A New Tool to Measure a Bank’s Risk of Failure

A recent paper argues that despite the revisions to the Basel capital framework (Basel III/IV), the fundamental flaw in the underlying methodology used for calculating the amount of capital a bank should hold remains, and that the true risks the bank may be subject to continue to go undetected. Entitled “ELPR: A New Approach to Measuring the Riskiness of Commercial Banks,” the paper has been written by Charles Lee from Stanford University Graduate School of Business, Yanruo Wang from Nipun Capital and Qinlin Zhong from the Renmin University of China School of Business. They put forward an alternate approach of measuring bank risk that appears to be a more accurate predictor of bank failures than the Basel capital framework.

Why is this important? Well, it has now been over 10 years since the Great Financial Crisis, the direct cost of which ran into the hundreds of billions. In the US alone it is estimated that it cost US$498 billion on a fair value basis, which was 3.5% of the US’s GDP in 2009. In the UK, a package amounting to £500 billion (which at that time was equivalent to c.US$850 billion) had to be put into place to rescue the banks. The indirect costs resulting from the ensuing lost output and lower investment run significantly higher, with the San Francisco Federal Reserve estimating this cost to be US$22.9 trillion in just the US alone!

It is not surprising therefore that a lot of regulatory focus has gone into ensuring that such widespread contagion is avoided and the taxpayer is not left on the hook again the next time a crisis occurs. One of the key issues the financial crisis highlighted was the ability for banks to arbitrage the internal risk models they developed under the Basel II framework to play down the riskiness of assets they held, and in turn reduce the amount of capital they were required to hold to guard against those assets souring. The reforms following the crisis included the new Basel III regulations. These set out to overcome the weaknesses of Basel II by enhancing the robustness and risk sensitivity of standardised approaches, constrain the use of the internal rating based (IRB)-based methods in some cases, and require floors for probability of default (PD) and loss given default (LGD) calculations where they continue to be used. In addition to this, a new expected loss-based accounting provisioning standard, IFRS 9, has been introduced. This requires banks to take more timely provisions against weakening loans, at the initial sign of any significant increase in credit risk rather than waiting until a risk event crystalises, and to make forward looking estimates on the performance of their loan portfolio.

The idea of these reforms was to strengthen reserves of capital banks hold against unforeseen events, as well as to increase the amount of provisions held against poorly performing assets, in order to reduce pro-cyclicality in the financial system. There is no doubt that these aims are being achieved to a degree.

Lee, Wang and Zhong, however, point to the fact that academic evidence to date (and I would add, real-life evidence following the financial crisis) suggests that the Risk Weighted Assets (RWA) metrics derived under the Basel guidelines do not in fact capture the true riskiness of banks. They propose another method of measuring bank risk, starting from the fact that the majority of a risk a bank is exposed to comes from its loan portfolio. The authors posit that there are a couple of fundamental flaws in the current RWA framework. Firstly, RWA calculations fail to properly account for the volatility of default losses over time (or second moment) in a bank’s loan portfolio. Whilst the current regulations focus on the average default rates in each loan category, the authors highlight that portfolio theory states that the variation in these default rates also gives rise to a need for higher levels of capital. So, it is the variance of default rates for each asset class from quarter to quarter that is important rather than merely the average default rate. The second weakness concerns the fact that the RWA framework does not directly account for concentrations of holdings or default correlations across asset classes, and particularly property holdings, the default of which were the cause of recent, and historical, bank failures. So, unsurprisingly, even if a bank has a well-diversified portfolio of loans on its book, if they are correlated then they will all tend to default when there is a recession or banking crisis.

What they are pointing out will not be a total surprise to regulators, particularly the latter point. It is an accepted fact that the Basel Pillar 1 framework does not capture concentration or correlation risks. This is precisely why under Pillar 2, for risks not captured under Pillar 1, supervisors are asked to estimate the concentration and correlation risks in a bank’s portfolio under the Supervisory Review and Evaluation Process (SREP). Depending on how severe these are, a related capital add-on is required. Each central bank/regulatory authority will therefore have their own methodology for carrying out such an assessment, which will dictate the level of any capital add-on required. The problem is that if this system was working as it should be, then they would have been able to pick out the heavy concentrations and correlations banks had built up to the real estate market prior to the financial crisis.  Moreover, if the Pillar 2 concentration risk adjustment was adding on sufficient capital to those banks with the most concentrated/correlated portfolios, then the methodology used by the authors would not have been able to show itself to be a better predictor of bank failures than the Tier 1 Capital Ratio (T1CR) measure that is currently in practice. Part of the problem is that the concentration risk assessment is a manual process, dependent on having regulatory staff with the right expertise who can look across all portfolios across all banks. This is usually a difficult ask, and invariably it means differences in approaches between the largest and smallest banks, and across jurisdictions with different regulators. Moreover, once concentration risks are identified, it is a judgement call as to how much more capital is required, which is again problematic.

This is where the approach proposed by the authors could be really useful. They put forward a Loan Portfolio Risk (LPR) variable that measures the time-varying volatility in default risk for a portfolio of bank loans. They split each bank’s loan portfolio into 14 asset classes. The riskiness of the bank portfolio as a whole is then a function of its exposure to each asset category, as well as the variance and cross-correlation structure in the delinquency rates across the 14 asset classes. LPR can therefore be thought of as the expected dollar losses for the loan portfolio as a whole, from a one standard deviation move in historical default rates, taking into account all default cross-correlations. Banks whose loan portfolios are concentrated in asset categories with highly volatile delinquency rates will have higher LPR scores, and banks whose holdings are in asset classes with low delinquency rate volatility will have lower LPR scores. They then propose a new capital adequacy measure that compares a bank’s adjusted book equity to its loan portfolio risk (LPR), which they term the Equity-to Loan-Portfolio-Risk ratio, or ELPR. They find this measure to be incrementally important in predicting bank failure up to five years in advance, even after controlling for all the CAMELS variables that regulators may have examined.

So, does this work in practice? They gathered data from more than 500 bank failures that took place during 2003 to 2017, and saw that the T1CR was able to anticipate only about 17 per cent of the variation in bank failures that occurred from one year to the next. Their measure of loan portfolio risk predicted almost 25 per cent of the variation. Their measure outperformed the T1CR even more in predicting the likelihood of bank failures two, three, or five years in the future.

Lee, Wang and Zhong conclude that some banks may actually need two or three times as much capital as the T1CR suggests. A couple of the specific examples of the approach are useful to examine. Silver State Bank (SSB) was a Nevada commercial bank with 17 branches in Las Vegas and Phoenix. It was deemed insolvent on September 5, 2008. SSB’s T1CR was 9.67 per cent and it was marked as Well Capitalised. SSB’s CAMELS score also looked strong, partly as SSB was very profitable in both 2006 Q4 and 2007 Q4. But from December 2006 the ELPR indicator would have flagged SSB as severely undercapitalised (with a probability of failing by Dec 2008 of 14.47 per cent; much higher than the base measure of 2 per cent). Another example concerns Oklahoma-based FNBL, a small national bank that started in 1902. The bank survived the crisis and is operational today. FNBL’s T1CR at the end of 2006 was 5.32 per cent, placing it in the riskiest one percent, with a 22 per cent implied probability of failure by the end of 2008. In contrast, FNBL’s ELPR measure at the end of 2006 was 2.21, around the 50th percentile, implying only a slim probability of failure by 2008 of around 0.07 per cent.

There is clearly still more work to do on this, but using publicly available data, they have managed to demonstrate quite a promising metric that captures a low frequency form of failure risk. Unlike static risk weights, their ELPR measure captures intertemporal variations in default risk over business cycles, and as it is based on the long-run variance of delinquency rates rather than recent quarterly estimates, it should reduce procyclicality concerns associated with IRB model-based estimates. Their approach is also more transparent and objective than the current IRB models approach. Whilst the authors have gone as far as to nominate their LPR approach as a new prototype Capital Adequacy Ratio measure for the Basel Committee to consider, I think that at the very least regulators may consider using it as part of their Pillar 2 concentration risk review. The methodology could be uniformly applied to all banks in the system, and then provide an additional triangulation point for particular banks that may be at risk and whose current T1CR may therefore not capture the full extent of risk they are exposed to.