Corporate Bonds: No Return to Higher Risk
Few theories have a higher reputation/evidence ratio than the theory that higher risk begets higher return.
At some level this statement is obviously true: over time you make more money in equities than government bonds, and in one’s career taking risks is necessary to garner large gains. But other than very general tendencies the theory has scant power. There is no return premium for higher risk stocks compared to lower risk stocks—all you get for sticking your neck out into high risk equities is more risk. The equity literature on this is pretty clear now, and only excessively subtle yet labored reasoning can make a risk premium appear.
John Cochrane, an otherwise excellent researcher at the University of Chicago, stated that CAPM (the Capital Asset Pricing Model, which states theory that expected returns are linear in beta) was “stunningly successful in a quarter century of empirical work.” But then he also notes we subsequently discovered that the beta-return correlation merely reflected the size-return correlation, and not the other way around (see New Facts in Finance, Economic Perspectives, Federal Reserve Bank of Chicago, 1999). That is, in the 1990’s we found that the empirical beta-return correlation was the result of unobserved variables, mainly size (market cap) and ‘value’ (book/market equity ratio). I don’t want to quibble on semantics, but it’s hard to say the CAPM was staggeringly successful if that success was only due to ephemeral ignorance. It’s like saying that the communism was a great economic success—until we saw the data.
But while the failure of the CAPM within equities has been well examined, a less well known, but even more transparent empirical counterexample to risk-return theory exists within corporate bonds. Almost all corporate bonds have ratings, which while imperfect, rather strongly separate risky from nonrisky bonds. This makes for a nice test of the risk-return theory, because while equity researchers have had trouble delineating risk, such a problem does not exist in corporate bonds. That is, among equities one has difficulties separating high and low risk stocks because of ambiguities measuring ‘the market’, among other issues. But ‘High Yield’ bonds have higher risk under any reasonable definition than investment grade bonds.
Data on lower rated bond prices are not reliable before the 1980’s. But now we have two decades of data, and perversely, the data suggest that not only is there no alpha to junk bonds, there’s no return premium at all. If you put $1 in the Merril Lynch US Corporate Index Index, and $1 in the Merril Lynch US High Yield Index, your total return over the past 16 years would have essentially the same. The volatility, however, would have been about 40% higher for the high yield portfolio.
How can this persist? Defaults, even for B-rated credits, are infrequent (6% on average) and highly cyclical. This makes them hard to put into a mental accounting the way one can amortize losses for, say, a pool of credit card defaults. A certain level of granularity seems necessary for people to apply statistical concepts, and defaults are simply too clustered by time and industry for people to avoid exempting major losses as irrelevant anomalies. I remember high yield CDO pitches from 1998, and all included the sincere belief that expected default rates for underlying collateral would be no more than 75% of its historical average. Take out a couple key industries in a few key years (Transportation in 1970, Hotels in 1990, Telecom in 2002), and average default rates are dramatically lower. It is very tempting to apply hindsight and think that such exclusions are rational, because there are always many particulars that make these outliers truly unique. Somewhat paradoxically, all bubbles are very different in their particulars yet still repeated.
If the risk-return theory doesn’t work as a description of reality, it’s perhaps even more useful as tool to tell you what you should and should not do. That is, if risk is not compensated by higher returns over time, the obvious implication is to avoid higher risk securities in the first place. The current (11/26/03) spread to Treasuries is a mere 325 basis points for B rated bonds. B-rated bonds have an average annualized default rate of 6% and an average recovery rate of 50% (which means a 300 basis point annualized loss expectation). Over the average lifetime of these loans this mean you are pretty much guaranteed no reward for your extra risk. Bad incentives, selective inference or simple ignorance underlies any decision to buy US corporate B-rated bonds today at these levels.
Eric Falkenstein 12/1/03