A Rehabilitation of Stochastic Discount Factor Methodology

In a recent Journal of Finance article, Kan and Zhou (1999) find that the 'Stochastic discount factor' methodology using GMM is markedly inferior to traditional maximum likelihood even in a simple test of the static CAPM with i.i.d. normal returns. This result has gained wide attention. However, as Jagannathan and Wang (2001) point out, this result flows from a strange assumption: Kan and Zhou allow the ML estimate to know the mean market return ex-ante. I show how this information advantage explains Kan and Zhou's results. In fact, when treated symmetrically, the discount factor - GMM and traditional methodologies behave almost identically in linear i.i.d. environments