7. Results
7.1 The countercyclical capital buffer on bank risk
Results for model specification (1) are presented in Table 3. In columns (I) and (IV), the two bank risk variables are regressed only on lagged dependent variables and bank-specific control variables. The coefficients for the instrumented lagged dependent variables do not suggest there exist self-mitigating forces for bank risk, which cause next period’s risk to grow more slowly than previous period. Instead, loan growth in the current period positively reacts to loan growth in the previous period, which implies that loan growth behaves pro-cyclically.
The coefficient for the lagged leverage growth rate on the other hand, is not statistically significant. The coefficients for bank specific variables confirm earlier expectations. First, banks with higher leverage ratios tend to have lower growth rates for the leverage ratio, and higher growth rates for outstanding loans. This relationship exists as banks that are better capitalised have more room to increase their risk profile. Second, banks which are more reliant on deposit funding are expected to decrease their leverage ratio and increase their lending activity more than wholesale funded banks. Thus, it seems that the latter banking category is less keen on increasing its risk profile due to negative market reactions.
Next, columns (II) and (V) consider the effects of individual macroprudential policy instruments. They show that the countercyclical capital buffer has been successful in increasing the growth rate of the leverage ratio, and reducing the growth rate of loans. More specifically, a one percent tightening of the CCyB increases the growth of the leverage ratio by 1.9 percentage points, and reduces the growth rate of loans by 2.3 percentage points. In columns (III) and (VI), the coefficients for the interaction between the CCyB and the lagged dependent variable are not statistically significant. This implies that the CCyB is not additionally effective when the financial cycle is more extreme. The results for the baseline regressions provide support for the first element of hypotheses H1 and H2, namely that the CCyB reduces bank risk. It does not provide support for the hypotheses’ second element however, as the CCyB does not seem to dampen the cyclical component of bank risk. When other macroprudential policies are considered, the LTV and DTI caps seem to be effective in increasing the growth of the leverage ratio, while they positively influence the growth of loans.
The latter effect is counterintuitive, and raises the concern of selection bias in the data sample.
31 After all, countries that have implemented borrower-based instrument might suffer more from an accumulation of risks in the banking sector. The characteristics of dummy variables enforce this bias, as it does not allow for policy variability. Dynamic provisioning has no significant impact on bank risk, however, it is expected to decrease the growth rate of the leverage ratio and outstanding loans whenever the financial cycle is more intense. Also, Table 3 shows no sign of second order autocorrelation and an overidentification of instrumental variables.
In Table 4, the results are presented for the baseline regressions in which the sample is split according to the phase of the financial cycle. It shows that the countercyclical capital buffer is only effective during the expansionary phase of the cycle. During an expansion, a one percent tightening of the CCyB increases the growth of the leverage ratio by 2.1 percentage points, and reduces the growth rate of loans by 3.1 percentage points. During a contraction, coefficients for the CCyB point towards the same direction, but are not statistically significant.
These results support hypothesis H3 and suggest that banks prefer holding on their voluntary buffers when market conditions are uncertain, while they weaken the notion that the CCyB is not a binding constraint during a boom. Again, Table 4 does not provide evidence for autocorrelation, but the p-values for the Hansen J-test in column (I) and (IV) fall outside Roodman’s confidence boundary. Nevertheless, in both cases the null hypothesis of instrument exogeneity is not rejected.
32
Table 3
Baseline regression results of macroprudential policies: 2006-2020
Explanatory variables Leverage growth Loan growth
(I) (II) (III) (IV) (V) (VI)
Lag dependent var. 0.067 Yes Yes 0.343*** Yes Yes
(0.054) (0.025)
Lag leverage ratio -1.249*** Yes Yes 0.518*** Yes Yes
(0.121) (0.100)
Lag deposit ratio -0.121*** Yes Yes 0.022** Yes Yes
(0.010) (0.011)
CCyB 1.856*** 1.561** -2.250*** -1.621***
(0.390) (0.357) (0.416) (0.326)
CcyB x lag dependent var. 0.071 -0.053
(0.098) (0.055)
LTV 2.386*** 2.047*** 1.579*** 0.837***
(0.564) (0.372) (0.342) (0.238)
LTV x lag dependent var. 0.032 0.102***
(0.076) (0.035)
DTI 2.104*** 1.137*** 1.796*** 1.089***
(0.679) (0.439) (0.350) (0.257)
DTI x lag dependent var. 0.399*** 0.039
(0.073) (0.039)
DP 1.411* 1.831*** 0.101 0.141
(0.807) (0.287) (0.572) (0.182)
DP x lag dependent var. -0.244*** -0.123***
(0.064) (0.033)
Constant 11.065 *** Yes Yes -1.516** Yes Yes
(0.686) (0.647)
AR(2) test 0.586 0.590 0.534 0.568 0.476 0.828
Hansen test of overid. 0.157 0.180 0.210 0.113 0.116 0.193
Observations 1,726 1,726 1,726 1,688 1,688 1,688
Notes: In these GMM regressions, lagged dependent variables, the lagged credit-to-GDP gap, time fixed effects and country fixed effects are used as instrumental variables for the lagged dependent variable. To save space, each macroprudential policy variable is regressed one at a time and results are stacked in one column. In the regressions where macroprudential policies are incorporated, if results are shown for the lagged dependent variable, bank-specific variables and model validity tests, they represent the regression which only uses CCyB as policy instrument variable. GMM robust standard errors are reported in the parentheses. The asterisks *, **
and *** stand for significance at the 0.1, 0.05 and 0.01 level, respectively.
33
Table 4
Differentiating macroprudential policy effectiveness by phase of the cycle Explanatory
variables
Leverage growth Loan growth
Expansion (I) Contraction (II) Expansion (III) Contraction (IV)
Lag dependent var. -0.023 0.102** 0.328*** 0.303***
(0.059) (0.046) (0.037) (0.033)
Lag leverage ratio -1.189*** -0.947*** 0.462*** 0.595***
(0.182) (0.131) (0.142) (0.160)
Lag deposit ratio -0.015 -0.050*** 0.027* 0.049***
(0.010) (0.013) (0.014) (0.017)
CCyB 2.019*** 0.716 -3.137*** 0.272
(0.493) (0.702) (0.593) (1.160)
LTV 0.756 3.356*** 3.807*** 0.891
(0.698) (0.578) (0.489) (0.574)
DTI 0.474 0.474 3.700*** 0.898
(0.846) (0.846) (0.455) (0.812)
DP 1.916 1.916 -2.739*** 0.202
(1.349) (1.349) (0.962) (0.700)
Constant 9.795*** 9.042*** -0.904 -3.585***
(0.756) (0.937) (0.630) (1.074)
AR(2) test 0.307 0.314 0.543 0.395
Hansen test of overid. 0.636 0.223 0.252 0.296
Observations 913 930 913 930
Notes: In these GMM regressions, lagged dependent variables, the lagged credit-to-GDP gap, time fixed effects and country fixed effects are used as instrumental variables for the lagged dependent variable. To save space, each macroprudential policy variable is regressed one at a time and results are stacked in one column. In the regressions where macroprudential policies are incorporated, if results are shown for the lagged dependent variable, bank-specific variables and model validity tests, they represent the regression which only uses CCyB as policy instrument variable. GMM robust standard errors are reported in the parentheses. The asterisks *, ** and *** stand for significance at the 0.1, 0.05 and 0.01 level, respectively.
34