Heuristic and Seawall Studies

The prominent analysis ofAdmati and Hellwig (2013) calls for exceptionally high capital requirements, although the authors do not attempt to calcu­late an optimal capital ratio. They rely heavily on the Modigliani-Miller (M&M) theorem to argue that higher equity capital would not increase costs.

Admati and Hellwig give a potentially misleading impression of the time trend of the equity ratios when they state that equity was 25 percent of assets “early in the 20th century” but declined to 6 to 8 percent “by the early 1990s” (p. 31). Hoenig (2013, chart 5) shows that this decline had already occurred by 1945, and that equity ratios changed little between then and the early 1990s. In most areas of economic policy, norms that prevailed in the prewar period are considered less relevant than those of the postwar period. Calomiris (2013, 19) observes more specifically that the evidence from the early 20th century is misleading because after the 1930s asset risk declined substantially as a consequence of the very large increase in bank holdings of cash. Moreover, elimination of restrictions on inter­state banking in the United States in the early 1990s (see chapter 5) would seem to constitute an important reason why lower equity ratios would have been needed than in earlier periods, when domination by “political elites at the state level...

Hanson, Kashyap, and Stein (2010) make a brief estimate of the “seawall-type. They argue that “the regulatory minimum in good times must substantially exceed the market-imposed standard in bad times.” They observe that in the first quarter of 2010, the four largest US banks held 8 percent of risk-weighted assets, considerably above the 6 percent level that was the precrisis regulatory standard for “well-capitalized” banks. They note that the IMF (2010) estimated cumulative credit losses at US banks from 2007 to 2010 at 7 percent of assets. On this basis, they suggest, “one could argue for a good-times regulatory minimum ratio of equity-to- assets of 15 percent” (p. 8).

A problem with this back-of-the-envelope estimate is that it mixes risk-weighted asset apples with total asset oranges. In the first quarter of 2010, the four banks noted by the authors had total assets of $7.70 trillion versus risk-weighted assets of $4.68 trillion.^[20] So although common equity tier 1 was 8.2 percent of risk-weighted assets for the four banks combined, it represented just 5.0 percent of total assets.

Moreover, the 7 percent loss figure appears to be overstated. A more recent study of the capital needs of global systemically important bank (G-SIBs) by the Financial Stability Board (FSB 2015a) examines losses by large international banks in the Great Recession and the Japanese banking crisis of the 1990s. It finds that “total losses have been up to almost 5 percent of total assets (Merrill Lynch, Wachovia), with half of the banks in a 2-4 percent range” (p. 6). Even the upper end of the loss range might thus more appropriately be set at 5 percent of total assets. On this basis, a restate­ment of the Hanson-Kashyap-Stein approach could reasonably set a target of common equity of 10 percent of total assets rather than 15 percent.

But a more basic question is why fully 5 percent of capital should be left in a crisis if there is an effective lender of last resort.

Goldstein (2017) uses a heuristic seawall approach to arrive at a range of 10 to 18 percent of total assets as the optimal ratio for bank capital requirements, with smaller US banks at the low end of this range and the largest G-SIBs at the upper end. By implication, he seeks a range of 18 to 32 percent for capital relative to risk-weighted assets.^[21] For the G-SIBs, the target would thus be about three times as large as the equity capital target in Basel III. It would be nearly twice as large even if one were to attribute complete equity equivalence to the TLAC requirement of 18 percent of risk-weighted assets.

Goldstein identifies 600 basis points as the needed increase in the capital-to-assets ratio, on a weighted-average basis. He estimates that such an increase would boost bank lending rates by 15 basis points. He invokes a rule of thumb that macro models show that an increase in the federal funds rate of 100 basis points reduces output by 100 basis points two years later. Illustratively positing that an increase in bank lending rates trans­lates to only a third as large an impact as a comparable increase in the federal funds rate, he implies that a 15-basis-point increase in bank lending rates would reduce output by only 5 basis points.

In contrast, in the model in chapter 4 a 600 basis point increase in the ratio of capital to total assets would raise the bank lending rate by 25 basis points.^[22] But a more important difference is my use of an aggregate production function, in which the long-term output effect of an increase of 100 basis points in the unit cost of capital to the economy is far greater than the 100-basis-point reduction Goldstein cites from macro models.^[23] One reason the macro-model example may understate growth impacts is that the change in the discount rate tends to be considerably greater than the change in the long-term interest rate, but the latter is what matters for investment, so a 100 basis point rise in the relevant long-term rate would imply a considerably larger rise in the federal funds rate.

More fundamentally, Goldstein's two-year horizon is far too short to evaluate the cumulative long-term impact on the stock of capital (consid­ering that depreciation is on a time scale more like 15 years). Overall, I find that a 100-basis-point increase in the capital-to-assets ratio would reduce long-term output by 15 basis points, so Goldstein's 600-basis-point incre­ment would cut long-term production by 90 basis points, not his 5 basis points.^[24]

Goldstein's target for capital loosely follows the seawall approach of Hanson, Kashyap, and Stein, amplified in a judgment call taking many factors into account. Citing the International Monetary Fund (IMF 2010), he states that US banks lost 7 percent of assets in credit write-downs in the 2007-09 crisis and indicates that the eight US G-SIBs had a leverage ratio of 8 percent in 2012. He therefore posits that the target ratio for capital relative to total assets should be 15 percent (8 percent plus another 7 percent against the next Great Recession).

The 7 percent write-down cited for the United States by both Hanson- Kashyap-Stein and Goldstein constitutes an estimated $885 billion in write-downs for 2007-10, of which about $200 billion was in mortgages, $400 billion in other loans, and $300 billion in securities, of which about two-thirds were residential mortgage securities (IMF 2010, 12). But over this four-year period, the 54 large US banks examined in chapter 3 reported cumulative positive net income of $177 billion (calculated from the Securities and Exchange Commission's 10-K dataset, discussed in chapter 3). Together they show a loss in only one year—2008 (when there was an aggregate $10 billion loss, the largest loss being the $27 billion by Citigroup). This dataset excludes Lehman Brothers, but if one applies the midpoint of the $100 billion to $200 billion hole in Lehman's balance sheet at the time of its bankruptcy (estimated in Cline and Gagnon 2013), adding Lehman would still leave the large bank net income slightly posi­tive for 2007-10.

Use of gross write-downs as the metric for bank stress thus gives a misleading picture by failing to consider that the banks had other profits that more than offset specific write-downs over this four-year period. Correspondingly, use of the 7 percent figure to calibrate the needed height of the seawall is a major overstatement. Moreover, it is not clear why the appropriate target for capital left after an extreme disaster should be as high as 8 percent of assets. A more moderate but still significant net capital position of, say, 3 percent would seem to be a more reasonable immediate postcrisis target. The banks would still be solvent, because assets would exceed liabilities.

Goldstein argues that bank losses in the Great Recession are actually too small to serve as the proper size of the seawall, because the losses would have been much greater if governments had not taken extraordinary inter­ventions to strengthen economies. This argument seems to explain why he does not take account of other bank profits that more than offset the credit write-downs. But this view seems to throw out the widely accepted principle that providing liquidity to solvent financial institutions is precisely what the central bank is supposed to do in a crisis. The extraordinary interventions in the Great Recession, such as Federal Deposit Insurance Corporation (FDIC) guarantees of new bank debt and guarantees for money market funds, amounted to massive lender-of-last-resort liquidity to the system precisely when such support was the appropriate policy response. Arguing that the proper metric for counterfactual bank losses in the Great Recession should add what banks' incremental losses would have been in the absence of such measures discards proper lender-of-last-resort policy. Goldstein implicitly assumes that the United States has tied its hands politically and would be unable to take such lender-of-last-resort measures in the next Great Recession. But suppressing potential output over the next several decades because of the assumption that it would be politically impossible to take the right action in the next 100-year flood seems misguided.

A recent study by researchers at the International Monetary Fund (Dagher et al. 2016) also applies a seawall framework to determine desirable capital requirements for banks. The authors use the same banking crisis database used in chapter 4 of the present study but focus on the observed levels of nonperforming loans (NPLs) recorded in that database. They then calculate what percent of loan losses bank capital would have covered if capital had been at alternative levels relative to risk-weighted assets. They assume either a central estimate of 50 percent loss given default on NPLs or a conservative estimate of 75 percent. They show a sharply nonlinear curve: Initially, additional capital covers large portions of losses, but as capital is raised still higher, additional loss coverage turns modest (similar to the benefits curve in figure 4.2 of chapter 4). For advanced economies, covering 85 percent of bank losses in banking crisis episodes would have required broadly defined capital ratios of15 percent of risk-weighted assets in the case of 50 percent loss given default and 23 percent in the case of 75 percent loss given default. Stripping out intangible equity, subordinated debt, and cyclical peak additional capital, this range corresponds to 9 to 17 percent of risk-weighted assets (5 to 10 percent of total assets).^[25]

Dagher et al. note that the top rate of the Basel III capital requirements schedule (for G-SIBs) plus a countercyclical buffer reaches 15.5 percent broadly defined capital relative to risk-weighted assets.^[26] Considering further that banks tend to hold more than the legally required minimum capital, they interpret their findings as being consistent with Basel III requirements for G-SIBs, as well as with the Financial Stability Board recommendations for TLAC. After adjusting for the different capital concepts, their 15 to 23 percent range is relatively close to the 12 to 14 percent range for tangible common equity relative to risk-weighted assets identified in this study.

King (2010) estimates the impact of higher capital requirements on lending rates. Using data for nearly 7,000 banks from 13 countries in the Organization for Economic Cooperation and Development (OECD) for the period 1993-2007, he identifies key profiles relevant to the lending cost impact. He finds that loans represented about half of total assets, investments and securities 1∕6th, interbank claims 1∕8th, and trading assets 1∕10th. Deposits represented 44 percent of assets, trading-related liabilities 15 percent, wholesale funding 14 percent, and interbank funding 13 percent. Total shareholders' equity was 5.3 percent of assets, and risk- weighted assets represented 53 percent of total assets. The return on equity was 15.5 percent.

King estimates average spreads on loans for the period at 275 basis points for US banks and 256 basis points for euro area banks. He estimates that the rates at which banks borrow average 100 basis points above the deposit rate for short-term debt and 200 basis points for debt with matu­rity of more than one year.

He assumes that higher capital requirements would necessitate the replacement of term debt by higher-cost equity. For a representative bank, an increase of 15 basis points in the lending spread would be necessary to offset an increase in the (risk-weighted) capital ratio by 1 percentage point.^[27] The calculation assumes no M&M offset in the reduction of the unit cost of equity. This estimate is almost the same as the estimates in BCBS (2010a) and Slovik and Cournede (2011) but somewhat higher than the 10 basis point median estimate across 10 studies in a recent BCBS (2016a) survey. The study estimates the corresponding impact on lending spreads from Basel III liquidity requirements (the net stable funding ratio) at 12 basis points.

Although the study does not recommend a particular level of bank capi­talization, it provides relevant estimates on the cost side as well as a useful financial profile of OECD banks in the precrisis period. In this regard, it warrants remark that the study finds that on average, deposits were less than half of liabilities. There is much emphasis in the literature on distor­tions from subsidized funding through government deposit guarantees. However, with deposits less than half of liabilities, and with a considerable portion of deposits too large to be included under guarantee ceilings (e.g., corporate deposits), the influence of the guarantee subsidy would seem to be much more modest than the literature might imply.