Discussion of: Banks Incentives and Quality of Internal Risk Models

Transcription

1 Discussion of: Banks Incentives and Quality of Internal Risk Models by Matthew C. Plosser and Joao A. C. Santos Philipp Schnabl 1 1 NYU Stern, NBER and CEPR Chicago University October 2, 2015

2 Motivation How should we regulate the risk-taking of financial institutions? One important tool: Capital Requirements Basel I (1988) - Assets assigned to broad risk categories have flat capital charge - Essentially same charge for corporate debt Basel II (2004) - Differentiate more between different asset risks - Allows banks to opt into internal ratings-based (IRB) approach - Assign a probability of default to each exposure, determines capital charge This paper considers IRB approach in the U.S. - Compare risk assessment across banks for the the same firm

3 Summary Hugely important question as we re-design financial regulation in the the aftermath of the financial crisis Papers make great use of a fantastic dataset Important paper with thought-provoking results Roadmap - Summary + Comments on Empirical Strategy - Takeaways for capital regulation

4 Empirical Strategy Question: Do banks have similar risk assessments? If not, why? Empirical challenge is to compare risk assessments across banks 1. Risk management not shared and standardized across firms 2. Different banks lend to different firms 3. Exposure differs across banks (e.g., collateral) IRB data allows comparison across banks 1. Data is standardized across firms 2. Focus on lending by different banks to the same firm 3. Focus on borrower-specific assessment Compare default estimate for the same firm across banks

5 Analyzing banks risk management Use sample of syndicated loans that are reported by at least two banks from 2009 to 2013 Focus on probability of default (PD) because it does not depend on bank-specific exposure Step 1: Estimate bank-specific deviation from other banks: PD ijt = PD ijt PD Median jt PD ijt is bank i s risk assessment of loan j at time t relative to its peers Step 2: Are there systematic differences across banks? PD ijt = α i + δ t + ɛ ijt

6 loss given default in percent. LGD and its dependent variables are reported before (Bef) and after (Aft) CRM. RW % is the risk-weight per dollar of EAD in percent; it is calculated using P D, LGDAft, and the maturity of the loan. Regressions include year-quarter fixed effects. The F -stat tests the hypothesis that bank fixed effects are equal. Standard errors are clustered by borrower and date but suppressed for brevity. ***, **, * indicate statistical significance at 1%, 5%, and 10%, respectively. Evaluating banks risk management (Table 3) Panel A (1) (2) (3) (4) (5) : P D LGDBef P D LGDBef P D LGDAft RW % Bank FE: *** 2.23*** 0.091*** 0.092*** 9.15*** *** 1.35*** 0.14*** 0.14*** 15.4*** *** -5.57*** -0.11*** -0.11*** -18.6*** ** *** 7.61*** 0.068*** 0.068*** 9.42*** *** *** *** -5.28*** *** ** *** -10.8*** *** -6.91*** 0.031** *** *** -3.15*** *** *** -4.23*** 0.12*** 0.11*** 3.21** *** *** 0.16*** 3.85*** *** -1.22*** 0.062*** 0.061*** 2.15*** *** -6.90*** 0.12*** 0.11*** *** *** *** -9.34*** *** 2.81*** 0.038*** 0.037*** 7.33*** F-Stat Observations 151, , , , ,625 R-squared Panel A (1) (2) (3) (4) (5) Difference % : in PDP Dbetween LGDBef least P D LGDBef and most P D LGDAft aggressive RW A% bank is 64 basis points (about 1 standard deviation). Bank FE: F-Stat Minor caveat: standard errors may be too small (15 clusters) Observations 151, , , , ,625 R-squared

7 What explains variation in banks risk management? Authors suspect regulatory arbitrage Banks may lower risk assessment to save on regulatory capital Are low-capital banks more aggressive? PD ijt = βtier1capital it + γ Controls it + δ t + ɛ ijt Identifying assumption: regulatory capital only affects risk assessment through incentives to underreport PD β > 0 suggests regulatory arbitrage Note that reverse causality goes the other way

8 Testing regulatory arbitrage (Figure 1) Figure 1: Weighted Average PD Deviations Relative to Tier 1 Gap 1 Weighted Deviation Tier 1 Gap Low-capital banks assign 70 basis point lower PD than Figure 1 plots the average deviation from median PD by bank quarter versus the Tier 1 Gap. The average is weighted by the share of utilized funds for that bank-quarter. Tier 1 Gap is the estimated residual from a regression high-capital of sample banks onbanks size, leverage, profitability and time-fixed effects. We obtain similar patterns with Tier 1 capital, however for confidentiality reasons we do not present that illustration. Great contribution to work on regulatory arbitrage (Acharya, Schnabl, and Suarez (2011))

9 What about variation in bank s risk-taking? BUT: omitted variable concern (risk attitude) Banks that perceive lower risk may report less risk and operate with higher leverage Less (perceived) risk higher leverage (lower Tier 1) + more aggressive risk management (lower PD)

10 What about variation in bank s risk-taking? 1. Test 1: Control for average portfolio risk 2. Test 2: Control for risk attitudes with bank fixed effects 3. Test 3: Estimate effect on large and risky loans 4. Test 4: Estimate whether PD predicts loans spreads

11 banks. Panel B, considers Tier 1 Gap, the estimated residual from a regression of sample banks on size, leverage, Portfolio PD is the average median PD for all credits in the bank s portfolio Participation is the percentage of outstanding credits the bank participated in that quarter. Regressions include year-quarter fixed effects. Standard errors reported in parentheses are clustered by bank and date. ***, **, * indicate statistical significance at 1%, 5%, and 10%, respectively. ROE and year-quarter effects. Test with more 1: than Control one reporter, for weighted average by utilization. portfolio risk (Table 7) Panel A (1) (2) (3) (4) (5) : P D P D P D P D P D Tier *** 0.044*** 0.049*** 0.052*** 0.078*** (0.018) (0.017) (0.017) (0.018) (0.025) ROE ** (0.74) (0.81) (0.74) (0.80) (0.89) log(assets) * (0.31) (0.31) (0.28) (0.68) (0.14) Foreign -0.95* (0.55) (0.56) (0.61) Portfolio PD 0.061** (0.026) Participation 1.43** (0.69) Estimates robust to controlling for average syndicate loan risk (Column 2) Foreign Only + U.S. Only + Year FE Bank FE Observations R-Squared BUT: banks may take other risks (not in syndicated loans) Observations R-Squared Panel B (1) (2) (3) (4) (5) : P D P D P D P D P D

12 is probability of default in percent. Panel A considers Tier 1, the most recent reported Tier 1 Capital ratio; log(assets) is the log of total assets; ROE is the most recent reported ROE; F oreign is a dummy for non-us banks. Panel B, considers Tier 1 Gap, the estimated residual from a regression of sample banks on size, leverage, Portfolio PD is the average median PD for all credits in the bank s portfolio Participation is the percentage of outstanding credits the bank participated in that quarter. Regressions include year-quarter fixed effects. Standard errors reported in parentheses are clustered by bank and date. ***, **, * indicate statistical significance at 1%, 5%, and 10%, respectively. Test 2: Control for risk attitude with bank fixed effects ROE and year-quarter effects. (Table with more 7) than one reporter, weighted by utilization. Panel A (1) (2) (3) (4) (5) : P D P D P D P D P D Tier *** 0.044*** 0.049*** 0.052*** 0.078*** (0.018) (0.017) (0.017) (0.018) (0.025) ROE ** (0.74) (0.81) (0.74) (0.80) (0.89) log(assets) * (0.31) (0.31) (0.28) (0.68) (0.14) Foreign -0.95* (0.55) (0.56) (0.61) Portfolio PD 0.061** (0.026) Participation 1.43** (0.69) Estimates robust to adding bank fixed effects (Column 1) Foreign Only + U.S. Only + Year FE Bank FE Observations R-Squared BUT: risk attitudes may change Observations R-Squared

13 held by the bank is in the top tercile for the bank-quarter. P ublic is a dummy equal to one for public fir Regressions include year-quarter fixed effects and the number of reporting banks for that credit-quarter. Stand errors reported in parentheses are clustered by bank and date. ***, **, * indicate statistical significance at 1 5%, and 10%, respectively. Test 3: Estimate effect on large and riskier loans (Table 8) Agents 1 st Q (1) (2) (3) (4) (5) (6) (7) : P D P D P D P D P D P D P D Tier 1 Gap 0.018* 0.023* 0.050*** 0.051*** 0.039*** 0.082*** 0.11** (0.0095) (0.012) (0.015) (0.016) (0.015) (0.017) (0.046) T1G Risky 0.14*** (0.021) Risky 0.12 (0.13) T1G Utilized 0.057*** (0.0094) Utilized (0.038) T1G Large 0.038*** 0.036*** (0.0057) (0.0058) Large 0.059*** 0.060*** (0.014) (0.013) T1G Share 0.046*** 0.044*** (0.0080) (0.0084) Share (0.041) (0.040) T1G Public *** (0.0075) Public ** (0.019) Effect is larger for risky and large loans BUT: there is also more room for disagreement Observations 151, , , , , ,888 3,139 R-squared

14 Test 4: Estimate PD on loan spreads (Figure 3) βlog PD.2 R Squared Tier 1 Gap Tier 1 Gap (a) βlogp D (b) R-Squared Figure 3: Bank Pricing Estimates vs. Tier 1 Gap PD predicts loans spreads better for high-capital firms Figure 3 plots estimates from eight bank pricing regressions versus their average Tier 1 Gap. The pricing models regress log credit spread on credit characteristics and log PD for each bank. (a) illustrates the elasticity of spreads with respect to PD based on estimated coefficients for each bank. (b) illustrates the R-squared of each bank s pricing regression. BUT: low-capital firms may do pricing for more opaque firms

15 What about variation in bank s risk-taking? 1. Test 1: Robust to controlling for average portfolio risk (other bank risk?) 2. Test 2: Robust to controlling for risk attitudes with bank fixed effects (changes in risk attitudes?) 3. Test 3: Larger effect for large and risky loans (more room for disagreement?) 4. Test 4: PD on loan spreads for newly issued loans is more informative for high-capital banks (different firms?) No test is perfect but taken jointly the tests are convincing 5. Suggestion: could also use information whether IRB data is used for computation of risk weights

16 banks. Panel B, considers Tier 1 Gap, the estimated residual from a regression of sample banks on size, leverage, ROE and year-quarter effects. Portfolio PD is the average median PD for all credits in the bank s portfolio Participation is the percentage of outstanding credits the bank participated in that quarter. Regressions include year-quarter fixed effects. Standard errors reported in parentheses are clustered by bank and date. ***, **, * indicate statistical significance at 1%, 5%, and 10%, respectively. with more than one reporter, weighted by utilization. Result with bank fixed effects (Table 7) Panel A (1) (2) (3) (4) (5) : P D P D P D P D P D Tier *** 0.044*** 0.049*** 0.052*** 0.078*** (0.018) (0.017) (0.017) (0.018) (0.025) ROE ** (0.74) (0.81) (0.74) (0.80) (0.89) log(assets) * (0.31) (0.31) (0.28) (0.68) (0.14) Foreign -0.95* (0.55) (0.56) (0.61) Portfolio PD 0.061** (0.026) Participation 1.43** (0.69) Results robust to variation within banks (Column 1) Foreign Only + U.S. Only + Year FE Bank FE Observations R-Squared Surprising given the short sample (mid-2009 to 2013) and that regulatory ratios move slowly Why did regulatory ratios change during this period? Observations R-Squared Panel B (1) (2) (3) (4) (5) : P D P D P D P D P D

17 The role of the regulator Some banks were strongly encouraged (or told) by regulators to raise more capital Banks are also told to improve risk management. They may also anticipate that PDs are benchmarked across banks Stronger regulatory intervention increase in capital (higher Tier 1 capital) and more conservative risk management (higher PD) Results may (partly) reflect impact of regulatory enforcement. Not obvious whether banks are too aggressive or too conservative. Suggestion: analyze whether results are partly caused by regulatory action

18 Implications for financial regulation Main result: Low-capital banks have less conservative risk management than high-capital banks Arguably due to regulatory arbitrage Suggests role for benchmarking in bank supervision BUT: You use it, you may lose it Need to exercise caution if capital requirements are costly and banks vary in risk perception Regulator also needs to be aware of its own role in enforcing regulation