THE DETERMINANTS OF COMMERCIAL BANK CREDIT SUPPLY TO THE BUSINESS SECTOR IN BOTSWANA: AN ARDL BOUNDS TESTING APPROACH

Transcription

1 Journal of Social and Economic Policy, Vol. 13, No. 2, December 2016, THE DETERMINANTS OF COMMERCIAL BANK CREDIT SUPPLY TO THE BUSINESS SECTOR IN BOTSWANA: AN ARDL BOUNDS TESTING APPROACH GOFAMODIMO SECHELE*, OBONYE GALEBOTSWE** AND MOGALE M. NTSOSA*** Abstract: Commercial banks in Botswana have over the years tended to hold low roortions of their assets as business credit. For a financial system that can be characterized as bankdeendent as is the case in Botswana, this can have non-trivial consequences on investment and economic growth. This study alies autoregressive distributed lag (ARDL) bounds testing aroach to Cointegration to investigate the main determinants of bank credit suly to the business sector in Botswana. The results show deosit levels, lending rate and Bank of Botswana Certificates (BoBCs) are the main determinants of bank credit suly to the business sector, both in the short- and long run. Increase in deosit levels and lending rates are associated with increases in bank credit suly. This indicates banks deendence on deosits as sources of loanable funds and the non-existence of credit rationing in the loans market. By contrast, increase in BoBCs is associated with declines in bank credit to the business sector. This confirms the view that BoBCs could be resonsible for low intermediation, esecially to the business sector. The study therefore recommends that efforts to romote business sector credit should target savings mobilization and reduction of BoBCs suly to the commercial banks. Keywords: Bank credit suly, business sector credit, ARDL Bounds-testing, Botswana INTRODUCTION Although it has been observed that economies are rone to cycles of boom and bust in credit markets, the 2007 financial crisis and the resultant economic recession has reminded the world that disrutions in credit markets can have significant effects on the real economy. Excessive bank credit contraction may damen economic activity and thus lead to unemloyment and loss of outut. Overexansion of credit may lead to mismatch of the loans maturity and the investments maturity, which may in turn result in failure to reay the loans. This may lead to a banking crisis and ultimately to a financial crisis. Therefore, credit develoment may contain imortant information for analyzing and forecasting economic activity. Studies carried out in develoed economies attributed financial crises to disrutions in credit markets. However, the olicy imlications of these results are not necessarily true for develoing economies because of the differences in the structure of the economies. Amongst other things, under develoment of caital markets in develoing economies imlies that bank credit remains the main source of investment funds for rivate entities. This means that develoing economies are more rone to financial crises if credit develoments are not well managed. * Deartment of Entrereneurshi, BA ISAGO University, Francistown, Botswana ** Deartment of Economics, University of Botswana, Gaborone, Botswana, obonye.galebotswe@moii.ub.bw *** Deartment of Economics, University of Botswana, Gaborone, Botswana

2 102 GOFAMODIMO SECHELE, OBONYE GALEBOTSWE AND MOGALE M. NTSOSA Notwithstanding, similar studies done in other develoing economies, articularly in Africa, have indicated that there are structural differences amongst develoing economies (see e.g., Guo and Steanyan, 2011). Therefore, the results arrived at are country-secific and not necessarily true for another country of the same income bracket. In this regard, it is imortant to carry out a study to investigate the factors that determine credit suly to the business sector in Botswana and not infer from the conclusions of studies done in other countries. The rest of the article is organized as follows. The next section gives an overview of the evolution of bank credit in Botswana. Section 3 discusses some of the literature on bank credit suly. This is followed by a discussion of the econometric methods used in the data analysis. Section 5 resents the estimation results and their analysis. The last section concludes the study and offers some olicy suggestions. OVERVIEW OF COMMERCIAL BANKS LOANS AND ADVANCES IN BOTSWANA Figure 1 shows the ercentage of commercial banks loans and advances by maturity for December 1994 and December From Figure 1 we notice that in 1994, the share of loans with short-term maturities (e.g., Overdrafts, 1-2yearsand 2-3years) was higher than the share of loans with long-term maturities. Notably, more than three quarters of commercial bank credit was issued on short-term basis in Overdrafts had the largest share followed by 1-2 years and 3-5 years maturity loans. Still in 1994, less than a quarter of credit was directed to long-term maturity loans. Loans with a smaller credit share are 5-7 years, 7-10 years and loans with maturities of 10 years and longer. On the basis of the foregoing, short-term credit dominated long-term credit in This may be the case that commercial banks referred lending on short-term basis than long-term, hence commercial banks were reluctant to lend to businesses since businesses require long-term funds for investment uroses. However in 2012, the share of credit for loans with short-term maturities decreased. A larger share of credit was directed to loans with the following maturities; 3-5 years, 5-7 years, and 10 years and longer, where 3-5 years maturities had the largest share. From Figure 1, we observe a decrease in the share of short-term maturity loans which in 1994 took the largest share of credit. Overdraft for examle, had the largest share of 31.3% in 1994, however in 2012 overdraft had a share of 11.6%. This gives an insight of the changes in the lending reference of commercial banks overtime. Since the share of loans with long-term maturity dominates the share of loans with short-term maturity, we can conclude that commercial banks refer lending to businesses since businesses generally require long-term credit for investment uroses. Figure 2 shows that the ercentage of domestic credit to GDP has generally been increasing in Botswana from 1994 to From 1994 until 1997, the ercentage of domestic credit over GDP was almost constant at 5%. After 1997, this ercentage grew raidly to around 31% in It then increased slightly from 2009 to The stee uward trajectory is then noticed from This indicates that the ercentage of domestic credit to GDP increased at a higher rate from 2011 onwards. However, it is worth noting that credit to GDP ratio remained below 50% between 1994 and 2012.

3 THE DETERMINANTS OF COMMERCIAL BANK CREDIT SUPPLY TO THE BUSINESS SECTOR Figure1: Percentage of Commercial Banks Loans and Advances by Maturity Source: Authors comutation based on data obtained from Bank of Botswana Annual Reorts Figure 2: Percentage of credit over GDP Source: Authors comutation based on data obtained from Bank of Botswana Annual Reorts REVIEW OF SOME RECENT EMPIRICAL LITERATURE Sharma and Gounder (2012) identified lending rates, inflation, total deosit, stock market develoment and GDP as the main determinants of bank credit suly to the business sector. This study was carried out in 6 South Pacific Island countries namely; Fiji, Paua New Guinea, Solomon Islands, Vanuatu, Samoa and Tango. According to Sharma and Gounder (2012) an increase in lending rates results in decreased credit suly to the business sector. Furthermore, an increase in deosits and develoment

4 104 GOFAMODIMO SECHELE, OBONYE GALEBOTSWE AND MOGALE M. NTSOSA of the stock market will lead to an increase in bank credit suly to the business sector. Brown et al (2012) suggested that the amount of bank deosits measure the lending caacity of the bank hence the suly of bank credit to the business sector. Using a Probit and Ordinary Least Squares (OLS) methods they found that high deosit rates influence savings ositively in Eastern and Western Euroe. In addition, they found a negative relationshi between cometition in the banking sector and bank credit suly to the business sector in Eastern Euroean countries. Herrera et al (2012) ointed out that the level of lending rates is in a way a credit rice on credit suly. Using a disequilibrium model, they found a negative relationshi between bank credit suly and lending rates in Egyt. Their results also show a negative relationshi between treasury-bills and bank credit suly. They argue that the negative relationshi between treasury-bills and bank credit indicates the tradeoff that exists between these two variables. According to Abuka and Egesa (2007) lending rates, credit to government and er caita GDP have a significant imact on the level of bank credit suly to the business sector. This study was carried out in the East African Community (Burundi, Kenya, Rwanda, Tanzania and Uganda). The study emloyed an Error Correction Model and found that high lending rates reduce the level of bank credit suly to the business sector. Government credit was identified to be negatively related to bank credit suly to the business sector. They argue that banks always refer to lend to the government because the robability of government loan default is lower comared to that of rivate entities. Per caita GDP for each of the 5 countries had a ositive relationshi with bank credit. Iyoboi (2008) used an ARDL bound test and an ECM reresentation of the ARDL to arrive at the conclusion that stock market develoment leads to increased credit suly in Nigeria. He argued that develoed stock markets have a significant influence on the lending interest rates and for that reason they remain a source of cometition to the banks. The study ointed out that a develoed stock market enhances liquidity in the economy through the everyday buying and selling of stocks and bonds. Therefore, in order to stay in business, banks will automatically increase credit suly to the business sector by lowering lending interest rates. ANALYTICAL FRAMEWORK AND ESTIMATION METHODOLOGY In examining the determinants of credit suly, economists have generally relied on the standard loan suly function in which loan suly is exlained by the following equation. L s = (, i,) D (1- ) (1) where L s is loan suly, is interest rate on loans, i is the interest rate on bonds, D is deosits, is the reserve requirement ratio. This model has often been secified in its log-linear form and extended to include other variables that ast studies have found to influence loan suly, such as outut and variables reresenting the structure of the

5 THE DETERMINANTS OF COMMERCIAL BANK CREDIT SUPPLY TO THE BUSINESS SECTOR banking sector or degree of financial deeening. This study follows Iyoboi (2008) and estimates the following log-linear model: Bcr t = GDP t + 2 Lr t + 3 De t + 4 Hcr t + 5 BOBC t + 6 Ms t + t (2) where 0 is the drift arameter, t is the white noise disturbance term, i: i = 1, 2, 3, 4, 5, 6 are the coefficients of the resective indeendent variables, Bcr is business credit, GDP is Gross domestic roduct, Lr is lending rate, De is deosits, Hcr is household credit, BOBC is Bank of Botswana Certificates, Ms is recirocal of numbers and it s a roxy for cometition in the banking sector. All variables are in real terms and natural logarithm excet lending rate and recirocal of numbers which are in ercentage and real terms, resectively. Business credit is the deendent variable in the model. It is measured as the amount of loans given out by commercial banks to businesses in millions of Pula 4. The reason for using bank credit is that banks are the main institutions for credit suly to businesses in Botswana (Bank of Botswana Annual Reort, 2012). Gross domestic roduct is measured in millions of Pula. An increase in GDP is exected to increase credit suly to the business sector. Therefore, a ositive sign is exected for this variable. This variable was used for examle by Hofmann (2001) for Euroe, and Abuka & Egesa (2007) for East African countries. Lending rate is the interest rate at which commercial banks lend to their rime business customers. Standard suly theory suggests that banks would increase credit suly in resonse to the increase in the lending rate. However, according to Bernanke and Blinder (1988) and Stiglitz and Weiss (1981) an increase in lending rates may lead to decrease in credit suly to the business sector because when lending rates are high banks emloy credit rationing measures which decrease the amount of credit suly to businesses, and vice-versa. Therefore the coefficient on lending rate may be negative or ositive. Deosit is the total amount of funds available for banks to loan out, quantified in millions of ula. Using a simlified balance sheet, Bernanke and Blinder (1988) showed that the amount of deosits indicate the lending caacity of the bank. This means that an increase in deosits is exected to increase credit suly, and vice-versa. Therefore a ositive sign is exected for deosits. Household credit is the amount of credit sulied by banks to households, measured in millions of ula. The trends in household credit and business credit show that there might be some level of substitution between business credit and household credit in Botswana. Therefore, a negative sign is exected for this variable. Bank of Botswana Certificates is the volume of Bank of Botswana Certificates (BOBCs) sold in the money markets, in millions of ula. In articular, the study uses the volume of BOBCs held by commercial banks as oosed to the volume held by rivate citizens or their total. According to Bernanke and Blinder (1988) there is a negative relationshi between loan suly and investment in government bonds. Therefore, BOBCs is exected to have a negative imact on business credit. Recirocal of numbers is used as a measure of bank concentration in Botswana. It is calculated as 1/n where n is the number of banks oerating in Botswana. It is exected to have a negative sign, suggesting that an increase in the number of commercial banks will increase banking sector cometition, resulting in an increase in credit suly.

6 106 GOFAMODIMO SECHELE, OBONYE GALEBOTSWE AND MOGALE M. NTSOSA Admittedly, this is not the commonly used measure of concentration mainly due to the fact that it attaches no weights to firm size, which is its main weakness. However the study adoted this measure because of insufficient data, which made it difficult to construct and use better measures of cometition such as the Hirschmann-Herfindahl Index. Equation 2 reresents the long run equilibrium relationshi between bank credit suly to the business sector and its main determinants. However, the main urose of this study is to analyze both long run and short run imacts of these variables on bank credit suly. In order to achieve this, the study adots the autoregressive distributed lag (ARDL) bounds testing aroach to co- integration develoed by Pesaran and Shin (1999). The ARDL has advantages over other methods of cointegration such as the two-ste Engle-Granger (1987) and the Johansen full Information Maximum likelihood-based aroach (1988) because it can be estimated regardless of the level of integration of the variables i.e. I(0) and I(1). In addition, ARDL aroach can be used in small samle sizes whereas the other methods mentioned above are unreliable in small samles (Pesaran et al, 2001). Furthermore, ARDL allows for different number of lags for different variables. We reformulate equation (2) as an ARDL model as follows: Bcr Bcr GDP k Lr De Hcr t t i t i i t i i t i i t i i t i i 1 i 0 i 0 i 0 i BoBC Ms Bcr GDP Lr De i t i i t i t t t t i 0 i 0 Hcr BoBC Ms (3) s t 1 6 t 1 7 t 1 t where P is the lag length, 0 is a drift arameter, i, i, i, i, i, i, and i are the short run dynamic coefficients, i : i = 1, 2, 3, 4, 5, 6, 7 are the long run multiliers, is the first difference oerator, t is the white noise disturbance term. The other variables are as defined before. Equation 3 is used to test for level relationshis amongst the model variables, which is carried out using the Wald test. The comuted F-value from the Wald test is comared with the critical values tabulated by Pesaran et al. (2001). We test the null hyothesis that the variables are not co-integrated against the alternative hyothesis of co-integration. That is; H o : 1 = 2 = = 7 = 0 null hyothesis, against H 1 : alternative hyothesis. If there is evidence of a long run relationshi we estimate the long run model using the following ARDL secification: Bcr Bcr GDP Lr De Hcr t t i t i i t i i t i i t i i t i i 1 i 0 i 0 i 0 i 0 BoBC Ms i t i i t i t i 0 i 0 (4)

7 THE DETERMINANTS OF COMMERCIAL BANK CREDIT SUPPLY TO THE BUSINESS SECTOR All variables in this model are as defined before. The error correction reresentation of the ARDL secification is then used to estimate the short run coefficients. It is secified as follows: Bcr Bcr GDP Lr De Hcr t t i t i i t i i t i i t i i t i i 1 i 0 i 0 i 0 i 0 (5) BoBC Ms ECT u i t i i t i t 1 t i 0 i 0 where t is a drift arameter, ECTt-1 is the error correction term, is the coefficient of adjustment to equilibrium. ut is the white noise disturbance term. All other variables are as defined before. It should be ointed out that since the ARDL is based on a single equation aroach, it may not be able to correct for the roblem of endogeneity, thus yielding inefficient estimates of both the short run and long run coefficients. In this study however, we do not exect this roblem since the effect of the deendent variables on indeendent variables is very small in Botswana, if it exists at all. That is, the indeendent variables are assumed to be weakly exogenous. Data and estimation results This study uses 1994 to 2012 quarterly time series data. The choice of the eriod of study is restricted by the fact that quarterly GDP data in constant 2006 rices is available starting from 1994 in Bank of Botswana Annual Reorts. The data sources include Bank of Botswana Annual Reorts and Statistics Botswana ublications. The data was analyzed using E-views statistical ackage. Stationary/Unit root test Since the alication of the bounds test rocedure requires that the underlying variables be either integrated of order zero I(0) or one I(1), and none be of order two I(2), an investigation of the time series roerties of the variables used in the model is carried out. The Augmented Dickey- Fuller (ADF) and the Phillis-Perron (PP) unit root tests were used to test for stationarity of the data. The ADF test is conducted under the assumtion that the error terms are correlated (Gujarati, 2009). The PP was develoed by Phillis & Perron (1988) and it uses nonarametric statistical methods to take care of the serial correlation in the error terms without adding lagged difference terms, hence it is believed to be more reliable than the ADF. However, the asymtotic distributions of the two tests are the same (Green, 2008). Table A1 in the aendices resents the unit root test results including intercet and trend whereas Table A2 resents the unit root test results including trend only. The ADF unit root test results including intercet and trend show that real GDP is the only stationary variable at levels, while all other variables are stationary at first difference. On the other hand, the PP unit root test results, also resented in Table A1, show that GDP and Lrate are stationary at levels whereas other variables are stationary at first difference. Although the results of the ADF and PP are different in terms of

8 108 GOFAMODIMO SECHELE, OBONYE GALEBOTSWE AND MOGALE M. NTSOSA magnitude, the end conclusion drawn from the results of the two aroaches is the same, excet for Lrate. Under the ADF Lrate is I(1) whereas under the PP Lrate is I(0). From Table A2, the conclusions on the ADF and PP unit root test results are consistent for all variables excet Hcred and Lrate. Hcred and Lrate are I(1) when using ADF and I(0) when using the PP. ARDL Bounds Testing and Cointegration After establishing that the model variables are a combination of I(0) and I(1), the next ste is to estimate equation (3) and then test for long run equilibrium relationshis. The equation was first estimated with lags of 2 as selected by the Schwarz Bayesian criterion. Then Hendry (1995) general-to-secific aroach was alied to obtain the following arsimonious model. To establish how well the model is erforming residual diagnostic tests were conducted. The Breusch-Godfrey LM test results show that there is no serial correlation in the residuals of the model. The results from the ARCH and the White heteroskedasticity test show that the error terms of the variables are consistent, imlying that there is no roblem of heteroskedasticity in the model. The model is also stable as shown by the cumulative sum of recursive residuals (CUSUM) and the CUSUM of squares (CUSUMSQ) tests resented in Figure A1 in the aendices. The results of the JB statistic resented in Figure A3 in aendices show that the JB statistic X 2 (2) = 0.91 is less than the critical value of the chi-square (5.99), therefore we fail to reject the null hyothesis of normality. From the above long run model obtained after the general-to-secific rocedure, we carry out the bound test to determine the existence of a long run relationshi between the remaining variables. The results of the Wald test are as resented in Table 2. Table 1 ARDL (1, 0, 0, 1) Deendent: D(LBUSCRED) Variable Coefficient t-statistic LBCR(-1) *** LHCR(-1) * LBOBC(-1) *** LDEPOS(-1) *** LR(-1) * MS(-1) *** D(LR) ** D(LBOBC) *** D(LDEP) ** D(LDEP(-1)) *** *,**,*** means significant at 10%, 5% and 1%, resectively Adjusted R-squared Breusch-Godfrey LM test F-statistics: (0.1918) ARCH test F-statistics: (0.0577) White test F-statistics: (0.2607) Jarque-Bera test JB statistic: (0.6349) Source: Author s comutation Note: Values in arenthesis () are robability values.

9 THE DETERMINANTS OF COMMERCIAL BANK CREDIT SUPPLY TO THE BUSINESS SECTOR Table 2 Wald test for Cointegration Wald Test: Equation: ARDL model Test Statistic Value df Probability F-statistic (5, 64) Chi-square The calculated F-value (9.93) is greater than the uer bound critical value (3.24) at 5% level of significance. The critical value is obtained from Table CI (i) Case I in Pesaran et al (2001) resented in the aendices. In this case the number of regressors (k) is six i.e. K= 6. The results suggest that we reject the null hyothesis of no co-integration and conclude that there is a long run relationshi between the variables. Long run relationshis After establishing existence of equilibrium relationshis, the next ste is to estimate the long run equilibrium relationshi (equation 4) and the results are resented in Table 3. Table 3 Estimation of long run relationshis Deendent: LBUSCRED Variable Coefficient t-statistic LBCR(-1) *** LR *** LR(-1) *** LBOBC *** LDEP ** LDEP(-1) *,**,*** means significant at 10%, 5% and 1%, resectively Source: Authors comutation FigureA1: ARDL model CUSUM of Squares

10 110 GOFAMODIMO SECHELE, OBONYE GALEBOTSWE AND MOGALE M. NTSOSA Figure A2: ECM model CUSUM of Squares Figure A3: ARDL model normality test From the long run coefficient results we calculate the long run equilibrium elasticities obtained by dividing i 0 ji by 1 i 1 1i where i 0 ji is the sum of the coefficients of the lags for the indeendent variables and i 1 is the sum of the 1i coefficients for the lags of the deendent variable.

11 THE DETERMINANTS OF COMMERCIAL BANK CREDIT SUPPLY TO THE BUSINESS SECTOR Table 4 Comutation of the long run equilibrium elasticities Variable LBCR LR LBOBC DEP Coefficients *** 0.055*** *** ** Long run elasticities *** *** ** *,**,*** means significant at 10%, 5% and 1%, resectively Source: Author s comutation The results show that a 1% increase in the lending rate will result in 0.20% increase in business credit. This result is consistent with standard economic theory which suggests that credit suly will increase when lending rate increases because lending rate is the rice of credit. The result differs from those of Sharma and Grounder (2012) for South Pacific Island and Abuka and Egesa (2007) for East Africa who found that an increase in lending rates decreases credit suly to the business sector. This suggests that credit rationing was not a roblem in Botswana s credit market during the samle eriod. The results also show that a 1% increase in Bank of Botswana Certificates held by commercial banks reduces credit suly to the business sector by 0.35%. The result is consistent with our rior exectations and the findings by Barajas and Steiner (2002) and Herrera et al (2012) for Latin America and Egyt, resectively. These studies found a negative relationshi between investment in government bonds by banks and bank credit suly, ointing out that government bonds crowd-out credit suly. In addition, the results suort the argument by Bernanke and Blinder (1988) who ointed out that there is a trade-off between loan suly and investment in bonds. Furthermore a 1% increase in deosits is associated with a 0.26% decrease in credit suly to the business sector. This result is counterintuitive as the exectation is that a rise in deosits, by increasing loanable funds, should motivate banks to lend out more to the business sector. Short run Dynamics The next ste considers the short run dynamic relationshis of the model s variables. The results of the error correction model after emloying the general-to-secific rocedure are given in Table 5. In order to insure that the results obtained from the analysis are reliable to make inference from, we carry out diagnostic tests to check the model for adequacy. The Breusch-Godfrey LM test show an F-statistic with robability value of 0.84, which means that we fail to reject the null hyothesis of no serial correlation, indicating that the model does not suffer from serial correlation. The ARCH test show F-statistic with robability value of 0.30 and the White test show F-statistic with robability value of Therefore, we fail to reject the null hyothesis of homoskedasticity and conclude that the model does not suffer from the roblem of heteroskedasticity (inconsistency of the variance of the error terms). Normality results resented in Figure A4 in aendices show that the JB statistic X 2 (2) = 3.10 is less than the critical value of the chi-square (5.99). Therefore we fail to

12 112 GOFAMODIMO SECHELE, OBONYE GALEBOTSWE AND MOGALE M. NTSOSA Figure A4: ECM normality test reject the null hyothesis of normality. In addition, the Ramsey RESET test does not reject the null hyothesis of no missecification as shown by robability of Furthermore, the CUSUM of squares test resented in Figure A2 in the aendices shows that the model is stable. The adjusted R-square shows that 52 % of the variation in the deendent variable is exlained by the variation in the indeendent variables. In addition, the error correction term (ECM (-1)) carries the correct sign and it is statistically significant. Table 5 Estimation of the ECM Deendent: D(LBCR) Variable Coefficient t- Statistic D(LBCR(-1)) *** D(LR) *** D(LR(-1)) *** D(LBOBC) *** D(LDEP) ** D(LDEP(-1)) *** ECM(-1) *** *,**,*** means significant at 10%, 5% and 1%, resectively Adjusted R-squared Breusch-Godfrey LM test F-statistic: (0.8446) ARCH test F-statistic: (0.3020) White test F-statistic: (0.3524) Ramsey RESET F-statistic: (0.1641) Jarque-Bera test JB statistic: (0.2118) Source: Note: Authors comutation Values in arenthesis ( ) are robability values.

13 THE DETERMINANTS OF COMMERCIAL BANK CREDIT SUPPLY TO THE BUSINESS SECTOR The sign and magnitude of the error correction term suggests that about 58% of the disequilibrium is corrected within the first quarter. The results show that business credit is ositively and significantly affected by its revious level. Furthermore, just like in the long run, increases in lending rate are associated with increases in business credit. The coefficient on Bank of Botswana Certificates is significant and shows a negative sign as was exected, suggesting that increase in the amount of Bank of Botswana Certificates held by commercial banks will result in a decrease in business credit. The nature of this relationshi shows the trade-off that exists between investment in treasury-bills and investment in loans. These results are consistent with both theoretical and emirical literature. Bernanke and Blinder (1988) ointed out that government bonds are somewhat default risk-free hence rovide incentive for commercial banks to urchase bonds leaving less funds available for credit. The result is also suorted by the findings by Barajas and Steiner (2002) for Latin America, Abuka and Egesa (2007) for East Africa, who share the view that government bonds are an oortunity cost for business sector credit suly. Unlike the results for the long run model, the short run relationshi between deosits and credit suly is consistent with theoretical exectations and results from much of the emirical literature surveyed. This suggests that an increase in the amount of deosits leads to increase in bank credit suly to the business sector. The fact that deosits are significant in the model shows that deosits are one of the main determinants of business sector credit suly in Botswana. CONCLUSIONS The main aim of this study was to investigate the main determinants of bank credit suly to Botswana business sector. In order to do this, the study alied the ARDL bounds testing aroach to co-integration to bank credit suly to businesses in Botswana using quarterly time series data. The results show that deosit levels, lending rate and Bank of Botswana Certificates (BoBCs) were the main determinants of bank credit suly to the business sector, both in the short- and long run. Gross domestic roduct (GDP), household credit and bank industry structure did not have any influence on changes in bank credit to the business sector. As exected, increases in deosit levels were associated with increases in bank credit to the business sector. This suggests that banks in Botswana deend on deosits as their source of loanable funds. Similarly, increases in lending rates are associated with increases in bank credit suly to the business sector. This suggests that credit rationing was not an issue during the samle eriod. By contrast, increases in BoBCs negatively affected the suly of bank credit to the business sector. This confirms the view that BoBCs, by offering commercial banks alternative lucrative and risk-free investment oortunities, discouraged bank intermediation, esecially to the business sector. The findings of this study suggest that efforts to romote business sector credit for investment uroses should be directed at savings mobilization and to reducing the suly of or attraction of BoBCs to commercial banks. NOTE 1. Pula is Botswana currency (US$1 H P 10.58)

14 114 GOFAMODIMO SECHELE, OBONYE GALEBOTSWE AND MOGALE M. NTSOSA References Abuka, C. A., & Egesa, K. A. (2007), An assessment of rivate sector credit evolution in the East African community: The candidates for a region wide strategy for the financial sector. Bank of Uganda. Asteriou, D. & Hall, S. G. (2007), Alied Econometrics. New York: Palgrave Macmillan. Bank of Botswana ( ), Bank of Botswana Annual Reorts. Gaborone Bank of Botswana (2012), Banking Suervision Annual Reort. Gaborone Bernanke, B. S., & Blinder, A. S. (1988), Credit, Money and Aggregate Demand. The American Economic Review, Brown, M., Ongena, S., Poov, A., & Yesin, P. (2012), Who Needs Credit & Who Gets Credit in Eastern Euroe? Euroean Central Bank. Botswana Stock Exchange (2012), Botswana Stock Exchange Annual Reort. Gaborone: Calza, A., Gartner, C., & Sousa, J. (2001), Modelling the Demand for Loans to the Private Sector in the Euro Area. Euroean Central Bank. Dickey, D. A., & Fuller, W. A. (1979), Distribution of The Estimator for Autoregressive Time series with a Unit Root. Journal of American Statistical Association, Egert, B., Backe, P., & Zumer, T. (2006), Private sector credit in Central & Eastern Euroe: New (over) shooting stars? William DAvidson Institute, Eita, J. H., & Jordaan, A. C. (2007), A causality analysis between economic growth and financial develoment in botswana. Pretoria: University of Pretoria. Engle, R. F., & Granger, C. W. (1987), Co-integration and Error Correction: Reresentation, Estimation and Testing. Econometrica, Fuller, W. A. (1976), Introduction to statistical time series. John Wiley & Sons. Granger, C. (1986), Develoment in the Study of Co-integrated Economic Variables. Oxford Bulletin of Economics and Statistics, Government. (2002), National Develoment Plan 9. Gaborone Greene, W. H. (2008), Econometric Analysis. New York: Pearson Education Inc. Gujarati, D. N. (2009). Basic Econometrics. New York: Mc Graw Hill. Guo, K., & Steanyan, V. (2011), Determinants of Bank Credit in Emerging Markets. Washington DC: International Monetary Fund. Hahn, A. L. (1949), The economics of illusion: A critical Analysis of Contemorary Economic Theory & Policy. Squier Publishing. Harvey, C. (1996), Banking Policy in Botswana: Orthodox But Untyical. Gaborone: Bank of Botswana. Hendry, D.F (1995), Dynamic Econometrics. Oxford: Oxford University Press Herrera, S., Hurlin, C., & Zaki, C. (2012). Why Don t Banks Lend to Egyt s Private Sector? Washington DC: The World Bank Hofmann, B. (2001), The Determinants of Private Sector Credit in Industrialised Countries: Do roerty rices matter? Basel: Bank of International Settlement.

15 THE DETERMINANTS OF COMMERCIAL BANK CREDIT SUPPLY TO THE BUSINESS SECTOR Igan, D., & Tamirisa, N. (2009), Credit growth and bank soundness: Evidence from emerging Euroe. Washington DC: International Monetary Fund. Innes, Mitchell. (1914), The Credit Theory of Money. Banking Law Journal, Imran, N., & Nishat, M. (2012), Determinants of Bank Credit Suly in Pakistan. Internation Business confrence in Business Management. International Monetary Fund (2008), Botswana: 2007 Article IV consultation- Staff Reort; Public Information Notice on the Executive Board Discussion; and statement by the Executive Director to Botswana. Washington DC: International Monetary Fund. Iyoboi, M. (2008), Bank and Non-Bank Financial Deeening and Economic Growth: The Nigerian Exerience ( ), John Wiley & Sons Ltd, Jacob, D., & Rayner, V. (2012), The Role of Credit Suly in the Australian Economy. Reserve Bank of Australia. Jefferis, K. (2010), BIFM Economic Review. Gaborone: Bifm Botswana Limited. Kaminsky, G. L. & Reinhart, C. M. (1999), The Twin Crises: The Causes of Banking and Balanceof- Payments Problems. The American Economic Review, Kennedy, P. (1998), A Guide to Econometrics. Oxford: Blackwell Publishers Ltd. Koo, G. (2005). Analysis of Economic Data. Chichester: John Wiley & Sons Ltd. Mosate, J. K. (2011), Determinants of Bank Credit Demand in Botswana. An unublished Undergraduate Dissertation. University of Botswana. National Develoment Plan (2007), Macroeconomic Outline and Policy Framework for NDP 10. Gaborone: Ministry of Finance & Develoment Planning. Non-Banking Financial Institutions Regulatory Authority (2012), NBFIRA Annual Reort. Gaborone: NBFIRA. Ntsosa, M. (2011), The Imact of the 2007 Global Financial Crisis on Botswana Economy. Asian- African Journal of Economics & Econometrics Pesaran, M. H., & Shin, Y. (1999), An autoregressive distributed lag modelling aroach to cointegration analysis. Cambridge University Press. Phillis, P. C., & Perron, P. (1988), Testing for a Unit Root in Time Series Regression. Biometrika, Sharma, P., & Gounder, N. (2012), Determinants of Bank Credit in Small Oen Economies: The case study of six Parcific Island Countries. Griffith University. Stiglitz, J., & Weiss, A. (1981), Credit rationing in markets with imerfect information. The American Economic Review, Studenmund, C. C. (1997), Using Econometrics: A Practical Guide. New York: Addison Wesley Longman Sultan, R. (2010), Short run and long run elasticities of gasoline demand in Mauritius: an ARDL bounds test aroach. Journal of Emerging Trends in Economics and Management Sciences, Tasic, N., & Valec, N. (2008), The maturity structure of Bank credit: Determinants & effects on economic growth. National Bank of Serbia.

16 116 GOFAMODIMO SECHELE, OBONYE GALEBOTSWE AND MOGALE M. NTSOSA APPENDICES Table A1 Unit root test (Intercet & Trend) Augmented Dickey-fuller (ADF) Phillis-erron (PP) Variable Levels First Integrated Levels First Integrated Difference order Difference order t-statistic t-statistic I(d) t-statistic t-statistic I(d) Buscred *** I(1) *** I(1) GDP *** I(0) *** I(0) Hcred *** I(1) *** I(1) Lrate *** I(1) ** I(0) BOBC *** I(1) *** I(1) Deos *** I(1) *** I(1) Note: All variables excet for lending rate and recirocal of numbers are exressed in natural logarithm.*, **, ***, means significant at 10%, 5% and 1%, resectively. I(0) means the variable is stationary at levels i.e. Integrated of order zero. I(1) means the variable is stationary at first difference i.e. Integrated of order one Table A2 Unit root test (Intercet only) Augmented Dickey-fuller (ADF) Phillis-erron (PP) Variable Levels First Integrated Levels First Integrated Difference order Difference order t-statistic t-statistic I(d) t-statistic t-statistic I(d) Buscred *** I(1) *** I(1) GDP *** I(1) *** I(1) Hcred *** I(1) ** I(0) Lrate *** I(1) *** I(0) BOBC *** I(1) *** I(1) Deos *** I(1) *** I(1) Note: All variables excet for lending rate and recirocal of numbers are exressed in natural logarithm.*, **, ***, means significant at 10%, 5% and 1%, resectively. I(0) means the variable is stationary at levels i.e. Integrated of order zero. I(1) means the variable is stationary at first difference i.e. Integrated of order one.