Every bank crisis is a good time for reevaluating how banks are measured, monitored, and managed. Recently, UBS produced a detailed Report to Shareholders that is remarkably informative description of how to lose $37B in one year. It would be great if every major corporation had the courage to generate this kind of informative mea culpa, and suggests there is little cognitive dissonance of this error. The report highlights several key issues that have applicability beyond the scope of the current subprime debacle, to a more general information systems issue.
I think the UBS experience highlights the importance of bank examiners being able to evaluate the business lines of a bank the way banks do, or should, look at their business lines. That is, large financial companies are composites of perhaps a hundred of lower line businesses, defined by a combination of currency, industry, instrument type, etc. The problem is, outsiders merely see top-line numbers for revenues and expenses, sometimes going down to lines like ‘trading’, ‘investment banking’, and ‘asset management’, still an incredibly high level of aggregation. Even Regulators and rating agency representatives are given very high, and selective, data. When a disaster hits it exposes a particular activity that was perhaps not given much thought: who knew UBS had $50 billion in US residential mortgages on their balance sheet?
The problem is that the most common information given out by banks is insufficient to see what is really going on. Assets, for example, are merely broken into large groupings based on Agency rating, maturity, and for non-rated assets whether they are performing or not. This is like assessing the health of someone by looking at their current body temperature and weight. VaR is applied selectively in a bank, leading to issues of adverse selection and moral hazard, as selective risky assets are rationalized as being on the ‘banking book’ instead of the ‘trading book’, so that the final top-down number excludes the majority of actual risk, and a biased majority at that. Another problem is that credit risk is still very difficult to measure in the VaR framework, so that given the particular collateral, obligor, and facility, can have a very different risk profile depending on details that banks are not obligated to supply, let alone validate. When you try to find comparables, you are acting more like a historian than a quant, and so you should expect these measures to be off by a factor of 2.
This focus would put more weight on revenues than direct risk measures. Now, a ‘best practices’ approach to internal profitability reporting uses what First Manhattan calls a Net Income After Capital Charge approach (NIACC). This combines the net income, and the return-on-equity, into a single number useful for capital allocation and performance evaluation. Net income alone ignores risk, while ROE is indifferent to the size of the effect; the NIACC combines them (see here for how NIACC, or EVA, is related to income).
When I was head of capital allocation at a regional bank, I discovered two insights not mentioned in risk management literature. First, profitability reporting was just as important as risk measures--for measuring risk! This is because you could pair the income generated with your bottom up risk measure, and see if it was reasonable. Often, your risk measure was off by a factor of 3 because you forgot certain assets had guarantees, or had off-balance sheet exposures. Most of these errors weren’t subtle, they merely omitted a significant factor. Most complex problems, like examining the risk of a large financial institution, are not really fundamentally complex, merely detailed, and so getting good approximations was more a function of asking for the right set of details, rather than working out the mathematics of a complex non-recombining lattice. Secondly, the aggregate numbers were pretty irrelevant. The bottom-up, aggregate required capital estimate was so dependent on uncertain assumptions about correlations, and the relation between our franchise value versus the value of assets on and off our balance sheet, that it was rightfully ignored.
The unfortunate reality is that we can’t measure risk nearly as good as we measure revenue. While we should work to make these risk measures better, in the meantime, looking at metrics based partially on revenues is very informative. Indeed, the average annualized volatility for the major 7 money center banks in the US had annualized 99% VaRs of about $1.8B in 2007, but still managed to lose an average of $42B in market value over the past 12 months. This suggests that, as a top-down number, reported VaR is about as useful as knowing the CEO’s favorite color.
The main problem in UBS's business model was that the senior RMBS were funded improperly, because if these securities were seen as having negative carry they would have been evaluated all their mortgages more stringently. UBS applied the bank funding rate to all their activities, and so Aaa rated mortgage paper had a positive carry. There was an internal arb in the bank, in that almost any paper funded at such rates generates positive carry, and given the infrequent (but massive) nature of credit losses, in combination with bonuses based on annual revenues, created an incentive to put these assets on the books.
The net result was therefore a combination of a credit risk, market risk, and operational risk, as incentives led to an inefficient allocation of resources that ended with a disaster.
If every bank showed its profitability reporting by line of business, an analyst would be able to flag these issues much better. That is, say there was a business group in charge of the RMBS on the bank’s balance sheet. Taking into account the capital applied to this business, and its funding rate, what was its NIACC? What was its ROE? If the bank reported that the NIACC, or Sharpe ratio, was significantly positive, one could note the asset class in question, and deduce something fishy that necessitated further inquiry. This is because such a large asset class simply does not have a large Sharpe ratio, and one could easily note this because no hedge funds warehouse large amounts of high-rated ABS. An assumption in the business model is clearly wrong, and given the amount of residential mortgages on the balance sheet, this error is significant.
One should expect financial institutions to have modest Sharpe ratios for the income generated by assets sitting on their balance sheets, because merely holding a common asset class simply does not have a Sharpe much above 0.25. The average Sharpe ratio of the stock market is about 0.4, and this is the ‘equity premium puzzle’, because it is so high. Thus, we should expect very few assets to be above this number, and if so, assumptions need to be examined.
There are confidentiality issues, in that some investments may have a proprietary business advantage, so it may be necessary to allow this informational reporting only for rating agencies and regulators, who sign nondisclosure agreements. The key is that specific net revenue, and capital allocation figures can lead to better questions by those outsiders like investors, and taxpayers, rely on to monitor these opaque institutions. Linking revenues to risk allows you to understand the bottom level risk numbers better, and find the landmines before they blow up.
At UBS, much of the growth in the Fixed Income group came from repackaging residential mortgages from the US into mezzanine (e.g., Ba rated) CDOs and reselling them. This generated fees of 125 to 150 basis points, compared to fees of only 40 basis points on senior tranches. But the 150 basis points on the mezzanine piece necessitated keeping 60% of the RMBS on their books, the senior pieces. These assets supported great trading revenue, but if there were properly funded and assessed the appropriate capital, it would have been obvious that while 150 basis points is great business, the costs generated by warehousing the supporting collateral put a limit on how much of this stuff was optimal. Instead, the residual assets warehoused seemed to have a positive NIACC (equal to an above-hurdle rate Sharpe), and thus at the margin added to the bank’s value, so there was effectively no limit to how much was optimal: as many as their mezzanine CDO group could sell. But no matter whether these senior piece were put in a separate business line, or kept within the group generating the restructuring fees, it would have set off red flags that a liquid and large asset class like RMBS was capable of generating significant alpha. Like taking 50 basis points out of a Treasury Bill trade, some familiarity with the spreads and returns of these assets suggests that some large tail risk is being assumed, because there just isn’t enough spread to generate this kind of profit without some error in transfer pricing or risk estimation.
Clearly, they rationalized the positive carry on super senior RMBS via the fat fees on the CDOs, but this had deleterious spill-over effects as well. UBS put on tens of billions of Aaa-rated ABS for things like autos and credit cards based on the same, flawed, funding model. After all, one could say, if you are going to fund Aaa RMBS at positive carry, why not Aaa auto loans? And so, when credit spreads for all structured finance widened, UBS took a considerable hit there as well, losing about $4B, all in an investment that never made sense to begin with. The origin of this loss was operational risk at its core.
The conventional corporate bond puzzle is that Investment Grade spreads are too high. The most conspicuous bond index captures US Baa and Aaa bond yields going back to 1919, which generates enough data to make it ‘the’ corporate spread measure, especially when looking at correlations with business cycles. Yet Baa bonds are still ‘investment grade’, and have only a 4.7% 10 year cumulative default rate after initial rating. As the recovery rate on defaulted bonds is around 50%, this annualizes to a mere 0.23% annualized loss rate. Since the spread between Baa and Aaa bonds has averaged around 1.2% since 1919, this generates an approximate 0.97% annualized excess return compared to the riskless Aaa yield, creating the puzzle that spreads are ‘too high’ for the risk incurred.
Though a puzzle, it would be a mistake for Aaa rated companies to actually assume this spread is investment alpha. The other two such return anomalies, the short end of the yield curve, and the equity premium , clearly do not imply one should take yield curve risk, or borrow to invest in equities. Public companies are not necessary for taking these bets, and so an inefficient way to address them even if an investor thought the represented merely ‘behavioral biases’. We may not fully understand these particular risk premiums, but they are not ‘arbitrage’. Banks would be wise to fund at best the A rate internally, to avoid this kind of gaming. When the UBS funded AAA rated assets at a positive carry, this error was essential for supporting the wrong track they went down.
A well-run bank should have income tied to a capital allocation based on economic risk, at the business line level. Thus it should be able to provide this information. An examiner would then see the various Sharpes, for the various business lines, and their assets on the balance sheet, and see if these make sense. A very high Sharpe invites investigation. Was it funded correctly? Are there off-balance sheet liabilities? Is the VaR, or capital allocation, correct? By having this on a specific asset class, such as super senior residential mortgages, or fixed-receive swaps in the Treasury account, the analyst can address the issue. With a top-down VaR, one has no way of asking relevant questions about how the VaR was estimated, but rather general questions about VaR methodology that are not likely to be informative.
If a bank is putting on a large amount of assets onto their balance sheet, or retaining exposure to off-balance sheet liabilities, there are two general paths to detect this: from a bottom up risk calculation, and through a top-down revenue examination. Most bank businesses are not amenable to a reasonably precise economic capital estimation based merely on the asset’s characteristics, and economic capital algorithms are applied inconsistently in the current framework. Thus, the revenues from these businesses add important information because in general we have good intuition on what reasonable risk-adjusted rates of return are, especially for prosaic asset classes like residential mortgages. A financial institution making exorbitant returns on these things implies some assumption within the bank is incorrect. Only through the obscurity of aggregation were these positions allowed to fester into the problems they became, and so at least regulators, but perhaps rating agency representatives, should be able to demand data on the profits, regulatory, and economic capital, at the lowest level line of business. If they say 'we manage at the 'asset management' level', allow the examiners to discuss in detail what they found. As shareholders learn of this, they will realize they are heading towards disaster, because you can't manage a bank at that level, so it's either reflects ignorance or mendacity.
1 Chen,L.; D.Lesmond; and J.Wei, 2007, Corporate Yield Spreads and Bond Liquidity. Journal of Finance, 62 (1), 119-149.
2 Mehra, Rajnish; Edward C. Prescott, 2003. "The Equity Premium Puzzle in Retrospect", in G.M. Constantinides, M. Harris and R. Stulz: Handbook of the Economics of Finance. Amsterdam: North Holland, 889-938. David Backus, A. Gregory, and Stanley Zin, 1989, Risk premiums in the term structure: Evidence from artificial economies,"Journal of Monetary Economics, 24, 371-399.