There are indeed things in this world which need regulation. One of them being the capital adequacy of the banking system, given the inherent fragility of anything based upon fractional reserve banking. However, when regulating it does aid if you start from a recognition of reality. Within banking capital structures and thus stress tests this means using real and market tests of value and liquidity. Not assuming, but studying. Which is, of course, exactly what wasn’t done:
How stress tests fail
Jeremy Bulow 08 May 2019
Bank stress tests in the US were an important tool for bailing out banks in the Great Recession. As this column points out, however, because the tests use regulatory rather than market measures of asset values and risk they have almost nothing to do with whether a bank will be economically solvent under test conditions. This column argues that the thousands of pages of post-crisis bank regulation have largely ignored perhaps the two most needed reforms: measuring asset values and risks in an economically realistic way. Reforming the stress tests is necessary for clearly and credibly placing responsibility for future banking losses in the private sector and for improving incentives for both managing old risks and for investing in new ones.
The US bank stress tests, namely the Dodd-Frank Act Stress Test (DFAST) and the Comprehensive Capital Analysis and Review (CCAR), supposedly require fortress balance sheets to withstand ‘severely adverse’ conditions: in 2018 defined as a 65% decline in equity prices, 30% and 40% drops in residential and commercial real estate, and 10% unemployment.1 But the tests use regulatory rather than market measures for both the value and riskiness of bank assets – measures that failed badly during the financial crisis.
Under the conditions described the Federal Reserve projects 5% losses second mortgages and home equity loans and 5.5% on student and auto loans.2 Nine of 35 banks make a regulatory profit– Wells Fargo makes $10 billion. In aggregate, net losses and other capital charges are just 1.3% of assets.3 The tests claim that even if the banks continue paying dividends and (in the CCAR version) implement planned share repurchases, they will have plenty of regulatory capital.
During the crisis, banks avoided realistic asset marking in two ways. First, they marked their ‘hold to maturity’ assets using proprietary models rather than market or fair values. For example, in 2008 the Federal Home Loan Bank of Atlanta estimated that its fair value losses on private label mortgages were $2.7 billion but its model losses were $44,000.4 The 2018 tests estimate losses using the historical relationship between model losses and fair value.5 Second, commercial banks were permitted to mark their supposedly fair valued ‘available for sale’ assets much more aggressively than the maligned investment banks. Citibank marked its alt-A portfolio at 85% of par while Lehman, the most aggressive investment bank, marked it at 39%. Wachovia marked its subprime collateralised debt obligations at 58% vs. 15% for Morgan Stanley.6
In 2008-11, over 400 banks in the US failed 90% of the imposed losses on the Federal Deposit Insurance Corporation of at least 14% of assets despite positive regulatory capital.7 For calibration, 6% regulatory capital was the standard for being “well capitalised”.
The tests also ignore the sharp increase in asset risk that occurs in a crisis. Say a borrower posts collateral of 120 against a non-recourse loan of 100. The riskless interest rate is zero and investors are risk-neutral. There is a 10% chance the collateral will fall in value, and if it does, it will drop by 25%, to 90. Then expected repayment is 99 and the loan can support 90 in riskless debt. But say the collateral is only 100 and there is still a 10% chance of a 25% decline. Then expected repayment falls to 97.5, but the bank’s claim is riskier and it can only support 75 in riskless debt. Required equity is 97.5-75=22.5 instead of 99-90=9, an increase of 150%.
What happens in the regulatory world? Capital requirements are proportional to risk-weighted assets. During the crisis, the ratio of risk-weighted assets to total assets crawled from 74.9% at the end of 2006 to a peak of 75.9% in the third quarter of 2007. In that tradition, the 2018 severely adverse scenario increases the ratio from 63% to 64%. An enormous increase in risk creates almost no stress in the tests. US banks raised negative net equity from the beginning of the financial crisis (mid-2007 or earlier) through the end of the third quarter of 2008, though two of the most sophisticated banks (JP Morgan and Goldman Sachs) did raise money in the weeks after Lehman and AIG collapsed. The stress tests institutionalise a similar path, or worse, for the next crisis.
Mismarking causes incentive problems that exacerbated the crisis. For example, take an asset marked at 90 with a capital requirement of 10%. Then, for regulatory purposes the asset is equal to 90 (1-.10) =81 in cash. Say the market value is 50. Then a sale would force the bank to recognise the loss and raise 31 in extra capital, even as the bank reduced its risks. On the other hand, selling equity to make new loans marked at fair value becomes much less profitable if the first priority for the returns from the new loans will be to pay for any unrecognised losses on old loans.
Using the market makes sense even if, as bankers and regulators often argue, market values might be wrong. If I try to borrow against a share of Facebook, selling for $180, the amount a bank will lend me depends on the current market value and risk. That the market might prove wrong and Facebook could ultimately be worth $80 or $280 doesn’t make market prices and ‘haircuts’ irrelevant.
The stress tests have made some improvements. Even as they assume lower losses than five years ago,8 test capital requirements for the large banks are higher than before the crisis. Also, some loopholes have been shrunk or closed. For example, many critics have attributed part of the crisis to the run-off of ‘shadow banking’ assets, particularly in the asset backed commercial paper market, in the second half of 2007. 74% of this market represented the off-balance sheet activities of commercial banks (and 1% of those of investment banks).9 Regulations now pay more attention to off-balance sheet financing.
But for all the thousands of pages of new banking regulation over the last ten years, the most important possible changes – moving to more realistic asset and risk marks – have largely been ignored.
Stress tests served a useful role in 2009. With banks in grave danger, the only ways to avoid bankruptcy were for the government to immediately inject enormous amounts of equity, to directly insure almost all short-term liabilities, or (politically much more palatable) to somehow persuade short-term creditors that it was safe to roll over their claims regardless of bank solvency. This required the regulators to promise lenders that the banks would be treated as though they were solvent enough to be kept in business no matter how insolvent they became. The tests, combined with the promises of regulators, made clear that no major bank would be allowed to fail before the end of 2010. So ‘uninsured’ debt coming due in 2009 could be rolled over for a year with the creditors assured that they could front-run the deposit insurer and get their money out before any bank closure, regardless of solvency. And in 2010 there would be another test, promising another year of safety.
In Europe today, many banks are still very weak and perhaps regulators feel compelled to maintain the same fictions about financial health that were so useful in the 2009 tests. But in the US, the ten-year recovery means that the major banks are currently solvent. It is time to move to a system that uses market-based capital requirements – a necessary step for clearly and credibly placing responsibility for future banking losses in the private sector and for improving incentives for both managing old risks and for investing in new ones.