By Giampaolo Gabbi, Professor of Risk Management Practice SDA Bocconi School of Management. Originally published at VoxEU.
The collapse of Silicon Valley Bank in 2023 highlighted the intertwined nature of multiple risk factors. This column outlines measures to correct some of the risk correlation limitations seen in the SVB case. It argues that models for determining asset correlations can be improved to better match empirical evidence on these relationships across firms and economic cycles. Furthermore, regulation and intervention measures should be focused on the correlations and causal relationships between risk drivers. Organisational measures should promote transparency and the protection of the value of financial stability.
“How did this happen?” According to Michael S. Barr, Vice Chair for Supervision at the Board of Governors of the Federal Reserve System, Silicon Valley Bank’s collapse is a “textbook case of mismanagement” (Financial Times 2023). But it is also a perfect case for a risk management textbook.
It is a perfect case because credit, market, interest rate, liquidity, and also conduct and operational risks are intertwined. If you then consider the contagion implications that were observed in the following days, systemic risk as well.
And it is crucial to try to understand how we have underestimated the correlation between so many factors.
The increasingly articulated banking regulatory framework has progressively added more risk factors, with new capital ratios, divided differently between Pillar I and Pillar II, but has not adequately addressed their interconnectedness, if one only traces the evolution from Basel I to Basel IV.
Analysing the structure of international banking regulation, one always gets the impression that it is organised in separate silos, like university textbook chapters on risk management.
But bank governance does not work that way (Dewatripont et al. 2023). And its supervision cannot be codified in such a mechanical way without taking into account the inevitable interactions between risk drivers.
We will not address the extension of international rules to regional banks. This is a relevant issue in certain countries that are not regionally based per se, such as the US, where a bank like Silicon Valley Bank (SVB), with a capitalisation between that of Bearn Sterns and Lehman Brothers, could hardly be considered regional.
This column will try to understand whether it is possible to correct some of the risk correlation limitations that the SVB case has highlighted most clearly, assuming that the question of the universal application of the rules can be resolved.
Asset Correlations
SVB’s loan portfolio was notoriously concentrated in technology start-ups. This specialisation enabled the bank to understand the financial (and perhaps industrial) needs of its counterparts. However, it created an unavoidable concentration risk (Gompers 2023).
This situation is not dissimilar to that of many other banks with limited territorial articulation operating in industrial districts. In their case, asset correlations are likely to be very high. They can create great value in positive market phases, but also reduce the diversification benefits in other phases.
A much-criticised aspect of regulation is that asset and default correlations fail to calibrate properly.
This is especially true when the economic cycle turns and interest rate changes become substantial.
Vozzella and Gabbi (2020) show how the empirical asset correlation (AC) increases with the probability of default. Second, the relationship between asset correlation and firm size shows an opposite pattern in different economic phases. In particular, the relationship has a U-shape during economic recoveries and an inverted U-shape when the economy enters a downturn. Finally, the results show that the empirical correlation values are much lower than those set by the regulator and that they vary across economic cycles. The results were obtained using a database of micro, small, and medium-sized enterprises (MSMEs), which are ideally comparable to the start-ups whose portfolio concentration contributed to the US regional banking crises.
It would therefore be desirable to revise the model for determining the asset correlations that affect portfolio risk-weighted assets (RWAs) and hence capital ratios. This could be in line with the macro-prudential logic of the countercyclical capital buffer: the decision by national monetary authorities to recalibrate bank capital by correcting for credit concentration risk. In line with the spirit of the targeted review of internal models (TRIM), it can therefore be suggested that a standard formula, defined by supervisors, be constructed that is more in line with the empirical evidence in countries where concentration risk emerges.
Measures to facilitate access to credit, which have already been tried for retail loans with the supporting factor, for infrastructure financing in the aftermath of the Covid-19 pandemic, and are currently being explored to accelerate the process of achieving sustainability targets under the EU Action Plan, may also be effective if the calculation of risk-weighted assets can be prevented from sterilising their impact.
As shown in Gabbi and Vozzella (2023), the likelihood of micro, small, and medium-sized enterprises having less access to credit than large firms remains high even after applying the supporting factor. Credit rationing mainly affects micro-enterprises, creating a crowding out effect and competitive inequality for countries where production and the business cycle are heavily dependent on smaller firms. The study of the effectiveness of the supporting factor introduced for European enterprises in 2013 (Regulation EU No 575, CRR) shows that banks have gained an overall advantage from small loan portfolios that may not have been passed on to the economic system. This is due to the risk of inadequate asset correlation measures, as described above. If the mechanisms underlying the regulatory formula introduced to improve access to credit for small and medium-sized enterprises are not corrected, credit support instruments such as those introduced by Article 501 of the CRR will not fully achieve their objectives.
In addition to the asset correlation issue, the SVB case also revealed a liability correlation issue: the funding of the bank was highly concentrated and in many cases exceeded the already high level of deposit insurance (at least compared to European banks) as discussed by Honohan (2023) in this debate. The flight of depositors thus turned into a run on the whole bank, with the consequences that we have observed. This raises the question of how different risk drivers can give rise to extreme scenarios, which are so far unexpected within the framework of banking regulation.
The Interconnections among Risks
The most critical element of the SVB case is to understand the mechanisms between the risks, and to highlight that it is not all due to unpredictable perfect storms. There are precise relationships that have also been observed in the past (e.g. the savings and loan association crisis), which were driven by partly similar factors and had dictated the revision of some rules, especially on maturity mismatch.
In the history of regulation, crises are often followed by a process of regulatory innovation and, following their effectiveness, by a debate that leads to their weakening. This is without taking into account that the absence of significant defaults may depend precisely on the effectiveness of these rules. This was the case with the interventions on the interest rate risk in the banking book and on the liquidity ratio prior to the financial crisis of 2008. Then the liquidity coverage ratio and the net stable funding ratio were introduced with Basel III.
But what is still lacking is an understanding of the risk network. How one risk can generate or intensify another, or worse, several others, as in the case of SVB.
The presence of government bonds in the bank’s trading book is often conditioned by the need to support government debt. The absence of credit risk (or the presumed risk-free rate) actually determines a market risk in the trading book, and rising interest rates, in the case of a positive duration gap, can lead to a reduction in the value of the bank’s assets. These signals are potentially exacerbated by the crisis of a sector to which one is exposed. Finally, the need for recapitalisation is a signal of imbalance, which determines a potential bank run.
And even winding-up operations, as the cases of Credit Suisse and First Republic have shown, do not prevent a run on depositors, which can lead to a systemic panic.
Paradoxically, all this is exacerbated by the heterogeneous application of market discipline (Pillar III). This is supposed to allow different stakeholders to understand the extent to which a bank is exposed to different risks. In the case of SVB, the reporting was incomplete. The disclosures on interest rate risk for non-trading activities under “Quantitative and Qualitative Disclosures on Market Risk – Interest Rate Risk Management” in Part II, Item 7A of SVBFG’s 2022 Form 10-K (SVB Financial Group 2022) did not include any indication of the impact that an increase in interest rates would have on the economic value of equity.
However, this is only one of the scenarios where risks interact and contribute to the banking crisis.
The element that urgently needs to change is the process of risk integration: today, risks are considered in isolation, ignoring the fact that extreme events (such as an unexpected rise in interest rates in 2022) are accompanied by effects on all possible exposures. Furthermore, the low and sometimes negative correlations observed in normal cases all jump and rise abruptly and the tails of the distributions cannot be ignored because that is where crises occur. Finally, banks cannot be allowed to calculate and report risk-weighted assets in so many different ways.
These correlations, and hopefully the causal relationships between risk drivers, need to be more organically captured by regulation. The system of reporting, internal controls, and supervision must be able to intervene with predictive logic once these mechanisms have been identified.
The intervention measures should be able to act as a dam to protect the credibility of individual institutions and of the banking system as a whole. A set of early warning systems based on risk linkages should be put in place. Once activated, these would require the management of recovery plans that are robust enough to avoid crises of confidence.
The lessons of this crisis must be the basis for a more organic system that is less concerned with individual risks and more concerned with crisis mechanisms. Effective measures must be in place, not only in terms of capital and liquidity ratios, but also in terms of organisational measures that promote transparency and the protection of the value of financial stability within a coherent risk appetite framework.
References available at the original.