)
print(rounded_correlation_matrix)
#VIF
vif_values <- car::vif(model1)
print(vif_values)

########Part d)
# Perform the Breusch-Pagan test for homoscedasticity
bptest(model1)

########Part e)

#HAC robust errors

library(sandwich)
model1_RobustSE <- coeftest(model1,vcov=vcovHC(model1,type='HC1'))
model2 <- model1_RobustSE
stargazer(model1,model2,type="text")

########Part f)
Data$GDPpc <- Data$`GDP, PPP (constant 2017 international $)`/Data$`Population, total`
Data$log_GDPpc <- log(Data$GDPpc)

model3 <- lm(log_GDPpc~Trade, data = Data)

model3_results <- summary(model3)
print(model3_results)

model4 <- lm(Trade~log_GDPpc, data = Data)

model4_results <- summary(model4)
print(model4_results)
stargazer(model3,model4, type="text")