18 th International Conference on Macroeconomic Analysis and International Finance, Rethymno, Greece Bad Management, Skimping, or Both? The Relationship between Cost Efficiency and Loan Qualy in Russian Banks Mikhail Mamonov, Center for Macroeconomic Analysis and Short-term Forecasting (CMASF), National Research Universy Higher School of Economics (NRU HSE) May 29, 2014
Motivation & Contribution 1/3 Fundamental questions: 1. Does higher cost efficiency always imply lower cred risk exposure of banks (the first case)? Or whether higher cost efficiency could mean insufficient spending on borrowers screening and lead thus to higher bank cred risks (the second case)? 2. Does the one case necessarily exclude the other? Or could they coexist? 3. What are possible motivations of banks managers in both cases? Why is important: 1. If the second case holds for some subsample of banks casts doubt on higher cost efficiency could be the reflection of best managerial practices. 2. Financial accounts falsifications undertaken by banks to satisfy official requirements. 3. Implementing of a proper kind of differential prudential regulation 4. Maintaining financial stabily 2
Motivation & Contribution 2/3 The main contribution: Skimpers vs. Bad managers: Comparative empirical analysis n n n New creria for skimping behavior identification compared to Berger & DeYoung (1997) s explanation replicated by Altunbus et al. (2007) and Fiordelisi et al. (2011). A two-step approach on estimating the effect of efficiency on risk separately for subsamples of banks defined through managers motivations to demonstrate higher efficiency levels: (a) preliminary effect: a panel Granger causaly test as a first step (b) final effect: controlling for a large set of micro- & macro- determinants of risk A new evidence on skimping nature: skimping doesn t always imply lower qualy of commercial loans as compared to non-skimping due to speed-of-lending effect (even during financial turmoils), skimping can provide more resilience to macroeconomic shocks but not all banks are able to use that and subject themselves to much higher cred risk as compared to banking system average 3
Motivation & Contribution 3/3 Why we choose Russian banking system for the analysis Before the bad debt crisis of 2008-2009 Russian banks concentrated their efforts on quick-and-easy prof extraction rather than on longer financial stabily issues (were too aggressive to pay due attention to borrowers screening procedures) Now, Russian banks suffer from the long lasting process of bad debts decreasing. It damages their balance sheets and forces them to keep higher provisions for losses Unfortunately, the pre-crisis story repeated in 2012-2013 Expected shocks from the macroeconomic side in 2014-2016 Under the condions described, we ask two practical questions: (a)whether managing cost efficiency by Russian banks could be a viable mechanism to control their loan qualy from the microeconomic side? (b)and whether such a mechanism is the same for different banks? 4
Related Lerature 1. Berger & DeYoung (1997) introduced the efficiency-risk hypotheses: «Bad management»: Efficiency (EFF) a signal of shortcomings in general managerial practices including inadequate borrowers screening NPL in the long run «Skimping»: in order to be more efficient in the short run a bank reduces expenses devoted to screening EFF and, at the same time, adverse selection of borrowers NPL in the long run «Bad luck»: deteriorations of macro- condions decreased credworthiness of borrowers NPL increasing expenses to the screening of new borrowers EFF 2. The Bad management effect was revealed in banking systems of: USA in Berger & DeYoung (1997), Eisenbeis (1997), EU in Williams (2004), Fiordelisi et al. (2011), Spain in Salas & Saurina (2002), Greece in Louzis et al. (2011), Russia in Mamonov (2012), Pestova & Mamonov (2013) 3. Quagliariello (2007) argues that the existence of the Bad management effect in Italy is spurius. 4. The Skimping effect was found in Berger & DeYoung (1997) for the subsample of US banks wh higher-than-median efficiency levels Altunbas et al. (2007) for EU banks as a whole 5
Methodology 1/6 Different approaches to efficiency estimations 1. Balance-sheet approach: operating cost-to-operating income ratios (CIR) 2. Econometric approach: Stochastic Frontier Analysis (SFA-scores) Distribution Free Approach (DFA-scores) Both are calculated on the basis of translog cost function wh loans, deposs, and offbalance sheet activies as banking outputs (Berger & DeYoung, 1997; Maudos & Fernandez de Guevara, 2007; Solis & Maudos, 2008): ln OC j1 lny j j, TREND TREND 1 0 3 2 2 1 2 3 k1 j1 3 j1 3 lny j kl lny j, k, lny l, TREND where for bank i at quarter t: Y outputs, P factor input prices (AFR, personnel, and physical capal), u inefficiency term to be estimated and transformed to ˆ u uˆ min, t SFA e 0, 1 DFA 1 uˆ ; uˆ 3 m1 3 m1 ln P m m m, ln P p, 1 2 3 3 r1 q1 rq TREND v ln P u r, ln P q, 3 min, t 3 s1 u1 lny su 0, s, 1 ln P u, 6
Methodology 2/6 How the Skimping was identified previously It s hardly expected that all banks whin a banking system are skimpers, i.e. skimp on risk-management (Berger & DeYoung, 1997) General framework of Berger & DeYoung (1997): skimpers are those «banks that willingly trade loan qualy for cost reductions, but manage the resulting loan qualy problems in a cost effective fashion» To test that they apply the higher-than-median efficiency filter for a sample of US banks From our standpoint, higher efficiency is necessary, but not sufficient condion. Banks wh best managerial practices vs. banks wh artificially increased efficiency. We need to distinguish between the two subgroups of efficient banks. 7
Methodology 3/6 New creria for Skimping identification We propose two explanations for skimping 1. Holding (expanding) the market share when competive posions are weakening. This objective may be achieved by extensively expanding loan supply. The easiest way to perform that is weakening lending standards. This results in increased cost efficiency in the short run, but is very likely to negatively affect the qualy of loans in a longer horizon. 2. Insufficiency of capal needed to achieve strategic objectives. To realize strategic objectives (expanding branch networks, entry into new markets, etc.) requires addional funds. Raising these funds may be achieved by increasing business profabily (ROA) or by attracting (expensive) retail deposs. The first one is easier for bank managers as they can just artificially cut the screening expenses. Shareholders will be satisfied in the short run, but loan qualy will obviously decline in the long run. 8
Methodology 4/6 New creria for Skimping identification We suggest two respective filters for the whole sample: 1. The annual growth rate of real loans is greater than 50th percentile level at least during the 4 previous quarters. 2. The capal-to-assets ratio is lesser than the 50 th or 25 th (for robustness) percentile level at least during the 4 previous quarters. We apply these filters separately combining each of them wh the Berger & DeYoung s higher efficiency condion. In our modification, we require efficiency being greater that the sample median during the 4, 8 or 12 quarters 9
Methodology 5/6 A two-step approach on estimating the effect of efficiency on risk: the first step Panel Granger causaly test (Berger & DeYoung, 1997; Fiordelisi et al., 2011): ODL t 4 k1 (1) k ODL tk where for bank i at quarter t: ODL is Overdue loans ratio (substute for NPL) reflecting the bank s loan qualy, EFF CIR-, SFA- or DFA-scores reflecting the bank s operating cost efficiency. Both equations are estimated for the whole sample and for the subsamples based on the higherthan-median efficiency filter by self and combined wh (b) the extensive growth condion or (c) the insufficient capal filter. 4 k k1 if ˆ (1) than provides arguments in favor of the Bad management effect; 4 k k1 0 if ˆ (1) than the Skimping effect; 0 4 if (2) than the Bad luck effect. k k1 0 4 k1 (1) k EFF tk Estimation technique: Two-step Difference GMM of Arellano & Bond (1991) (1) t EFF t 4 k1 (2) k ODL tk 4 k1 (2) k EFF tk (2) t 10
Methodology 6/6 A two-step approach on estimating the effect of efficiency on risk: the second step Empirical equation for loan qualy: ODL t ODL t1 EFF t1 N 1 j1 BSF where for bank i at quarter t: ODL is Overdue loans ratio (substute for NPL) reflecting the bank s loan qualy (exposure to cred risk), EFF CIR-, SFA- or DFA-scores reflecting the bank s operating cost efficiency, BSF and MACRO j-th bank-specific and l-th macro determinants of cred risk, respectively j j, q N 2 l1 MACRO l l, tk t The main purpose is to identify to what extent the estimates of β for skimpers and for the whole sample are robust to variation in BSF and MACRO variables 11
Data 1. The Bank of Russia web-se (www.cbr.ru): Bank-specific characteristics (BSF) monthly balance sheets of banks (Form 101); quarterly prof and loss accounts (Form 102). 2. The Federal State Statistics Service web-se (www.gks.ru): Macroeconomic controls (MACRO) Time period: Q1 2005 Q3 2012 (35 quarters) Number of banks (depending on quarter): in original sample: 705-1024; in adjusted sample: 650-997 In order to eliminate the negative impact of outliers, we excluded from the original sample all those observations for which: the real interest rate exceeded 200% annually (0.1% of the inial volume of data); the loans-to-deposs ratio was above 1000% (2% of the inial volume of data); the ratio of liquid assets to deposs exceeded 305% (1% of the inial volume of data). 12
Estimation results 1/6 A two-step approach on estimating the effect of efficiency on risk: the first step Panel Granger causaly test : the full (adjusted) sample of banks The Bad management effect is supported, while the Skimping effect was not identified The Bad luck effect was found to be significant ***, ** and * a coefficient is significant at the 1%, 5%, and 10% level, respectively. Robust standard errors are reported in parentheses under the coefficients. 13
Estimation results 2/6 A two-step approach on estimating the effect of efficiency on risk: the first step Panel Granger causaly test : the full (adjusted) sample filtered by higher-than-median efficiency condion alone (М4, M5, and М6 4, 8, and 12 quarters, respectively) combined wh the insufficient capal condion (М7, М8 capal-to-assets < 50 th, 25 th perc., resp.) combined wh the extensive growth condion (М9 annual growth of real loans > 50 th perc.) combined wh the oppose to extensive growth condion (М10 annual growth of real loans<50 th perc.) 1) Skimping does exist whin Russian banking system 2) It was found for banks wh higher cost efficiency levels AND Aggressive lending behavior ***, ** and * a coefficient is significant at the 1%, 5%, and 10% level, respectively. Robust standard errors are reported in parentheses under the coefficients. 14
Estimation results 3/6 A two-step approach on estimating the effect of efficiency on risk: the second step Bad management is confirmed, to some extent Skimping is robustly confirmed for the subsample of highly efficient banks wh aggressive lending behavior 15
Estimation results 4/6 Addional findings: differences in senetivy to macroeconomic shocks (Robust OLS estimation results) Skimpers are less sensive to macroeconomic shocks as compared to bad managers 16
Estimation results 5/6 Addional findings: differences in profabily Skimpers are more profable as compared to bad managers regardless the state of business cycle 17
Estimation results 6/6 The growing scale of Skimping and their higher cred risk exposures Findings: 1) About ¼ of all skimpers (approx. 100 banks) are exposed to higher cred risks compared to the median non-skimper. 3) Skimpers held 1.6% of market for loans in Q1 2010 and up to 16.4% in Q3 2012. 18
Conclusion & Policy implications 1/2 1. We contribute to the lerature on efficiency-risk analysis by (a) describing bank managers motivation for skimping and developing new creria for skimping identification, and (b) first applying this methodology to Russian banks on the quarterly basis 2. The bad management behavior holds on average for the Russian banks. The skimping behavior s relevant for those Russian banks that are: (a) not just highly cost efficient, as predicted by Berger and DeYoung for US banks, but (b) that at the same time pursue aggressive strategies in the market for loans to households and non-financial firms 3. The Skimping is not the case for those Russian banks that demonstrate lower capal-to-assets ratio and that are highly cost efficient at the same time. 19
Conclusion & Policy implications 2/2 4. Median skimper sustainably demonstrates lower overdue loans ratio compared to the median non-skimper. 5. About ¼ of all skimpers (approx. 100 banks) are exposed to higher cred risks compared to the median non-skimper. 6. Skimpers held 1.6% of market for loans in Q1 2010 and 16.4% in Q3 2012. What is their role in providing inter-bank loans and what could be the negative chain effect to systemic liquidy risk if they go bankrupt are both opened questions. 7. Recommendations 1) introduction of increased requirements to capal adequacy ratios of skimpers by the Bank of Russia and 2) increasing quarterly payments to the Russian Depos Insurance Agency - both depending on subgroups of skimpers ODL exceeding the average of banking system. 20
Thank you! 21
Appendix: Full Estimation results 1/2 A two-step approach on estimating the effect of efficiency on risk: the second step 22
Appendix: Full Estimation results 2/2 A two-step approach on estimating the effect of efficiency on risk: the second step 23