CRC Credit Research Centre The credibility of European banks risk-weighted capital: structural differences or national segmentations? Brunella Bruno, Bocconi Giacomo Nocera, Audencia Nantes École de Management Andrea Resti, Bocconi (S) @andrearesti andrea.resti@unibocconi.it
Outline of this talk Background Related literature Sample and key variables Results Why do risk weights differ across banks? Why do some banks use internal models more? Do investors believe in Basel s risk weights? Preliminary evidence from multilevel models Final remarks
Outline of this talk Background Related literature Sample and key variables Results Why do risk weights differ across banks? Why do some banks use internal models more? Do investors believe in Basel s risk weights? Preliminary evidence from multilevel models Final remarks
Background: Wide distrust of banks RWs Market participants have questioned the reliability and comparability of RWs, and focused on adjusted capital ratios Regulators have endorsed the use of backstops against the opportunistic use of risk-weighted measures Plain, un-weighted capital ratios Output floors (CP306) and Input floors (Sweden, CP362) on RWAs generated with internal models
Outline of this talk Background Related literature Sample and key variables Results Why do risk weights differ across banks? Why do some banks use internal models more? Do investors believe in Basel s risk weights? Preliminary evidence from multilevel models Final remarks
Related literature Practitioners and supervisors e.g. BIS 2013-16; EBA 2013-14: RWAs show considerable differences both across banks and across countries Academic literature Vallascas and Hagendorff (2013), show that riskweights are ill-calibrated to a market measure of bank portfolio risk (i.e., bank asset volatility) Mariathasan and Merrouche (2013) find that average risk-weights decrease once internal models gain regulatory approval, and this is especially true for capital-constrained banks
Outline of this talk Background Related literature Sample and key variables Results Why do risk weights differ across banks? Why do some banks use internal models more? Do investors believe in Basel s risk weights? Preliminary evidence from multilevel models Final remarks
Sample and data sources 50 largest banks by total assets (and with pillar 3 information available) 48 listed 49 adopt IFRS 17 countries: EU, Norway and Switzerland Unbalanced panel over 2008-2014 Source of data Pillar 3 reports (Banks websites) Compulsory, but not standardized Financial statements (Bankscope) Market-based risk measures (Bloomberg) Macroeconomic variables (World Bank and OECD reports) Belgium 2 Denmark 2 Finland 1 France 5 Germany 3 Greece 4 Hungary 1 Ireland 2 Italy 8 Netherlands 2 Norway 1 Poland 1 Portugal 2 Spain 5 Sweden 4 Switzerland 2 United Kingdom 5 All 50
Dependent variables, risk weights, have been decreasing over 2008-2014 RWA_TA RWA_EAD Average risk weight (assetweighted) Average weight for credit risk (includes OBS) Lloyds TSB+ HBOS
Main explanatory variables Name Description SIZE Natural log of total assets DEPOSITS Customer deposits / total assets LOANS Loans / total assets CORPORATE Corporate loans / total loans RETAIL Retail loans / total loans ROA Return on assets (winsorised) GDP_GROWTH Annual real growth in national GDP BANKCONC Share of top 3 banks on total assets BANKONGDP Bank total assets on GDP STANDARD Loans under standardised approach / total loans IRB_LOANS Loans under internal ratings-based approach / total loans HIGH_RETCORP_IRB Dummy for heavy users of internal ratings-based models CDSSPREAD CDS spread WACC Weighted average cost of capital ASSETVOL Asset volatility ZSCORE Z-score (measure of credit risk) IMPAIRED_LOANS Impaired loans / total loans TIER1 Tier 1 capital / risk-weighted assets RATING_A Average rating (higher values mean higher quality)
Outline of this talk Background Related literature Sample and key variables Results Why do risk weights differ across banks? Why do some banks use internal models more? Do investors believe in Basel s risk weights? Preliminary evidence from multilevel models Final remarks
Univariate statistics / 1 Business models Correlation with Mean Median Max Min Sigma Obs RWA_TA RWA_EAD SIZE 12.5 12.5 14.7 10.3 1.3 349-62.9%*** -50.4%*** DEPOSITS 50.1 50.7 94.7 4.0 15.6 349 42.7%*** 34.8%*** LOANS 53.2 57.9 81.7 12.2 17.2 349 71.2%*** 55.0%*** RETAIL 30.5 31.8 60.6 0.0 12.3 348 22.0%*** 15.6%*** CORPORATE 34.4 34.3 54.9 3.2 9.8 348 0.4% 2.7% Institutions / total loans (%) 11.6 8.8 48.2 0.9 9.3 348-27.3%*** -24.9%*** GOVERNMENT 13.0 13.0 33.6 0.0 6.1 348-17.3%*** -16.0%*** ROA 0.1 0.3 4.4-12.4 1.3 349-6.9% -3.0% Banks showing higher risk weights are smaller, more involved in the traditional loans/deposits business, more exposed to retail portfolios (as opposed to governments and financial institutions)
Univariate statistics / 2 Risk models Correlation with Mean Median Max Min Sigma Obs RWA_TA RWA_EAD STANDARD_LOANS 44.0 35.5 100.0 1.9 30.1 348 74.0%*** 71.0%*** IRB_LOANS 56.1 64.5 98.1 0.0 30.1 348-74.1%*** -71.0%*** FIRB_LOANS 9.3 0.0 92.0 0.0 17.0 348-9.9%* -11.9%** AIRB_LOANS 45.8 51.2 98.1 0.0 30.5 348-66.1%*** -61.6%*** Banks showing lower risk weights are heavy users of IRB models, especially advanced ones. This is consistent with the incentives created by the Basel II Accord.
RWA / TA Univariate statistics / 3 Capital and risk Correlation with Mean Median Max Min Sigma Obs RWA_TA RWA_EAD TIER1_RWA 11.7 11.6 22.4-6.7 3.6 348-55.7%*** -57.7%*** Stock return volatility (%) 3.4 3.1 10.3 1.0 1.5 240 1.6% 5.9% CDS spreads (bps) 256.5 134.8 1999.4 28.7 321.3 307 38.7%*** 30.6%*** WACC (%) 3.9 3.6 13.4 0.4 2.1 240 25.1%*** 23.5%*** IMPAIRED_LOANS 7.2 4.9 44.9 0.4 7.5 349 34.2%*** 30.9%*** ASSET_VOL 14.9 9.7 168.9 0.2 19.8 333 19.8%*** 19.0%*** Z Score 2.5 2.4 12.4-1.5 2.1 286-23.7%*** -30.1%*** RATING_A 4.7 5.0 6.0 1.0 1.2 348-37.02*** -29.81*** Risk-weight based indicators are not inconsistent with most market-based risk measures (CDS, WACC, ASSETVOL, ZSCORE), ratings and the banks actual bad loans experience (impairment ratio). Tier 1 / TA = k A negative link with risk-weighted capital suggests that investors/supervisors look at un-weighted leverage, too. Tier 1 / RWA
Why do RWA differ across banks? odds ratio x_or = ln[x/(1-x)] RWA_TA_OR RWA_EAD_OR Coef Std. Coef. Coef Std. Coef. Constant 0.173 Higher -0.104 RWs for small SIZE -0.106*** -0.169 banks, -0.025 strongly geared -0.058 DEPOSITS 0.009*** 0.176 towards 0.004 loans and 0.114 LOANS 0.017*** 0.378 traditional 0.008*** portfolios 0.271 CORPORATE 0.013*** 0.159 0.008*** 0.150 IRB_LOANS -0.006** -0.238 Low -0.008*** RWs for heavy -0.432 IRB TIER1_RWA -0.050*** -0.231 users -0.043*** and wellcapitalised banks -0.294 Dummy France 0.023 0.118 Dummy Germany National -0.228-0.156 Dummy Greece segmentations, 0.112-0.100 Dummy Italy especially 0.037-0.115 Dummy Spain outside SSM 0.031-0.045 Dummy Sweden -0.422*** -0.256** Dummy UK 0.236* 0.275** Adj. R-square 0.818 0.686 Joint F 120.995 59.415
Outline of this talk Background Related literature Sample and key variables Results Why do risk weights differ across banks? Why do some banks use internal models more? Do investors believe in Basel s risk weights? Preliminary evidence from multilevel models Final remarks
What drives IRB adoption? Large, well capitalised banks use IRB more IRB_LOANS_OR IRB_LOANS_OR IRB_LOANS_OR Coef Std. Coef. Coef Std. Coef. Coef Std. Coef. Constant -7.397*** -6.645*** -11.02*** SIZE 0.536*** 0.394 0.601*** 0.442 0.582*** 0.428 DEPOSITS -0.025** -0.229-0.025** -0.227-0.012-0.113 LOANS -0.022* -0.226-0.017-0.170-0.013-0.130 CORPORATE 0.019* 0.110 0.019* 0.107 0.012 0.070 RETAIL 0.034*** 0.246 0.022 0.157 0.020 0.145 Tier 1_RWA 0.141*** 0.301 0.067** 0.144 0.077*** 0.165 Dummy France -0.965* Dummy Germany -1.111** but national Dummy Greece supervisory -1.142* Dummy Italy -1.641*** practices do Dummy Spain -0.438 matter Dummy Sweden -0.043 Dummy UK -0.467 which in turn are driven by the banking industry s lobbying power BANKCONC 0.037*** 0.282 BANKGDP 0.183*** 0.207 Adj. R-square 0.583 0.674 0.678 Joint F 14.550 16.593 16.266
Outline of this talk Background Related literature Sample and key variables Results Why do risk weights differ across banks? Why do some banks use internal models more? Do investors believe in Basel s risk weights? Preliminary evidence from multilevel models Final remarks
Multivariate results: Do investors believe in Basel s RWs? WACC WACC CDSSpread CDSSpread Asset Vol Asset Vol RWs affect market-based measures of risk Constant 3.174 7.471*** 164.167 172.063 0.110 0.087 RWATA 0.099** 0.051*** 4.646** 4.903*** 0.006** 0.006*** RWAEAD_O -0.042-0.570 - -0.002 - SIZE 0.663-24.806* 24.889** -0.008 - DEPOSITS 0.015-1.622** -1.873** 0.001 - Traditional lending businesses are perceived as less risky LOANS -0.043-0.910 - -0.004* -0.004* CORPORATE -0.067*** -0.047*** -1.064-0.005* -0.005** RETAIL -0.033* -0.033*** -0.628 - -0.000 - Asset profitability matters, write-downs are less clear EQUITY_RATIO 0.374** 0.349*** -0.659-0.005 - IMPAIRED_LOANS -0.043-12.910*** -12.914*** 0.020*** 0.022*** ROA -1.268*** -1.031*** -157.70*** -154.97*** -0.023 - GDP_GROWTH -0.435*** -0.419*** -51.339*** -51.307*** 0.004 RATING_A -1.155** -0.672*** -71.778*** -75.872*** 0.057** 0.049** Time and country dummies are significant Adj. R2 0.790 0.783 0.846 0.841 0.436 0.427 F on countries 14.990*** 13.400*** 23.640*** 22.050*** 2.340* 3.690** F on years 15.060*** 19.560*** 22.230*** 23.900*** 9.830*** 10.310***
Outline of this talk Background Related literature Sample and key variables Results Why do risk weights differ across banks? Why do some banks use internal models more? Do investors believe in Basel s risk weights? Preliminary evidence from multilevel models Final remarks
Experimenting with multilevel models 2-level model: 3-level model: Level 3: countries Level 2: countries Level 2: banks RWA x x it i i1 i1 in in Level 1: observations RWA x x it i i1 i1 in in Level 1: observations
2-level models: what evidence of national segmentations? 2-level model: LR* Pvalue* Level 2: countries SIZE 33.22 0.0000 DEPOSITS 46.34 0.0000 LOANS 54.24 0.0000 CORPORATE 35.47 0.0000 IRB_LOANS 33.44 0.0000 *Base case: RE model RWA x x it i i1 i1 in in Level 1: observations Slopes for the main variables of the RWATA model differ significantly across countries
How did national segmentations evolve over time? LR P-value 2 levels (a) Level 2: countries* 2.36 0.3076 (b) Level 2: banks 79.32 0.0000 (c) 3 levels Level 2: banks Level 3: countries 41.36 0.0000 (c) vs. (b) 11.91 0.0077 *after accounting for changes in GDP
Outline of this talk Background Related literature Sample and key variables Results Why do risk weights differ across banks? Why do some banks use internal models more? Do investors believe in Basel s risk weights? Final remarks
Final remarks Risk weights are affected by the banks size, business model and asset mix The adoption of internal ratings based (IRB) approaches is (as expected) a powerful driver of bank risk-weighted assets Lower risk weights are positively linked to the banks capital cushion IRB adoption is more widespread in countries where supervisory capture is potentially stronger Regulatory risk weights are not disconnected from market-based measures of bank risk
CRC Credit Research Centre The credibility of European banks risk-weighted capital: structural differences or national segmentations? Brunella Bruno, Bocconi Giacomo Nocera, Audencia Nantes École de Management Andrea Resti, Bocconi (S) @andrearesti andrea.resti@unibocconi.it