Monday, May 14, 2012

Risk and Return Confusion

If you listen to the first 20 seconds of this clip, you'll hear a simple proposition about risk and return from a top-rate finance professor, John Geanakoplos. The problem is, though it isn't really clear, he is talking about a risk akin to a default rate, and a return like a stated yield.  It's not really central to this particular lecture, but its the kind of throw-away assertion that highlights experts take the risk premium for granted, even though it's as common as a jackalope for your average investor.


Stephen Cecchetti’s textbook Money, Banking, and Financial Markets also presents the seemingly straightforward example of how bonds with higher default rates have higher yields: risk and expected return are positively correlated. Yet this is purely an anticipation of the default rates so is not risk in the sense of something priced. BBB bonds have, over time, the same total return as B-rated bonds, in spite of the fact that yields for B-rated bonds are always higher than for BBB-rated bonds. One must subtract the expected defaults and the resulting losses from a stated yield regardless of one’s risk tolerance. Confusing stated yield with return is a simple, obvious error. This is the kind of rationalization people make when they are certain their big picture is correct. The distinction between the amortized expected loss from defaults and priced risk is a fundamental distinction in modern risk-return theory. The usage of expected loss risk as opposed to 'risk premium' risk when it generates intuitive support at 30,000 feet suggests that financial professionals have a strong, active bias toward the big idea: risk begets average returns.

Geanakoplos's  theory, btw, is best highlighted by his takeaway: write down the principal on all problem mortgages by 50%.  This saves money because banks expect to get only 25% in the current morass.  The moral hazard of this solution is rather stark, as its unclear why anyone current on a mortgage should not immediate cease paying until he gets the 50% gift.  He has a hedge fund.  


Dan said...

Unfortunately with corporate bonds (especially junk), there is not good historical data. But,the historical spread difference of BBB - B is 5% - 1.8% = 3.2%

The default rate on the B is 4.5% with recovery of 40%, which gives you 1.8%.

If you look at actual returns you won't see much differences from AAA to A. But if you go down to B, you'll also notice differences in actual returns.

Eric Falkenstein said...

Junk bonds, have default rates around 5 percent annually and are hybrids between equities and bonds because they are not merely a function of interest rates but of the stock price. Altman and Bana (2004) and Kozhemiakin (2007) noted there is no premium to high-yield portfolios relative to investment-grade portfolios, a set of bonds with a 3.84 percent average annual default rate from 1970 to 2001. Altman and Stonberg (2006) noted that a bankrupt bond portfolio underperforms investment-grade bonds.

Both high-yield and bankrupt bonds have more volatility and cyclicality than investment-grade bonds and do their worst when returns are most valued, in bad times. Junk bonds are intuitively and empirically risky.

Data from the Merrill Lynch High Yield Index show a 8.27 percent annualized return relative to the 7.63 percent return of their investment-grade index from March 1987 through December 2011. The risk premium is seen to be a modest 0.64 percent annually.

Yet the indexes are really an understatement of the anomaly here because of how such indexes have a systematic bias when portraying illiquid or unaudited asset classes. Even today many times a junk bond’s bid-ask spread is 5 points wide for a 70 dollar bond. Transaction costs eat away at returns derived from closing prices.

Several ETFs now allow investors access to junk and investment-grade bond returns with reasonable liquidity. The illiquidity of the ETF collateral, however, shows up in its returns. Looking at two large high-yield ETFs (JNK and HYG) and three large investment-grade ETFs (LQD, BND, and CIU) and comparing them to the Merrill Index returns over similar time periodswhere the the high-yield ETFs underperformed their indexes by quite a bit more than the investment grade ETFs, and offset any 0.64 percent annualized premium in the indexes.

Mercury said...

“One must subtract the expected defaults and the resulting losses from a stated yield regardless of one’s risk tolerance.”

Don’t you mean *actual* defaults? If you’re considering BBB corporates you want to look at the stated yield and consider the longer term loss rate for all triple Bs. Even then I’m sure you can find periods where yields are relatively high and defaults are low…until things revert to the longer term mean but the farther back you go to find out what that longer term average is you probably run into lower quality/less comprehensive data.

At the risk of trying to fit the model to the data here is it possible that the Moody’s/S&P ratings scale isn’t the best way to define an asset class?

Eric Falkenstein said...

Well, ex ante, expected, ex post, actual. In the long run they should be the same. The key is that on average returns to investors are not significantly higher for junk bonds, though stated yield are always higher by 350 basis points on average.

Campbell Dawson said...

The only yield data by rating grade I have seen is from ML and dates back to 1996. From what we have done, BBs will provide higher returns (depending on your starting and ending data point) but at clearly higher risk. If anything, investment grade behaviour (low risk, stability of default rates) goes down to BB+ and possibly BB. Below that you get the variability of default rates. You should also realise that the duration of the BBB sector is much longer than the BB sector so any index/ETF returns will include a portion of interest rate effect as well as coupon and default

Max said...

The analysis is more complicated than just default rate - many bonds are callable, and a high yield bond is particularly likely to be called (if it doesn't default).