COVID-19 RESOURCES AND INFORMATION: See the Atlanta Fed's list of publications, information, and resources for help navigating through these uncertain times. Also listen to our special Pandemic Response webinar series.

About


The Atlanta Fed's macroblog provides commentary and analysis on economic topics including monetary policy, macroeconomic developments, inflation, labor economics, and financial issues.

Authors for macroblog are Dave Altig, John Robertson, and other Atlanta Fed economists and researchers.

Comment Standards:
Comments are moderated and will not appear until the moderator has approved them.

Please submit appropriate comments. Inappropriate comments include content that is abusive, harassing, or threatening; obscene, vulgar, or profane; an attack of a personal nature; or overtly political.

In addition, no off-topic remarks or spam is permitted.

January 4, 2018

Financial Regulation: Fit for New Technologies?

In a recent interview, the computer scientist Andrew Ng said, "Just as electricity transformed almost everything 100 years ago, today I actually have a hard time thinking of an industry that I don't think AI [artificial intelligence] will transform in the next several years." Whether AI effects such widespread change so soon remains to be seen, but the financial services industry is clearly in the early stages of being transformed—with implications not only for market participants but also for financial supervision.

Some of the implications of this transformation were discussed in a panel at a recent workshop titled "Financial Regulation: Fit for the Future?" The event was hosted by the Atlanta Fed and cosponsored by the Center for the Economic Analysis of Risk at Georgia State University (you can see more on the workshop here and here). The presentations included an overview of some of AI's implications for financial supervision and regulation, a discussion of some AI-related issues from a supervisory perspective, and some discussion of the application of AI to loan evaluation.

As a part of the panel titled "Financial Regulation: Fit for New Technologies?," I gave a presentation based on a paper  I wrote that explains AI and discusses some of its implications for bank supervision and regulation. In the paper, I point out that AI is capable of very good pattern recognition—one of its major strengths. The ability to recognize patterns has a variety of applications including credit risk measurement, fraud detection, investment decisions and order execution, and regulatory compliance.

Conversely, I observed that machine learning (ML), the more popular part of AI, has some important weaknesses. In particular, ML can be considered a form of statistics and thus suffers from the same limitations as statistics. For example, ML can provide information only about phenomena already present in the data. Another limitation is that although machine learning can identify correlations in the data, it cannot prove the existence of causality.

This combination of strengths and weaknesses implies that ML might provide new insights about the working of the financial system to supervisors, who can use other information to evaluate these insights. However, ML's inability to attribute causality suggests that machine learning cannot be naively applied to the writing of binding regulations.

John O'Keefe from the Federal Deposit Insurance Corporation (FDIC) focused on some particular challenges and opportunities raised by AI for banking supervision. Among the challenges O'Keefe discussed is how supervisors should give guidance on and evaluate the application of ML models by banks, given the speed of developments in this area.

On the other hand, O'Keefe observed that ML could assist supervisors in performing certain tasks, such as off-site identification of insider abuse and bank fraud, a topic he explores in a paper  with Chiwon Yom, also at the FDIC. The paper explores two ML techniques: neural networks and Benford's Digit Analysis. The premise underlying Benford's Digit Analysis is that the digits resulting from a nonrandom number selection may differ significantly from expected frequency distributions. Thus, if a bank is committing fraud, the accounting numbers it reports may deviate significantly from what would otherwise be expected. Their preliminary analysis found that Benford's Digit Analysis could help bank supervisors identify fraudulent banks.

Financial firms have been increasingly employing ML in their business areas, including consumer lending, according to the third participant in the panel, Julapa Jagtiani from the Philadelphia Fed. One consequence of this use of ML is that it has allowed both traditional banks and nonbank fintech firms to become important providers of loans to both consumers and small businesses in markets in which they do not have a physical presence.

Potentially, ML also more effectively measures a borrower's credit risk than a consumer credit rating (such as a FICO score) alone allows. In a paper  with Catharine Lemieux from the Chicago Fed, Jagtiani explores the credit ratings produced by the Lending Club, an online lender that that has become the largest lender for personal unsecured installment loans in the United States. They find that the correlation between FICO scores and Lending Club rating grades has steadily declined from around 80 percent in 2007 to a little over 35 percent in 2015.

It appears that the Lending Club is increasingly taking advantage of alternative data sources and ML algorithms to evaluate credit risk. As a result, the Lending Club can more accurately price a loan's risk than a simple FICO score-based model would allow. Taken together, the presenters made clear that AI is likely to also transform many aspects of the financial sector.

January 3, 2018

Is Macroprudential Supervision Ready for the Future?

Virtually everyone agrees that systemic financial crises are bad not only for the financial system but even more importantly for the real economy. Where the disagreements arise is how best to reduce the risk and costliness of future crises. One important area of disagreement is whether macroprudential supervision alone is sufficient to maintain financial stability or whether monetary policy should also play an important role.

In an earlier Notes from the Vault post, I discussed some of the reasons why many monetary policymakers would rather not take on the added responsibility. For example, policymakers would have to determine the appropriate measure of the risk of financial instability and how a change in monetary policy would affect that risk. However, I also noted that many of the same problems also plague the implementation of macroprudential policies.

Since that September 2014 post, additional work has been done on macroprudential supervision. Some of that work was the topic of a recent workshop, "Financial Regulation: Fit for the Future?," hosted by the Atlanta Fed and cosponsored by the Center for the Economic Analysis of Risk at Georgia State University. In particular, the workshop looked at three important issues related to macroprudential supervision: governance of macroprudential tools, measures of when to deploy macroprudential tools, and the effectiveness of macroprudential supervision. This macroblog post discusses some of the contributions of three presentations at the conference.

The question of how to determine when to deploy a macroprudential tool is the subject of a paper  by economists Scott Brave (from the Chicago Fed) and José A. Lopez (from the San Francisco Fed). The tool they consider is countercyclical capital buffers, which are supplements to normal capital requirements that are put into place during boom periods to dampen excessive credit growth and provide banks with larger buffers to absorb losses during a downturn.

Brave and Lopez start with existing financial conditions indices and use these to estimate the probability that the economy will transition from economic growth to falling gross domestic product (GDP) (and vice versa), using the indices to predict a transition from a recession to growth. Their model predicted a very high probability of transition to a path of falling GDP in the fourth quarter of 2007, a low probability of transitioning to a falling path in the fourth quarter of 2011, and a low but slightly higher probability in the fourth quarter of 2015.

Brave and Lopez then put these probabilities into a model of the costs and benefits associated with countercyclical capital buffers. Looking back at the fourth quarter of 2007, their results suggest that supervisors should immediately adopt an increase in capital requirements of 25 basis points. In contrast, in the fourth quarters of both 2011 and 2015, their results indicated that no immediate change was needed but that an increase in capital requirements of 25 basis points might be need to be adopted within the next six or seven quarters.

The related question—who should determine when to deploy countercyclical capital buffers—was the subject of a paper  by Nellie Liang, an economist at the Brookings Institution and former head of the Federal Reserve Board's Division of Financial Stability, and Federal Reserve Board economist Rochelle M. Edge. They find that most countries have a financial stability committee, which has an average of four or more members and is primarily responsible for developing macroprudential policies. Moreover, these committees rarely have the ability to adopt countercyclical macroprudential policies on their own. Indeed, in most cases, all the financial stability committee can do is recommend policies. The committee cannot even compel the competent regulatory authority in its country to either take action or explain why it chose not to act.

Implicit in the two aforementioned papers is the belief that countercyclical macroprudential tools will effectively reduce risks. Federal Reserve Board economist Matteo Crosignani presented a paper  he coauthored looking at the recent effectiveness of two such tools in Ireland.

In February 2015, the Irish government watched as housing prices climbed from their postcrisis lows at a potentially unsafe rate. In an attempt to limit the flow of funds into risky mortgage loans, the government imposed limits on the maximum permissible loan-to-value (LTV) ratio and loan-to-income ratio (LTI) for new mortgages. These regulations became effective immediately upon their announcement and prevented the Irish banks from making loans that violated either the LTV or LTI requirements.

Crosignani and his coauthors were able to measure a large decline in loans that did not conform to the new requirements. However, they also find that a sharp increase in mortgage loans that conformed to the requirements largely offset this drop. Additionally, Crosignani and his coauthors find that the banks that were most exposed to the LTV and LTI requirements sought to recoup the lost income by making riskier commercial loans and buying greater quantities of risky securities. Their findings suggest that the regulations may have stopped higher-risk mortgage lending but that other changes in their portfolio at least partially undid the effect on banks' risk exposure.

August 11, 2016

Forecasting Loan Losses for Stress Tests

Bank capital requirements are back in the news with the recent announcements of the results of U.S. stress tests by the Federal Reserve and the European Union (E.U.) stress tests by the European Banking Authority (EBA). The Federal Reserve found that all 33 of the bank holding companies participating in its test would have continued to meet the applicable capital requirements. The EBA found progress among the 51 banks in its test, but it did not define a pass/fail threshold. In summarizing the results, EBA Chairman Andrea Enria is widely quoted as saying, "Whilst we recognise the extensive capital raising done so far, this is not a clean bill of health," and that there remains work to do.

The results of the stress tests do not mean that banks could survive any possible future macroeconomic shock. That standard would be an extraordinarily high one and would require each bank to hold capital equal to its total assets (or maybe even more if the bank held derivatives). However, the U.S. approach to scenario design is intended to make sure that the "severely adverse" scenario is indeed a very bad recession.

The Federal Reserve's Policy Statement on the Scenario Design Framework for Stress Testing indicates that the severely adverse scenario will have an unemployment increase of between 3 and 5 percentage points or a level of 10 percent overall. That statement observes that during the last half century, the United States has seen four severe recessions with that large of an increase in the unemployment rate, with the rate peaking at more than 10 percent in last three severe recessions.

To forecast the losses from such a severe recession, the banks need to estimate loss models for each of their portfolios. In these models, the bank estimates the expected loss associated with a portfolio of loans as a function of the variables in the scenario. In estimating these models, banks often have a very large number of loans with which to estimate losses in their various portfolios, especially the consumer and small business portfolios. However, they have very few opportunities to observe how the loans perform in a downturn. Indeed, in almost all cases, banks started keeping detailed loan loss data only in the late 1990s and, in many cases, later than that. Thus, for many types of loans, banks might have at best data for only the relatively mild recession of 2001–02 and the severe recession of 2007–09.

Perhaps the small number of recessions—especially severe recessions—would not be a big problem if recessions differed only in their depth and not their breadth. However, even comparably severe recessions are likely to hit different parts of the economy with varying degrees of severity. As a result, a given loan portfolio may suffer only small losses in one recession but take very large losses in the next recession.

With the potential for models to underestimate losses given there are so few downturns to calibrate to, the stress testing process allows humans to make judgmental changes (or overlays) to model estimates when the model estimates seem implausible. However, the Federal Reserve requires that bank holding companies should have a "transparent, repeatable, well-supported process" for the use of such overlays.

My colleague Mark Jensen recently made some suggestions about how stress test modelers could reduce the uncertainty around projected losses because of limited data from directly comparable scenarios. He recommends using estimation procedures based on a probability theorem attributed to Reverend Thomas Bayes. When applied to stress testing, Bayes' theorem describes how to incorporate additional empirical information into an initial understanding of how losses are distributed in order to update and refine loss predictions.

One of the benefits of using techniques based on this theorem is that it allows the incorporation of any relevant data into the forecasted losses. He gives the example of using foreign data to help model the distribution of losses U.S. banks would incur if U.S. interest rates become negative. We have no experience with negative interest rates, but Sweden has recently been accumulating experience that could help in predicting such losses in the United States. Jensen argues that Bayesian techniques allow banks and bank supervisors to better account for the uncertainty around their loss forecasts in extreme scenarios.

Additionally, I have previously argued that the existing capital standards provide further way of mitigating the weaknesses in the stress tests. The large banks that participate in the stress tests are also in the process of becoming subject to a risk-based capital requirement commonly called Basel III that was approved by an international committee of banking supervisors after the financial crisis. Basel III uses a different methodology to estimate losses in a severe event, one where the historical losses in a loan portfolio provide the parameters to a loss distribution. While Basel III faces the same problem of limited loan loss data—so it almost surely underestimates some risks—those errors are likely to be somewhat different from those produced by the stress tests. Hence, the use of both measures is likely to somewhat reduce the possibility that supervisors end up requiring too little capital for some types of loans.

Both the stress tests and risk-based models of the Basel III type face the unavoidable problem of inaccurately measuring risk because we have limited data from extreme events. The use of improved estimation techniques and multiple ways of measuring risk may help mitigate this problem. But the only way to solve the problem of limited data is to have a greater number of extreme stress events. Given that alternative, I am happy to live with imperfect measures of bank risk.

Author's note: I want to thank the Atlanta Fed's Dave Altig and Mark Jensen for helpful comments.


April 13, 2016

Putting the MetLife Decision into an Economic Context

In a recently released decision, a U.S. district court has ruled that the Financial Stability Oversight Council's (FSOC's) decision to designate MetLife as a potential threat to financial stability was "arbitrary and capricious" and rescinded that designation. This decision raises many questions, among them:

  • Why did MetLife sue to end its status as a too-big-to-fail (TBTF) firm?
  • How will this decision affect the Federal Reserve's regulation of nonbank financial firms?
  • What else can be done to reduce the risk of crisis arising from nonbank financial firms?

Why does MetLife want to end its TBTF status?
An often-expressed concern is that market participants will consider FSOC-designated firms too big to fail, and investors will accord these firms lower risk premiums  (see, for example, Peter J. Wallison). The result is that FSOC-designated firms will gain a competitive advantage. If so, why did MetLife sue to have the designation rescinded? And why did the announcement of the court's determination result in an immediate 5 percent increase in the MetLife's stock price?

One possible explanation is that the FSOC's designation guarantees the firm will be subject to higher regulatory costs, but it only marginally changes the likelihood it would receive a government bailout. The Dodd-Frank Act (DFA) requires that FSOC-designated firms be subject to consolidated prudential supervision by the Federal Reserve using standards that are more stringent than the requirements for other nonbank financial firms.

Moreover, the argument that such designation automatically conveys a competitive advantage has at least two weaknesses. First, although Title II of the DFA authorizes the Federal Deposit Insurance Corporation (FDIC) to resolve a failing nonbank firm in certain circumstances, DFA does not provide FDIC insurance for any of the nonbank firm's liabilities, nor does it provide the FDIC with funds to undertake a bailout. The FDIC is supposed to recover its costs from the failed firm's assets. Admittedly, DFA does allow for the possibility that the FDIC would need to assess other designated firms for part of the cost of a resolution. However, MetLife could as easily have been assessed to pay for another firm as it could have been the beneficiary of assessments on other systemically important firms.

A second potential weakness in the competitive advantage argument is that the U.S. Treasury Secretary decides to invoke FDIC resolution only after receiving a recommendation from the Federal Reserve Board and one other federal financial regulatory agency (depending upon the type of failing firm). Invocation of resolution is not automatic. Moreover, a part of any decision authorizing FDIC resolution are findings that at the time of authorization:

  • the firm is in default or in danger of default,
  • resolution under other applicable law (bankruptcy statutes) would have "serious adverse consequences" on financial stability, and
  • those adverse effects could be avoided or mitigated by FDIC resolution.

Although it would seem logical that FSOC-designated firms are more likely to satisfy these criteria than other financial firms, the Title II criteria for FDIC resolution are the same for both types of firms.

How does this affect the Fed's regulation of nonbank firms?
Secretary of the Treasury Jack Lew has indicated his strong disagreement with the district court's decision, and the U.S. Treasury has said it will appeal. Suppose, however, that FSOC designation ultimately does become far more difficult. How significantly would that affect the Federal Reserve's regulatory power over nonbank financial firms?

Although the obvious answer would be that it would greatly reduce the Fed's regulatory power, recent experience casts some doubt on this view. Nonbank financial firms appear to regard FSOC designation as imposing costly burdens that substantially exceed any benefits they receive. Indeed, GE Capital viewed the costs as so significant that it had been selling large parts of its operations and recently petitioned the FSOC to rescind its designation. Unless systemically important activities are a core part of the firm's business model, nonbank financial firms may decide to avoid undertaking activities that would risk FSOC designation.

Thus, a plausible set of future scenarios is that the Federal Reserve would be supervising few, if any, nonbank financial firms regardless of the result of the MetLife case. Rather, ultimate resolution of the case may have more of an impact on whether large nonbank financial firms conduct systemically important activities (if designation becomes much harder) or the activities are conducted by some combination of smaller nonbank financial firms and by banks that are already subject to Fed regulation (if the ruling does not prevent future designations).

Lessons learned?
Regardless of how the courts and the FSOC respond to this recent judicial decision, the financial crisis should have taught us valuable lessons about the importance of the nonbank financial sector to financial stability. However, those lessons should go beyond merely the need to impose prudential supervision on any firms that are systemically important.

The cause of the financial crisis was not the failure of one or two large nonbank financial firms. Rather, the cause was that almost the entire financial system stood on the brink of collapse because almost all the major participants were heavily exposed to the weak credit standards that were pervasive in the residential real estate business. Yet if the real problem was the risk of multiple failures as a result of correlated exposures to a single large market, perhaps we ought to invest more effort in evaluating the riskiness of markets that could have systemic consequences.

In an article in Notes from the Vault and other forums, I have called for systematic end-to-end reviews of major financial markets starting with the origination of the risks and ending with the ultimate holder(s) of the risks. This analysis would involve both quantitative analysis of risk measures and qualitative analysis of the safeguards designed to reduce risk.

The primary goal would be to identify and try to correct weaknesses in the markets. A secondary goal would be to give the authorities a better sense of where problems are likely to arise if a market does encounter problems.