University of Wollongong Thesis Collections University of Wollongong Thesis Collection University of Wollongong Year 2001 Financial deregulation, banking development, and the likelihood of banking fragility: the case of Indonesia Siti Astiyah University of Wollongong Astiyah, Siti, Financial deregulation, banking development, and the likelihood of banking fragility: the case of Indonesia, Doctor of Philosophy thesis, Faculty of Commerce, University of Wollongong, 2001. http://ro.uow.edu.au/theses/1744 This paper is posted at Research Online.
CHAPTER 6 COINTEGRATION TEST AND ERROR CORRECTION MECHANISM FOR SAVING, CREDIT, AND PRIVATE INVESTMENT: THE INDONESIAN CASE 6.1. Introduction The objective of this chapter is to investigate the existence of the long run relationship among the variables in the system of each equation by using the Johansen cointegration test. As discussed earlier, the multivariate Johansen cointegration method is used for testing and analysing cointegration by linking with the Vector Auto-Regression (VAR) model. The estimation of the Johansen cointegration test assumes the Gaussian error term. Therefore, the determination of lag-length of VAR for estimating the Johansen cointegration test is important to obtain the Gaussian error term. If the result of the Johansen cointegration test suggests that there is long mn relationship among the specified variables in the system of equations, the error correction mechanism is valid. Before estimating the Johansen cointegration test, there are two major steps of econometric testing that should be carried out. First, the unit root test for all related variables to examine whether the variable is stationary or non-stationary. Second, econometric testing is used to determine the lag-length of VAR for estimating the Johansen cointegration test. The selection of the lag-length of VAR is related to obtaining the serial uncorrelated of the error term. Moreover, after determining the lag-length of the VAR, the Johansen cointegration test will be carried out by using the 162
lag-length of VAR that has been determined previously. If the result of the Johansen cointegration test indicates that there is a long mn relationship among the specified variables, the long mn equation that is consistent with economic theory is estimated. Furthermore, the error correction mechanism is derived from its long mn equation. The organisation of this chapter is as follows. Section two presents the Johansen cointegration test for real saving through the banking sector and its long mn equation. Section three presents the Johansen cointegration test for real credit from banks. This section is divided into three sub-sections, which is the estimation of the Johansen cointegration test for real credit from total banks, state banks, and national private banks respectively. In addition, the discussion also covers the long run equation of real credit from total banks, state banks, and national private banks respectively. Section four will present the Johansen cointegration test for real private investment and its long mn equation. Section five presents the error correction mechanism for saving through the banking sector, credit from banks, and private investment. The last section is the conclusion of this chapter. 6.2. Real Saving through the Banking Sector As discussed previously, prior to estimating the cointegration test for real saving through the banking sector, each variable in the equation of real saving through the banking sector is examined to determine whether the variable is stationary or nonstationary. The unit root test for each variable is carried out by using the Dickey-Fuller (DF) and Augmented Dickey-Fuller (ADF) tests. In addition, the unit root test for each variable in the equation of real saving through the banking sector is also carried out by considering the stmctural break related to the existence of the 1997 financial and banking crises. 163
6.2.1. Unit Root Test for Variables Related to Real Saving through the Banking Sector Unit root test for each variable related to real saving through the banking sector which is expressed in equation (16) is carried out for the sample period 1983:1-1999:2"^ The resuft of the DF and ADF tests using equations (31) and (35) for each variable related to the real saving is reported in Table 6.1. Table 6.1: Unit Root Test for Variables Related to Real Saving through the Ban ring Sector Variable Selected Test statistics lag-length DF ADF Level: LFSR LGDPR RDR RDIF First difference: DLFSR DLGDPR DLRD DLINF Critical value at 5% level is -2.91. 1 4 4 6 1 5 3 3 0.65 0.69-1.65-1.80-2.59-2.53-1.64-2.59-8.16-6.18-10.30-3.81-6.44-5.35-3.20-5.32 "^ The sample period is started from the first quarter of 1983 (1983:1) due to the availability and consistency data in quarterly basis. 164
Table 6.1 indicates that the null hypothesis of a unit root cannot be rejected at the 5% level for all variables. It suggests that all variables are non-stationary in the level'^*^. As the variable is non-stationary in the level, the stationary test of those variables is carried out for the first difference of the variable. Test statistics of the DF and ADF tests for the first difference of each variable related to real saving through the banking sector indicate that the null hypothesis of a unit root for all variables can be rejected at the 5% level. It suggests that all variables related to real saving through the banking sector are stationary at the first difference. In other word, all variables are 1(1). The estimation of the ADF test has covered the period of the 1997 financial and banking crises. It could be argued that the crisis has contributed to changing the major monetary and banking sectors variables. The power of the tiaditional ADF test, however, does not consider the existence of stmctural break such as the 1997 financial and banking crises. Therefore, the unit root test by considering the stmctural break associated with the 1997 financial and banking crises is carried out for each variable related to real saving through the banking sector. To estimate the unit root test by considering the existence of the 1997 financial and banking crises, we will use the Perron approach with the assumption that the time of break is known (Perron (1989)). ^ When the SBC selection criteria is used, the appropriate lag-length of RDIF is 3 and it indicates that the null hypothesis can be rejected. But selection of higher order indicates that the null hypothesis cannot be rejected at the 5% level. The selection of higher order also supported by other criteria which is select 6. As there is a contrast resuh, the F-test is carried out and the result is indicating that lag-length 6 can be selected. 165
The time break is related to the "big shock" of the 1997 financial and banking crises and the break time is assumed to be the second quarter of 1997*'^^ The model allows a one-time exogenous change in both intercept and slope (Model C of Perron (1989)) To test the stationary properties of this model, however, requires the critical value that has been provided by Perron in table VLB (p. 1377, Perron (1989)). The result of the unit root test conditional on the presence of stmctural break for variables related to real saving through the banking sector using procedure in equation (48) is reported in Table 6.2. LFSR LGDPR RDR RDIF Table 6.2: Unit Root Test and Stmctural Break for Variables Related to Real Saving through the Banking Sector k >" 6 fi 7 d a 4 3.22 1.91 0.01-0.03-0.25-0.33 6 4 4 (2.50) 6.22 (3.15) 4.15 (2.50) -3.24 (-0.73) (1.65) 1.60 (1.99) -124.42 (-2.72) -965.65 (-3.90) (2.42) 0.01 (3.20) -0.02 (-0.81) 0.16 (1.16) (-1.51) -0.03 (-2.14) 1.91 (2.59) 14.96 (3.69) (-2.88) 0.07 (1.76) 21.41 (5.92) 62.45 (3.71) -0.59-0.36-0.54 ta -2.46-3.13-2.60-2.53 Figures in the bracket are t-statistics. T 58 A = = = 0.88, critical value of A = 0.9 at 5% level is -3.80. T 66 As discussed in the previous chapter, as the currency crisis started in the middle of July 1997 and rapidly trigger to the fmancial and banking crises, therefore, the break time is determined at the second quarter of 1997 (1997:2) as the data is on quarterly basis. By using 1997:2 as the time break, the model allows the change of intercept and slope in the third quarter of 1997 (1997:3). '^^ The unit root test by allowing a one-time exogenous change in both intercept and slope and also using 1997:2 as a known break time is also appued for all variables related to real credit from banks and real private investment. '^^ The unit root tests in the presence of structural break are applied for the level of variables. As discussed earlier, these tests are applied in order to get the possibility of higher power. As the results of the conventional ADF test indicate that all variables are stationary in the first differenced, therefore the unit root test in the presence of structural break for the first differenced for the related variables is not estimated in this research and it can be done for further research. 166
Table 6.2 suggests that test statistics for the unit root test by allowing a onetime exogenous change for all variables related to real saving through the banking sector are non-stationary. It is consistent with the result of the ADF test. Consequently, all variables related to real saving through the banking sector are 1(1) as shown in the previous ADF test. As all variables related to real saving through the banking sector are 1(1), the Johansen cointegration test for real saving through the banking sector can be carried out. 6.2.2. Johansen Cointegration Test for Real Saving through the Banking Sector The cointegration test for real saving through the banking sector will be carried out for two sample periods: 1983:1-1997:2 and 1983:1-1999:2. The first sample period does not cover the 1997financialand banking crises, while the second sample period has covered thefinancialand banking crises period. As discussed in the earlier chapter, the determination of lag-length of VAR for estimating the Johansen cointegration test is important for obtaining the Gaussian error term. The selection of lag-length of VAR is determined by the selection model criteria that is the Akaike Information Criteria (AIC) and Schwarz Bayesian Criteria (SBC) and the Likelihood Ratio (LR) test. The LR test and the selection model criteria of estimation of various lag-length of VAR of the level of variables for the sample period 1983:1-1997:2 are reported in Table 6.3. 167
Table 6.3: Test Statistics and Model Selection Criteria of VAR: LFSR LGDPR RDR RDIF lag-length 5 4 Sample: 1983:1-1997:2 Deterministic and/or exogenous variables: C^^'* LL AIC SBC 10.45-10.65-78.65-145.64 LR test ;^^(16)= 42.21 3-46.12-98.12-149.35 ;^^(16) = 113.14 2-58.44-94.44-129.91 /(48) =137.79 1-68.95-88.95-108.65 %\64) = 158.79 0-284.13-288.13-292.07 ;if'(80) =589.15 Table 6.3 suggests that when the AIC selection criteria is used, the appropriate lag-length for the Johansen cointegration test is 4. However, when the SBC selection criteria is used, the appropriate lag-length for the Johansen cointegration test is I. As discussed earlier, the lower lag-length is preferred as the sample period is small. In addition, test statistics of 64 restrictions of moving from VAR with lag-length 5 (after this it will be denoted as VAR(5)) to VAR(l) is 158.8 which is higher than the critical value of ;^^(64)at the 5% level which is about 90.53'^^. It indicates that moving to VAR(l) can not be rejected. To confirm that VAR(l) is acceptable in terms of residual whiteness, the Lagrange Multiplier (LM) test is carried out by using VAR(l) for each equation of the level of VAR for the sample period 1983:1-1997:2. The F-test statistic of the LM test for individual equations by using VAR(l) indicates that there is serial correlation. To '^'* C = intercept '^^ If there is difference selection of lag-length based on AIC and SBC, the lower lag-length is selected. The selected lag-length, however, should be serially uncorrelated. As discussed in chapter 6, the higher lag-length might be required to obtain the Gaussian error term. '^^ If the critical value of the Chi-square of the exactly degree of freedom is not available, the closer degree of freedom will be used. For example, as ;^^ (64) is not available, the % (70) is used. 168
reduce the serial correlation problem, VAR(3) is selected'^^. The LM test for individual equations by using VAR(3) indicates that there is no serial correlation for the individual equation. Furthermore VAR(3) is preferred instead of VAR(l). In addition, test statistics of 32 restrictions of movingfi-omvar(5) to VAR(3) is 113.14 which is higher than the critical value of ;if^(32) at 5% level which is about 43.77. It suggests that moving to VAR(3) cannot be rejected. As a result, the VAR(3) is selected for estimating the Johansen cointegration test for saving through the banking sector for the sample period 1983:1-1997:2. When the sample period is extended to 1999:2, the variable to determine the lag-length of VAR for estimating the Johansen cointegration test for real saving through the banking sector is added, by including a dummy variable for the 1997 financial and banking crises (variable DUM). The DUM variable has value one for 1997:3-1999:2 and zero otherwise. The test statistics and the model selection criteria of estimation of various lag-length of VAR for the sample period 1983:1-1999:2 is reported in Table 6.4. Table 6.4 suggests that when AIC selection criteria are used, the appropriate lag-length for estimating the Johansen cointegration test for real saving through the banking sector for the sample period 1983:1-1999:2 is 4. However, when the SBC selection criteria is used, the appropriate lag-length for estimating the Johansen cointegration test is 1. '^^ The selection of VAR(2) suggests that the LM test for individual equation by using VAR(2) has serial correlation problem. Moreover VAR(3) is selected. '^* Table 6. la in Appendix. 169
hi addition, test statistics of 64 restrictions of moving fi-om VAR(5) to VAR(l) is 239.1 which is higher than the critical value of ;tf^(64)at 5% level. It indicates that moving to VAR(l) can not be rejected. Table 6.4: Test Statistics and Model Selection Criteria of VAR: LFSR LGDPR RDR RDIF lag-length 5 4 Sample 1983:1-1999:2 Deterministic and/or exogenous variables: C DUM LL AIC SBC -46.41-73.97-145.97-221.96 LR test ;^'(16)= 55.13 3-119.95-175.95-235.05 ;if'(32) = 147.08 2-137.60-177.60-219.82 /(48) =182.39 1-165.97-189.97-215.30 j'(64) =239.13 0-449.03-457.03-465.47 ;^'(80) =805.25 To confirm that VAR(l) is acceptable in terms of residual whiteness, the LM test is carried out by using VAR(l) for each equation of the level of VAR for the sample period 1983:1-1999:2. The F-test statistic of the LM test for individual equations by using VAR(l) indicates that there is no serial correlation. Therefore, VAR(l) can be selected to estimate the Johansen cointegration test for real saving through the banking sector for the sample period 1983:1-1999:2. The results of the Johansen cointegration test for real saving through the banking sector for the sample period 1983:1-1997:2 and 1983:1-1999:2 are reported in Tables 6.5 and 6.6 respectively. Table 6.5 indicates that the test statistics, based on maximal eigenvalue, can reject the null hypothesis of no cointegrating vector (Ho:r = 0) at the 5% level. '^' Table 6.2a in Appendix 170
The test statistics also can reject the null hypothesis of no cointegrating vector at most equal to one and two (Ho:r<\ and Ho:r<2) at the 5% level respectively. It indicates that there are three cointegrating vectors. As a result, there is a long run relationship among the specified variables for the sample period 1983:1-1997:2. The existence of a long mn relationship among the specified variable is also found when the sample period is extended to 1999:2. Table 6.6 indicates that the null hypothesis of no cointegrating vector for the sample period 1983:1-1999:2 can be rejected at the 5% level. Besides, the next null hypothesis of no cointegrating vector at most equal to one and two are also can be rejected at the 5% level. Therefore, by using the sample period 1983:1 to 1999:2, there are three cointegrating vectors. It suggests that there is a consistent finding for the existence of a long mn relationship among the specified variables. If the number of cointegrating vectors is more than one, the selection of the cointegrating vector should be based on economic theory. In other words, the appropriate cointegrating vectors selected should be consistent with economic theory and by normalising the logarithm for real saving through the banking sector, it will be obtained the long mn equation for real saving through the banking sector. 171
Table 6.5: Cointegration Test for Real Saving through the Banking Sector Sample: 1983:1-1997:2 Variables in the cointegrated system: LFSR LGDPR RDR RDIF Maximal Eigenvalue of Stochastic Matrix Ho: Ha: Statistic Critical Value r = 0 r<l r<2 r<3 r=\ r = 2 r = 3 r = 4 37.08 23.28 17.14 8.30 95% 28.27 22.04 15.87 9.16 90% 25.80 19.86 13.81 7.53 Trace of Stochastic Matrix Ho: Ha: Statistic Critical Value r = 0 r<l r<2 r<3 r>l r>2 r>3 r>4 85.80 48.72 25.44 8.30 95% 53.48 34.87 20.18 9.16 90% 49.95 31.93 17.88 7.53 172
Table 6.6: Cointegration Test for Real Savmg through the Banking Sector Sample: 1983:1-1999:2 Variables in the co-integrated system: LFSR LGDPR RDR RDIF 1(0) included in the VAR: DUM Maximal Eigenvalue of Stochastic Matrix Ho: Ha: Statistic Critical Value r = 0 r<l r<2 r<3 r = l r = 2 r = 3 r = 4 87.54 47.86 30.79 7.46 95% 28.27 22.04 15.87 9.16 90% 25.80 19.86 13.81 7.53 Trace of Stochastic Matrix Ho: Ha: Statistic Critical Value r = 0 r<\ r = 2 r<3 r>l r>2 r>3 r>4 173.65 86.11 38.25 7.46 95% 53.48 34.87 20.18 9.16 90% 49.95 31.93 17.88 7.53 173
6.2.3. Long Run Equation for Real Saving through the Banking Sector The result of the long mn equation for real saving through the banking sector, which is expressed in equation (16), based on the Johansen approach is reported in Table 6.7. Table 6.7: Long Run Equation for Real Saving through the Banking Sector LGDPR RDR RDIF Intercept 1983:1-1997:2 2.25 (0.07) 0.01 (0.006) 0.002 (0.003) 14.00 (0.77) 1983:1-1999:2 3.01 (2.50) 0.06 (0.22) -0.003 (0.03) 27.54 (41.56) Figures in brackets are standard error. Table 6.7 indicates that real income has a significant impact on real saving through the banking sector for the sample period 1983:1-1997:2. The sign of the parameter for real income is consistent with the hypothesis, which is positive. This finding implies that an increase in real income (real GDP) will increase real saving. The long mn impact of real income on real saving for the sample period 1983:1-1999:2, however, is not significant even though the sign of the parameter is consistently positive. '^ When the differential between domestic and foreign interest rates, and intercept are restricted, the long run impact of real income on real saving is 0.97. 174
The important variable to influence real savmg under the financial deregulation hypothesis is real deposit rates. The restriction test for real deposit rates indicates that in the long mn real deposit rates have a positive and significant unpact on real saving through the banking sector at the 10% level for the sample period 1983:1-1997:2. The significance and positive impact of real deposit rates are consistent with the hypothesis. The positive impact of real interest rates on real saving through the banking sector suggests that fmancial deregulation policy contributed to encouraging fiind mobilisation into the banking system. The positive and significant impact of real interest rates can be used as a policy variable to mobilise funds through the banking sector. The finding suggests that an increase of 1 % in real interest rates is associated with an increase in real saving through the banking sector in the long run of about Rp.1.02'^' billions for the sample period 1983:1-1997:2. The finding of the positive relationship between real deposit rates and real saving through the banking sector is consistent with the finding of Warman and Thirlwall (1994) for Mexico, Athukorala (1996) for hidia, and Darsono (1999) for hidonesia'^l The policy implication of the benefits of high interest rates on real saving through the banking sector, however, has to be related to policy credibility. Very high interest rates might be associated with an "incredible" policy that might influence the power of interest rates on mobilisation offiindsthrough the banking sector. Very high '^' An increase of 1% in real deposits rates will increase an antilog (0.01) of real saving (as the real saving is denoted in the logarithm), and the result is 1.02. "^ Darsono (p.l 14, 1999) found that interest rates are positive and significantly influence saving through the banking sector for the sample period 1983:3-1996:4. The saving fimction of Darsono is as follows: S = f(y, R, P, E), where S= saving through the banking sector, Y=real income (GDP), R= nominal deposit rates, P= price level, E= exchange rates. In addition, the methodology to obtain the long run equation of real saving through the banking sector is based on the Phillips- Hansen and Ordinary Least Squares (OLS). The OLS for the long run saving equation in Darsono (1999) study is associated with standard residual based test for the cointegration in the saving fimction. Therefore, the saving fimction and the methodology is differ with this thesis. This thesis included the effective differential between domestic and foreign interest rates and the long run equation is obtained based on the result of the Johansen cointegration test. 175
interest rates might be related to the lack of credibility of economic policy which might reflect such factors as the lack of a sound financial stincture. Fry (1997) argues that very high interest rates will reduce national saving since very high interest rates are associated with increased risk and uncertainty. In addition, the impact of real interest rates on saving through the banking sector is not significant when the sample period is extended to 1999:2. The interest rates increased substantially after the second quarter of 1997. However, a high interest rate was followed by a high inflation rates. As a result, real interest rates even became negative. The high interest rates followed by high inflation rates indicate unsoundness and high risk in the financial sector. Therefore, the finding suggests that moderate real interest rates accompanied by the perception of a credible policy has a significant impact on the mobilisation of funds through the banking sector in the long mn. It implies that the finding supports the view that moderate real interest rates is a potent instmment for increasing real saving through the banking sector. Besides, the interest rates policy should be accompanied by stable macroeconomic policy to reduce risk and uncertainty. The significant impact of real interest rates on real saving through the banking sector implicitly indicates that inflation rates have a significant negative impact on real saving through the banking sector in the long mn. The negative relationship between inflation rates and real saving through the banking sector is through real interest rates. The negative relationship between inflation rates and real saving indicates that high inflation rates will encourage depositors to switch fi-om saving through the banking sector to other kinds of saving such as real assets. Consequentiy, a higher expectation of inflation induces the movement of saving through the banking sector to other kinds of saving. Therefore, the policy implication of the positive impact of real interest rates on saving can also be interpreted by maintaining inflation at low or at reasonable rates. 176
On the other hand, the long mn impact of an effective differential between domestic and foreign interest rates is not significant in either sample period. The insignificance of the effective differential between domestic and foreign interest rates might be related to the stability of the expected depreciation of the domestic currency for most of the time of the study. Before the onset of the currency crisis in 1997, depreciation of the domestic currency could be predicted as the government was committed to maintaining a stable foreign exchange rate. In addition, the insignificance of the effective differential between domestic and foreign interest rates to real saving implies that real saving through the banking sector is not sensitive to the differential rates. If real saving through the banking sector is not responsive to interest rate differentials, the movement of capital especially capital inflows before the onset of the 1997 financial and banking crises might be related to the perceived profitability of domestic investment from the foreign investors' perception following financial deregulation policy (Edwards (1986)). It implies that an increase in domestic interest rates relative to foreign interest rates, assuming that the depreciation of domestic currency is constant, is not a potent policy to increase real saving through the banking sector. It implies that capital inflow before the onset of the 1997 financial and banking crises might not flow through the banking sector but to other kind of investment such as investment in the capital market. Hence, the changes of real saving through the banking sector in hidonesia with respect to interest rates changes are mainly associated with the domestic response to interest rates changes. '" Darsono (1999) found that nominal exchange rates have negative and significant influence on saving through the banking sector for the sample period 1983:3-1996:4. However, it is argued that the nominal exchange rates was relatively constant during tiiat period except for die 1986 due to the devaluation policy. 177
This finding contrast with the Mexico, the changes in financial saving with respect to interest rates changes in Mexico is associated with the domestic response to interest rates changes and the effect of capital flight (Warman and Thirlwall (1994)). In addition, it is hypothesised that variable real saving through the banking sector is an important variable for financing credit from banks. 6.3. Real Credit from Banks Real saving through the banking sector is an important variable to determine the supply of credit from banks in the financial deregulation hypothesis for developing countries. The estimation of the real credit fiinction is disaggregated into the estimation of real credit from total banks, real credit from state banks, and real credit from national private banks. In addition, before any estimation is carried out, all variables related to real credit from banks are investigated to examine whether they are stationary or non-stationary. 6.3.1. Unit Root Test for Variables Related to Real Creditfi"omBanks Unit root tests for each variable related to real credit from banks (which cover total banks, state banks, and national private banks) which are expressed in equations (22), (23), and (24) are carried out for the sample period 1984:l-1999:2^^^ '^* The sample period is started from the fu-st quarter of 1984 (1984:1) is mainly related to the availability of data. Data for lending interest rates of non-priority economic sector, for example, is only available after the third quarter of 1983 (1983:3). Data for interest rates of Certificate of Bank Indonesia (SBI) is available since 1984:1. Therefore, the sample period is started from 1984:1. 178
The result of the DF and ADF tests by applying procedure in equations (31) and (35) for each variable related to real credit from banks is reported in Table 6.8. Table 6.8: Unit Root Test for Variables Related to Real Credit from Banks Variables Selected Test-Statistic Lag-length DF ADF Level: Total Banks: LCREDITR LFSAR LBIR CAPER RL INF RSBI State Banks: LCREDITSR LFSASR LBISR CAPERS RLS Nat.Priv.Banks: LCREDITNR LFSANR LBINR CAPERN RLN First Differenced: Total Banks: DLCREDITR DLFSAR DLBIR DCAPER DRL DINF DRSBI State Banks: DLCREDITSR DLFSASR DLBISR DCAPERS DRLS Nat.Priv.Banks: DLCREDITNR DLFSANR DLBINR DCAPERN DRLN 1 1 1 3 1 5 5 3 1 1 3 1 1 1 1 2 1 4 4 1 2 1 4 4 2 4 1 2 1 2 1 1 6 1-1.46 0.76-1.40 0.99-1.30-1.67-2.23-1.20 1.02-2.14 4.14-2.64-2.06-1.45-0.58-3.20-1.97-4.50-7.19-4.95-4.60-5.63-3.16-5.89-5.36-5.86-8.16-3.80-7.21-6.34-8.06-4.35-4.16-7.43-1.48 0.75-2.50-1.57-1.97-1.20-1.62-0.94 0.47-2.01-2.12-2.74-1.76-1.54-1.54-2.06-2.08-4.00-3.65-3.72-12.75-4.48-3.30-6.74-4.58-3.86-6.89-13.22-5.57-4.44-4.74-3.38-4.64-5.27 The critical value at 5% level is -2.91 179
Table 6,8 suggests that the results of the DF and ADF tests for the level of each variable related to real credit from banks indicate that the null hypothesis of a unit root cannot be rejected at the 5% level for most variables. It indicates that all variables related to real credit from total banks, state banks, and national private banks are non-stationary in the level, except the variable of real capital equity of national private banks (CAPERN). The result of the DF test for the real capital equity of national private banks indicates that the null hypothesis of a unit root can be rejected at the 5% level. It suggests that real capital equity of national private banks is stationary by using the DF test. The ADF test for variable real capital equity of national private banks, however, suggests that the null hypothesis of a unit root cannot be rejected at the 5% level. Consequentiy, based on the ADF test, the variable real capital equity of national private banks is not stationary in the level. If the variables are non-stationary in the level, the unit root tests for the first difference of those variables are carried out. Table 6.8 indicates that the results of the DF and ADF tests for the first difference of each variable related to real credit from total banks, state banks, and national private banks indicates that the null hypothesis of a unit root can be rejected at the 5% level for all variables. It suggests that the first difference of each variable is stationary. As a result, all variables related to real credit from total bank, state banks, and national private banks are 1(1). As discussed earlier, the sample period covers the "big shock" which is associated with the 1997 financial and banking crises, therefore the unit root test conditional on the presence of a stmctural break should be carried out. The result of the unit root test conditional on the presence of a stmctural break by applying equation (48) for each variable related to real credit from total banks, real credit from state banks, and real credit from national private banks is reported in Table 6.9. 180
Table 6.9: Unit Root Test and Structural Break for Variables Related to LCREDITR LFSAR LBIR CAPER RL INF RSBI RLR RSBIR LCREDITSR LFSASR LBISR CAPERS RLSR LCREDITNR LFSANR LBINR CAPERN RLNR k 4 4 4 6 6 6 6 6 6 4 4 4 4 6 4 4 4 6 6 fi 1.91 (1.65) 2.14 (2.07) 4.04 (2.75) -0.01 (-0.008) 5.35 (2.30) 8.78 (4.32) 9.48 (1.84) 8.47 (2.72) 3.79 (1.61) 1.16 (1.46) 1.78 (1.70) 4.59 (3.27) 3525.3 (1.53) 6.21 (2.24) 1.19 (1.44) 1.26 (2.45) 1.55 (2.12) 446.33 (0.51) 5.49 (1.92) 6 6.2 (5.32) 2.72 (2.78) 6.38 (1.52) 952.81 (7.72) 2.2 (0.16) -267.54 (-2.75) 88.8 (0.74) -2.04 (-0.03) 37.43 (0.77) 5.78 (5.14) 4.45 (2.86) 0.05 (0.02) 570165 (4.96) 18.09 (0.22) 6.77 (5.27) 0.29 (0.40) 11.29 (2.77) 603106.2 (8.76) -102.52 (-1.60) y9 0.008 (1.59) 0.009 (1.95) -0.005 (-2.07) 0.31 (1.55) -0.02 (-1.67) 0.03 (1.34) -0.05 (-1.08) -0.06 (-1.86) -0.04 (-1.0) 0.002 (0.97) 0.005 (1.46) -0.01 (-3.30) 105.62 (1.51) -0.06 (-1.52) 0.009 (1.35) 0.009 (1.97) 0.008 (1.76) 42.91 (0.68) -0.03 (-1.08) 7-0.11 (-5.35) -0.04 (-2.65) -0.1 (-1.34) -16.62 (-7.75) -0.01 (-0.05) 4.91 (2.87) -1.37 (-0.65) -0.14 (-0.10) -0.7 (-0.84) -O.I (-5.05) -0.07 (-2.68) 0.007 (0.13) -9970.1 (-5.01) -0.51 (-0.35) -0.12 (-5.40) -0.005 (-0.42) -0.19 (-2.63) -10583.9 (-8.84) 1.66 (1.50) Real Credit from Banks d -0.33 (-3.04) -0.25 (-3.00) -0.59 (-2.13) -30.71 (-3.48) 2.64 (1.98) -3.07 (-0.77) -1.91 (-0.29) 13.19 (3.43) 11.69 (2.84) -0.39 (-3.97) -0.42 (-3.97) -0.32 (-1.09) -16551 (-1.99) 11.06 (2.71) -0.32 (-2.26) -0.09 (-1.01) -0.33 (-1.04) -17770 (-4.01) 20.18 (5.51) a -0.19-0.21-0.41-0.76-0.26-1.26-0.55-0.62-0.41-0.1! -0.18-0.46-1.15-0.46-0.14-0.14-0.24-0.19-0.35 ta -1.63-2.03-2.74-1.41-2.3-4.83-1.85-2.87-2.03-1.39-1.67-3.24-1.74-2.46-1.40-2.25-2.06-0.70-2.1 Figures in the bracket are t-statistic 181
Table 6.9 indicates that all variables, except for inflation rates, are nonstationary in the level. These are consistent with the conventional results of the ADF test. The exception is associated with inflation rates. The stationary test conditional on the presence of a stmctural break of inflation rates (INF) suggests that inflation rates are stationary in the level. This result is in contrast to the result of the previous DF and ADF tests. As a result, the inflation rates is 1(0) and consequently inflation rates cannot be included in the cointegration test. Moreover, inflation rates and lending rate variables cannot be tested separately. Therefore, the lending rates should be presented in real terms. In line with this, the SBI rates (RSBI) are also presented in real terms (RSBIR). Consequently, real lending rates of total banks (RLR), real lending rates of state banks (RLSR), and real lending rates of national private banks (RLNR) are introduced in the unit root test by considering the existence of a stmctural break. The result suggests that those three variables are non-stationary in the level and it is consistent with the DF and ADF tests. By changing the presentation of lending rates and the SBI rates into real terms, all variables related to real credit from total banks, real credit from state banks, and real credit from national private banks are non-stationary in the level. But as discussed earlier, all variables became 1(1). The critical value is as follows: T 54 /I = -^ = = 0.87, critical value of A = 0.9 at 5% and 10% level are -3.80 and -3.46 T 62 respectively. '^* The DF and ADF tests for RLR, RLSR, and RLNR indicate that they are 1(1) respectively. 182
As all variables are 1(1), the Johansen cointegration test for real credit from total banks, real credit from state banks, and real credit from national private banks can be estimated. 6.3.2. Johansen Cointegration Test for Real Credit from Total Banks The cointegration test for real credit from total banks is carried out for two sample periods, 1984:1-1997:2 and 1984:1-1999:2. The lag-length of VAR for estimating the Johansen cointegration test is determined by the model selection criteria of the result of the estimation of various lag-lengths of the level and the LR test. The model selection criteria and the LR test of the level variable: LCREDITR LFSAR LBIR CAPER RLR RSBIR for the sample period 1984:1-1997:2 are reported in Table 6.10. lag-length 5 4 Table 6.10: Test Statistic and Model Selection Criteria of VAR: LCREDITR LFSAR LBIR CAPER RLR RSBIR Sample: 1984:1-1997:2 Deterministic and/or exogenous variables: C LL 175.53 114.42 AIC -35.58 SBC -177.46 LR test ;{f'(36) = 122.22 3 71.25-42.75-150.59 ;if'(72) = 208.57 2 42.32-35.68-109.46 ;}f'(108) = 266.43 1 9.78-32.22-71.94 /(144) = 331.50 0-237.44-243.44-249.11 ;if'(180) = 825.95 Table 6.10 suggests that when the AIC and the SBC selection criteria are used, the appropriate lag-length of VAR for the Johansen cointegration test is 1. In addition, the LR statistics of moving from VAR(5) to VAR(l) is 331.50 is higher than the 183
critical value of ;i;^(144)at the 5% significance level which is 172.72^". It mdicates that moving to VAR(l) cannot be rejected. To confirm that VAR(l) is acceptable in terms of residual whiteness, the Lagrange Multiplier (LM) test is carried out by using VAR(l) for each equation of the levels VAR. The F-test statistic of the LM test for individual equations using VAR(l) indicates that there is no serial correlation for individual equations^^^. As a result, VAR(l) can be selected for estimating the Johansen cointegration test for real credit from total banks for the sample period 1984:1-1997:2. When the sample period is extended to 1999:2, the estimation of VAR of the level of those variables is added to by a dummy variable for the 1997 financial and banking crises (DUM). The model selection criteria and the LR test for sample period 1984:1-1999:2 is reported in Table 6.11. Table 6.11: Test Statistic and Model Selection Criteria of VAR: LCREDITR LFSAR LBIR CAPER RLR RSBIR Sample: 1984:1-1999:2 Deterministic and/or exogenous variables: C DUM lag-length 5 4 LL 27.29-14.98 AIC -170.98 SBC -330.34 LR test j^ (36) = 84.54 3-83.52-203.52-326.10 ;^'(72) = 221.61 2-138.93-222.93-308.74 ;jf'(108) = 332.45 1-249.17-297.17-346.20 /'(144) = 552.92 0-582.74-594.74-606.99 j'(180) = 1220.0 137 The Chi-Square distribution for the degree of freedom more than 100 is calculated as follows: X^ = - {x -I-.^(2V - 1)} ^ (p.576, Ramanathan (1992)). This formula will be applied to calculate the Chi-square {X ) distribution for the degree of freedom exceeded than 100. '^* Table 6.3a in Appendix. 184
Table 6.11 suggests that when the AIC selection criteria are used, the appropriate lag-length for the Johansen cointegration test for real credit from total banks for the sample period 1984:1-1999:2 is 4. However, when the SBC is used, the appropriate lag-length for the Johansen cointegration test is 2. In addition, the LR test statistic of 108 restrictions of moving from VAR(5) to VAR(2) is 332.45 which is higher than the critical value of ;)r^(108) at the 5% level which is about 132.97. It indicates that moving to VAR(2) cannot be rejected. To confirm that VAR(2) is acceptable in terms of residual whiteness, the LM test is carried out by using VAR(2) in each equation for the level of VAR for the sample period 1984:1-1999:2. The LM test indicates that there is a serial correlation problem by using VAR(2). To reduce the serial correlation problem, VAR(4) is selected. By using VAR(4), there is no indication of serial correlation for individual equations. In addition, the LR test of 36 restrictions of moving from VAR(5) to VAR(4) is 84.5 is higher than the critical value at the 5% level. Therefore, moving to VAR(4) cannot be rejected. As a result, VAR(4) is selected to estimate the Johansen cointegration test for real credit from total banks for the sample period 1984:1-1999:2. The result of the Johansen cointegration test for real credit from total banks is reported in Tables 6.12 and 6.13 for the sample period 1984:1-1997:2 and 1984:1-1999:2 respectively. Table 6.12 indicates that the null hypothesis of no cointegrating vector can be rejected at the 5% level. The next null hypothesis of no cointegrating vector at most equal to one, based on maximal eigenvalue, also can be rejected at the 5% level. It suggests that there are two cointegrating vectors. As a result, there is a long run relationship among the specified variables for the sample period 1984:1- '^^ Table 6.4a in Appendix 185
1997:2. When the sample period is extended to 1999:2, the long mn relationship among those variables still exists. Table 6.13 indicates that the null hypothesis of no cointegrating vector for the sample period 1984:1-1999:2 can be rejected at the 5% level. The next null hypothesis of no cointegrating vector at most equal to one and two are also can be rejected at the 5% level. It indicates that there are three cointegrating vectors for the sample period 1984:1-1999:2. Therefore, there is a long mn relationship among the specified variables for the whole sample period. In addition, as discussed in the previous chapter, the major share of total banks is associated with state banks and national private banks. The behaviour of real credit from state banks might differ from the real credit from national private banks. Therefore, the Johansen cointegration test is also carried out for real credit from state banks and real credit from national private banks respectively. 186
Table 6.12: Cointegration Test for Real Credit from Total Banks Sample 1984:1-1997:2 Variables in the co-integrated system: LCREDITR LFSAR LBIR CAPER RLR RSBIR Maximal Eigenvalue of Stochastic Matrix Ho: r = 0 r<\ r<2 r<3 r<4 r<5 Ha: r = l r = 2 r = 3 r = 4 r = 5 r = 6 Statistic 48.11 38.77 24.42 17.07 12.15 5.20 Trace of Stochastic Matrix Ho: r = 0 r<\ r<2 r<3 r<4 r<5 Ha: r>l r>2 r>3 r>4 r>5 r = 6 Statistic 145.73 97.62 58.85 34.42 17.35 5.20 Critical Value 95% 90% 40.53 37.65 34.40 31.73 28.27 25.80 22.04 19.86 15.87 13.81 9.16 7.53 Critical Value 95% 102.56 75.98 53.48 34.87 20.18 9.16 90% 97.87 71.81 49.95 31.93 17.88 7.53 187
Table 6.13: Cointegration Test for Real Credit from Total Banks Sample: 1984:1-1999:2 Variables in the co-integrated system: LCREDITR LFSAR LBIR CAPER RLR RSBIR Variables 1(0) in the VAR: DUM Maximal Eigenvalue of Stochastic Matrix Ho: r = 0 r<\ r<2 r<3 r<4 r<5 Ha: r = l r = 2 r = 3 r = 4 r = 5 r = 6 Statistic 87.50 43.19 32.93 14.16 8.11 2.43 Trace of Stochastic Matrix Ho: r = 0 r<l r<2 r<3 r<4 r<5 Ha: r>\ r>2 r>3 r>4 r>5 r = 6 Statistic 188.32 100.82 57.63 24.69 10.54 2.43 Critical Value 95% 90% 40.53 37.65 34.40 31.73 28.27 25.80 22.04 19.86 15.87 13.81 9.16 7.53 Critical Value 95% 90% 102.56 97.87 75.98 71.81 53.48 49.95 34.87 31.93 20.18 17.88 9.16 7.53
6.3.3. Johansen Cointegration Test for Real Creditfi-omState Banks Estimation of the Johansen cointegration test for real credit from state banks is carried out for the sample periods 1984:1-1997:2 and 1984:1-1999:2. As discussed earlier, the selection of lag-length of VAR for estimating the Johansen cointegration test is determined by the model selection criteria and the LR test of the estimation of various lag-lengths of VAR of the level variable. The model selection criteria and the LR test of the estimation of various lag-length of VAR of the level of variables related to real credit from state banks is reported in Table 6.14. lag-length 5 4 Table 6.14: Test Statistics and Model Selection Criteria of VAR: LCREDITSR LFSASR LBISR CAPERS RLSR RSBIR Sample: 1984:1-1997:2 Deterministic and/or exogenous variables: C LL -199.63-273.63 AIC -417.63 SBC -553.84 LR test statistic ;ir'(36) = 148.01 3-310.94-418.94-521.09 ;)r'(72) =222.61 2-334.85-406.85-474.96 ;(f'(108) =270.45 1-360.66-396.66-430.71 ;^'(144) =322.05 0-782.75-782.75-782.75 ;}r'(180) = 1166.3 Table 6.14 suggests that when the AIC and the SBC selection criteria are used, the appropriate lag-length of VAR for the Johansen cointegration is 1. The test statistic of 144 restrictions of moving from VAR(5) to VAR(l) is 322.05 which is higher than the critical value of ;}f^(144) at the 5% level which is about 172.72. It indicates that moving to VAR(l) cannot be rejected. To confirm that VAR(l) is acceptable in terms of residual whiteness, the LM test is carried out using VAR(l) in 189
each equation of the level of VAR for the sample period 1984:1-1997:2. Estimation of the LM test for individual equations by using VAR(l) indicates that there is no serial correlation of the individual equations^"^^. Therefore VAR(l) can be selected for estimating the Johansen cointegration test for real credit from state banks for the sample period 1984:1-1997:2. When the sample period is extended to 1999:2, the estimation of VAR is added by a dummy variable for the 1997 financial and banking crises (DUM). Estimation of various lag-lengths of VAR for estimating the lag-length of VAR for the model of real credit from state banks for the sample period 1984:1-1999:2 is reported in Table 6.15. Table 6.15: Test Statistics and Model Selection Criteria of VAR: LCREDITSR LFSASR LBISR CAPERS RLSR RSBIR Sample: 1984:1-1999:2 Deterministic and/or exogenous variables : C DUM lag-length 5 4 LL -405.01-469.27 AIC -625.27 SBC -784.63 LR test statistic ;i:'(36) = 128.52 3-503.21-623.21-745.80 j'(72) = 196.41 2-566.48-650.48-736.29 ;^'(108) =322.95 1-668.54-716.54-765.57 ;ir^ (144) =527.06 0-951.73-963.73-975.98 ;}f'(180) = 1093.4 Table 6.15 indicates that when the AIC selection criteria are used, the appropriate lag-length of VAR for estimating the Johansen cointegration test is 3. However, when the SBC selection criteria is used, the appropriate lag-length for the Johansen cointegration test is 2. In addition, test statistics of 108 restrictions of moving from VAR(5) to VAR(2) is 322.95 which is higher than the critical value of ''* Table 6.5a in Appendix 190
;i;^(108)at the 5% level which is about 132.97. It indicates that movmg to VAR(2) cannot be rejected. To examine whether VAR(2) is acceptable in terms of residual whiteness, the LM test is carried out using VAR(2) for each equation of the level of VAR for the sample period 1984:1-1999:2. Estimation of the LM test for individual equations by using VAR(2) indicates that there is serial correlation for real capital equity of state banks''*^ By using a higher order of LM, however, the serial correlation of the real capital equation is lower. On the other hand, the LM test of higher order of VAR indicates that there is a stronger serial correlation. It is argued that the existence of serial correlation in the equation for real capital equity in the sample period 1984:1-1999:2 might be related to the turbulent banking conditions following the 1997 fmancial and banking crises. As residual correlation of individual equations is minimised by using VAR(2), the VAR(2) is selected for estimating the Johansen cointegration test for real credit from state banks for the sample period 1984:1-1999:2. The results of the Johansen cointegration test for real credit from state banks for the sample period 1984:1-1997:2 and 1984:1-1999:2 are reported in Tables 6.16 and 6.17. Table 6.16 indicates that the null hypothesis of no cointegrating vector can be rejected at the 10%) level. However, based on the trace statistic, the null hypothesis of no cointegrating vector can be rejected at the 5% level. In addition, the next null hypothesis based on maximal eigenvalue of no cointegrating vector at most equal to one also can be rejected at the 10% level. It indicates that, based on maximal eigenvalue, there are two cointegrating vectors at the 10% level. Therefore, the long run relationship among the specified variables exists at the 10% level. The long mn ''" Table 6.6a in Appendix 191
relationship among the specified variable also exists for the sample period up to 1999:2. Table 6.17 indicates that the null hypothesis of no cointegrating vector for the sample period 1984:1-1999:2 can be rejected at the 5% level. The next null hypothesis of no cointegrating vector at most equal to one also can be rejected at the 5% level. It indicates that there are two cointegrating vectors by using the sample period 1984:1-1999:2. It suggests that there is a long ran relationship among the specified variables. In addition, the long mn relationship also exists in the estimation of the Johansen cointegration test for real credit from national private banks. 192
Table 6.16: Cointegration Test for Real Credit from State Banks Sample: 1984:1-1997:2 Variables in the cointegrated system: LCREDITSR LFSASR LBISR CAPERS RLSR RSBIR Maximal Eigenvalue of Stochastic Matrix Ho: r = 0 r<l r<2 r<3 r<4 r<5 Ha: r = \ r = 2 r = 3 r = 4 r = 5 r = 6 Trace of Stochastic Matrix Ho: r = 0 r<l r<2 r<3 r<4 r<5 Ha: r>l r>2 r>3 r>4 r>5 r = 6 Statistic 37.66 32.03 21.82 12.76 11.02 5.74 Statistic 121.02 83.37 51.34 29.52 16.76 5.74 Critical Value 95% 40.53 34.40 28.27 22.04 15.87 9.16 Critical Value 95% 102.56 75.98 53.48 34.87 20.18 9.16 90% 37.65 31.73 25.80 19.86 13.81 7.53 90% 97.87 71.81 49.95 31.93 17.88 7.53 193
Table 6.17: Cointegration Test for Real Credit from State Banks Sample: 1984:1-1999:2 Variables in the cointegrated system: LCREDITSR LFSASR LBISR CAPERS RLSR RSBIR Variable 1(0) included in the VAR: DUM Maximal Eigenvalue of Stochastic Matrix Ho: r = 0 r<l r<2 r<3 r<4 r<5 Ha: r = \ r = 2 r = 3 r = 4 r = 5 r = 6 Trace of Stochastic Matrix Ho: r = 0 r<l r<2 r<3 r<4 r<5 Ha: r>l r>2 r>3 r>4 r>5 r = 6 Statistic 123.55 39.41 27.48 14.73 10.67 0.33 Statistic 216.17 92.62 53.21 25.73 11.00 0.33 Critical Value 95% 40.53 34.40 28.27 22.04 15.87 9.16 Critical Value 95% 102.56 75.98 53.48 34.87 20.18 9.16 90% 37.65 31.73 25.80 19.86 13.81 7.53 90% 97.87 71.81 49.95 31.93 17.88 7.53 194
6.3.4. Johansen Cointegration Test for Real Credit from National Private Banks The cointegration test for real credit from national private banks is carried out for the sample period 1984:1-1997:2 and 1984:1-1999:2. The selection of lag-lengtii of VAR for estimating the Johansen cointegration test is determined by the model selection criteria and the LR test of various lag-lengths of VAR of level variables: LCREDITNR LFSANR LBINR CAPERN RLNR RSBIR, which are reported in Table 6.18. lag-length 5 4 Table 6.18: Test Statistic and Model Selection Criteria of VAR: LCREDITNR LFSANR LBINR CAPERN RLNR RSBIR Sample: 1984:1-1997:2 Deterministic and/or exogenous variables: C LL -133.29-197.26 AIC -341.26 SBC -477.47 LR test statistic ;i;'(36) = 127.93 3-243.35-351.35-453.51 ;(f'(72) =220.11 2-286.38-358.38-426.49 ;if'(108) =306.18 1-313.39-349.39-383.45 ;^^ (144) =360.20 0-816.37-816.37-816.37 ;^^(180) = 1366.2 Table 6.18 indicates that when the AIC selection criteria are used, the appropriate lag-length of VAR for estimating the Johansen cointegration test is 4. However, when the SBC selection criteria are used, the appropriate lag-length of VAR for the Johansen cointegration test is 1. In addition, the statistics of 144 restrictions of moving from VAR(5) to VAR(l) is 360.2 which is higher than the critical value of 195
;jf^(144)at the 5% level which is about 172.72. It suggests that movuig to VAR(l) cannot be rejected. To analyse whether VAR(l) is acceptable in terms of residual whiteness, the LM test is carried out using VAR(l) for each equation of the level of VAR for the sample period 1984:1-1997:2. The LM test for individual equations by using VAR(l), however, indicates the existence of serious serial correlation. To reduce the serial correlation problem, VAR(2) is selected. The LM test, by using VAR(2), for individual equations of the level of variable VAR indicates that there is no serial correlation problem for all individual equations'"^^. In addition, test statistics of 108 restrictions of moving from VAR(5) to VAR(2) is 306.2 which is higher than the critical value of 2"^(108) at the 5% level which is 132.97. Therefore, moving to VAR(2) cannot be rejected and VAR(2) can be selected for estimating the Johansen cointegration test for real credit from national private banks for the sample period 1984:1-1997:2. When the sample period is extended to 1999:2, the estimation of VAR is added by a dummy variable for the 1997 financial and banking crises. The model selection criteria and the LM test of the estimation of various lag-lengths of VAR for the sample period 1984:1-1999:2 are reported in Table 6.19. Table 6.19 suggests that when the AIC selection criteria are used, the appropriate lag-length of VAR for estimating the Johansen cointegration test for real credit from national private banks is 4. However, when the SBC is used, the appropriate lag-length of VAR for the Johansen cointegration test is 2. In addition, the test statistic for 108 restiictions moving from VAR(5) to VAR(2) is 403.63 which is '''^ Table 6.7a in Appendix 196