國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
17
Chapter 3: Methodology and Test Results 3.1 Methodology and Data
Using the model described by Bernanke and Blinder (1988), as shown in Chapter 2, a test was performed to see how interest rates would change during a change in monetary policy. The data within the model consists of the real GDP, reserve money supply, loan interest rate (INTEREST), and the overnight money rate7
After performing the necessary adjustments of the data, a cointegration test is performed with EViews to determine if there were any cointegrating equations. The final cointegration result showed that there were no cointegrating equations, therefore a vector autoregression (VAR) equation was set up. The VAR setup had GDPSA and MONEYRATE, and INTEREST as endogenous variables, while the exogenous policy variable is RMSA. From this set up, we will be able to see how RMSA will affect all other endogenous variables by looking at the coefficient’s sign and whether the resulting coefficient is statistically significant or not. The RMSA coefficient for MONEYRATE is of particular interest in this thesis, as it will give a clear indication of whether the change is positively related or inversely related. From this result, we are able to determine whether Taiwan’s credit channel follows the theoretical model or is different than in theory.
(MONEYRATE) from Taiwan between the first quarter of 1992 to the fourth quarter of 2009. The overnight money rate was chosen as a substitute to the bond rate described in Chapter 2 as Taiwan does not have an exact match of the bond rate as seen in the US or Europe. Real GDP, in million NT, is seasonally adjusted (GDPSA). Also, since the unit of reserve money was in million NT, it is seasonally adjusted (RMSA) to be consistent with GDP. Figures 3.1-3.4 show the trend line graph of GDPSA, RMSA, MONEYRATE, and INTEREST respectively.
7 隔夜拆款利率
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
18
Figure 3.1. Line graph of GDPSA.
Figure 3.2. Line graph of RMSA.
1500000 2000000 2500000 3000000 3500000
92 94 96 98 00 02 04 06 08
GDPSA
1200 1400 1600 1800 2000 2200 2400
92 94 96 98 00 02 04 06 08 RMSA
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
19
Figure 3.3. Line graph of MONEYRATE.
Figure 3.4. Line graph of INTEREST.
0 1 2 3 4 5 6 7 8
92 94 96 98 00 02 04 06 08
MONEYRATE
1 2 3 4 5 6 7 8 9 10
92 94 96 98 00 02 04 06 08
INTEREST
‧
The model involves using RMSA as an exogenous variable. After testing for the best lag length to use and whether there were any cointegrating equations, the conclusion was reached that for this model in particular, 3 lags would be used since the AIC value is the lowest at the 3rd lag level. Since there are no cointegrating equations, a VAR was used for this model. The final VAR equation is equaled to:
� GDPSAt
3.2.2. Effect on MONEYRATE
Because RMSA is an exogenous variable, we are unable to see the impulse response of MONEYRATE to RMSA. Due to this, we cannot see by impulse response how the overnight money rate will change when a monetary policy is implemented.
However, when looking at the coefficient of RMSA, we can see that it is positive (0.0001) and statistically significant at the 5% level. When considering the three possible results as described in equation (2.18), we can then conclude that a change in monetary policy will have a positive impact on the overnight rate, showing that if there were an expansionary monetary policy, the overnight money rate would rise.
This is similar to the case described in Figure 2.3, where the interest rate will rise.
3.2.3. Effect on GDPSA
Because RMSA is an exogenous variable, we are unable to see the impulse response of GDPSA to RMSA. Due to this, we cannot see by impulse response how total output will change when a monetary policy is implemented. However, equation (2.8) shows that in theory a change in the monetary policy will have a positive effect
‧
on total output. This positive effect is shown in the VAR output, as the coefficient of RMSA is positive (14.277). Therefore, this change does follow the theoretical model.
However, even though the coefficient is consistent in theory, it is not statistically significant at the 5% level. From the final results, we can conclude that while the theoretical model shows that there will be a positive impact on total output and real data does show this same effect, the data used in this test does not show significance in the change.
3.2.4. Effect on INTEREST
Because RMSA is an exogenous variable, we are unable to see the impulse response of INTEREST to RMSA. Due to this, we cannot see by impulse response how the loan rate will change when a monetary policy is implemented. However, when looking at the coefficient of RMSA, we can see that it is positive (0.0001), but not statistically significant at the 5% level. This is similar to the case described in Figure 2.3. This shows that while the effect is positive, the data does not show a significant result.
3.2.5. Considering 1997 Asian Financial Crisis and 2008 Global Financial Crisis
The date range of this model includes two major financial crises, the 1997 Asian Financial Crisis, and the 2008 Global Financial Crisis. To see the effects of both crises, two dummy variables, one for the 1997 Asian Financial Crisis (D97) and the other for the 2008 Global Crisis (D08), were placed in the model as exogenous variables. When these dummy variables were added into the model, it showed that D97 is statistically significant for MONEYRATE only, D08 is statistically significant for GDPSA only, and D97 and D08 are not significant for INTEREST. This shows that the 1997 Asian Financial Crisis had a positive effect on the change in money rate, while the 2008 Global Financial Crisis had a negative effect on GDP. Also, the addition of these dummy variables into the model did not have a change on the statistical significance of RMSA to the dependent variables, as the coefficient for MONEYRATE is stillstatistically significant at the 5% level, while the coefficients for GDPSA and INTEREST were not statistically significant at the 5% level.
3.3. Summary
Table 3.1 shows the results of the coefficients, T-statistics and whether the coefficient is significant for the respective variables to RMSA:
‧
Table 3.1. Dependent Variable Effects to Monetary Policy Shift
Coefficient T-statistic Significant?8
MONEYRATE 0.0001 2.952 Yes
GDPSA 14.227 0.714 No
INTEREST 0.0001 1.077 No
Source: EViews results
From the above results, when there is a monetary policy shock, the money rate change is positive. This is similar to the scenario described in Figure 2.3, which is opposite of what the theoretical model showed. As theory has shown that for either the bond or loan rates to increase, a monetary tightening policy has to be
implemented. However, actual data shows that the change is positive, meaning that a monetary expansion policy will lead to an increase in the overnight money rate.
This result shows that Taiwan does not necessarily follow the traditional theory as a monetary policy change has a positive effect on the overnight money rate. Also, when looking at the change in the loan rate, the result shows that while the change is positive, it is not statistically significant. Although the data still shows that the change in positive, meaning that a monetary expansion policy will increase the loan rate which also contradicts the theoretical model, the change in not statistically significant at the 5% level. Although a monetary policy change will have a positive effect on the change in output, which is the same as the theoretical model, the data does not show significance at the 5% level.
Table 3.2 shows the results of the coefficients, T-statistic, and significance for RMSA and the respective dummy variables to the other dependent variables:
Table 3.2. Dummy Variables Results to Monetary Policy Shift
MONEYRATE GDPSA INTEREST
D97 1.082/4.197/Y910 10366.50/0.451/N -0.001/-0.016/N D08 0.004/0.016/N -84547.53/-3.392/Y 0.041/0.624/N RMSA 0.001/2.749/Y 27.567/0.714/N 0.0001/0.974/N Source: EViews results
After adding dummy variables for the two major financial crises during the model’s time frame, the 1997 crisis shows a statistically significant impact on the
8 5% Level
9 Coefficient/T-statistic/Significant at 5% level
10 Y=yes, N=no
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
23
money rate, the 2008 crisis shows a statistically significant impact of total output, but neither shows a statistically significant impact on the loan rate. By looking at the coefficients, we can also see that the 1997 Crisis has a positive effect on the
overnight money rate, showing that this crisis led to a jump in the money rate. Also, the 2008 Crisis has a negative effect on totally output, showing that this crisis led to a significant drop in total output. These results also show that the 1997 Crisis had a positive effect on total output and a negative effect on the loan rate, and the 2008 Crisis had a positive effect on the money rate and a positive effect on the loan rate, but these changes are not statistically significant. Also, even though the dummy variables have been added to the model, the signs of the coefficients and the significance levels of the policy variable are the same as the previous model when the dummy variables were not present.
‧
Bernanke and Blinder (1988) showed in their paper that it is possible that the bond rate would increase in the event that income increases. Their model showed that the traditional IS-LM model that was already in place is inadequate in
determining the exact effects of the bond rate when bank reserves increase. Once the loan market was placed into the IS-LM model to become their CC-LM model, a monetary expansionary policy would shift both the LM and CC curves, rendering the bond rate change undetermined. As described, there are cases where the bond rate could increase, decrease, or remain constant. These cases all depend on whether the LM curve shift is greater than the CC curve, the other way around, or if both curves shifts are equal to each other. Also, after solving the equations, Bernanke and Blinder also showed that the same monetary policy and resulting possible shifts in the LM and CC curves could result in different changes in the loan rate, too.
With reserve money set as the exogenous policy variable, the model showed that a change in monetary policy will have a positive impact on the overnight rate. In previous theories, such a change would have an inverse impact on the overnight rate, but what is shown here is that Taiwan’s credit channel doesn’t necessarily follow the previous theoretical model. Also, the model showed that while there is a positive change in the loan rate, this change is insignificant. A possible explanation is that in Taiwan’s credit channel, the loan rate is unaffected by any monetary policy change.
This means that any monetary policy change will only affect total output and the loan rate will remain constant. Finally, as the date range of this model covers the 1997 Asian Financial Crisis and the 2008 Global Financial Crisis, setting up a dummy variable to test the impact showed that the 1997 Crisis had a more significant impact on the money rate, while the 2008 Crisis had a more significant impact on total output. Even so, when looking at the changes in the economy based on an IS-LM model, it showed that when there is a monetary policy change, the changes in the money rate and output levels are of the same signs as is under the CC-LM model, suggesting that in Taiwan, the IS-LM changes also contradict the theoretical model.
However, as this paper dealt with the overnight money rate as a substitute to the bond rate, it is possible that other close substitutes would show a different effect.
Also, from Bernanke and Blinder’s paper, their money market equation includes a reserve requirement ratio, which can also change the bond rate when the ratio changes. Future studies can place an emphasis on this aspect, or a close substitute to the bond rate to determine how the credit channel operates in Taiwan.