I have been reading an interesting paper by Tobias Adrian, Erkko Etula and Tyler Muir proposing a single factor asset pricing model that is based on shocks to securities broker-dealer leverage. The performance of this single factor model in pricing the Fama-French and momentum portfolios seems to be as good as that of the four factor model that includes the three Fama-French factors (market, size and value) and the momentum factor. In addition, the leverage factor model prices risk free bond portfolios as well as the four factor model augmented with a factor for interest rate level.
The results seem too good to be true and Bayesian theory teaches us that surprising results are likely to be false even if they are published in a top notch peer reviewed journal (see for example here or here). (I do recall the incident a couple of years ago when the Chen-Zhang q-factor papers became “defunct” after a timing error was identified in the initial work.) Having said that, the Adrian-Etula-Muir paper has been around since 2008 and was last revised in March 2012. Maybe, it has survived long enough to be taken seriously.
Another possible criticism is that the Adrian-Etula-Muir paper does all the empirical analysis using the Fama-French style size-value-momentum portfolios and not on the individual stocks themselves. Falkenblog goes so far as to say “What I suspect, though I haven’t done the experiment, is that if you regress individual stocks against this factor there will be a zero correlation with returns.” My own intuition is that the effect would not weaken so dramatically in going from portfolios to individual stocks. In any case, asset pricing tests have to be based on portfolios to obtain statistical power – the correct question to ask is whether the correlation with a random well diversified portfolio is likely to be high.
Adrian-Etula-Muir motivate their finding with the argument that broker-dealer leverage proxies for the health of the financial sector as a whole, and that because of limited participation and other factors, the wealth of the financial intermediaries matters more than that of the representative household in forming the aggregate Stochastic Discount Factor (SDF). This appears to me to be a stretch because even if we focus on intermediaries, leverage is not the same thing as wealth.
My initial reaction was that the leverage factor is actually a liquidity factor, but their results show that leverage shocks are largely uncorrelated with the shocks to the Pastor-Stambaugh (2003) liquidity factor.
I wonder whether the leverage factor may be a very elegant way of picking up time varying risk aversion so that the single factor model is close to the CAPM with time varying risk aversion. The empirical results show that the leverage factor mimicking portfolio is very close to being mean variance efficient. If this is so, then we may have a partial return to the cosy world from which Fama and French evicted us a couple of decades ago.