When Are Risk Premia Not About Risk?

9/03/2012 05:08:00 PM
Back in early 2010, I coauthored a little paper about subprime mortgage meltdown. During the course of that research I came across a book written by one of the financial industry "quants" who had inside access to all the financial data about the riskiness of the mortgages being issued by one of the major investment banks at the time. If I remember correctly, that was Eric Falkenstein, who had written the book Seeking Alpha, published in 2009.

One thing that has stuck out in my mind ever since was how Falkenstein had actually looked at the financial data--data that is not available to economists or the public--and found that in many cases the real returns from subprime mortgages was actually positively correlated with the credit worthiness of the borrower. He described how he had actually presented his findings to a group of economists, and was essentially shouted down because none of the economists could believe it was true--the higher the risk (the lower the credit-worthiness), the economic theory goes, the higher the risk premium that investors will demand. In those economists' defense, Falkenstein was unable to actually hand over the data (it was proprietary information that belonged to the investment bank, not to him), but if it is true that this kind of relationship is pervasive in finance, then it means that the conventional economic theories about asset pricing are wrong. Very wrong.

The conventional theory goes like this: investments in assets like mortgages are risky. That means that there is a difference between the nominal interest rate--the rate that gets published in the mortgage agreement--and the "real return" which is what we would expect the investment to earn on average. So to a rough approximation, the nominal return is what the borrower cares about (rough approximation because the borrower also cares about some types of risk, such as inflation), while the real return is what the investor cares about. Now if the investor is risk-neutral, then he won't care about the level of risk involved in an investment--he would be just as willing to invest in a risk-free asset (like, say, a treasury bond) as he would be to invest in a high-risk subprime mortgage that promises the same real return. In reality few, if any, investors are risk neutral, so instead they will demand a higher real return from the high-risk investment than they would from a lower-risk investment. This is called risk aversion.

All of the macro and micro-level evidence says that investors are risk-averse. So we should expect that the real return on subprime mortgages to be negatively correlated with their credit-worthiness (which is a measure of the riskiness of the investment). Falkenstein, however, found the opposite.

My hypothesis is that Falkenstein discovered that in fact the spread between nominal and real returns of assets is not just about risk premia. In addition, it is about non-competitive markets. Investors face a catch-22 when it comes to mortgages. On the one hand, they want higher returns from riskier borrowers because they are riskier, while on the other hand riskier borrowers are less able to afford these higher-return mortgages. The existence of credit ratings, then, means that the mortgage is segmented not just according to riskiness, but also according to willingness-to-pay. Those with the best credit scores have a higher willingness-to-pay for mortgage financing by virtue of the fact that they can more easily afford to pay more, while those with the worst credit scores--the subprime borrowers--have a lower willingness-to-pay because they are less able to afford a house, and are far more likely to be willing to rent rather than buy a place. If mortgage lenders are non-competitive (and the evidence shows that they are becoming less and less competitive with time, due to massive consolidations) then they are free to charge different risk premia from the different types of borrowers as they please. The economics of segmented, non-competitive markets says that lenders should, all else equal, charge higher prices to those with higher willingness-to-pay than others. All is not equal, of course: the the different types of borrowers also have different riskiness. Taken together, we would expect that the nominal interest rate on subprime mortgages to be negatively correlated with the credit-worthiness, but the spread between the nominal and real return would be positively correlated with credit-worthiness. That is, the risk premium demanded for borrowers with better credit scores would actually be higher than for the others, because they are more willing to pay the higher return. If the difference in willingness-to-pay is large enough, then this could actually cause the real return to be positively correlated with the credit-worthiness, which is exactly what Falkenstein observed in his data.

I am not familiar enough with the (extensive) economic research literature on finance to know where this theory fits in. When it comes to truly micro-founded financial models (ie, consumption-based finance), it is very hard to capture the real-world notion of default risk, since agents in these models generally never actually default, and in models where they do default, it is usually a voluntary or "strategic" default, which misses a lot of what happens in the real world. My thinking is somewhat along the lines of the "complexity" Simon Wren-Lewis has written about here. Namely, it may not be possible to realistically describe default risk in analytically-derived micro-founded models: true default risk comes from the unforeseen inability of household's to finance their debt obligations, which never happens in rational expectations that ignore, for example, income uncertainty of heterogeneous, finitely-lived households.

Mostly as a note to myself, just want to point out that maybe this problem could be addressed through mechanism design techniques. Essentially, non-competitive mortgage lenders are like central planners who engage in redistribution between different types of heterogeneous households, subject to an incentive compatibility constraint. So we could derive the investment banks optimal (non-linear) pricing schedule using credit-worthiness as a random variable conveying information about household heterogeneity. This would possibly solve a number of problems in asset pricing encountered by macroeconomists, who assume that prices must be linear in their models.