May 2008. Comments on Robin Greenwood and Dimitri Vayanos’ paper for the IGM “Beyond Liquidity ” conference at the GSB Gleacher center, May 9-10 2008. The paper from Dimitri’s website . I learned two important lessons in reading and thinking about this paper. 1) When arbitrageurs are limited by risk-bearing capacity, “downward-sloping” demands depend on correlations. The paper and my comments have a lovely example in which arbitrageurs are asked to hold more long-term bonds and less short-term bonds. The result is that all yields go up! Why don’t long yields go up and short yields go down? Because risks are described by a one-factor model, so all that matters is how much overall duration risk arbitrageurs have to hold.  2) We’re probably doing a bad job of correcting for serial correlation in all predictive regressions. Typically, we think expected returns move slowly over time. The right hand variable also moves slowly over time, but doesn’t capture all of the expected return variation. This situation means that residuals have a slow-moving AR(1) plus an unforecastable component, which is the same thing as an ARMA(1,1).  This structure will be very poorly captured by standard “nonparametric” procedures such as Newey-West, since you’re unlikely to put in enough lags to capture the long-run component, and also poorly captured by parametric procedures like fitting an AR(1). “Short” samples make the problem worse. More in the comments. 

Read More >