METHODOLOGY FOR DETERMINING THE RELATIONSHIP BETWEEN STOCK MARKET A...
4.1 Research objectives:
The objective of this thesis is to find out the existence of relationship between stock market development, banking development and economic growth based on panel data.
4.1.1 Research questions:
The overriding questions that this thesis will purport to answer are:
To determine whether there is a relationship between stock market development, banking development and economic growth for developing and developed economies.
To determine whether the impact on economic growth is positive or negative.
To check the robustness of the relationship by the use of different proxies.
To compare where the relationship is more significant; developed or developing economies.
4.1.2 Research hypothesis:
Ho: Stock market impacts on economic growth.
H1: Stock market does not influence economic growth.
4.2 Research design:
In this thesis we assess the relationship between stock market development and economic growth in three groups which comprise of developed countries group, developing countries group and the pooled sample of countries. It is noteworthy that the sample of countries included in a particular type of economy has been derived from International Monetary Fund. The data set covers the period from 1991 to 2008.
4.2.2 Data and measures:
There are basically two types of data; primary and secondary data. Primary data is raw data extracted through observations, surveys and interviews, while secondary data relates to information which has already been gathered by other researches and such data is to be used for this thesis. The data source for the dependent variable, percentage change in GDP per capita, has been derived from International Monetary Fund (IMF). The main source for our independent variables has been obtained from the World Bank’s ‘World Development Indicators’.
Y = f (SMDEX, BANKGDP, CPI, EDUC, EXGDP) (1)
In the above equation, proxy for the dependent variable, Y, has been the per capita growth rate.
SMDEX refers to the stock market development index, where such index comprises of two measures, namely SIZE and LIQUIDITY. These measures of the index have been widely used in previous studies including (Levine, 1996; Rousseau and Wachtel, 2000; and Seetanah, 2008 amongst others). SIZE comprises of the concentration ratio.
Concentration ratio is measured by dividing market capitalisation over GDP. If a market is dominated by only few companies, there is likely to be manipulation in the price formation process. Highly concentrated markets are mostly found in poor economies. Thus market concentration is assumed to be negatively correlated with market size and market liquidity.
LIQUIDITY includes trading value ratio and turnover ratio. The former, which equals the value of listed shares divided by GDP, is taken as the indicator for stock market development. This ratio measures the stock market size, and shows ability to mobilise the capital through diversification of risk. It represents organised trading of firm equity as a share of national output and therefore should positively reflect liquidity in the economy (Garcia, 1999).
The turnover ratio, second component of liquidity, is measured as the value of total shares traded divided by market capitalisation. As per Levine and Zervos (1996), the liquidity of the stock markets may impact on economic activity such that investors are unwilling to relinquish control of their saving for long periods. This ratio measures the ability to trade economically significant positions on the stock market.
BANKGDP shows banking development and our proxy used, domestic credit provided by financial intermediaries to the private sector over GDP, has been followed from Beck and Levine (2003). This indicator has been popularly used since it excludes credit to the public sector and thereby measures more specifically the contribution of financial institution in funding private sector. The study of Boyd and Smith (1996) suggests that banks and stock markets may be complements rather than substitutes to each other, and that of Demirguc-Kunt and Levine (1996a) show that the degree of stock market development is positively related to bank development. Thus it is expected that there is a positive correlation between them and that they grow simultaneously.
Inflation, proxy as the consumer price index, reflects the annual percentage change in the cost to the average consumer of acquiring a basket of goods and services that may be fixed or changed at particular time intervals. Various studies including Fischer (1993) and Barro (1996) concluded that inflation is not good for long term economic growth. Thus it is estimated to have a negative correlation between inflation and economic growth. EDUC demonstrates the quality of human capital by having secondary enrolment ratio as the proxy. Secondary education aims at laying the basics for lifelong learning and human development, regardless of age. Higher education level is expected to bring in more skilled resources and thus better prospects to the country’s growth. EXGDP is the total of export divided by the GDP of the country. It has been used as the proxy for trade openness. This export amount includes the value of merchandise, freight, insurance, transport, travel, royalties, license fees, and other services, such as communication, construction, financial, information, business, personal, and government services. It is noted that this value does not consider compensation of employees and investment income (formerly called factor services) and transfer payments.
yit = β0 + β1smdex it + β2bankgdp it + β3cpi it + β4educ it + β5xgdp it + ε it (2)
where i stands for the different countries in the sample, t denotes the time dimension and µ is the error term. The small letters denotes the natural logarithm of the variables. The model used in this thesis is a linear-logarithmic one. The available panel data set for the twenty two countries is balanced.
4.3.1 Static Panel:
To overcome the limitations in the use of single-equation OLS cross sectional regression model and pooled OLS, Kennedy (2003) advises the use of panel data techniques. In this thesis, there has been the use of panel data analysis which endows regression analysis with both a spatial (twenty two countries) and temporal dimension (eighteen years). Among the main types of panel data analytic models, it includes the fixed effect and random effect models which have been both considered.
184.108.40.206 Fixed effects:
A fixed effects technique is a one that represents the studied quantities in terms of independent variables such that the quantities are treated as non-random. It examines if intercepts vary across groups or time periods. The assumption made under the fixed effects estimates is that the individual specific effect is correlated with the independent variables. Such a model is defined as per the following regression:
yit = αi + β′xit + ε it i = 1,...,N ; t = 1,...,Ti (3)
αi, where i=1,...,N, denotes the constant coefficients specific for each country. They demonstrate that the differences across the studied countries can be explained by differences in the constant term. The validity of the fixed effect will lie in the αi, i=1,...,N, being unequal. This will thereby confirm the existence of significant heterogeneity across countries. That is, it will show that each country’s situation is different from each other.
220.127.116.11 Random effects:
In case of random effects modelling, there includes the presence of a random constant term in the regression. Some of the independent variables are assumed to be rising from random causes. It analyses the differences in error variances. In this case, the model is as follows:
yit = β′xit + ε it i = 1,...,N ; t = 1,...,Ti (4)
ε it reflects the error component disturbances (μi + υit ). The individual specific effects are independent of υit and are normally distributed. μi is specific to a particular observation. For the validity of the random effect, υit which shows the deviation from the constant of the cross-sectional unit (country) has to be uncorrelated with the errors of the variables. The benefit of such model is that it allows for time-invariant variables to be included among the regressors.
18.104.22.168 Hausman specification test:
The main purpose of the Hausman specification test is to determine whether to use the fixed or random effects model. The question lies to whether there is significant correlation between the unobserved person-specific random effects and the explanatory variables. In case there is no such correlation, it will be most appropriate to use the random effects model. However, if there is such a correlation, the random effects model would be inconsistently estimated, and therefore the fixed effects model will be more powerful. The null hypothesis is that there is no correlation.
4.3.2 Dynamic Panel:
Dynamic Panel Data Model is used for the analysis since the GDP per capita depends on its previous values. The idea behind the use of dynamic panel data is not only to control for unobserved cross country heterogeneity but also it allows the inspection of dynamic relations. To examine the relationship between stock market development, bank development and economic growth, the Generalised-Method-of Moments (GMM) estimators developed for dynamic panel models by Arrellano and Bond (1991) has been used. The reason for the use of the GMM is that it, according to Green (1997), yields consistent and efficient estimates in the presence of arbitrary heteroskedasticity. Moreover, the estimates of the Ordinary Least Square (OLS) technique have not been considered because its methods are biased and even inconsistent (Gujarati, 2003).
The growth regression can be modelled as follows:
yi,t - yi,t-1 = αyi,t-1 + β′Xi,t + ηi + εit i = 1,...,N ; t = 1,...,Ti , (5)
where yit denotes the per capita real GDP rate; xit represents the vector of explanatory variables, αt is the period specific intercept terms to capture changes common to all countries. η is an unobserved country-specific effect, ε is the error term, and the subscripts i and t represent country and time period, respectively.
According to Arrellano and Bond (1991), the equation (5) should be differenced such that it becomes as follows:
(yi,t - yi,t-1) – (yi,t-1 - yi,t-2) = α (yi,t-1 - yi,t-2) + β′ (Xi,t - Xi,t-1) + (εi,t - εi,t-1) (6)
In this specification, the country specific effect is dropped out, but a new kind of bias arises since the new term (εi,t - εi,t-1) is correlated with the lagged dependent variable, (yi,t-1 - yi,t-2). Hence, Arellano and Bond (1991) proposed the following moment conditions:
E [ yi,t-s (εi,t - εit-1) ] = 0 for s≥ 2; t = 3,...,T (7)
E [ xi,t-s (εi,t - εit-1) ] = 0 for s≥ 2; t = 3,...,T (8)
The equation (7) and (8) are based under the assumptions that the error term, ε, is not serially correlated, and that the independent variables, X, are weakly exogenous. Bearing these conditions in mind, Arellano and Bond (1991) propose a two-step GMM estimator. Firstly, the error terms are assumed to be independent and homoskedastic across countries and over time. Secondly, the residuals retained in the first step are used to construct a consistent estimate of the variance-covariance matrix, thereby relaxing the assumptions of independence and homoskedasticity. However, the first step GMM estimator will be used since it has demonstrated to yield to more reliable inferences. The asymptotic standards errors from the two step GMM estimator have been found to have a downward bias (Blundell and Bond 1998).
It is noteworthy that the above specification has not been estimated through the use of OLS since it might cause a problem of endogeneity if yt-1 is endogeneous to the error terms through εi,t-1.
22.214.171.124 Sargan specification test:
For the GMM estimator to be consistent, both the instruments chosen from the lagged endogenous and explanatory variables and the assumption that the error terms do not exhibit serial correlation have to be valid. To address these issues, the Sargan specification test suggested by Arellano and Bond (1991), Arellano and Bover (1995), and Blundell and Bond (1998) has been employed. It is a test of over-identifying restrictions such that it tests the overall validity of the instruments by analyzing the sample analogue of the moment conditions used in the estimation process. The Sargan test has the belief that the residuals should be uncorrelated with the set of exogenous variables if the instruments are truly exogenous.
126.96.36.199 Arellano-Bond autocorrelation test:
It is obvious that Sargan test is testing more than just autocorrelation in the errors. Thus, no autocorrelation is necessary but not sufficient to pass this test with reasonable certainty. According to Arellano and Bond (1991) the Sargan test is not as sensitive to autocorrelation as is their autocorrelation test. This implies that the two tests sometimes disagree, with the Sargan test being sensitive to other types of violations of assumptions, but also being less sensitive to particular violations associated with autocorrelation. Pertaining to this, the Arellano-Bond autocorrelation test has been used to check for the presence of any residual autocorrelation. This autocorrelation test is valid under many forms of dynamic panel model estimation, even though it has been presented through a two-step robust estimation by Arellano and Bond.
4.4 Limitations of study:
Theory emphasises that stock markets and banks may play in reducing informational asymmetries and lowering transactions costs. However, there is no direct measures of the degree to which markets and banks in a broad cross-section of countries ameliorate information and transactions costs. Thus, based on the absence of a direct link between theory and measurement, proxy measures of stock market size and liquidity and banking development activity are used to gauge cross-country differences in stock market and bank development.
Furthermore, there has been the use of GMM estimator. However, this difference estimator may contain statistical shortcomings. Our main study lies in examining the cross-country relationship between stock market development and economic growth, which is in fact eliminated in the difference estimator. In cases of persistent explanatory variables, Blundell and Bond (1998) have demonstrated that lagged levels of these variables are weak instruments for the regression equation in differences. This affects the asymptotic and small-sample performance of the difference estimator. It goes on further to mention that differencing may worsen the bias due to measurement errors in variables.
Finally, data being used in our thesis is of secondary type and thus all the shortcomings of such type of data can be noted with, though websites from which the data has been collected is presupposed to be reliable.
Need help with your dissertation methodology?
Our qualified researchers are here to help. Click on the button below to find out more:
In addition to the example methodology above we also have a range of free study materials to help you with your own dissertation: