The impact of macroprudential policies on bank risk under economic uncertainty: Evidence from emerging Asian economies

Abstract

This paper examines the impact of macroprudential policies on the nexus between economic uncertainty and bank risk in emerging Asian economies. We construct our index of economic uncertainty by applying the Generalized Autoregressive Conditional Heteroskedasticity (GARCH)-in-mean model to a series of important macroeconomic variables, and borrow macroprudential policy indices from Cerutti et al. (2017) and Alam et al. (2019). Using bank-level panel data from approximately 600 commercial banks in 11 emerging Asian economies during the period 2000–2016, we find consistent evidence that bank risk increases with economic uncertainty, while macroprudential measures play an ameliorative role in uncertainty-induced bank risk. Our baseline findings are largely driven by macroprudential measures that aim to dampen the credit cycle more than those that target increasing the resilience of the banking sector. Our further analyses of the heterogeneous impacts of macroprudential measures by specific type on the risk of banks show that liquidity-based instruments, reserve requirements and currency instruments play a more conspicuous role in the economic uncertainty‒bank risk nexus than capital-based and asset-side macroprudential instruments.

Keywords:

• Economic uncertainty
• bank risk
• macroprudential policy
• emerging Asia

JEL classification:

• G21
• G15

Acknowledgement

Minghua Chen thanks the financial support by the National Social Science Fund of China (No. 18BJY245).

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 Various forms of macroprudential actions have been implemented more frequently in emerging economies than in advanced countries (Cerutti et al. 2017; Altunbas, Binici, and Gambacorta 2018; Alam et al. 2019). Evidence of relatively larger average effects in emerging market countries for liquidity and housing macroprudential policies compared with advanced and low-income countries has been reported in the literature (for example, Araujo et al. 2020). Kim, Kim, and Mehrotra (2019) reports that the frequency of macroprudential measures has generally increased after the 2007-2008 Global Financial Crisis in Asian countries.

2 A developing economy is conventionally classified as an emerging economy if it has some characteristics of developed countries. Typically, emerging economies own some, even though not all, of the following features: (1) A rapid transmission from low-income, less developed, pre-industrial economies to relatively high income, modern and industrialized economies; (2) Promising potential for further development; (3) Increasingly integrated with the global economy, in terms of not only their goods market but also financial markets. Due to the lack of official classification or definition to emerging economies, we select our sample of emerging Asian economies by following some prior works (e.g., Soedarmono, Machrouh, and Tarazi (2013) and Wu et al. (2020)).

3 For example, Cukierman and Meltzer (1986) and Grier and Perry (2000).

4 Numerous works have addressed the economic relevance of uncertainty that is sourced from output growth, inflation, and exchange rate (e.g., Fountas and Karanasos (2007), Caporale, Menla Ali, and Spagnolo (2015), and many others). We measure output growth as the monthly difference of the logarithm of industrial production index, inflation rate as that of the logarithm of consumer price index, and currency depreciation rate as that of the logarithm of the domestic currencies’ foreign exchange rate against the U.S. dollar. All these variations are seasonally adjusted and annualized.

5 We ensure the stationarity of the three series by using the Augmented Dickey-Fuller test.

6 To determine the optimal lags, N, for each variable’s mean equation, we use the Akaike Information Criterion (AIC) and the Schwarz Criterion (SC), and then adjust the lags to secure clean residuals that pass various diagnostic tests. First, we conduct the Lagrange Multiplier (LM) test proposed by Engle (1982) for the presence of conditional volatility in each of interested variables in each sample economy and find favorable results overall, suggesting the GARCH model as a reasonable choice for these series. Second, we check if there is evidence for any remaining patterns in the residuals by calculating the Ljung-Box Q-statistics for up to 6 and 12 lags of the levels of the standardized residuals in the estimated GARCH-in-mean systems for each variable in each country. Insignificant Q(6) and Q(12) statistics, as we find generally, suggest that adequate number of lags are included in our specifications such that the standardized residuals are not serially correlated. Third, we compute the Ljung-Box Q²(6) and Q²(12) statistics, which test for the sixth- and twelfth-order serial correlation in the squares of standardized residuals. We find again that these statistics are insignificant in almost all cases, which is interpreted as that our models adequately capture the conditional heteroskedasticity in the process of output growth, inflation and exchange rate depreciation. The specific results are available upon request.

7 It is noteworthy that we do not argue that our indicator of economic uncertainty outperforms other uncertainty indicators in prior works. To construct a superior indicator of economic uncertainty is not the purpose of this paper. All extant uncertainty indicators have their own pros and cons and it would be hard to conclude, and we also believe it should be concluded, which one is better than the others.

8 Economic uncertainty in some Asian economies (e.g., Korea and Indonesia) in the early years of 2000s is found higher than their long-term norm, probably due to the aftershock of Asian financial crisis.

9 See, for example, Laeven and Levine (2009) and Demirgüç-Kunt and Huizinga (2010).

10 To check the robustness of our main results, we use two alternative series of MPI, based on Cerutti et al. (2017) and Alam et al. (2019), respectively, when we conduct different regressions.

11 Using one-year lagged observations of the uncertainty indicator is in line with the practice of Wu et al. (2020), who assume that it would take some time for bank risk to respond to economic uncertainty.

12 Before we conduct our estimation, we conduct the Fisher-type test to check the stationarity of our variables. For all variables, our results reject the null hypothesis that the series under examination is non-stationary.

13 We examined the heterogeneous impact of macroprudential policies on the economic uncertainty-bank risk nexus across different economies. See Appendix 2 for details.

14 Our classification of macroprudential instruments is slightly different from the practice of Ely et al. (2021), which clubs macroprudential tools into four categories, namely, structural tools, borrower-based, capital-based and asset-based tools.

15 We also examined the impact of individual macroprudential tools on the nexus between economic uncertainty and bank risk. As there were in total 16 macroprudential measures employed by central banks and financial regulators in our sampled economies, we add the indicators for each macroprudential measures and their interaction with the indicator of economic uncertainty in our regressions. We find some evidence for that, reserve requirements, two foreign currency instruments, dynamic loan loss provisions and limits on credit growth buffers the adverse impact of economic uncertainty on bank stability, while liquidity requirements, limits on bank leverage and limits on the loan-to-deposit ratios seemingly cause higher risk when economic uncertainty arises. Our results are qualitatively consistent with some earlier works which find heterogeneous, even competing, impact of different macroprudential policies on bank risk. The detailed results for our above examination are available upon request.

16 As the fixed effects estimator would yield inconsistent estimators when the time dimension is limited in such dynamic specifications (Nickell 1981), we alternatively use the dynamic Generalized Method of Moments (GMM) panel estimator. The lagged dependent variable is treated as endogenous.