Internalizing the Too Big to Fail (TBTF) Risk Subsidy

The premise that there are major subsidies and distortions to bank behavior caused by the TBTF dynamic appears to underpin much of the public policy discussion of TLAC. The underlying proposition is that if banks are forced to maintain much larger subordinate and CoCo debt, they will behave in a less risky fashion, and creditors rather than taxpayers will be the ones to pick up the bill in the event of bank failures.

There does not seem to be much literature examining whether CEOs of large banks do in fact think and act in an especially risky fashion. In prin­ciple the issue involves the need to pursue sectoral regulation to temper oth­erwise bad behavior of corporate actors (such as the disguising of emissions test results in the automobile sector). In practice, in the United States at least, it was the highly leveraged investment banks (Bear Stearns, Lehman), a hedge-fund-like London unit of an insurance company (AIG), and the government-sponsored mortgage agencies (Fannie Mae and Freddie Mac) that were the core causes of the Great Recession financial crisis, not the commercial banks. Indeed, some of the largest commercial banks played stabilizing roles by absorbing large failing financial firms (JPMorgan Chase absorbed Bear Sterns, Bank of America absorbed Merrill Lynch, Wells Fargo absorbed Wachovia; Cline 2010). Moreover, the public did not wind up losing any money on the support to the banking system from the Troubled Asset Relief Program (TARP) (Cline and Gagnon 2013). Nor would classic central banking behavior following Bagehot (1873) lead to TBTF distor­tions, because in a panic the central bank would provide support only to solvent institutions, large or small.^[156] These considerations suggest that a dose of caution, if not skepticism, may well be warranted in approaching the literature on the TBTF implications for TLAC.

Two researchers at the Bank of England provide a useful review of ap­proaches to estimating the TBTF subsidy, along with new estimates for the United Kingdom (Noss and Sowerbutts 2012). They identify two main approaches: “funding advantage” and “contingent claims.” Within the first approach are two subcategories. “Size-based” estimates examine the differential in borrowing costs of large banks and all other banks as the measure of implicit government subsidy from TBTF. “Ratings-based” esti­mates use the differential between the major ratings agencies “stand-alone” and “support” ratings on banks (the “ratings uplift” for TBTF banks). The ratings approach takes account of different business models as well as judg­ment about the probability of official support.

The second broad approach conceptualizes public support as a put option offered to the banking system on its assets. The volatility of banking stocks is used as a basis for projecting the future volatility of banking system assets. A resulting probability distribution of asset values then provides the basis for estimating the expected shortfall of the system's asset values below what is consistent with meeting minimum capital ratios and hence the ex­pected official sector support to the banking system. The “option price” subcategory uses current-year stock option prices for banks to estimate future asset volatility. The “historical” subcategory uses past share prices to estimate expected volatility.

The different approaches yield an extremely wide range of estimates of TBTF subsidies for the United Kingdom. The temporary steepening in the relationship of ratings to spreads boosted the estimated subsidy in the ratings uplift approach from £25 billion in 2008 to £120 (about 8 percent of GDP) in 2009, returning to £30 billion in 2010. The options-based ap­proach places the 2010 subsidy at £120 billion; the historical contingent claims approach places it at £20 billion. Even within the same subapproach (historical contingent claims), the estimate for 2008 soars to £350 billion under year-by-year calibration (reflecting extreme risk aversion in markets that year) but only £15 billion using longer-term through-the-cycle calibra­tion.

Nguyen (2013) examines whether bank risk taking is affected by sub­ordinated debt. As the (inverse) indicator of risk, he uses the natural loga­rithm of the “z-score”—that is, z = (ROA + CAR) /g_roa, where ROA is return on assets, CAR is the ratio of equity capital to assets, and σ_f,₀A is the stan­dard deviation of the return on assets. He cites other studies showing that the logarithm of this “distance from insolvency” ratio has a strictly mono­tonic negative relationship with the probability of insolvency. He applies data on 727 banks in 71 countries (with nearly half of the sample located in the United States) for 2002-08, resulting in about 4,400 firm-year ob­servations. He uses the lagged equity-to-assets ratio as the instrument for the subordinated debt ratio. In this first-stage regression there is always a negative sign, indicating that banks issue subordinated debt to raise tier 2 capital (i.e., to substitute for additional equity). In his main results, the (instrumented) ratio of subordinated debt to risk-weighted assets (or alter­natively to total liabilities) shows a strong positive and highly significant impact on the (log) z ratio. These results support the general proposition that subordinated debt has a disciplining effect.

Nguyen's other results pose a major problem for enthusiasts of TLAC for large banks, however. When a dummy variable is included for TBTF banks (defined alternatively as meeting a threshold of 10, 15, or 20 percent of the country's total deposits), there is a significant negative coefficient on the interaction of the dummy and the subordinated debt ratio that is larger than the coefficient on the simple subdebt ratio in the equation (p. 138). As a consequence, the disciplining influence of subordinated debt is completely absent for the TBTF banks.

Marques, Correa, and Sapriza (2013) examine whether the prospect of government support causes banks to take more risk. They frame the analysis as testing two opposing hypotheses. Under the “market discipline” hypothesis, for which they cite Merton (1974) and Flannery and Sorescu (1996), government support decreases the incentive of investors and de­positors to monitor or influence bank risk taking, thereby inducing greater risk. Under the “charter value” hypothesis, for which they cite Keeley (1990), government support reduces bank risk taking by increasing banks' charter values and hence reducing the gambling incentive. To measure prospective government support, the authors use the difference between the Moody's bank financial strength ratings, which exclude any external support a bank may receive from its parent or the government, and Moody's bank deposit ratings, which incorporate external support.

Marques, Correa, and Sapriza also use the z-score as the measure of risk, noting that its “distance to insolvency” property stems from the fact that when losses exceed equity, the numerator turns negative. The inverse of the z-score is the probability that losses will exceed equity capital. Using data on about 340 banks in 56 countries, they estimate the influence of (potential) government support, with a one-year lag, on the z-scores ofbanks in 2003-04 and 2009-10. Their results support the market discipline hypothesis rather than the charter value hypothesis.

An important result is that bank size per se does not influence riskiness, which Marques, Correa, and Sapriza interpret to mean that the TBTF influ­ence is offset by the lower volatility of returns on assets as a consequence of greater diversification. In tests examining the additional influence of regulatory stringency, they find that the term for interaction of government support with “activity restrictions” is positive and significant in the period of recent crises, although not with a large enough coefficient to offset fully the risk-inducing influence of government support. They interpret their overall findings to mean that policy measures that “increase the incentives by depositors, small shareholders, and subordinated creditors to monitor or influence banks' attitudes toward risks should decrease the moral hazard associated with government support to the financial system,” including limits on the amount of support governments can pledge (p. 19).

Afonso, Santos, and Traina (2014) examine the influence of potential government support on risk taking by banks using Fitch “support rating floors” (SRF). Begun in March 2007, this Fitch rating is unique in reflecting only potential sovereign support, not other external support (such as support from a parent holding company). The authors use data for more than 200 banks in 45 countries. They find a statistically significant posi­tive influence of the SRF on two alternative measures of (ex post realized) bank risk taking: impaired loans and net charge-offs. The significance stretches back through 11 quarters of lags (the full period of availability of the ratings), providing some reassurance that the causation is not running in the other direction (from current loss to current government support).^[157] They also find that when they interact the SRF with the issuer default rating as a measure of general riskiness of the bank, the coefficient is negative and significant, indicating that banks with lower general credit ratings have a larger impact on increased risk from greater potential government support.

Laeven, Ratnovski, and Tong (2014) examine the influence of bank size on risk taking as well as systemic risk. They emphasize that the world's largest banks experienced rapid asset expansion in the decade preceding the financial crisis, with the typical size of the top three US and top three European banks surging from about $700 billion in assets in 1999 to about $2 trillion to $3 trillion by 2008, with growing concentration of the banking sector. Their business models shifted away from loans toward market-based activities, in part in response to deregulation (the Gramm-Leach-Bliley Act in the United States, the Big Bang in the United Kingdom, and the creation of a single financial market in the European Union). Lower capitalization, more fragile funding, and increased organizational complexity accompa­nied this transformation.

The authors show that there is a significant negative correlation between bank size (the logarithm of assets) and the tier 1 capital ratio, the deposits-to-assets ratio, and the loans-to-assets ratio and a positive corre­lation with short-term funding. They contrast a benign interpretation of economies of scale and benefits of diversification with malign alternative explanations (TBTF subsidies, managerial empire building).

They then carry out tests using data for 370 banks with assets of $10 billion or more in 52 countries. Taking advantage of the natural experiment of the financial crisis, they use the change in bank stock prices from July 2007 to December 2008 as the indicator of bank riskiness. They regress this change on bank size, capitalization (the tier 1 capital ratio or leverage ratio), an index of funding fragility, activity (the share of loans in total assets or share of noninterest income), and complexity (measured by the number of subsidiaries). They find that larger banks are riskier than smaller ones, that lower capitalization in larger banks increases risk, and that unstable funding is associated with more bank risk. At the level of individual banks, however, neither more market-based activity nor organizational complexity increased risk. The average bank stock return was -45 percent, and the return for banks with at least $50 billion in assets was -55 percent. A five­fold increase in assets would have reduced the return another 7 percentage points, a 20 percentage point increase in reliance on funding from deposits would have increased returns by 12 percentage points, and an increase in capital ratios by 2.5 percentage points would have increased returns by 13 percentage points.

The study then examines systemic risk, measured by “SRISK.” This measure calculates a bank's contribution to the deterioration of capitaliza­tion of the financial system as a whole during a crisis (Acharya, Engle, and Richardson 2012). Their estimate is based on “the dollar value of capital shortfall that a bank is expected to suffer when the US financial stock index falls by 40 percent over six months” (p. 15). This measure implicitly assumes that stock market prices accurately reflect the long-term values of assets held by the banks rather than overshooting in a panic, and thus seems likely to be overstated in a crisis period. SRISK is determined by bank stock vola­tility in times of distress, the covariance of the bank's stock with the market in such periods, bank leverage, and bank size.

Their list of top 10 SRISK banks places the Royal Bank of Scotland at number 1 and Bank of America at number 10. Most of the institutions on the list received government support in the crisis. Their tests show a significant influence of bank size and complexity on SRISK. When bank size is interacted with structural variables, the tests show that large banks contribute more to systemic risk when they have less capital, have fewer de­posits, and engage more in market-based activities. The authors conclude that there should be more stringent capital requirements for large banks because of systemic risk, consistent with the G-SIB surcharges in Basel III. They also infer a case for CoCos and for restrictions on risky trading activi­ties but worry that sole reliance on higher capital requirements may push risks to unregulated sectors and may impose a cost from loss of economies of scale.

Siegert and Willison (2015) provide a useful survey of studies estimating TBTF subsidies. They opine that “the existence of the TBTF problem is now widely accepted by academics, politicians and regulators across the world... [namely the problem from] systemic risk and moral hazard” (p. 4). This characterization is probably accurate, but it would seem to dismiss lightly such considerations as the alternative economies of scale explanation of lower funding costs of large banks and (with regard to moral hazard) the inconsistency of undue risk-taking incentive with the fact that in an insol­vency shareholders are expected to be wiped out even if creditors are not. The authors do not mention the “charter value” argument at all. They do cite Davies and Tracey (2014), who note that once TBTF status is included as a control, tests no longer show economies of scale. They do not cite con­trary estimates (such as those discussed below). They review event studies (the reaction of bank stock prices to the 1984 testimony of the comptroller of the currency admitting after the bailout of Continental Illinois that 11 US banks were TBTF, spreads for large banks following the Long-Term Capital Management rescue, and the opposing impact on spreads following the Bear Sterns rescue versus the Lehman Brothers collapse). Mergers pro­vide another area for event analysis.

Among analyses of the size effect, they emphasize the findings of Acharya, Anginer, and Warburton (2014), who examine the influence of bank size on spreads after controlling for risk using the distance to default measure discussed above. Acharya, Anginer, and Warburton find that in 1990-2012, banks in the top decile of the size distribution enjoyed bond spreads 30 basis points lower than those facing other banks, and the advan­tage reached 100 basis points in 2009. However, by 2013 the differential had swung to a small disadvantage (8 basis points).

The survey authors cite Hindlian et al. (2013) as obtaining a similar negative TBTF effect (10 basis points) by 2013 for the six largest US banks (and a low funding advantage of only 6 basis points between 1999 and mid- 2007). The negative effect by 2013 in both studies is provocative, because it suggests that the introduction of policies for resolving large banks may have sharply reduced or eliminated the previous TBTF funding advantage.

A study by the US Government Accountability Office (GAO 2014) also finds that large US banks no longer enjoy a TBTF funding advantage. It applies 42 models for the period 2006-13. All of them find that funding costs for larger bank holding companies were lower than those for smaller ones in 2008-09. However, in 2011-13 more than half of the models find that larger bank holding companies had higher bond funding costs than smaller ones, given the average level of credit risk.

Siegert and Willison (2015) also survey the ratings-based (support rating versus stand-alone rating) studies of the funding advantage of TBTF banks. They note that using this method, the International Monetary Fund (IMF 2014b) placed the advantage of being a systemically important bank at 5 basis points in the United States and 20 basis points in the euro area before the crisis. These figures rose to 30 basis points and 80 basis points, respectively, by 2010 and were still at 15 and 60 basis points, respectively, by 2013. Siegert and Willison emphasize the pattern of rising TBTF benefit in the crisis and note that the Noss-Sowerbutts (2012) estimates translate into a funding cost advantage of an extraordinary 630 basis points for the four large UK banks in 2009. They also survey estimates of the TBTF subsidy using equity price volatility but suggest that this approach may overestimate the cyclicality of default risk and hence of implicit subsidies. Regarding the type of debt, they note the findings of Araten and Turner (2012) that if federal funds, noninterest deposits, and repo financing are excluded, the TBTF funding advantage is modest (18 basis points) in com­parison with ranges identified in other studies for bonds (-6 to +80 basis points), general wholesale funding (60 to 100 basis points), and deposits (17 to 80 basis points; p. 16).