Based on the theoretical arguments described as aboved, our study mainly investigates the interaction of financial development and economic growth via the channel of capital accumulation.
In line with the objectives as presented in this research, the estimated regression equations on examining the causal finance-growth relationship are suggested as follows:
- First regression function: Effect of financial development (FD) on economic growth (EG):
Examining whether financial development (FD) impacts on economic growth (EG):
EGit = αo + α1FDit + α2FDIit+ α3GOVit+ α4TRAOit + α5GDPGRTHit-1 + εit
- Second regression function: Effect of economic growth (EG) on financial development (FD):
Page 37
Investigating whether economic growth (EG) affects on financial development (FD):
FDit = βo + β1EGit + β2M2/GDPit + β3BANKCREDIT/GDPit + β4 DOMCREDIT/GDPit + β5 GROSSDSAit + σit
It should be further noted that Masten et al., (2008) suggested that the lagged GDP per capita growth should be included in the right hand side of the growth estimation equation in order to capture the persistence of economic growth (Masten et al., 2008).
The explanatory variables and dependent variables are defined in Table 3 as follow:
Table 3: Variable Description:
Variables: Description: Unit
EG Growth rate of real GDP per capita percentage growth rate of GDP GDPGRTHt-1 Lagged one year GDP per capita growth Annual percentage
growth rate of GDP
FD Financial development. Percent of GDP
M2/GDP Ratio of Broad money to GDP Percent of GDP BANKCREDIT/GDP Ratio of credit offered by bank to private
sector to GDP
Percent of GDP DOMCREDIT/GDP Ratio of domestic credit to private sector
to GDP
Percent of GDP
GROSSDSA Gross domestic saving as percentage of GDP
Percent of GDP
FDI Foreign direct investment Percent of GDP
GOV General government consumption Percent of GDP
TRAO Trade openness measured by the sum of import and export of goods
Percent of GDP
Page 38
Country sample (N=22): developing countries from different regions (See Appendix B).
Observed period is from 1990 to 2011 3.2.2 Econometric methodology:
Many approaches are used for investigating the causal relationship between financial development and economic growth. For example, the generalized method of moments with pure cross sectional instrument variables are employed in Levine et. al., (2000);
Zagorchev et. al., (2011); Habibullah and Eng (2006); Granger test within an error correction model (ECM) in Khalifa Al-Yousif (2002); Geweke decomposition test in Calderón and Liu (2003); various multivariate time series analysis in Hassan et al., (2011); dynamic OLS and vector error correction model in Lee and Chang (2009), and method of panel unit root and cointegration tests in Christopoulos and Tsionas (2004).
Most these empirical studies suggested that the variables used in investigating the causal relationship between financial development and economic growth are considered as endogenous (Levine et. al., 2000; Zagorchev et. al., 2011; Habibullah and Eng 2006; Masten et al., 2008). Zagorchev et. al., (2011) and Masten et al., (2008) have further explained that the generalized method of moment (GMM) with instrumental variables can correct for biases caused by inconsistent regression coefficient estimates, which simple OLS may fail to correct in the presence of endogenous regressors. The failure is due to the presence of violation of the orthogonality condition between the error term and explanatory variables (Zagorchev et. al., 2011; Masten et al., 2008).
Following above empirical studies, we use both the generalized method of moment with instrumental variables and the pooled ordinary least squares (OLS) technique to analyze the causal finance-growth relationship in seleted developing countries. The observation period is from 1990 to 2011.
Page 39
We use the panel data because the panel data have two advantages. The first advantage is that the panel data allows us to control for the variables that we cannot observe or measure, and the variables that change over time but not across entities. This means that it accounts for the factor of the individual heterogeneity. The second advantage is that the panel data provides more information on data. By combining time series observation and cross sectional observation, panel data gives more degrees of freedom and more efficiency.
The generalized method of moment was addressed by L. Hansen in his paper published in 1982 (Baum, C. F., Schaffer, M. E., & Stillman, S., 2003). The generalized method of moment with instrumental variables was widely used and a very well known tool among empirical researchers in the last twenty years in controlling for the endogeneity problem (Baum, C. F., Schaffer, M. E., & Stillman, S., 2003; Baum, C. F., Schaffer, M. E., & Stillman, S., 2007). The generalized method of moment with instrumental variable estimation does not require that the error terms in the equations are distributed. This method mainly relies on the assumptions that there is no association between instrumental regressors and the error terms of the equations (Baum, C. F., Schaffer, M. E., & Stillman, S., 2003; Baum, C. F., Schaffer, M. E., &
Stillman, S., 2007; Cameron, A. C., & Trivedi, P. K. 2009).
With an observed annual panel data setting, We begin with the equations on investigating the estimated causal relationship between economic growth and financial development as below:
- First regression function: Effect of financial development (FD) on growth (EG):
Examining whether financial development impacts on economic growth:
EGit= αo + α1FDit+ α2FDIit+ α3GOVit+ α4TRAOit + α5GDPGRTHit-1+ εit (1) - Second regression function: Effect of Economic growth (EG) on Financial
development (FD):
Investigating whether economic growth affects on financial development:
Page 40
FDit = βo + β1EGit + β2M2/GDPit + β3BANKCREDIT/GDPit + β4 DOMCREDIT/GDPit
+ β5 GROSSDSAit+ σit (2) 3.2.2.1 Examining the effect of economic growth on finanacial development
The equation (2) can be expressed in the general form as:
Yi,t = α + βXit + γFit + εit (3) In which,
- Y is finanacial development. This indicator is measured by the sum of the ratio of Broad money to GDP (M2/GDP), Ratio of credit offered by bank to private sector to GDP (BANKCREDIT/GDP), Ratio of domestic credit to private sector to GDP (DOMCREDIT/GDP), and Gross domestic savings as percentage of GDP (GROSSDSA) in country i and year t.
- X is the set of economic growth determinants, including Foreign direct investment (FDI), General government consumption (GOV), Trade openness measured by the sum of import and export of goods (TRAO), and one year lagged GDP per capita growth (GDPGRTHit-1).
- F is the determinants of financial development, including Ratio of Broad money to GDP (M2/GDP), Ratio of credit offered by bank to private sector to GDP (BANKCREDIT/GDP), Ratio of domestic credit to private sector to GDP (DOMCREDIT/GDP), and Gross domestic savings as percentage of GDP (GROSSDSA). ε is the error term, and i and t is represented for country and time, respectively.
In estimating the equation (3), Hao, C. (2006) and Zagorchev et. al., 2011 show that the error term εit is not only assumed to be uncorrelated with Fit, but it also is correlated with regressor Xit. This correlation gives the result from the ordinary least square (OLS) estimator being biased estimates and inconsistent for coefficient estimations (Masten et al., 2008, Cameron, A. C., & Trivedi, P. K. 2009). Therefore, the potential endogeneity of explanatory variables is arised.
Page 41
Regarding to the equation (1) and equation (2), we further notice that above equations are treated as a multiple equations in which a dependent variable is a function of independent variables. The first equation has a group of variables (i.e. FDI, GOV, TRAO, one year lagged GDP growth), but it also has a dependent variable (FD) on the right hand side. Similarly, the second equation has the dependent variable (EG) as the right hand side variable. The variables which appear in the right hand side of the equation, are referred to as exogenous variables. The variables, appearing the right hand side and also have their own equations, are called endogenous variables. For example, in the equation (1), financial development (FD) variable is suspected to be endogenous to economic growth (EG) variable. In the second equation (2), economic growth (EG) variable is is suspected to be endogenous to financial development (FD).
This leads to endogeneity bias problems (Zagorchev et. al., 2011; Habibullah and Eng 2006; Masten et al., 2008). In the presence of endogeneity bias, the ordinary least square (OLS) can produce inconsistent and bias regression coefficient estimations.
Then, OLS estimator can no longer efficiently interpret the causal relationship and the marginal effect on the change of exogenous variables.
Therefore, in order to control for the endogeneity bias, we estimate the equations on the causal relationship between economic growth and financial development with the GMM approach using instrumental variables as suggested by Zagorchev et. al., (2011), Masten et al., (2008) Baum, C. F., Schaffer, M. E., & Stillman, S., 2003, Baum, C. F., Schaffer, M. E., & Stillman, S., 2007, Cameron, A. C., & Trivedi, P. K.
2009. These authors proposed that the generalized method of moments (GMM) with the instrumental variable approach is an optimal solution for controlling the endogeneity bias difficulty, which is arisen in estimating the causal relation between financial development and economic growth (Zagorchev et. al., 2011; Habibullah and Eng 2006; Masten et al., 2008; Hao, C., 2006; Cameron, A. C., & Trivedi, P. K. 2009).
Furthermore
Page 42
Hao, C. (2006) and Zagorchev et. al., 2011 show that a robust and consistent GMM estimation does not require the assumption of the exact distribution of the error terms.
The consistency of GMM estimation mainly depends on the validity and relevance of instruments (Baum et al., 2003; 2007; Cameron, A. C., & Trivedi, P. K. 2009; Levine et al., 2000). The valid and relevant instrument variables (z) should satisfy two assumptions as follows:
- The instruments (z) are correlated with the endogenous explanatory regressors (Cov (z|x) ≠ 0).
- The instruments (z) are not correlated with the error terms in the second stage regression (Cov (z|u)= 0).
According to Levine et al., (2000), the above requirements mean that the changes in instrumental variables (z) lead to the changes in dependent variables through the changes in the explanatory variables. To make it simple, the path diagram can be illustrated as below:
In the path diagram, we suppose y is dependent variables, x is explanatory variables, z is introduced as instrumental variables and the error term u embodies all other elements that determine the dependent variables y (Cameron, A. C., & Trivedi, P. K. 2009).
To investigate the causal impact between financial development and economic growth by using GMM system estimation with instrument variables for controlling the endogeneity problem, there are some econometric specification tests should be done as follows:
- Testing for endogeneity of regressors
z x y
u
Page 43
Baum et.al., (2003) and (2007) propose that it is essential to test of whether ordinary least square (OLS) estimation technique is appropriate and Generalized method of moment with instrument variables approach is an efficient solution for controlling the endogeneity problem. This test is implemented with the Durbin Wu Hausman test statistic. The null hypothesis (H0) indicates that the variables are exogenous. If the result of Durbin-Wu-Hausman test significantly reject the null hypothesis (H0) that the variables are exogenous, this gives the conclusion that the variables are endogenous. Hence, we conclude that there is endogeneity occurred.
It is suggested that the ordinary least square (OLS) estimation technique is no longer appropriate due to biased and inconsistent OLS estimators. Hence, Generalized method of moment with instrument variables approach is more efficient than OLS estimation technique (Baum et.al., 2003; 2007; Cameron, A. C.,
& Trivedi, P. K. 2009).
- Testing for the validity and relevance of instruments
We test the validity of the instrumental variables by using the indicator Hansen- Sargan test of over-identifying restrictions. The null hypothesis (H0) indicates that all instrumental variables are valid. If failure to reject the null hypothesis (H0), we come to the conclusion that the instrument variables are appropriate or valid (Baum et.al., 2003; 2007; Cameron, A. C., & Trivedi, P. K. 2009).
Additionally, we regress the GMM with instrumental variable estimation in this paper by using ivregress commands. These commands are introduced in Stata 12. We also use vce(robust) option at all stages of the estimation process in order to control for the serial correlation and heteroskedasticity problem.
3.2.2.2 Examiming the effect of finanacial development on economic growth To investigate the relationship between financial development and economic growth, we use the estimated equation (1) as follows:
EGit= αo + α1FDit+ α2FDIit+ α3GOVit+ α4TRAOit + α5GDPGRTHit-1+ εit
Page 44
This estimated equation can be expressed in the general form as:
Xi,t = α + βSit + γFit + εit
In which,
- X is denoted as the economic growth measured by the growth rate of real GDP per capita in the country i and year t.
- S is the set of economic growth determinants, including Foreign direct investment (FDI), General government consumption (GOV), Trade openness measured by the sum of import and export of goods (TRAO), and one year lagged GDP per capita growth.
- F is the set of indicators of financial development, including the Ratio of Broad money to GDP (M2/GDP), Ratio of credit offered by bank to the private sector to GDP (BANKCREDIT/GDP), Ratio of domestic credit to private sector to GDP (DOMCREDIT/GDP), and Gross domestic savings as percentage of GDP (GROSSDSA).
We perform the test for whether Generalized method of moment with instrumental variables approach is still an efficient solution for controlling the endogeneity problem.
This test is implemented with the Durbin Wu Hausman (DWH) test statistic. The null hypothesis (H0) indicates that the variables are exogenous. If the result of Durbin-Wu- Hausman test significantly reject the null hypothesis (H0) that the variables are exogenous, this gives the conclusion that the variables are endogenous. However, the result is a robustified DW-h test. p-value (0.7358) > 0.05. we do not reject the null hypothesis (H0) that the variables are exogenous. This indicates that there is no endogeneity occurred. Therefore, it is suggested that Generalized method of moment with instrument variables approach is less efficient than OLS estimators. Hence, OLS estimation technique in this case is more appropriate than Generalized method of moment with instrument variables approach (Baum et.al., 2003; 2007; Cameron, A. C.,
& Trivedi, P. K. 2009).
Page 45
Moreover, due to the Ratio of Broad money to GDP (M2/GDP), Ratio of credit offered by bank to the private sector to GDP (BANKCREDIT/GDP), Ratio of domestic credit to private sector to GDP (DOMCREDIT/GDP), and Gross domestic savings as percentage of GDP (GROSSDSA) are highly correlated amongs themselves for most developing countries in the sample, so we intend to estimate four separate regressions to test the impact of financial development on economic growth as suggested by Hansan et al., (2011).
Additionally, we implement the Ordinary least square estimation in this paper by using regress commands. These commands are introduced in Stata 12. We also use robust and cluster option at all stages of the estimation process in order to control for the serial correlation and heteroskedasticity problem.