FOR DIFFERENT BANK TYPOLOGIES?
6.4 STATISTICAL TESTS FOR DIFFERENCE AMONG BANK TYPOLOGIES
Two types of statistical tests were carried out in addition to the graphical representation above: non-parametric and parametric tests. The non- parametric test applied was the Wilcoxon rank-sum test whereas the t-test was applied as the parametric test43.
The results of the Wilcoxon rank-sum test across these bank typologies were given below with the null hypothesis that there was no difference between two groups. Here, total lending was the ranking variable which was measured by the gross loan as a ratio of GDP.
The Wilcoxon rank-sum test results showed that the null hypothesis was rejected implying that there was difference between public and private banks in terms of total lending. Similar findings were observed for both size and age typology suggesting that there were differences between large and small banks as well as between new and old banks. However, the null for the mode of operation typology was not rejected implying that there was no significant difference between Islamic and conventional banks in terms of total lending.
Table 6.1: Wilcoxon rank-sum test results for bank typologies of ownership, size, mode of operation and age
Typology Ownership Observation Rank
sum Expected Hdifference 0: no between two (unmatched) groups Ownership Private 30 500 570 -2.714 (0.0066) Public 7 203 133 Size Large 6 207 114 z = -3.832 (0.0001) Small 31 496 589 Mode of operation Islamic 30 609 570 1.512 (0.1304) Conventional 7 94 133 Age New 16 426 304 z = 3.740 (0.0002) Old 21 277 399
The results of the t-test also supported the findings of the Wilcoxon rank- sum test, showing that there were differences for most of the bank-specific characteristics. The results of this test were provided here. The results showed that the coefficient of ownership, size and age typologies were significant at 1% level while it was not for the mode of operation.
Table 6.2: t-test results for ownership, size, mode of operation and age
Gross loan Coefficient Standard
error t p > | t | 95% confidence interval Ownership 1.665 0.097 17.08 0.000 1.473 1.856 Size 2.348 0.080 29.32 0.000 2.191 2.505 Mode of operation -0.293 0.120 -2.45 0.015 -0.529 -0.579 Age -1.015 0.084 -12.05 0.000 -1.181 -0.850 6.5 METHODOLOGY
This study used panel data. It was logical to assume that the lending behaviour of banks would be influenced by its past lending and therefore a dynamic model specification was more appropriate to use. Based on the methodologies used before in this area of research and also because of its
advantages over other panel methods (already discussed before), two-step system GMM was considered the most appropriate method of estimation for this type of model. For robustness, the Hausmann test was applied to see whether the fixed effects or the random effects method was more appropriate and then the appropriate method was applied.
The main equation of total lending to be estimated in this study could be written as:
GLit = α0 + α1GLi,t-1 + β1GDPt + β2INTit + β3(FLt)+ β4(FLtBTit)+ εit (6.1)
The above equation explained effect at bank-level on lending where GL was representing gross lending, GDP was showing economic growth, interest rate was given by INT, FL was expressing the financial liberalisation index and BT was showing different bank typologies (ownership, size, mode of operation and age). The interaction terms of FL and BT showed bank typologies based on bank-specific characteristics interacted with the financial liberalisation index. Banks were represented by subscript i and t was showing year. The variables of lagged dependent variable, economic growth and interest rate were the most common variables applied in most of the earlier studies on lending44.
6.6 DATA
The data of this study comprised bank-level information of the banking sector in Bangladesh with annual data for the period of 1997-2011. Version 13.1 of STATA (StataCorp, 2013) was used for the estimation of system GMM to an original panel dataset of 555 observations (NT = 37 15).
6.6.1 Dependent Variable
The aim of this study was to examine the effect of bank-specific characteristics on lending. The lending in real terms was used in this study to reflect the actual scenario. Lending in real terms rather than in nominal
44Some studies have also used inflation but this variable is not used in this study due to
terms has also been used by others before (Hofmann, 2001; Calza et al., 2003; Hulsewig et al., 2004; Brzoza-Brzezina, 2005).
6.6.2 Explanatory Variables
Different studies have used different sets of explanatory variables. Some of them were more common while some were used less frequently across studies. The three most common explanatory variables used in the earlier studies were: economic growth, interest rate and the lagged dependent variable. The definition of all these variables and their measurement were given in detail in Appendix 6.1.
Economic Growth: It was expected that if there was economic growth, there would be higher demand for investment and also increased demand for loan. This was mainly due to the fact of favourable economic conditions. Therefore, economic growth should affect lending positively. This was also observed in earlier empirical studies (Cottarelli et al., 2003; Kiss et al., 2006; Kraft, 2006; Gattin-Turkalj et al., 2007; Brissimis et al., 2014). To capture economic growth, real GDP was used in this study.
Interest Rate: The rate of interest was another variable that was frequently employed in studies of lending. It was expected to have a negative relationship with lending since lower interest rate should increase the demand for credit and vice versa (Egert et al., 2006). In this study, to capture the effect of interest rate, interest rate in real terms was taken which was calculated by deducting the current inflation from the nominal interest rate. To convert interest rate into real terms, both CPI and GDP deflator were used. Results using the real interest rate using CPI are presented in the main text while the other measure of real interest rate is given in the appendix (in Appendix 6.3).
Lagged Dependent Variable: Lag of the dependent variable was included in this model with an aim to capture and account for the persistence of lending from the earlier period. It was expected to have a positive relationship with the dependent variable of lending. This was also
employed in earlier studies and was found to be positively affecting lending (e.g. Gattin-Turkalj et al., 2007).
Financial Liberalisation: Since financial liberalisation took place in most of the economies around the 1990s, the impact of it was part of some of the studies of lending. As the liberalisation process was initiated at the backdrop of financial repression and was proposed to remove various credit restrictions to ensure the free flow of credit, it was expected that there would be a positive relationship between liberalisation and lending. Since it was a continuous and multi-faceted process (Bandiera et al., 2000), the results could be misleading if a dummy variable or only a single variable was used to represent this versatile process.
Therefore, as described in the previous chapters, to address the process in a more comprehensive way, an index of financial liberalisation was created on the basis of the earlier works. The index used in this study was mainly based on the work of Abiad et al. (2010). Although most studies had either used a dummy or a single indicator of liberalisation, the use of index to appropriately capture the process of liberalisation was not uncommon. Cottarelli et al. (2003) used a similar index in their study of CEEC countries.
Bank-specific Characteristics: Different bank-specific characteristics could play a role in lending. These included bank ownership, size, mode of operation and age (discussed in detail in section 6.2). Summarily it could be said that there could be differences in the lending behaviour of banks according to these characteristics and it would be interesting and worthwhile to see if and how significantly these characteristics affected bank lending.
6.6.3 Sources of Data
Like the previous empirical chapters, Bankscope was the main source of data of this chapter. Data of all banks were not always available for full 15 years (detailed description of data availability has been given earlier in