With stress tests, less is more
Dynamic stress testing — or using econometric models to forecast a financial institution’s income and regulatory capital through hypothetical economic shocks — is a good idea. But the Federal Reserve’s rules need to keep improving to make them less onerous and more reliable.
Fortunately, the Fed seems to agree, having taken several recent steps in that direction, among them the March 28 publication of 80 pages of new data explaining model assumptions and analytics. Vice Chairman Randal Quarles speaks openly about the issues that stress tests raise, which in itself is a breath of fresh air. Still, while the Fed has now disclosed more than it ever has about its stress tests since he was appointed, the information is not as granular as institutions would like, and timing of important assumptions still seems to be an issue. The delivery of this latest data came just one week before banks were scheduled to make their annual Comprehensive Capital Analysis and Review, or CCAR, submissions.
That said, it’s worth giving the Fed credit for signaling that it appreciates the relationship between risk management and profitability, and for taking the first steps to make its stress testing process more transparent and efficient. While some may feel that it is of no consequence if large banks are inconvenienced by cumbersome capital and stress testing requirements, they couldn’t be more misinformed. Stress test requirements directly impact the economy and the wallets of every American.
Capital requirements and stress tests assign risk to each asset on the books of a bank. Those risk weightings determine how much capital an institution must hold, which in turn influences how credit is allocated throughout the country.
In addition, economic and timing uncertainties in the process cause banks to hold capital buffers beyond what the law requires just to avoid being out of compliance when the assumptions of the model shift, as they can each year. Every extra dollar of capital that is locked away in a bank’s vault to pad this surplus directly reduces multiples of that dollar that could otherwise be lent to American consumers and businesses.
In its March 28 release, the Fed states that the stress test process will become more effective as its mysteries are reduced. As it moves in that direction, it is balancing the benefits of complete transparency and the risk of disclosing so much that banks can “game the system.”
Having been a bank regulator whose agency initially employed a black box model to determine the winning bids for hundreds of failed institutions, I can appreciate the attraction of opaque assumptions. But we learned, as the Fed suggests, that transparency improves the model and engenders credibility. If providing too much information makes reaching capital adequacy a more effective process, it is hard to see how the system gets gamed.
Critics reflexively argue that improving the stress test and bank capital requirements is a camouflaged way of reducing capital requirements that will pave the road to economic perdition. They have likely never tried to calculate a capital requirement or worked through a stress test. Even Fed governors admit that the current 24 different measurements of capital adequacy need slimming down.
The more straightforward that regulation is, the more effective it will be. The dog and the Frisbee analogy noted by the economist Andrew G. Haldane of the Bank of England posits that a dog catches a Frisbee by simplifying velocity, trajectory and gravity and simply runs so that his gaze remains constant to the path of the Frisbee. Mr. Haldane suggests that financial crisis catchers — regulators — fail because there are increasingly too many of them creating complex regulation that that is costly, cumbersome and unlikely to predict or control a crisis.
Models are tools, and as such, are driven by the reliability of their assumptions. Michael Lewis explains in his 2008 book "Panic" why the highly regarded Black Scholes options pricing model did not perform in the 1987 stock market. It simply failed to anticipate that selling short doesn’t work when no one in the market is buying.
When models regulate to the average, as they typically do, they can actually create the problem they are meant to prevent. A one-size-fits-all stress test creates a financial monoculture of risk that can deepen and prolong economic downturns as every bank’s balance sheet reacts the same way in a crisis. That effect is exacerbated by the fact that, according to a 2019 survey by Deloitte Global Risk Management, capital stress testing is increasingly driving the operating decisions of financial institutions.
Finally, models and can be seductively deceptive. Too often, increasing reliance on them encourages a false sense of security by users and leads to underutilization of human judgment and instinct. A paper released on April 1 by Paul H. Kupiec, a resident scholar at the American Enterprise Institute, concludes that how much capital a financial institution requires using a stress test model is largely a “guess” based on the expert judgment of the regulators.
The performance of a model is ultimately a function of the quality and amount of data it relies upon. Financial and industrial companies are using artificial intelligence and super intelligent computers to simplify vast amounts of big data to increase predictability, reduce risk and optimize efficiencies. Bank regulators are, unfortunately, way behind in the use of AI to help them regulate.
I believe an answer to better stress testing and better regulation overall is better technology in the hands of the regulators. Artificial intelligence and its ability to marshal big data may eventually help them create stress tests that are highly reliable and less intrusive. Until then, as Andrew Haldane noted, asking today’s regulators to prevent tomorrow’s crisis with yesterday’s tools is like asking a dog to “catch a Frisbee by first applying Newton’s Law of Gravity.”