Negative interest rates policy and banks’ risk taking Empirical evidence 1 Negative Interest Rates Policy and Banks'''' Risk Taking Empirical Evidence Whelsy Boungou* 4 October 2019 Abstract Using a pane[.]
Version of Record: https://www.sciencedirect.com/science/article/pii/S0165176519303817 Manuscript_ef6b8982e9a8b56ac71bb20754b27c9f Negative Interest Rates Policy and Banks' Risk-Taking: Empirical Evidence Whelsy Boungou* October 2019 Abstract Using a panel dataset of 9,421 banks from 59 countries over the period 2009-2018 and a Difference-in-Differences estimator, this paper aims to assess the effects of negative interest rates on banks' risk-taking We find that banks' risk-taking has been lower in countries where negative rates have been implemented This effect depends on the characteristics of a country's banking system, namely the level of capitalization and size Keywords: Negative Interest Rates, Bank Risk-Taking, Difference-in-Differences JEL: E43, E52, E58, G21 * University of Bordeaux – LAREFI E-mail address: whelsy.boungou@u-bordeaux.fr © 2019 published by Elsevier This manuscript is made available under the Elsevier user license https://www.elsevier.com/open-access/userlicense/1.0/ « Negative rates will not provoke the collapse of the financial system »1 (Mario Draghi) Introduction Since 2012, several central banks in Europe and the Bank of Japan have gradually introduced a negative interest rate policy (NIRP) The NIRP complements asset purchases and forward guidance implemented since the Global Financial Crisis to ensure that the economy is sufficiently stimulated According to Cœuré (2016), the main reason for the introduction of negative rates is cash NIRP aims to increase the supply of credit by taxing banks' excess reserves at the central bank and, in fine, to support growth These effects on growth can be achieved through the portfolio rebalancing channel Jobst and Lin (2016) argue that the portfolio rebalancing with negative rates reduces term and credit risk premia, eases financial conditions and supports credit and economic activity As a result, NIRP improve borrowers' creditworthiness while reducing loan losses provisions However, a prolonged period of negative rates could also pose financial stability problems, by encouraging banks to seek more profitable assets other than credit, consistent with a risktaking channel (Borio and Zhu, 2012) Against this backdrop, the purpose of this paper is to evaluate the effects of negative rates on banks' risk-taking Following Altunbas et al, (2012), we examine whether these effects on risk-taking would depend on the characteristics of banks (their level of capitalization and size) The literature on negative interest rates is burgeoning and has focused on their effects on bank profitability (Molyneux et al, forthcoming), the lending channel (Eggertsson et al, 2019; Heider et al, 2019), and on exchange rate (Thornton and Vasilakis, 2019) We contribute to this literature, by investigating first the impact of negative rates on banks' risk-taking and, second how bank characteristics can influence the transmission of negative rates to risktaking This paper differs from existing contributions in two ways First, while Nucera et al, (2017) focus on banks' systemic risk, we focus on banks' individual risk, which has not been explored until now A second novelty is the use of a unique large panel dataset covering 9,421 banks located in 59 countries for the period 2009-2018 This goes beyond existing studies which typically look at single countries in a domestic context (Nucera et al, 2017; Eggertsson et al, 2019; Heider et al, 2019) We measure banks' risk-taking, using three complementary measures related to banks' balance sheets: Risk-weighted assets, Z-score and provisions Using the Difference-inDifferences methodology, the main result is that banks located in countries where negative rates have been implemented have taken less risk in the years following the introduction of negative interest rates In addition, our results show that the effects of NIRP on risk-taking have been stronger for banks that are small and well capitalized ECB Press Conference at Frankfurt am Main on 12 September 2019 The remainder of this paper is organized as follows Section presents our empirical methodology, including the data Section summarizes the empirical findings The last section concludes Data and empirical methodology 2.1 Data Using a large unbalanced panel dataset over the period 2009-2018, we examine the effects of NIRP on banks’ risk-taking We assemble a dataset from several sources (Fitch Connect, central banks, OECD and Datastream) Our final database consists of non-consolidated data from 9,421 banks operating in 59 advanced and emerging countries Our group of countries is divided into two subgroups (treated and control) Treated is the group of banks operating in countries where negative rates have been implemented (Bulgaria, Denmark, Euro Area, Hungary, Japan, Norway, Sweden and Switzerland) and conversely, control is the group of banks operating in countries that have not applied negative rates We have sorted our database by deleting missing bank data and winsorizing the data at the 1st and 99th percentile level to ensure that outliers not bias our estimates (e.g., when assets are less than zero or provisions are below zero) To measure banks’ risk-taking, we focus on three variables widely used in the banking literature: the ratio of risk-weighted assets to total assets (RWA/Total assets), the changes in log-transformed Z-score (a higher Z-score means that the bank takes less risk.)2, and the ratio of loan loss provisions to gross loans (Provisions) A high ratio would indicate an increase in the banks’ risk-taking We acknowledge that these measures could also be driven by decisions before 2008 such that they might not only reflect risk-taking that occurred during the sample period However, any changes in these measures should - at least in part - reflect changes in new activities and therefore in risk-taking behavior3 We control for a set of bank-specific factors that are well known to influence banks' risktaking To account for the liquidity level of banks, we use liquid assets to total assets (Liquidity) As other banking characteristics, we use banks capitalization as equity to assets ratio (Capitalization) We include deposits as customer deposits to total assets (Deposits) Finally, we introduce the natural logarithm of total assets to control for banks’ size (Size) The banking literature suggests that the environment in which banks operate may have effects on their behavior In order to control for the banking market structure, we use the Herfindhal-Hirschman Index (HHI) We also control for inflation and real GDP 2.2 Empirical methodology The Z-score is computed as return on assets (ROA) plus the capital ratio divided by the standard deviation of ROA As Beck et al, (2013), we use a three-year rolling window to compute the standard deviation of ROA Using the volatility of equity, the natural logarithm of the Z-score (proxy with the return on equity) and the logarithm of risk-adjusted assets profits (measured as the return on assets divided by the volatility of assets) in robustness, we find similar results (available on request) To examine the effect of negative interest rates on banks’ risk-taking we use a Difference-inDifferences (DiD) methodology We compare the effects of negative rates on risk-taking for a treatment group of banks (Treated) with a control group of banks (Control) unaffected by NIRP Equation (1) summarizes our baseline model: , , = + , + , + , ∗ , + , + + +Ɛ, , (1) Where , , is the risk-taking measures for the bank i in country j at year t , is a dummy variable equal to if bank i in country j is affected by NIRP, otherwise , is a dummy variable equal to in years following implementation of NIRP The coefficient of is our DiD estimator in average banks’ risk-taking behavior between treated and control groups4 , refers to both bank-specific and country-specific controls , and Ɛ , , are respectively time fixed-effect, time invariant bank fixed-effects and idiosyncratic error In addition, based on the analysis of Rosenbaum and Rubin (1983), we check the robustness of our results by combining DiD and Propensity Score Matching and find similar results (available on request)5 Main Findings Table presents our main results Standard errors are robust and clustered at bank level We use three complementary measures of banks' risk-taking Our DiD estimator (denoted in the tables as the NIRP-Effect) is found to be significant across these measures Overall, our results show that the increase in risk-taking has been lower among banks operating in countries where negative rates have been implemented, compared to the group of banks not affected by negative rates6 These results are consistent with those of Nucera et al, (2017), which analyze the reactions of euro area banks’ systemic risk reactions to rate cuts in negative territory7 Following Altunbas et al, (2012), we analyze whether the effects of negative rates on risktaking differ according to bank characteristics (including capitalization and size) Table reports the results of the evaluation of the effects of negative interest rates on banks' risktaking according to the capitalization (less or more) and the size (small or large) We show that small banks, located in countries where NIRP have been implemented, have taken relatively less risk compared to large banks (see Panel A and B) In addition, better capitalized banks have relatively taken less risk compared to less capitalized banks (Panel C and D) As a result, both small and better capitalized banks have been successful in mitigating the effects of NIRP on risk-taking This issue of bank heterogeneity corroborates the results of Heider et al, (2019) that banks specific characteristics can influence the effects of negative rates on banks' risk-taking behavior Using the Variance Inflation Factor (VIF), we test the control variables for multicollinearity A mean VIF of 1.85 suggests that our control variables are not highly correlated Our results are robust to considering the level of country-specific policy rates or the real interest rates It should be noted that our DiD estimator only represents the difference between the average risk-taking of banks in the post-NIRP period and in the pre-NIRP period This average effect may also result from other differences such as differences in bank business models Bongiovanni et al, (2019), also found a reduction in banks' holdings of risky assets, in countries where negative rates have been implemented 4 Table 1: Effect of negative NIRP on risk-taking RWA/Total assets Log(Z-score) Provisions 0.201*** -0.153 *** (0.06) (0.05) -0.504*** 0.836 *** (0.14) (0.29) 0.015*** -0.008 (0.00) (0.01) -0.239*** 1.623 *** (1.69) (0.07) (0.34) 0.786 -0.240*** -0.327 *** (0.51) (0.05) (0.11) -10.605 -2.311* 1.506 (10.68) (1.20) (1.26) 0.149 *** -0.025*** -0.001 (0.05) (0.01) (0.01) -0.041 -0.014 -0.065 *** (0.05) (0.01) (0.01) 58.356 *** 6.464*** 2.009 ** (4.17) (0.36) (0.89) Observations 38272 49698 53979 Number of banks 6679 9026 9421 2009-2018 2011-2018 2009-2018 0.119 0.006 0.011 Year FE Yes Yes Yes Bank FE Yes Yes Yes NIRP-Effect -5.319 *** (0.31) Liquidity -29.007 *** (1.66) Capitalization 0.200 *** (0.05) Deposit Size HHI Inflation GDP Constant Sample period R-squared (within) -4.842 *** Notes: Robust standard errors clustered by banks in parenthesis *** p