Abstract
This paper examines the predictability of size portfolio returns using a new database constructed from the London Stock Exchange for the period 1985–1995. Predictability of returns, both adjusted and unadjusted for risk, are examined and, because evidence of nonlinearity and nonnormality is found in these series, conventional autocorrelation analysis is supplemented with analysis using Markov chain processes. It is found that predictabilities appear for the largest size portfolios rather than the smallest, so that, although a size effect remains in the market, it is rather different to that which is usually thought to hold.