Market proxies and the conditional prediction of returns

Ex post efficient proxies for the market portfolio are tested against the equal weight proxy. The equal weight proxy outperforms the others when the criterion is squared error of conditional prediction of returns. The ex post efficient proxies use maximum likelihood estimates of return. Stein estimates of return will generally be different from the maximum likelihood estimates and they necessarily correspond to market proxies which are not efficient ex post. In other words, there generally exists a better, inefficient, proxy than an ex post efficient proxy when the criterion is squared error of conditional prediction of return.