There is a perception held by some that banks can’t be trusted to make assessments of their risk profile for fear that they will always underreport their level of risk and “game the system.” This view is often used to support the use of risk-blind capital requirements as well as the use of regulatory risk models instead of bank risk models in the stress tests. More broadly, this view supports a general notion that large bank regulatory capital should be fully prescribed by regulators without any input from banks. In this post, we show that the evidence points to the exact opposite — bank risk models often show higher levels of risk than that of regulatory risk models. In addition, using the experience of the recent pandemic, we show that bank risk models respond swiftly to increases in risk and uncertainty. These data convincingly show that bank risk assessments respond to increases in volatility. Accordingly, regulators and the public should give greater consideration to bank risk assessments in the regulatory framework.
Standardized vs. Advanced Risk-Weighted Assets
Large banks, such as Financial Services Forum members, are subject to capital requirements that depend on firms’ risk-weighted assets (RWA). RWAs are intended to reflect the relative riskiness of different assets so that the amount of capital required on, say, a U.S. Treasury bond is lower than the amount of capital required on a loan to a fledgling tech start up. RWAs are computed according to two methods: a standardized approach and an advanced approach.
The standardized approach is essentially a very large and complex look-up table that assigns a specific risk weight to each asset category. These risk weights do not vary with the risk profile of the underlying asset and are generally quite conservative. As an example, under the standardized approach, all corporate loans receive a 100% risk weight regardless of the history and payment experience of the corporate borrower. Accordingly, a loan to a stable blue-chip company is perceived to be as risky as a loan to a start up company. In addition, banks also compute RWAs according to the advanced approach, which uses bank internal models to assess the risk of an asset. In the case of a corporate loan, the advanced approach considers the financial health and historical experience of a corporate borrower so that borrowers with a demonstrably lower risk profile receive a lower risk weight. What’s more, advanced approach RWA calculations are updated over time as risk conditions change, while standardized risk weights are not. This is important because it implies that advanced approach RWAs have the capacity to respond to a changing risk environment.
Standardized vs. Advanced RWA: What Do the Data Show?
We consider the difference between advanced and standardized RWAs for the Global Systemically Important Banks based in the U.S. – the Forum members — over the past five years: 2016-2020. As there are eight Forum members and 20 quarters over the past five years, there are 160 bank-quarter observations. In Table 1 we show the number of bank-quarters in which advanced approach RWA exceeds standardized RWA and the number of bank-quarters in which standardized RWA exceeds advanced RWA. As shown in the table, Forum members report advanced RWAs that are larger than standardized RWAs in 73 of 160 bank-quarters. This finding is important as it clearly shows that bank risk assessments are often more conservative than standardized, regulatory risk assessments. Accordingly, a view that bank risk models “always underreport risk” is simply not consonant with the data. Further, the finding that bank risk models are more conservative than regulatory risk assessments nearly half the time is somewhat surprising and illuminating in light of the substantial conservatism built into regulatory risk assessments.
How do Bank Risk Assessments Respond to a Changing Risk Environment?
The oft-used quip that “the only constant in life is change” couldn’t be more true in the context of risk monitoring. Risk changes and often without warning. As a result, it is important that risk assessments maintain sufficient flexibility to respond to changing risk conditions. The recent experience of the pandemic provides an interesting natural experiment that can be used to assess the sensitivity of bank risk models to changing risk conditions.
In Figure 1 we show the number of Forum members who reported advanced RWA that were higher than the standardized RWA from the fourth quarter of 2019 through the fourth quarter of 2020. We also show a well-accepted measure of uncertainty – dispersion in professional GDP forecasts from the Survey of Professional Forecasters. When professional forecasters disagree more about the future trajectory of the economy, there is greater uncertainty in the economy. This series is useful for identifying how much risk and uncertainty changed over the course of 2020.
As shown in Figure 1, uncertainty clearly rose in the second quarter of 2020 as the virus took hold. As the figure also shows, however, risk and uncertainty have abated since that time. At the end of 2020, risk was somewhat elevated relative to the end of 2019, but not by an unprecedented amount.
Looking at the data on bank risk models, we see that the number of Forum members for which the advanced RWA exceeded the standardized amount closely tracks the change in risk and uncertainty. Specifically, in the fourth quarter of 2019, two members exhibited an advanced RWA in excess of the standardized RWA. But in the second quarter of 2020 – when uncertainty was at its peak – this number more than doubled to five of the eight Forum firms. Accordingly, these data clearly show that bank risk models are sensitive to changing risk conditions and bank risk assessments increase when overall risk and uncertainty increases.
Large bank risk assessments are often more conservative than standardized regulatory risk assessments. This in itself is worth noting in light of the intentionally conservative nature of regulatory risk assessments. In addition, large bank risk assessments are sensitive to changing risk conditions. These data clearly show that the notion that bank risk assessments are artificially low and are meant to game the system is incorrect. The data clearly show otherwise. This finding has important implications for how regulators and the public should think about the use of bank models in the regulatory capital framework. Rather than replacing bank risk assessments with standardized regulatory models, regulators should review and consider how bank driven risk assessments can inform and improve the large bank regulatory framework.