Uploaded by Mehnaz Mashreen

ECO465 Final paper

Methodology:
The Economic model:
The aim of our paper is to find out the contribution of stock market capitalization and investment to the
growth rate of GDP. Since all our variables are continuous variables, and our model is linear, we are
using OLS method.
GDP growth rate = β0+β1Stockmarket Capitalization+ β2Investment to purchasing power parity ratio + ε
This equation shows the model that we are going to estimate. We know the stock market and
investment sector in Bangladesh is not quite developed. We want to find out regardless of this fact how
much these sectors are contributing to the growth of this country.
Data collection and variables:
The data we have collected is secondary data from FRED (Federal Reserve Economic Data). Our
dependent variable is GDP growth rate and independent variables are Investment and Stock market
capitalization. We could not collect many observations since Bangladesh Economy was not very good
before 1990s. Here are some details about the variables,
Variable
Obs
Mean
InvestPPP
gdpgr
SMCGDP
52
58
25
12.31159
4.271062
11.51529
Std. Dev.
6.745164
3.779178
9.804513
Min
Max
1.765136
-13.97373
1.38023
23.80312
10.95279
34.3251
Table 1: Brief detail of the variables.
In the table we can see that InvestPPP is the data of investment of Bangladesh as a share of the
purchasing power parity. We used this data because it was the best fit for our model. All other
investment data had some shortcomings for example: the investment sector consists of financial and
non-financial investments, of which financial investments are not very prominent in Bangladesh. So,
until recently the investment in financial sector was not significant. So, rather than taking bits of data
from many sources we chose this data. And we can see the highest share of investment in PPP is 23.8%
whereas mean is 12.31%. The other two variables are SMCGDP which is for stock market capitalization
which has a mean of 11.52 and gdpgr represents GDP growth rate which has a mean of 4.27%, 10.95%
being the highest and -13.97% being the lowest (due to liberation war our economy plunged)
Findings:
This following table is our regression output. reg gdpgr SMCGDP InvestPPP
Source
SS
df
MS
Model
Residual
7.40024186
6.13981919
2
15
3.70012093
.409321279
Total
13.5400611
17
.796474179
gdpgr
Coef.
SMCGDP
InvestPPP
_cons
-.0373804
.2118276
1.375186
Std. Err.
.0264398
.0498815
.9230485
t
-1.41
4.25
1.49
Number of obs
F(2, 15)
Prob > F
R-squared
Adj R-squared
Root MSE
P>|t|
0.178
0.001
0.157
=
=
=
=
=
=
18
9.04
0.0027
0.5465
0.4861
.63978
[95% Conf. Interval]
-.0937356
.1055077
-.5922454
.0189748
.3181474
3.342617
Table 2: Regression output.
From the table we can see the value of F statistics is 9.04 which is basically the ratio of MSM and MSR.
The p value is 0.0027 which is less than our alpha value 0.05 that means our independent variables are
reliable for predicting the dependent variable. The value of R squared is 0.5465 that means around
54.65% of the variance in the dependent variable which is GDP growth rate can be explained by our
independent variables which are investment and stock market capitalization. The value of our β0, β1, β2 is
1.375186, -0.0373804 and 0.02118276. We can see our one variable which is investment is significant
with a standard error of 0.0498815. And stock market capitalization is statistically insignificant with a
standard error of 0.0264398.
Now we will do a few tests to see whether the results we got have any heteroskedasticity, autocollinearity, multicollinearity or not.
1. Heteroskedasticity:
Graph Plot for heteroskedasticity:
.2
.1
0
Residuals
-.1
-.2
-.3
1.4
1.5
1.6
Fitted values
1.7
1.8
Fig 1: Check for heteroskedasticity.
On the y axis we have residuals. We don’t see any cluster of data so data is homoscedastic as per this
graphical test.
Breusch-Pagan test:
. hettest , rhs fstat
Breusch-Pagan / Cook-Weisberg test for heteroskedasticity
Ho: Constant variance
Variables: loginv logsmc
F(2 , 15)
Prob > F
=
=
0.10
0.9013
Table 3: Breusch-Pagan test for heteroskedasticity.
The null hypothesis of OLS is constant variance. So here the null hypothesis is constant variance that
means the variance is constant for all data.
We can see P value>0.05 so we cannot reject the null. Hence data is homoscedastic.
2. Auto collinearity:
0
-.1
-.2
-.3
Residuals
.1
.2
We generated residuals using Stata for testing auto-collinearity. In the diagram we plotted residuals
over the year keeping yline at 0.
1990
1995
2000
year
2005
2010
Fig 2: Plotting residuals.
We can see a little positive autocorrelation here, as the error terms are following the pattern of previous
year. For being absolutely sure we will do another test which is Durbin-Watson test.
Durbin-Watson test:
. estat dwatson
Durbin-Watson d-statistic(
3,
18) =
1.227876
Table 4: Durbin-Watson test.
We can see that the value of d is close to 2 so the value of rho ρ would be almost equal to 0, means a
positive autocorrelation.
Solution for auto-correlation:
White-het robust variance:
We do this for eliminating any kind of auto-collinearity and heteroskedasticity in data. We do not have
heteroskedasticity but we have auto-collinearity.
. newey gdpgr SMCGDP InvestPPP, lag(2)
Regression with Newey-West standard errors
maximum lag: 2
gdpgr
Coef.
SMCGDP
InvestPPP
_cons
-.0373804
.2118276
1.375186
Newey-West
Std. Err.
.0201369
.0563276
.9531284
Number of obs
=
F( 2,
15) =
Prob > F
=
t
P>|t|
-1.86
3.76
1.44
0.083
0.002
0.170
18
7.17
0.0065
[95% Conf. Interval]
-.0803013
.0917682
-.6563593
.0055404
.331887
3.406731
Table 5: White -het robust.
We can see after using white het robust the p value for stock market capitalization is 0.082 whereas in
tabe-1 it was 0.178. That means the insignificance f this variable has lessened.
3. Multicollinearity:
. vif
. corr SMCGDP InvestPPP gdpgr
(obs=18)
SMCGDP Invest~P
SMCGDP
InvestPPP
gdpgr
1.0000
0.3794
0.0371
1.0000
0.6972
Variable
VIF
1/VIF
InvestPPP
SMCGDP
1.17
1.17
0.856048
0.856048
Mean VIF
1.17
gdpgr
1.0000
Table 6: Test for Multi-collinearity.
VIF is variance inflation factor. We can see the value of VIF is less than 10, that means there is
no multicollinearity. Also, the values are correlation between the values is very less, there is no
multicollinearity in the data. That means our independent variables are not related to each other.
They do not have any significant impact on the other one.