The risk sensitivity of Basel risk weights and loan loss provisions: evidence from European banks

Abstract

Recent literature suggests that regulatory risk measures do not adequately capture the actual economic risk of bank portfolios. We shed new light on this issue by analyzing both regulatory and accounting standards, i.e. capital requirements and loan loss provisions. Examining a sample of large European banks for the years 2002–2015, we show that regulatory risk sensitivity, i.e. the response of Basel risk weights to asset volatility as our measure of a bank's asset portfolio risk, is substantially higher than what has been shown so far in the literature. Despite the occasionally bad reputation that risk weights have, we provide new evidence that they are adequately calibrated for banks with low or medium levels of risk. For crisis periods and for high-risk banks, however, risk weights still do not adequately reflect the actual portfolio risk. This results in insufficient capital, even with the stricter Basel III minimum capital requirements. Regarding loan loss provisions, we establish a theoretical link between expected losses and asset volatility. We document a strong empirical association, which fits well to the theoretical model. Overall, we find no indication that the risk sensitivity of loan loss provisions has been insufficient, at least since the financial crisis.

Keywords:

• Risk weights
• loan loss provisioning
• banking regulation
• capital requirements
• Basel II
• Basel III

JEL Classification Codes:

• G21
• G28

Acknowledgments

The authors thank Ben Ranisch, Peter Reus, participants of the Southern Finance Association 2017 Annual Meeting, two anonymous referees, the associate editor, and the editor (Chris Adcock) for their valuable comments and suggestions.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 In the original model, this fixed loss is a total loss. We will use an extended version to cover non-trivial values for loss given default.

2 Vasicek (1987) derives this formula for LGD = 1. The extension is straightforward.

3 Following Ronn and Verma (1986), we assume that the face value of the debt, $D_{i,t}$, is the present value of the default point (discounted by the risk-free rate).

4 For example, according to Basel Committee on Banking Supervision (2013a), up to three-quarters of the variability in risk weights in the banking book is driven by differences in underlying risk arising from banks' asset composition. The remaining variation relates to supervisory choices at the national level and banks' modeling choices under the IRB approach.

5 Consider again the example above, where RWD varies across banks with different asset risks but not across time, due to a non-cyclical risk weighting scheme (see Figure 3b). To concretize, let ${\sigma }_{A,0}=0.01$ and ${\mathit{\text{RWD}}}_{A,0}=0.4$ (Bank A) and ${\sigma }_{B,0}=0.02$ and ${\mathit{\text{RWD}}}_{B,0}=0.8$ (Bank B). Based on these two observations, the standard is perfectly risk-sensitive with ${\beta }_0=40$. Regarding changes to the next period, let ${\sigma }_{A,1}=0.02$ and ${\mathit{\text{RWD}}}_{A,1}=0.4$ (Bank A) and ${\sigma }_{B,1}=0.04$ and ${\mathit{\text{RWD}}}_{B,0}=0.8$ (Bank B). The standard is thus fully non-cyclical (no changes in RWD from t = 0 to t = 1), yet still perfectly risk-sensitive, but with a decreased risk sensitivity of ${\beta }_1=20$. Our approach according to Equation (8(8) ${\mathit{\text{RWD}}}_{i,t}=c+{\beta }_t{\sigma }_{i,t}+{\delta }^{\prime }{\theta }_{i,t}+{ϵ}_{i,t},$(8) ) would exactly yield those beta estimates. A dynamic panel data model, ${\mathit{\text{RWD}}}_{i,t}=c+\gamma {\mathit{\text{RWD}}}_{i,t-1}+\beta {\sigma }_{i,t}+{ϵ}_{i,t},$however, would map the behavior of the capital standard to a factor loading of $\gamma =1$ for the lagged RWD, while the measure of risk sensitivity, β, would take on a value of zero.

6 To facilitate comparability between our various specifications of (9(9) ${\mathit{\text{RWD}}}_{i,t}=c+{\beta }_t{\sigma }_{i,t}+(c^L+{\beta }_t^L{\sigma }_{i,t})1_{\{{\sigma }_{i,t}<{\sigma }^{\ast }\}}+{\delta }^{\prime }{\theta }_{i,t}+{ϵ}_{i,t},$(9) ), we choose to employ a uniform value of ${\sigma }^{\ast }$. Since the threshold is rather insensitive to different specifications of (9(9) ${\mathit{\text{RWD}}}_{i,t}=c+{\beta }_t{\sigma }_{i,t}+(c^L+{\beta }_t^L{\sigma }_{i,t})1_{\{{\sigma }_{i,t}<{\sigma }^{\ast }\}}+{\delta }^{\prime }{\theta }_{i,t}+{ϵ}_{i,t},$(9) ), our main results and conclusions do not change when individually optimized values of ${\sigma }^{\ast }$ are used.

7 On average, credit risk accounts for 85% of total risk-weighted assets, while market risk only accounts for 5%. The remaining 10% are induced by operational risk.

8 As a further variable, for IRB banks, we also investigated the difference between LLP on the balance sheet and expected losses according to the IRB model. However, we found no significant relationship to asset volatility, and we therefore refrain from reporting the results.

9 We also estimated a specification similar to Model (III) with Low/High Risk. As expected, results do not show a significantly higher risk sensitivity for the low risk area.

10 Worldscope Field 02275 ‘Reserve for Loan Losses’ represents the allowance end of year (calculated as allowance, beginning of year, plus provision for loan losses net charge-offs for the respective year).

11 Blundell-Wignall and Roulet (2013), Mariathasan and Merrouche (2014) and Baule and Tallau (2016) present similar RWD developments for different samples.

12 Another possible explanation is based on the theoretical association of expected losses and asset volatility, which increases with lower correlation of the loans (see Figure 1). Hence, the estimated higher risk sensitivity may be caused by more diversified loan portfolios after the crisis.

13 Similar to the results of the multivariate analyses, the difference in LLP is not significant for the total sample period. We do, however, obtain significantly higher LLP for high-risk banks for the sub-period after 2010.

14 For the sake of brevity, we do not report estimates for instrumental variables for all models. Results confirm the OLS estimates and can be obtained from the authors on request.

15 Regulatory risk weights are bounded below. For example, the Standardized Approach for credit risk requires a minimum risk weight of 20% for claims on banks or corporates. Furthermore, for the IRB Approach, PD estimates are subject to a floor of 0.03%, leading to a minimum risk weight of about 15% for claims on corporates according to the IRB formula (Basel Committee on Banking Supervision 2004).

16 Of course, banks may hold capital buffers above the 8.5% minimum Tier 1 regulatory capital leading to lower actual PDs (the mean Tier 1 ratio for our sample is 14.6% for 2014).