Thursday, December 19, 2013
Risk premia or behavioral craziness?
John Cochrane is quite critical of Robert Shiller's Nobel lecture.
Cochrane wishes Shiller would give a more rigorous definition of "bubble" (I couldn't agree more), and he also thinks that Shiller is trying to make finance less quantitative and more literary (I somehow doubt this, given that Shiller is first and foremost an econometrician, and not that literary of a guy).
But the most interesting criticism is about Shiller's interpretation of his own work. Shiller showed that, over long time horizons, stock prices mean-revert. He interprets this as meaning that the market is inefficient and irrational - in other words, he attributes mean reversion to what I call "behavioral craziness". But others - such as Gene Fama - interpret long-run predictability as being due to predictable, slow swings in risk premia.
Who is right? As Cochrane astutely notes, we can't tell who is right just by looking at the markets themselves. We have to have some other kind of corroborating evidence. If it's behavioral craziness, then we should be able to observe evidence of the craziness elsewhere in the world. If it's predictably varying risk premia, then we should be able to measure risk premia using some independent data source. Simply appealing to plausibility - i.e., throwing up your hands and saying "Oh, come on!" - is not good enough to resolve the puzzle.
Personally I suspect that it's behavioral craziness, given the fact that experimental asset markets exhibit highly predictable and significant craziness that looks suspiciously like real-world "bubble" episodes. But six undergrads trading tens of dollars in a computer lab is hardly a perfect proxy for the U.S. stock market, so the experimental evidence is suggestive rather than decisive.
There have been some models that have tried to explain how there could be slow, predictable variations in risk premia. The ones I've seen are DSGE-type models that typically involve either consumption habit formation, or Kreps-Porteus preferences (if you don't know what those are, don't ask). Some people swear that the latest of these models have resolved the puzzle. Cochrane is not so sure about the habit-formation models, a sentiment shared by many finance profs I've talked to. (Also, the models of this type that I've seen tend to use RBC-style productivity shocks as the source of aggregate uncertainty, making me pretty skeptical of them right off the bat...but that's just me.)
As for direct evidence of real-world behavioral craziness, this has been scarce...until recently, perhaps. Check out this paper by Robin Greenwood and Andrei Shleifer. They collate six different data sets that ask investors about their expectations of stock returns. The six series are highly correlated, meaning that they are really capturing some general phenomenon in the market. The stated expectations all seem to be "extrapolative", meaning that when returns have been good recently, people think they will continue to be good. But the stated expectations are usually wrong; when people think returns will be high, returns tend to fall soon after. What's more, that fall could be predicted by a simple asset pricing model. In other words, these stated expectations are not rational expectations.
So if these stated expectations really do represent investors' beliefs, then we have direct evidence of behavioral craziness. Now, Cochrane has suggested that people responding to these surveys are not reporting their true beliefs, but rather their "risk-neutral probabilities" - in other words, he thinks they are letting some of their risk aversion creep into their statements about their beliefs. If that's true, then these surveys wouldn't be good evidence of behavioral craziness.
So the issue has not been decided yet. But progress has been made. In my opinion, the available evidence is suggestive of the "behavioral craziness" explanation, but not conclusive. The important thing is that this is not one of those "this will never be resolved" sorts of debates. This is a debate that can and will be resolved, as better and better data becomes available. Science progresses.
(Note: This blog post was edited to remove a term that I found out is actually a slur in the UK. No offense was intended.)