11. REVISION BIBLIOGRAFICA
2.1. MATERIA PRIMA (Jergón Sacha)
2.1.1.1. Nombres comunes
Based on Indonesian data from 1981.Q1 to 2016.Q2, the Gregory and Hansen (1996) cointegration test with three variant models is estimated for investigating the long-run relationship between each of the various monetary aggregates with their determinants (see Table 2.20). The selection of lag length is chosen based on routine of Eviews’ Gregory and Hansen separable program which is maximum 8 and is checked downwards to find the best lag using Akaike criteria (AIC) at the 5% significance level. The results show that all of the real monetary aggregates (RM1, RM2, RDIVM1 and RDIVM2) are cointegrated with using C/T model (level shift with trend) for both logarithm models and log-log models. The study uses the Gregory-Hansen cointegration tests with model C/T (constant and trend) assuming that there is no existence of regime shift due to intercept and slope change for Indonesia data during the period of study. The result of the break dates is very useful for anchoring the dummy
73 variables of structural break in the analysis of ARDL bounds testing in the next section. The break dates are consistent with the financial deregulation and the Asian Financial Crisis in Indonesia.
Table 2.20 Gregory-Hansen Cointegration Test Results with Structural Breaks
Models Money
The Gregory and Hansen (1996) are Model C (level shift), Model C/T (level shift with trend), and Model C/S (intercept and slope shift)
74 2.7.8 ARDL Bounds Test
Having observed that all of the real money aggregates (RM1 and RM2) and divisia money (RDIVM1 and RDIVM2) are cointegrated within the endogenous structural breaks for both logarithm models and log-log models, the study proceeds to estimate the cointegrating equation by using the ARDL method. This can eliminate apprehensions that the stationary degree among the variables is not the same at the first difference I(1). Before examining the ARDL Bounds Test, it is necessary to ensure that the unit root of any variables should not be integrated at I(2).
Based on Table 2.7 and 2.8, by using ADF and PP unit root test for first difference variable, all the variables are integrated at first difference. Therefore, there will be no variables that are stationary at I(2). The specification of the ARDL model for both logarithm models and log-log models is divided into four variants. Specifications 1 and 2 use the standard features of unrestricted constant and restricted time trend. However, Specification 2 includes an additional dummy variable of structural break. Meanwhile, Specifications 3 and 4 use the standard features of restricted constant and no time trend with a dummy variable of structural break, which is multiplied by the time trend in Specification 4. Dummy variables of structural break for both logarithm models and log-log models with respect to real monetary aggregates (RM1 and RM2) and divisia money (RDIVM1 and RDIVM2) are based on the result of the Gregory-Hansen cointegration tests with model C/T (constant and trend) discussed in Table 2.20 in the previous section, assuming that there is no existence of regime shift due to intercept and slope change for Indonesia data during the period of study.
After taking into account the dummy variables of structural break in the model, cointegration is examined using the Bounds Test unrestricted error correction model. To find the optimal lag, the Schwarz Information Criterion (SIC) is used, and the result of the estimation Bounds Test shown for logarithm models and for log-log models. The logarithm models with all variants with respect to real monetary aggregates and divisia money are shown in Table 2.21.
75 Table 2.21 The Results of the ARDL Cointegration of Logarithm Models (Bounds Test)
Specifications Types ARDL F-Statistics Results
RM1, LREALGDP, IDEP1M, LER, IUSTB3M: For Specification 1 and 2: Bounds F-test Critical Value at 10%, 5% and 1% levels, 3.53%, 3.97% and 4.92%, respectively
For Specification 3 and 4: Bounds F-test Critical Value at 10%, 5% and 1% levels, 3.09%, 3.49% and 4.37%, respectively
76 In determining the existence of a long-run relation of unrestricted error correction estimation of the models, the F-statistics should be higher, in particular the Bounds F-test Critical Value.
In regard to Specifications 1 and 2, the (upper bound) Bounds F-test Critical Value at the 10%, 5% and 1% levels is 3.53%, 3.97% and 4.92%, respectively, while for Specifications 3 and 4, the Bounds F-test Critical Value at the 10%, 5% and 1% levels is 3.09%, 3.49% and 4.37%, respectively. The study found that based on the Bounds F-test Critical Value, all of the real monetary aggregates (RM1 and RM2) and divisia money (RDIVM1 and RDIVM2) are cointegrated with their determinants which are real GDP 1 month deposit rate, exchange rate, and 3 months US Treasury Bills with at least 1% confidence level except for divisia broad money (RDIVM2) with variants Specifications 1 and 2 recorded at 5%. Meanwhile, the log-log models with all variants regarding real monetary aggregates (RM1 and RM2) and divisia money (RDIVM1 and RDIVM2) are shown in Table 2.22.
Similar to the logarithm model, the (upper) Bounds F-test Critical Value at the 10%, 5% and 1% levels is 3.53%, 3.97% and 4.92%, respectively and for Specifications 3 and 4, the Bounds F-test Critical Value at the 10%, 5% and 1% levels is 3.09%, 3.49% and 4.37%, respectively.
Regarding the Bounds F-test Critical Value, all of the real monetary aggregates (RM1 and RM2) and divisia money (RDIVM1 and RDIVM2) are cointegrated with their determinants which are real GDP 1 month deposit rate, exchange rate, and 3 months US Treasury Bills with 1% confidence level except for divisia broad money (RDIVM2) with variant Specification 2 noted at 10%.
After discovering that based on the ARDL method, both logarithm and log-log models provide solid evidence that cointegration exists for the relation between real monetary aggregates (RM1 and RM2) and divisia money (RDIVM1 and RDIVM2) and their determinants, the study proceeds to conduct a long-run estimation of the models. Table 2.23 shows the real narrow money (RM1) as the dependent variable.
77 Table 2.22 The Results of the ARDL Cointegration of Log-log Models (Bounds Test)
Specifications Types ARDL F-Statistics Results
RM1, LREALGDP, LDEP1M, LER, LUSTB3M: For Specification 1 and 2: Bounds F-test Critical Value at 10%, 5% and 1% levels, 3.53%, 3.97% and 4.92%, respectively
For Specification 3 and 4: Bounds F-test Critical Value at 10%, 5% and 1% levels, 3.09%, 3.49% and 4.37%, respectively
78 Table 2.23 The Long-run Linear ARDL Estimation of RM1 (Dependent Variable)
Long-run Estimation of Logarithm Models (Semi-Elasticities) INDEPENDENT
VARIABLES Specification 1 Specification 2 Specification 3 Specification 4
LREALGDP Coefficient 2.006 2.066 1.408 1.474
Asterisks (*), (**) and (***) indicate significant at 10%, 5% and 1% levels, respectively
It is found that within the logarithm models, the outperforming model variants are for Specifications 3 and 4 due to the appropriate coefficient and are statistically significant. In detail, the coefficient elasticity of real GDP is relatively close to unity and positive signs (around 1.4). Additionally, the 1 month deposit rate (domestic) and 3 months US Treasury Bills also show the correct sign based on theory but the magnitude is relatively low for all variants (around 0.01). Unfortunately, for the nominal exchange rate, the coefficient is positive which is contradictory to the theory. Meanwhile, within the log-log models, the coefficient elasticity
79 of real GDP for Specifications 3 and 4 is consistent (around 1.5). The sign of the other variables is in line theoretically, for example the coefficient for 1 month deposit rate and exchange rate is around 0.1 and has a negative sign. Additionally, 3 months US Treasury Bills also shows the correct sign based on theory but the magnitude is relatively low for all variants (around 0.04).
Real broad money (RM2) as the dependent variable is tabulated in Table 2.24.
Table 2.24 The Long-run Linear ARDL Estimation of RM2 (Dependent Variable)
Long-run Estimation of Logarithm Models (Semi-Elasticities) INDEPENDENT
VARIABLES Specification 1 Specification 2 Specification 3 Specification 4
LREALGDP Coefficient 16.536 2.564 1.687 -7.511
[t-statistics] [0.235] [1.514] [3.464]*** [-0.197]
Asterisks (*), (**) and (***) indicate significant at 10%, 5% and 1% levels, respectively
80 It is found that for both logarithm models and log-log models, the results are not satisfactory because all the variables are not statistically significant except for real GDP that has a coefficient elasticity around 1.6. Table 2.25 shows the real divisia narrow money (RDIVM1) as the dependent variable.
Table 2.25 The Long-run Linear ARDL Estimation of RDIVM1 (Dependent Variable)
Long-run Estimation of Logarithm Models (Semi-Elasticities) INDEPENDENT
VARIABLES Specification 1 Specification 2 Specification 3 Specification 4
LREALGDP Coefficient 0.975 1.086 1.280 1.163 Asterisks (*), (**) and (***) indicate significant at 10%, 5% and 1% levels, respectively
It is found that within the logarithm models, the outperforming model variants are for Specifications 3 and 4 due to the appropriate coefficient and are statistically significant except for exchange rate.
81 In detail, the coefficient elasticity of real GDP is relatively close to unity and positive sign (around 1.2). Additionally, the 1 month deposit rate (domestic) also shows the correct sign based on theory but the magnitude is relatively low for all variants (around 0.01). Meanwhile, within the log-log models, the coefficient elasticity of real GDP for Specifications 3 and 4 is consistent (around 1.3). Unfortunately, the sign of nominal exchange rate does not follow the theory. For example, the coefficient for 1 month deposit rate is around 0.14, while the 3 months US Treasury Bills also shows the correct sign based on theory but it is not statistically significant (around 0.01).
Table 2.26 shows the real divisia broad money (RDIVM2) as the dependent variable. It is found that in the logarithm models, the best performance of the model variants is for Specification 4, supported by the appropriate coefficient and is statistically important, except for 3 months US Treasury Bills. Furthermore, the coefficient elasticity of real GDP is positive at 1.9. Also, the 1 month deposit rate (domestic) also displays the correct sign based on theory but the magnitude is relatively low for all variants (around 0.01). Unfortunately, for the nominal exchange rate, the coefficient is positive which is contradictory to the theory.
Meanwhile, within the log-log models, the coefficient elasticity of real GDP for Specification 4 is above 2. The sign of the other variables is in line theoretically, for example the coefficient for 1 month deposit rate and exchange rate is around 0.1 and has a negative sign. Meanwhile, the short-run estimation of both the logarithm and log-log models with real narrow money (RM1) as the dependent variable has shown the important role of 1 month deposit rate and nominal exchange rate.
82 Table 2.26 The Long-run Linear ARDL Estimation of RDIVM2 (Dependent Variable)
Long-run Estimation of Logarithm Models (Semi-Elasticities) INDEPENDENT
VARIABLES Specification 1 Specification 2 Specification 3 Specification 4
LREALGDP Coefficient 3.505 3.868 1.703 1.892
Asterisks (*), (**) and (***) indicate significant at 10%, 5% and 1% levels, respectively
Nevertheless, the sign of the exchange rate in the short run is positive 0.1, indicating that in the short run currency substitution does not occur in Indonesia (see Table 2.27).
83 Table 2.27 The Short-run Linear ARDL Estimation of RM1 (Dependent Variable)
Short-run Estimation of Logarithm Models (Semi-Elasticities) INDEPENDENT
VARIABLES Specification 1 Specification 2 Specification 3 Specification 4
D (IDEP1M) Coefficient -0.006 -0.006 -0.006 -0.007
Constant Coefficient -4.475 -4.706
- -
Constant Coefficient -3.111 -4.805
- -
Asterisks (*), (**) and (***) indicate significant at 10%, 5% and 1% levels, respectively
84 For the error correction term, all variants have a negative value (-0.25) which is in line with theory, indicating that the speed adjustment to equilibrium is relatively fast. Meanwhile, the short-run estimation of the logarithm with real broad money (RM2) as the dependent variable has shown the important role of 1 month deposit rate and nominal exchange rate, including its first lag (see Table 2.28). Nevertheless, the coefficient of 1 month deposit rate is relatively low (0.003). The coefficient of first lag exchange rate is around -0.09, indicating that it takes time for households to do currency substitution due to the depreciation of local currency. In terms of the error correction term, all variants have a negative value (less than -0.06) which is in line with theory, indicating that the speed adjustment of equilibrium is relatively low when hit by a shock.
In addition, the short-run estimation of the logarithm with real divisia narrow money (RDIVM1) as the dependent variable has shown the important role of nominal exchange rate and 3 months US Treasury Bills rate (see Table 2.29). Meanwhile, the log-log model only depends on the nominal exchange rate. Nevertheless, the nominal exchange rate in the short run does not show that the currency substitution and 3 months US Treasury Bills rate is statistically insignificant. In terms of the error correction term, all variants have a negative value (above -0.3) which is in line with theory, indicating that the speed adjustment of equilibrium is relatively faster when hit by a disturbance. Moreover, the short-run estimation of the logarithm with real divisia broad money (RDIVM2) as the dependent variable has shown the important role of its lag 3, 1 month deposit rate and its first lag, nominal exchange rate, and 3 months US Treasury Bill rate (see Table 2.30). Meanwhile, the log-log model only depends on the 1 month deposit rate and its first lag as well as nominal exchange rate with its first lag.
The coefficient of first lag exchange rate is around -0.07, indicating that it takes time for households to do currency substitution due to the depreciation of local currency. In terms of
85 the error correction term, all variants have a negative value (above -0.1) which is in line with theory, indicating the slowing speed adjustment of equilibrium when hit by a disturbance.
Table 2.28 The Short-run Linear ARDL Estimation of RM2 (Dependent Variable) Short-run Estimation of Logarithm Models (Semi-Elasticities)
INDEPENDENT
VARIABLES Specification 1 Specification 2 Specification 3 Specification 4
D (IDEP1M) Coefficient -0.002 -0.002 -0.002 -0.002
Asterisks (*), (**) and (***) indicate significant at 10%, 5% and 1% levels, respectively
86 Table 2.29 The Short-run Linear ARDL Estimation of RDIVM1 (Dependent Variable) Short-run Estimation of Logarithm Models (Semi-Elasticities)
INDEPENDENT
VARIABLES Specification 1 Specification 2 Specification 3 Specification 4
D (LER) Coefficient 0.145 0.146 0.149 0.145
Constant Coefficient -1.64 -2.209
- -
Specification 1 Specification 2 Specification 3 Specification 4
D (LER) Coefficient 0.145 0.147 0.147 0.146
[t-statistics] [4.987]*** [5.141]*** [5.234]*** [5.121]***
Constant Coefficient -1.683 -3.665
- -
Asterisks (*), (**) and (***) indicate significant at 10%, 5% and 1% levels, respectively
87 Table 2.30 The Short-run Linear ARDL Estimation of RDIVM2 (Dependent Variable)
Short-run Estimation of Logarithm Models (Semi-Elasticities)
INDEPENDENT VARIABLES Specification 1 Specification 2 Specification 3 Specification 4
D (RDIVM2(-1)) Coefficient -0.055 -0.034 -0.064 -0.061
[t-statistics] [-0.817] [-0.491] [-0.985] [-0.977]
D (RDIVM2(-2)) Coefficient 0.301 0.335 0.242 0.238
[t-statistics] [4.924]*** [5.275]*** [3.922]*** [3.995]***
D (RDIVM2(-3)) Coefficient 0.162 0.202
- -
Constant Coefficient -5.398 -7.752
- -
Constant Coefficient -4.849 -5.351
- -
Asterisks (*), (**) and (***) indicate significant at 10%, 5% and 1% levels, respectively
88 2.7.9 Parameter Stability
The study applies CUSUM and CUSUMQ in order to examine the stability of the ARDL model. This approach uses ARDL estimation residual. Table 2.31 shows that the logarithm model does not have a serial correlation problem and heteroscedasticity problem, whereas the log-log model has a serial correlation problem except for Specification 2. In addition to that, in terms of the CUSUM and CUSUMQ square test, the parameter stability results are good for all variants.
Table 2.31 The Residual of ARDL Estimation and Parameter Stability Test (RM1)
Logarithm Models (Semi-Elasticities)
89 Table 2.32 shows that the logarithm model does not have serial correlation and heteroscedasticity problems except for Specification 1 and Specification 4, whereas the log-log model has a serial correlation problem except for Specification 2. Furthermore, the log-log-log-log model has a serial correlation problem and heteroscedasticity problem except for Specification 2 which has no heteroscedasticity problem. In addition to that, in terms of the CUSUM and CUSUMQ square test, the parameter stability results are not promising especially for CUSUM square test.
Table 2.32 The Residual of ARDL Estimation and Parameter Stability Test (RM2)
Logarithm Models (Semi-Elasticities)
90 Table 2.33 shows that the logarithm model does not have a serial correlation problem and heteroscedasticity problem. Similarly, the log-log model has no serial correlation problem and heteroscedasticity problem except for Specification 4. In addition to that, in terms of the CUSUM and CUSUMQ square test, the parameter stability results are good for all variants.
Table 2.33 The Residual of ARDL Estimation and Parameter Stability Test (RDIVM1)
Logarithm Models (Semi-Elasticities)
91 Table 2.34 The Residual of ARDL Estimation and Parameter Stability Test (RDIVM2) Logarithm Models (Semi-Elasticities) Specifications 3 and 4 for logarithm models. In addition to that, in terms of the CUSUM and CUSUMQ square test, the parameter stability results are not promising especially for CUSUM square test.
92 2.8 Conclusion
The chapter revisits divisia money for Indonesia following Habibullah (1999a) by adjusting the measurement of divisia money based on the Indonesian central bank’s official definition of simple-sum narrow and broad money, expanding quarterly data from 1981.Q1 to 2016.Q2 and presenting monthly data from 2003.M12 to 2016.M6. Many central banks have produced this money divisia regularly and use it in their monetary policy formulation and decision. In the case of Indonesia, the use of divisia money was not the main concern since Bank Indonesia still relies on simple-sum monetary aggregates. The nominal divisia money shows an upward trend after financial deregulation in the 1990s and the Asian Financial Crisis in 1997 despite a small difference between the simple-sum (M1) and divisia narrow money (DM1) due to the highest substitutability representing the narrow money components. Additionally, this might occur because the financial deepening in Indonesia not only relies on commercial banks but also non-bank financial institutions. The movement of divisia money is used as an information variable due to the pro-growth policy, as well as for projecting future economic prospects where the contractionary economy during crisis can be revealed through a significant increase in money supply, representing the uncertainty and public confidence in the financial system.
This chapter also re-examines and evaluates the connections between money, prices and economic activity in Indonesia. A money-in-utility model is calibrated to fit Indonesian data since the models can be applied to examine the determinants of the money demand function.
The simulated impulse responses are consistent with the main property of this model – money superneutrality. Consistent with previous studies, the results show that the growth of money supply shocks do not have any real impact and productivity shocks mostly influence the real variables, even though the property of superneutrality is experimented with various money aggregates (both simple-sum narrow and broad money, and divisia narrow and broad money).
93 The simulation also shows the impulse responses of various nominal money aggregates, nominal interest rate and inflation to productivity and each of the money growth shocks.
For the second objective, the chapter examines the Granger causality between the various real money aggregates, real income, the short-term nominal interest rate, the nominal exchange rate, and the foreign interest rate (the 3 months US Treasury Bills rate). The chapter used an array of cointegration tests and also incorporated an endogenous structural break and bounds test under the autoregressive distributed lag (ARDL) method. The test results show that the Fisher equation holds in Indonesia, indicating there is no need to incorporate the Fisher equation.
The results show that the standard Johansen test, the G-H test and ARDL bounds test (with a single structural break) found at least one cointegrating vector confirming that the real simple-sum narrow and broad money, and the real divisia narrow and broad money have a long-run relationship and that real money demand in Indonesia is stable over the sample period.
However, the results are mixed depending on the combination of model specifications with various measurements of monetary aggregates and the parameter stability tests. Nevertheless, the simple-sum M1 and divisia money M1 are still the superior choice for money demand in Indonesia in comparison to their M2 counterparts. This solid evidence concludes that divisia money can be used in the policy mix where merely relying on price stability will not yield optimal results. In particular, the explanation power of divisia money in Indonesia can support Bank Indonesia to formulate the optimal policy through the liquidity information in the financial market. Since optimal results cannot be obtained from merely relying on price stability, this solid evidence determines that money divisia can be used in the policy mix. In formulating the optimal policy through the financial market’s liquidity information, Bank Indonesia can be supported by the explanation power of divisia money in Indonesia.
94 Regardless of the property of superneutrality test results, assuming non-separability in the utility function will potentially change the conclusion and improve the analysis. Additionally, adding a labour-leisure choice might possibly eliminate the puzzle. Having evidence that the money demand in Indonesia is stable based on a single equation has to be examined for robustness by checking in the DSGE model or in the vector autoregressive system.
95