Comments and Discussion.

MetriCk, Andrew ; Tarullo, Daniel K.

Bank capital regulation is a policy area that has grown to be extremely technical and complex. In this paper, Robin Greenwood, Samuel Hanson, Jeremy Stein, and Adi Sunderam venture into this morass and make an important contribution, providing an intuitive benchmark model for thinking about capital requirements. Using this benchmark model, they propose three core principles, which I paraphrase as (i) multiple bank capital requirements should be collapsed into one; (ii) concerns about regulatory arbitrage are best met with regulatory discretion, not additional ex ante rules; and (iii) after a shock to bank capital ratios, regulators should force banks to increase their capital (the numerator), and not to decrease their risk-weighted assets (the denominator).

Greenwood and his colleagues provide one of the cleanest expositions that I have seen of the main issues in bank capital regulation. Along with some previous work by subsets of these authors, the present paper joins a small group that should start the reading list of anyone interested in the topic. But the specific conclusions are, I think, special to the benchmark model, and I believe it is premature to extend them into real-world policy recommendations. To explain my reticence, I need to start with some background on the tangled web of bank capital. In this comment, I attempt to answer four background questions. With these questions answered, I then provide four observations about the paper.

QUESTION 1: WHAT IS BANK CAPITAL, AND WHY WOULD BANKS WANT TO HOLD IT? In a first approximation, "bank capital" is the same as "equity on the balance sheet," and the bank capital ratio is equity divided by some weighted average of assets, where weights are defined by the "riskiness" of the asset. When all weights are the same, then the bank capital ratio is the same as the "leverage ratio." Things get more complicated when we allow for various flavors of equity (and junior debt) to also count as "capital," and when the weights are built by complex models or are implicitly defined in stress tests. But to get the flavor of what bank capital actually does, it is fine to just think of this first approximation.

In principle, it would be logical to think that the higher bank capital is, the lower the probability of distress and failure. If a bank gets a negative shock that reduces the value of its assets by 15 percent, then that bank will still be solvent if it had 20 percent equity to begin with, but it would be insolvent if it had started with only 10 percent equity. For the moment, we assume that distress is costly and that holding capital is also costly (but more on that below). Then, a profit-maximizing bank would make a trade-off, holding costly capital against the reduced probability of distress. With this trade-off, we can see why a bank might have some optimal positive level of capital.

QUESTION 2: WHY SHOULD THE GOVERNMENT CARE ABOUT BANK CAPITAL? The cost-benefit calculation for capital would also hold for any firm in any industry, not just for banks. But the government does not regulate the capital levels of almost any other industry. We have no government minimums for capital ratios in Internet companies, bookstores, or restaurants. If you want to open a McDonald's franchise, then your investors and lenders may insist on some minimum equity investment, but the government stays out of it.

But banks are different, for a few reasons. First, most governments offer some kind of insurance for small depositors. Then, if a bank becomes insolvent, the government could be on the hook to reimburse those depositors. From this alone, there will be a government cost for bank failure, and a government interest in capital regulation.

Even without deposit insurance, however, governments may be concerned about externalities of bank failure, and with spillovers to other financial institutions. If a McDonald's franchise fails, this might be good for the Burger King down the street. But if a bank fails, there may be fire sales of those banking assets during liquidation, fears that the bank down the street might also be or become insolvent, and a general loss of trust (a public good) in the financial and monetary system. With these externalities, we can see why governments might have an interest in bank solvency (and thus bank capital), even above and beyond the private incentives of individual banks.

QUESTION 3: WHAT ARE THE COSTS OF HAVING BANK CAPITAL REQUIREMENTS? If capital makes distress less likely, and if distress is (socially) costly, then there are social benefits to capital. But what about the costs? There are many answers to the cost question floating around in industry, politics, and the press. Most are wrong. And the confusion about costs tends to cloud the policy debate.

These fallacious arguments have several flavors. A detailed debunking is done by Anat Admati and Martin Hellwig (2013), and readers looking for a thorough treatment are pointed to them. But the core fallacy is worth a short discussion. This core fallacy is that bank capital is costly because equity requires a higher return than does debt. Then, if banks need to fund themselves with more expensive equity, they will pass these costs along to their customers, with a resulting hit to investment and economic growth.

The starting point for this fallacy is the simple observation that average equity returns are higher than average debt returns. And indeed, this empirical fact is also supported by financial theory. Proponents of this extra "cost" of equity can then do simple calculations comparing these relative costs and calibrating the relatively large effects of higher bank capital requirements.

However, this argument is flawed, and the flaw was pointed out almost 60 years ago in perhaps the most famous paper in all of corporate finance. Franco Modigliani and Merton Miller (1958) demonstrated that, in an economy without frictions, the value of a firm should be independent of the form of its financing. The modern interpretation of Modigliani-Miller is that though it is true that, ceteris paribus, equity is more costly then debt, it is also true that as a firm gets more leveraged (thus having relatively more debt than equity), the required return on both equity and debt will rise, so that the weighted average cost of capital will stay the same. The mathematics of this equivalence is straightforward, and the intuition is even more so: The overall value of the cash flows to the assets on the left side of the balance sheet should not depend at all on how those cash flows are divided between different claimholders (debt and equity) on the right side of the balance sheet.'

Of course, we do not live in a frictionless world, and frictions can spoil the perfect Modigliani-Miller equivalence. Some of these frictions are purely at the level of the private firm, such as the tax deductibility of debt, which would matter for all firms. But the impact of these frictions on most firms would be relatively small. Returning again to our fast food example, the parent company McDonald's Corporation had a capital ratio (equity to assets, in this example) of about 20 percent at the end of 2015 (McDonald's Corporation 2016). Suppose the government came in with a new law requiring all restaurant companies to hold at least 25 percent equity. How much would this cost McDonald's? Under perfect ModiglianiMiller conditions, it would cost nothing. With the friction from the tax advantage of debt, we can estimate a cost of between 10 and 15 basis points per year. (2) This is not zero, but compared with things like corporate tax rates, it is a drop in the bucket. It should not make the CEO of McDonald's claim that the burger business will come to a screeching halt.

But CEOs of big banks can get much more exercised. (3) Why is this? These are highly capable, profit-maximizing bankers. It seems unlikely that they would care so much about an issue unless they truly believed it had an impact on their bottom line. And even if we think the arguments made to support this position are sometimes fallacious, this does not mean there is no "real" reason lurking somewhere. And indeed, there is one--and it goes to the heart of what it means to "be a bank." This brings us to the last question.

QUESTION 4: IF COSTS ARE LOW, WHY DO BANKS CARE SO MUCH? AND WHY SHOULD THE GOVERNMENT CARE AT ALL? The Modigliani-Miller benchmark begins with the thought experiment of an all-equity firm, moving to then consider changes as debt is added to the capital structure. But the notion of an all-equity bank is missing a central purpose of banking: the liquidity transformation that leads to money creation. Bank deposits are the bank's debt, which is backed by long-term assets, and which can be used as "money" (checking accounts) by depositors. An all-equity bank is an oxymoron, because this bank could not accept deposits. So we cannot really start with the Modigliani-Miller benchmark.

From this point on, we quickly get to a real social cost friction for Modigliani-Miller: The use of bank deposits as money implies a transaction ("convenience") value for deposits. Because depositors get some convenience value from this transaction's use, they are willing to accept a lower return on deposits than they would on an equivalent debt instrument without such convenience. The resulting convenience yield makes debt cheaper for banks than for nonbank firms, and it forms the basis for the business model of banking: Banks manufacture liquid claims from illiquid assets, and compete with each other (in part) by doing this manufacturing in an efficient way. Models of banking based on the convenience yield are the driving force of the models of costly bank capital used by Skander Van den Heuvel (2008) and Harry DeAngelo and Rene Stulz (2015).

Seen from this perspective, even relatively small cost differences can translate into large competitive changes--the kind of competitive changes that drive the industrial organization of banking and the migration of banking activities into the shadow sector. Two key papers that make these points are those by Anil Kashyap, Jeremy Stein, and Samuel Hanson (2010) and by Adi Sunderam (2015). So, clearly, I am not saying anything here that Greenwood and his colleagues do not already know. Put another way, if the business of banking is to produce money-like debt instruments, then requiring more capital (equity), and thus less debt, is like direct regulation on the product mix of banks. Thus, the right analogy is not to McDonald's capital structure, but instead if regulation required that McDonald's get some minimum fraction of its sales from Chicken McNuggets. This kind of rule might indeed have major competitive implications.

To summarize, both banks and governments have a legitimate interest in both the benefits and costs of capital rules--but for somewhat different reasons. This tension drives my four main observations about the paper, which follow.

OBSERVATION 1: A SINGLE RULE MAY NOT BE OPTIMAL WHEN RISK WEIGHTS ARE UNCERTAIN Greenwood and his colleagues provide a benchmark model where we currently have none. That is surely the right first step. But in this benchmark model, the "correct" risk weights are known. Would their first result--that a single rule is optimal--still hold if the correct risk weights were uncertain? In subsection II.C of their paper, the authors argue that this kind of uncertainty would still imply a single rule: "In summary, a Knightian form of model uncertainty suggests the use of a single riskbased capital requirement that averages across plausible risk models." This conclusion does not seem obvious to me. Instead, I believe that a complete answer to this question would require specification of a few key parameters embedded in a more complex model. I think we would need to know the expected efficiency costs of multiple constraints, the potential cost of having the "wrong" rule, and the flexibility to change rules in the future as we learn the true model or as the world changes.

As an illustration, consider the simple problem of a pedestrian crossing a busy street. We start by assuming no jaywalking--all crossing will be at a corner, where there is a traffic light timed perfectly and unfailingly to a "Walk / Don't Walk" signal. Our task is to set a rule for pedestrians. One possible rule (no. 1) is to focus exclusively on the Walk / Don't Walk signal, and act accordingly. Another possible rule (no. 2) is to ignore the sign and the traffic lights and simply look both ways before crossing, only crossing when there are no cars, or when cars are at a full stop at the intersection. Rule no. 3 is to observe both rules no. 1 and no. 2. If drivers are infallible, then rule no. 1 would be strictly preferred to rule no. 3--why waste time looking both ways and waiting for cars to stop? But many of us (I think) would insist that our children follow rule no. 3; the efficiency gains of rule no. 1 seem small compared with the catastrophic losses if we have the wrong "model" of driver behavior. (4) This example gets closer to bank capital if we imagine that driver behavior might change over time, and also that rules are sticky, so that once we set a rule for pedestrians it would take congressional legislation to change it.

Although Greenwood and his colleagues make a general point about a single constraint being better than multiple constraints, their specific objection is to the leverage requirement, because it seems far-fetched to imagine a world where all risk weights would be the same. But including a (seemingly) extreme rule can sometimes be optimal in an uncertain world. There is a large engineering literature on this topic of "robust control," with a translation into economics given by Lars Hansen and Thomas Sargent (2001). It is the reason we build redundancy into things where failure can be catastrophic, and it can be a rational strategy for dealing with radical uncertainty.

Note that robust control is not an argument for the optimality of the current rules. Greenwood and his colleagues are surely correct that the current system is suboptimal, and it has been constructed piecemeal over time by interested and influenced agents, and not all at once by a disinterested control engineer. Nevertheless, it is still plausible that the current clunky system--imperfect as it may be--is superior to a clean system that is only optimal under certainty. I think we need more flexible models to answer this question.

OBSERVATION 2: REGULATOR DISCRETION IS A TWO-EDGED SWORD Greenwood and his colleagues' second principle is that ex post regulatory discretion is better than additional ex ante rules. In this case, I believe the principle might not hold if we had a richer objective function for regulators and the social planner. As a benchmark, it makes sense to assume that regulators are trying to maximize social welfare. But my casual observation suggests that it might be better to treat regulators as agents, with principals using legislation (and rules) to constrain them. This is not an indictment of just one side of the political spectrum; I think each side would want to constrain the other's regulatory discretion. And even someone without any particular political preference might want to have stability in the regulatory system, so that banks know how to plan for the future. Stability, and also adherence to legislative intent, are both weakened if we allow too much regulatory discretion.

One can also make the same argument using the language of robust control, as used above for observation 1. If regulatory discretion is considered a moving part in the overall financial system, we might also want to make rules that are sticky, because allowing rapid changes to rules, in a world where correct risk weights will always be uncertain, is itself an additional source of risk.

OBSERVATION 3: DYNAMIC RESILIENCE IS AN IMPORTANT POINT, BUT IT ALSO SHOWS WHY RULES MIGHT BE BETTER THAN DISCRETION Greenwood and his colleagues' third principle concerns "dynamic resilience." The main idea here is that, after a shock, banks should be required to maintain capital ratios by increasing the denominator, not by decreasing the numerator. It differs from "static" capital ratio rules, because after a shock we do not want banks to shrink their balance sheets and worsen the prevailing credit conditions.

The point about dynamic resilience is important, and was a crucial reason for the success of the stress tests in the United States in 2009.1 do not have any objection to this principle. Instead, I observe that dynamic resilience is likely to work best as a rule, outside the easy discretion of the regulator. After a shock, banks will not want to raise equity, and regulators will be under pressure to grant exceptions. Not all regulators will be in a position to resist this pressure--and it may be optimal to give them no choice, so that regulators can honestly say that their hands are tied.

If this observation is correct--and it would take a model to make it more precise--then I would argue that this is a lesson for all kinds of bank regulation, and that it reinforces my previous two observations. Overall, optimal regulation might need more rules and less discretion.

OBSERVATION 4: THE PAPER ABSTRACTS AWAY FROM THE "MIGRATION PROBLEM" TO SHADOW BANKING, AND I FEAR THIS MAY LEAD ITS CONCLUSIONS TO BE MISINTERPRETED Greenwood and his colleagues explicitly restrict the scope of their paper to not include any debate about the optimal level of bank capital. Instead, they focus on modeling the optimal policy design to reach any specific steady-state level. Furthermore, the analysis is general enough that one could apply it to any kind of financial institution, including shadow banks (where legal authority exists to regulate them).

These restrictions are important, because they enable the authors to abstract away from the "migration problem," where even small costs of capital requirements might lead to a movement of banking activity outside the regulated part of the system.

The authors are well aware of the migration problem, as they have (in various subsets) written many of the most important papers about it. (5) But this still cannot let us off the hook here. This is--I think--the single most important problem we need to deal with in the bank capital literature. And if we make policy recommendations based on models that ignore migration, it is important to put big red warning disclaimers on these recommendations.

To illustrate this point by example, I return to Greenwood and his colleagues' useful analogy about marginal tax rates. In their footnote 5, they point out that it would be inefficient if different firms faced different marginal tax rates for the same activity. The analogy here is to having multiple types of capital requirements, so that different banks can face different constraints on the margin. And the analogy is apt.

But it is even worse if marginal tax rates differ by an arbitrary choice of corporate form or industry classification. Certainly, we would not want to tax McDonald's differently from Joe's Corner Burger Joint, simply because (hypothetically) McDonald's was labeled as "fast food, diversified" and Joe's as "fast food, burgers." But if capital requirements hold for "banks" and not for "shadow banks," then our multiple marginal rates problem could quickly become severe. (6)

CONCLUSION Greenwood, Hanson, Stein, and Sunderam have written a clear paper on an important topic. Their benchmark model boils the bank capital problem down to its essence, and enables them to propose three principles for the design of bank capital policy.

This benchmark model is an excellent and sorely needed step. But I argue that it is premature to draw policy principles from it. My argument takes the form of observations about the limitations of the analysis, rather than any specific criticisms of the model. These limitations boil down to two main things. First, in a world with uncertainty about risk weights, regulator incentives, and policy flexibility, our rules for bank capital need to be robust to a wide variety of possible realities--and this robustness can lead (perhaps) to multiple and overlapping rules, and to skepticism about regulator discretion. Second, we should not debate bank capital rules without everywhere and always considering the effect of these rules on the migration of banking activities to the shadow banking sector.

REFERENCES FOR THE METRICK COMMENT

Admati, Anat, and Martin Hellwig. 2013. The Bankers' New Clothes: What's Wrong with Banking and What to Do About It. Princeton University Press.

DeAngelo, Harry, and Rene M. Stulz. 2015. "Liquid-Claim Production. Risk Management, and Bank Capital Structure: Why High Leverage Is Optimal for Banks." Journal of Financial Economics 116, no. 2: 219-36.

Dimon, Jamie. 2017. "Dear Fellow Shareholders." Letter, JPMorgan Chase, April 4. https://www.jpmorganchase.com/co[phi]orate/investor-relations/document/ar2016-ceolettershareholders.pdf

Hansen, Lars Peter, and Thomas J. Sargent. 2001. "Robust Control and Model Uncertainty." American Economic Review 91, no. 2: 60-66.

Kashyap. Anil K., Jeremy C. Stein, and Samuel Hanson. 2010. "An Analysis of the Impact of 'Substantially Heightened' Capital Requirements on Large Financial Institutions." Working paper. http://www.hbs.edu/faculty/Pages/item.aspx?num=41199

McDonald's Corporation. 2016. "Annual Report for the Fiscal Year Ended December 31, 2015." SEC Form 10-K. http://dllge852tjjqow.cloudfront.net/CIK-0000063908/1b39131b-56e9-4467-8256-5d7b5cefa385.pdf

Modigliani, Franco, and Merton H. Miller. 1958. "The Cost of Capital, Corporation Finance and the Theory of Investment." American Economic Review 48, no. 3: 261-97.

Sunderam, Adi. 2015. "Money Creation and the Shadow Banking System." Review of Financial Studies 28, no. 4: 939-77.

Van den Heuval, Skander J. 2008. "The Welfare Cost of Bank Capital Requirements." Journal of Monetary Economics 55, no. 2: 298-320.

(1.) The fallacy is sometimes stated from a different direction, with a claim that investors "demand" a certain return on equity, and a decrease in leverage will make that level harder to obtain. But lower leverage should decrease investor demands, as the risk to equity would then be lower as well.

(2.) See Kashyap, Stein, and Hanson (2010) for an exposition of this type of calculation.

(3.) See, for example, Jamie Dimon's (2017) letter to JPMorgan Chase shareholders.

(4.) The problem with following only rule no. 2 all by itself is that you do not know exactly when the light is going to change!

(5.) In their abstract, Kashyap. Stein, and Hanson (2010) write that "due to the unique nature of competition in financial services, even these modest effects raise significant concerns about migration of credit-creation activity to the shadow-banking sector, and the potential for increased fragility of the overall financial system that this might bring."

(6.) But of course we do sometimes have different rates for different corporate forms, because some corporate forms can be defined as pass-through tax entities. And, I would agree, this is indeed a problem when those different corporate forms can be direct competitors in a specific sector.

Although this paper by Robin Greenwood, Samuel Hanson, Jeremy Stein, and Adi Sunderam ranges broadly over the capital regulatory regime, its focus is understandably on the model that the authors have developed. Most interesting were the steady-state version's policy implications, arising from what the authors identify as the problem of multiple tax regimes. Specifically, the model suggests that the current regime's requirement that banks maintain minimum capital ratios based on different risk metrics leads to economic inefficiency in the allocation of credit--not just the administrative inconvenience and costs that have often been mentioned in the past. Although I also found the dynamic version of the model interesting and important, its policy implication that banks should be required to raise additional equity in bad states of the world is consistent with arguments that have been made in the past.

In this comment, I concentrate on what the authors describe as the logic of the steady-state model: that the leverage ratio requirement in U.S. law be eliminated. This is a particularly interesting policy issue, because it raises the more general question of how to incorporate the insights that come from elegant economic modeling into functioning regulatory regimes, which often have significant institutional constraints and involve noneconomic objectives, such as accountability. This question can involve choices from among blunter or more sophisticated rules or, as with the current capital regime, a combination of the two.

Before addressing this question, I want to make one point about the assumption in the model that banks have inherently different productivity functions, which the authors refer to as the "natural competitive advantage" of each bank. This assumption plays a key role in the conclusion that a regime with multiple capital rules--such as one with both a leverage ratio and a risk-based requirement--leads to distortions in lending behavior. Although the model is convincing in explaining that multiple rules can lead to short-term inefficiencies in lending decisions (though the relative size of these inefficiencies is not so clear), I was less convinced that there is something inherent or natural in each bank's productivity function.

Today's banks are the result of many past decisions that were themselves informed by strategic, business, and regulatory considerations. The transition costs to a higher productivity function for a given activity may or may not be substantial. Additionally, if taken another step, the implication of this part of the model might be that having the most systemically important institutions be less diversified--and maybe be even closer to looking like monoline operations--would be desirable. This seems to me a much thornier issue, pitting the potential advantages of minimizing asset correlation and avoiding the serious mistakes that can come when entering new businesses, on one side, against the potential disadvantages of less diversification of earnings and risk within each bank and the diminution in competition among the largest banks, on the other side.

In the remainder of my comment, I put this question aside, stipulate the utility of the model, and turn to the implications of its steady-state version. The logic of this model argues for a policy that rounds out the menu of approaches to capital regulation. For the last 30 years, the United States has had both leverage ratios and risk-based capital requirements. For the last 7 years, there have been multiple forms of risk-weighted requirements applicable to the largest banks--point-in-time requirements based on standardized risk weights; requirements based on a supervisory stress test; and, at least for the time being, bank-by-bank point-in-time requirements based on the internal models version of Basel II. (1) Since the global financial crisis, there have been numerous proposals to substantially increase the leverage ratio, either effectively or formally making it the only binding capital requirement. The model developed by Greenwood and his colleagues argues instead for eliminating the leverage ratio and establishing a single, risk-weighted minimum capital requirement, as determined by a stress test.

Note that I say the "model" would eliminate the leverage ratio. The authors themselves suggest only that it be "dialed back." This recommendation takes account of numerous practical considerations. However, despite their explicit proposal only to dial back, they seem to remain ambivalent, because a good part of their analysis is directed toward suggesting alternatives to the leverage ratio for combating arbitrage. The divergence between the policy implied by the model (embellished with some of the paper's analysis) and the policy actually proposed provides a useful opportunity to explore the above-mentioned issue of how to incorporate the learning that comes from economic models into operational regulatory systems. As such, this case study may have heuristic value for thinking about other financial regulations and, perhaps, even for other regulatory areas.

There are two kinds of reasons why it may not be optimal to fully incorporate a model into regulation. First, in practice, institutional considerations may impede the model's full implementation and its policy implications. Second, the model--though internally powerful--may be missing some salient economic factor. In truth, it may sometimes be hard to distinguish between the two types of reasons. For example, is the institutional inability of an agency to implement a model organizationally remediable, or is the model flawed in assuming that there could be any reasonable capacity to acquire certain knowledge or make certain distinctions? Asking if the fault lies in the model or the administrative agency may not always be the most productive route to determining good policy. Accordingly, here I parse the issue slightly differently: Will the balance of benefits and costs favor, on one hand, making the basic regulation conform to the model and trying to accommodate institutional limitations with various second-best devices? Or, on the other hand, will they favor taking a blunter or hybrid approach, but using the insights from the model in shaping the regulatory regime to make it more efficient?

One virtue of including a relatively simple rule, such as the leverage ratio in a capital regime, relates to the agency of the Federal Reserve Board as recipient of delegated congressional authority. A simple rule can provide some comfort to Congress and the public that banking regulators do not intentionally or inadvertently endanger financial stability by becoming too lax in their implementation of granular, highly risk-sensitive requirements, which by their nature will be at least somewhat opaque.

Monitoring the enforcement of a leverage ratio is not a strictly mechanical exercise, because judgments do need to be made about the treatment of off-balance-sheet and certain other assets in the denominator. But it is a lot easier than monitoring the internal ratings-based approach to capital requirements or the supervisory stress test. This is not a dispositive point in the present context, but it is a relevant factor in considering the costs and benefits of the two regulatory approaches outlined above.

Moving to the substance of the regime implied by the model, the authors note that it assumes precise and proper ex ante risk weighting of bank balance sheets, an assumption that is concededly unrealistic. They go on to note the consequent opportunities for regulatory arbitrage, owing to what we might call the second-mover advantage of the banks, which can exploit any identified inaccuracies or lacunae in ex ante risk weights. Rather than a blunt instrument like the leverage ratio, however, the authors prefer to give the regulators another move in the game, by using the stress test to impose a higher loss function on those assets with which "banks are loading up to an unexpected degree." The approach they suggest is one that would apply the core policy insight coming from the model--that there should be only one form of capital requirement--and then try to make institutional adjustments to deal with the practical limitations of realizing some of the model's assumptions.

The Federal Reserve already does a bit of this. When the macroeconomic scenarios are considered internally each year, the staff will offer to the board members various optional add-ons--that is, certain assumptions of higher losses that do not follow ineluctably from the chosen macro-economic scenarios. Sometimes these potential add-ons are motivated by rapid increases in the growth of an asset class, usually where valuations near or above the historical range are also observed. But this process, which in my experience occasions considerable discussion of each possible add-on, is nothing like what the authors contemplate--that is, a systematic variation in loss functions based on "supervisors' observations of granular changes in the composition of bank portfolios."

I have my doubts as to how feasible such a shift would be. Clearly, this would not be something for the governors to decide on an asset-by-asset basis. To date--and quite appropriately, I think--the Board of Governors has not involved itself in deciding the details of the supervisory model. But the proposal could well require either many asset-by-asset decisions or a basic change in the supervisory model, so that all loss functions would be made more sensitive to yearly shifts in the banks' collective holdings of each asset. The former looks a lot like again putting the regulators in a reactive position, chasing bank behavioral shifts in the regulatory backand-forth, and doing so in a particularly inefficient way (that is, through a series of debates by the Board of Governors over the meaning of increases in bank holdings of certain assets). The latter would require a new set of judgments, which would need to be quantified, on how much of an increase in an asset class would be considered arbitrage, and what the resulting increases in the loss function should be.

As the authors suggest, it might be worthwhile to have Federal Reserve staff undertake an internal exercise to see how this might be done (though with quantitative expertise at a premium, my priority would be an exercise to inco[phi]orate second-order loss effects into the stress test). But it is certainly not an approach ready for a rollout anytime soon. Even if this enhancement of the stress test eventually proved feasible, I wonder whether it might not be more of a complement to, rather than a substitute for, the leverage ratio. Here is why.

The leverage ratio is directed not only, and perhaps not even principally, at arbitrage in the conventional sense of conscious exploitation of the blunt edges of any risk-weighted system. The notion of the leverage ratio as a complementary capital requirement is that it is a countercyclical instrument that should capture--and, in theory, constrain--the presence on the balance sheet of all assets whose risk seems low in normal times. It is directed at more than the observed current inappropriateness of risk weights based on changes in the acquisition of certain kinds of assets. It is meant to guard against the ignorance of both banks and regulators as to which seemingly safe assets could become otherwise essentially overnight and thus occasion runs or other financial disruptions. Additionally, even a very ingenious method of adjusting the stress test loss functions might fail to capture the systemic effects from rapid, substantial deleveraging in periods of stress. (2)

In short, the bluntness of the leverage ratio can be construed as an admission of almost Knightian uncertainty about the financial system over time. It is, in this sense, directed at bad states of the world. So I am not surprised that it fits uncomfortably into the steady-state version of the authors' model. The authors readily admit that no single model captures all risks. But they suggest--as a better safeguard against Knightian uncertainty--a single, risk-based capital requirement that averages across plausible risk models. Later they suggest an alternative backstop capital rule that would push up risk weights for asset categories that are given low risk weights under the primary regime, but would stop short of leveling all risk weighting as under a leverage ratio.

The former suggestion, in assuming that regulators could identify all plausible risk models, seems a bit at odds with the observation of a state of (near) Knightian uncertainty. It is precisely the unexpected and unanticipated that should worry us. One might be skeptical that even the best, most conscientious regulators can identify all plausible risk models, quite apart from the challenges in trying to implement such a regime.

The latter alternative of a backstop that increases low risk weights is more responsive to the uncertainty associated with tail risks in the financial system. Of course, the alternative is inconsistent with a literal implementation of the model developed by the authors in the paper. Again, however, the question is whether, as a practical matter, it is substantially superior to a leverage ratio and thus worth the time and effort of regulators to fashion such a substitute. Without a more specified proposal, it is of course hard to know. But my intuition is that some modifications to the leverage ratio would probably yield most of the benefits of a different backstop rule, while requiring considerably less regulatory effort.

As I have explained elsewhere (Tarullo 2017), the supplementary leverage ratio for the eight global, systemically important banks (G-SIBs) could be made proportional to the risk-weighted G-SIB surcharge for each bank, rather than imposing a 2 percentage point higher leverage ratio requirement on all eight banks, as under the current regulation. Furthermore, the idea developed at the Fed last year to integrate point-in-time capital requirements with the stress-testing regime would add the G-SIB surcharge to the poststress minimum requirement. This change would decrease the likelihood that the leverage ratio will be binding in normal times for banks pursuing sensible and useful business strategies. (1)

In essence, my favored policy would take the insight produced by the authors' model as an additional argument for modifying the leverage ratio itself. This stands in contrast to the logic of the model, as complemented by some of the authors' analysis and suggestions, that the leverage ratio be jettisoned and that other institutional mechanisms be used to compensate for what the model may lack. My policy inclination is in turn based on (i) my suspicion that the costs identified in the authors' model for normal-time lending would be fairly modest with the changes I contemplate, and (ii) my belief that the alternative regime implied by the model would be vulnerable to the predictive and practical limitations of regulators.

Before closing, let me briefly address the authors' other policy recommendations. I can state very quickly that I agree with most of them, such as consolidating the number of capital constraints, eliminating the use of the internal models approach to capital regulation, and integrating the current point-in-time capital requirements with the current stress-testing regime. I do have one observation on their proposals for more transparency--with which I fully agree.

To date, complaints about opacity in the supervisory stress test have come from some banks, which would like to be able to arbitrage the regime even before capital requirements are set. But the public has an interest in not having the Fed dilute stress-testing requirements by, for example, quietly weakening the loss functions applied to one or more significant asset classes or adopting more optimistic assumptions about bank revenue during stress periods. This interest would be furthered by the authors' proposal that the Fed explain post hoc material changes in each year's supervisory test, as well as how the assumptions in its supervisory model stack up against market information, such as stock prices and credit default swap spreads. I should add that the public would be well served if outside experts (like these authors) were to regularly evaluate the rigor of, and changes to, the stress test scenarios and, to the degree possible, loss functions and revenue assumptions.

REFERENCES FOR THE TARULLO COMMENT

Bank of England. 2014. "The Financial Policy Committee's Review of the Leverage Ratio." Consultation Paper, July. London, http://www.bankofengland.co.uk/financialstability/Documents/fpc/fs_cp.pdf

Tarullo, Daniel K. 2017. "Departing Thoughts." Speech given at Princeton University, April 4.

(1.) The intention of the negotiators of Basel II was to apply only the internal models-based requirement to the largest banks, but the financial crisis altered those plans. The inclusion of the so-called Collins Amendment in the Dodd-Frank Act required that all banks meet the standardized, risk-weighted requirement, regardless of whether the internal models alternative was being imposed on them. In fact, the diversion of bank and supervisory resources by the financial crisis and its aftermath slowed down the process of banks' internal models being validated by supervisors and, as noted below, the value of retaining this separate requirement is questionable. The Federal Reserve's creation of Comprehensive Capital Analysis and Review, based on its supervisory stress test, added the third requirement, which is based on banks maintaining the minimum standardized, risk-weighted ratios, assuming the losses under stress projected by the Fed's supervisory model.

(2.) For a discussion of this and related points, see Bank of England (2014, pp. 14-16).

(3.) The authors endorse the integration of point-in-time requirements with the stresstesting regime. This step would also be consistent with the authors' recommendation to consider increasing the G-SIB surcharges, in line with their view that current requirements for the largest banks are toward the low end of a reasonable range. It should be noted that integrating point-in-time requirements with the stress test would, for the first time, incorporate the current surcharges as a part of the poststress capital expectations and thus increase requirements de facto. In the approach designed last year by the Federal Reserve, the impact of this increase would be partially offset by relaxing assumptions on balance sheet growth and paring back the scope of the requirement for "prefunding" all potential capital distributions during the capital-planning horizon. The recent recommendations by the Treasury Department on regulatory change appear to propose that the integration be done without any increase in capital requirements for the G-SIBs.

GENERAL DISCUSSION Phillip Swagel saw the paper as insightful in its explanation of why banks with different business models look to converge, and appreciated its raising of the question of whether diversity in bank business models is good for financial stability or growth. He worried, however, about the conclusion drawn by the authors to focus on ex post policy--that is, on cleaning up the mess. This ex post approach has the familiar problem inherent in discretion: Political pressures make it difficult to take necessary actions. The Federal Reserve faces this challenge with its focus on total loss-absorbing capacity (TLAC) as a way to ensure that banks have an adequate capital buffer. It could be difficult to impose losses if the holders of TLAC securities are politically powerful or sympathetic. Europe's experiment with contingent convertible bonds that can be converted to equity is evidence of this challenge; so far, they have not worked as intended.

Swagel agreed with the authors that the stress tests had been valuable, but thought the paper missed the point of why the initial test in 2009 was so successful in a way that made the episode less instructive for their purpose. The situation in 2009 was that capital raising by banks stopped as policymakers discussed in public whether large institutions should be nationalized. The first stress test results in May 2009 provided an "all clear" signal that it was safe to invest in the industry again. That is, the stress tests offset a policy error. In the other direction, the ex post criticism that policymakers during the crisis should have required firms to stop paying dividends to shareholders is incomplete. (The Troubled Asset Relief Program did not freeze dividend payouts, but only prohibited increases in dividends.) Had dividends been frozen, it could actually have made it harder for firms to raise new capital--dividends were a key component of crucial capital raises, such as Warren Buffett's September 2008 investment in Goldman Sachs. The broad point is that ex post regulation poses its own difficulties.

As noted in the paper and emphasized by Daniel Tarullo in his comment, higher capital requirements would make other regulations, such as the supplementary leverage ratio, less important. Swagel thought it would be useful for the Federal Reserve to make clear that all owners of TLAC securities will face losses in a crisis; saying this publicly and repeatedly ahead of the next financial crisis would make it more likely to happen when needed. Similarly, Swagel thought the Federal Deposit Insurance Corporation should continue to develop its framework for the use of the orderly liquidation authority in Title II of the Dodd-Frank Act, which also allows for losses to be imposed on investors in the process of the unwinding of a large financial firm.

Donald Kohn discussed his main takeaways from the paper, notably that regulations could be eased in certain circumstances, and that some streamlining of regulation seemed efficient. He also took away from the paper that the leverage ratio requirements on banks should be pulled back somewhat, and instead regulators should improve the application of risk-based capital requirements. If regulators can estimate the correct risk weights, this seemed to be the right approach. Notably, the leverage ratio avoids the procyclicality that can result from using banks' risk models to determine capital, but combining better risk weights with well-designed stress tests could achieve the same purpose without the need for the supplementary leverage ratio, thereby eliminating its distortionary effects. However, this framework would depend heavily on regulators calculating the correct risk weights, not the banks themselves. Kohn believes this would be a prerequisite for dialing back the leverage ratio, because it would prevent the banks from finding workarounds. Kohn noted his particular affinity for the ex post regulator adjustments to stress tests to check banks' propensities for modeling that lowers their capital requirements, adding that the United Kingdom does something similar. He expressed concern that recent discussions in the United Stated have been moving away from this framework, instead favoring less frequent stress tests and a greater reliance on banks' models of risk, rather than regulators'. He much preferred the idea broached in the paper that more responsibility should be placed on the regulators, and less on the banks.

Robert Hall inquired as to why capital requirements should be necessary at all. As the 2008 crisis showed, banks' stock values showed signs of stress, but their capital positions were still high, indicating that capital is not always the best measure. Hall instead argued that banks should be more resilient overall, and that the focus on capital and equity was misplaced. Resilience ultimately comes down to how flexible banks' sources of funding are, which can be addressed in ways other than forcing them to hold more capital. In mutual funds, for example, funding is contingent, meaning that they do not need to hold capital.

Natasha Sarin stated that in her reading of the paper, the main assumption made is that specialization in banking is a good thing. She asked whether it was obvious that such an assumption was true. For example, she noted that diversification tends to lead to a lower probability of bank failure. So though the paper certainly shows that convergence in banking has emerged as a result of regulation, Sarin was not convinced that this is a bad thing.

Ben Bernanke suggested that the optimal set of capital requirements could be nonlinear. One reason why appropriate capital requirements could be higher for some firms than others is that adding assets in a category in which the firm already has a concentrated position could indeed be riskier at the margin. Fixed risk weights, however, imply that required capital is linear, that is, capital per unit of assets is independent of the size of the existing position. That the marginal risk of adding to already large positions can be large--especially in a very adverse situation--is what stress tests are designed to take into account. Jon Steinsson agreed with Bernanke that perhaps the capital framework should be nonlinear. The discussion and the paper seemed to assume that linear risk-weighted capital requirements are the right approach, a point on which Steinsson was not convinced.

James Hines noted that if varying regulatory constraints are binding for different institutions, this would seem to imply an incentive for banks to merge. He wondered if this was indeed the case--and, if not, whether this implies that the costs of the regulatory constraints are not actually that high. Alan Auerbach asked whether the determination of capital itself, and the weights used to determine what capital actually is, may be a concern vis-a-vis whether new financial regulations are useful.

Alan Blinder disagreed with the notion that capital charges act mainly as a sort of tax on banks, a point implicit in Hall's remark and that had permeated the discussion. Blinder stated that taxes are imposed in normal times to raise revenues, whereas capital requirements are there to deal with nonnormal situations. Building on Swagel's comments, Blinder stated that the point of capital is to provide loss-absorbing capacity during an adverse scenario. For this purpose, the type of capital held by banks is very important. In the mid-1990s, which coincided with a long period without a crisis, the assets that supervisors counted as capital included some that were quite risky and low in quality. This is a "belt and suspenders" argument for the imposition of capital charges alongside other regulatory safeguards like stress tests. If capital served as a kind of tax, this would not be the type of framework regulators would use, he concluded.

Aaron Klein used chopsticks as a metaphor to help conceptualize the leverage ratio and stress tests: "With two chopsticks, you can eat elegantly. With one, you're just stabbing something." In this vein, he argued that both the leverage ratio and risk-based capital requirements are necessary, together. He noted that critics often are uncomfortable when one of the two regulations serves as a binding constraint on banks that limit their behavior--complaining that they have inappropriately universal risk weights, in the case of the leverage ratio; and that stress tests are imperfect hypothetical scenarios that regulators could get wrong, in the case of risk-based capital requirements. In reality, Klein noted that the point of regulation is to allow all the constraints to bind. The paper emphasizes a preference for stress tests, but this assumes that an all-seeing, all-knowing regulator can appropriately and accurately design the stress tests--a difficult task. Instead, he wondered if regulators could incorporate a consistent and symmetric error factor into stress tests to make them stronger and more powerful.

Jeremy Bulow clarified that the leverage ratio is essentially another set of risk-weighted capital requirements that simply and ignorantly sets all the weights on different asset classes equal to 1. Notably, if risk weights were double their current levels and overall capital requirements were half their current levels, there would be no real change in the risk-based capital held by banks. In this scenario, however, Bulow believes that fewer people would probably favor the leverage ratio as a good complement to risk-based capital.

Bulow also believes regulators should rely more on markets to determine capital requirements. As banks become more complicated, it becomes more important to separately determine the capital requirements for their different assets. One way to do this is to look at how much a bank pays to borrow on a nonrecourse basis, using a particular asset as collateral. This could help determine the risk weight on that asset; and regulators could also add some amount to the capital requirement, based on the overall strength of the institution holding it. Combining this type of information with the expertise of regulators would probably be more accurate and efficient than simply allowing regulators to set capital requirements totally independently.

Adi Sunderam was the first author to respond. He noted that, of course, neither the authors nor regulators know for sure what the proper risk weights for capital are. When there is uncertainty about these weights, the natural inclination is to set them equal to 1, which is what the leverage ratio does. However, Sunderam stated that the central idea of their paper was to incorporate this uncertainty about the risk weights into a single regulatory requirement, rather than two--namely, risk-based capital requirements and the leverage ratio. This would allow the ratios to apply more evenly to banks with different business models. Otherwise, one single rule, like the leverage ratio, would naturally apply in varying ways to different banks.

Sunderam added that it is important to think about the overall value of stress tests. During the crisis, stress tests were used to mark the assets on bank balance sheets to market prices. However, during normal times, they have a different purpose; they act to impose a separate set of de facto capital requirements on top of the leverage ratio and the risk-based capital requirements. The present paper argues that stress tests should be used not to impose additional capital requirements, but in an ex post manner to place regulatory emphasis on specific risks that are growing at banks.

Samuel Hanson affirmed the uncertainty about the risk weights of certain asset classes. He made reference to "Knightian uncertainty," a term popularized by Frank Knight in the 19th century that distinguishes between uncertainty and actual risk, even though the two are nearly impossible to distinguish in practice. (1) Such uncertainty in regulation tends to lead to a blending or averaging of risk weights across asset classes. The leverage ratio, however, applies the same risk weight to all asset classes, which Hanson believes does not seem to be a good way to solve Knightian uncertainty. Hanson agreed with Bernanke and Steinsson that risk-based capital perhaps inappropriately assumes that risk weights should be applied uniformly or linearly. He referenced the work of Michael Gordy that set this standard. (2) Hanson noted, however, that the data on bank balance sheets do not currently indicate sufficient variation in the concentration of positions across banks to make the linear assumption for risk weights a problem. Therefore, he thinks the criticism of linear risk weights is less of a concern.

Jeremy Stein conceptualized the present paper's idea by hypothesizing a new regime where regulators start from scratch. In such a regime, regulators could start by choosing how to design the ideal capital rule for a representative bank; this rule could be a linear combination of existing rules--such as the conventional risk-based capital requirement and the leverage ratio. He clarified that the paper argues simply that such a rule should apply in the same way across all banks, regardless of their business model.

Stein's reading of Tarullo's comment was that the effective relevance of the leverage ratio should be time varying, in that it would be more tightly binding during a credit boom, but not in normal times. Although Stein conceded that this point was a valid argument, under this premise, the leverage ratio acts not as an ex ante constraint on behavior but rather as an ex post flag that risky behavior has already been taking place--in the sense that when the leverage ratio binds, regulators know there is a problem. This once again puts the ball in the court of regulators.

On Bulow's comment that regulators should rely more on markets, Stein noted that the present paper explores this issue, highlighting stock prices of credit default swaps as potentially valuable sources of information. Stein was uncomfortable with the notion of writing an ex ante rule conditional on stock prices, but he did note that regulators doing stress tests should be required to explain ex post how the stress test scenario takes into account changes in financial market prices.

(1.) Frank H. Knight, Risk, Uncertainty and Profit (Boston: Houghton Mifflin, 1921).

(2.) Michael B. Gordy, "A Risk-Factor Model Foundation for Ratings-Based Bank Capital Rules," Journal of Financial Intermediation 12, no. 3 (2003): 199-232.

COMMENT BY

ANDREW METRiCK

COMMENT BY

DANIEL K. TARULLO