Bank risk in a decade of low interest rates

The paper uses the ratio of non-performing loans, the ratio of loan loss reserves and loan loss provision to measure banks’ credit risk.

The risk-weighted off-balance-sheet assets including financial derivatives and other off-balance-sheet assets are calculated based on 0 %, 20 %, 50 % and 100 % conversion factors regulated by the two-step approach of the regulatory capital rule. For instance, unused commitments with an original maturity of one year or less are in the 0 % credit conversion factor group; a commercial letter of credit is in the 20 % credit conversion factor group; unused portions of commitments, including home equity lines of credit, with an original maturity exceeding one year or that are unconditionally cancellable are in the 50 % credit conversion factor group; guarantees or financial guarantee-type standby letters of credit, recourse obligations and direct credit substitutes, and forward agreements with a certain drawdown are in the 100 % credit conversion factor group. A financial institution must determine the credit equivalent amount of its off-balance sheet financial derivative contracts that are not subject to qualifying bilateral netting contracts.

One anonymous reviewer noted that the correlation between interest income and non-interest income is higher than 0.9, as shown in Table 2. We acknowledge the issue and explore the relationship between contemporaneous value of one variables against a one-period lag of the other variable. The result which is not presented in Table 2 suggests a weaker correlation, close to −.35. Therefore, collinearity issue is unlikely to be significant.

The difference between rate-sensitive assets and rate-sensitive liabilities.

We also perform a Zivot-Andrews (1992) unit root test and the result only suggests one structural break at 2006q1, which does not fit the pattern of 10-year T-Bond rate shown in Panel B of Figure 1.

Arellano and Bover (1995) also discussed dynamic panel data models concerning the use of predetermined instrumental variables.

We performed two-step GMM estimations. Two-step GMM estimation fails to reject the null hypothesis of zero serial correlation up to the fifth order, therefore up to 5 periods of lags of the dependent variable are included in the GMM model.

We adopted an anonymous reviewer’s suggestion to use the difference between individual observations from its overall average for the variables contained in the interaction terms. This allows us to isolate the impact of dummy variable (Crisis Period and Post-2011) on the key variables of 10 year Treasury rate and GDP growth rate. We thank the anonymous reviewer for this suggestion.

We also included an interaction term of lagged off-balance-sheet item and time dummy variables, as one reviewer’s suggestion based on Balli and Sorensen (2013) that we should consider the impact of regulatory change during the financial crisis period on some off-balance-sheet activities. The results of these additions are not materially different from those presented in Table 4.

We change the order of the variables included in the system and find the results do not change materially. Results are available upon request.

Due to space limitations, we do not include the results of the 6-variable Panel VAR model and impulse response functions. Regression results are available upon request.

Those results are not presented here but are available upon request.