The main purpose of this paper is to estimate portfolio risk and investigate the link between risk measures and the predictability of markets. A subperiod analysis is implemented for a group of four Latin American emerging markets from July 1997 to November 2006. Both value at risk and expected shortfall are used to do analysis in this paper. We use a semiparametric approach that estimates a time-varying conditional volatility model and updates residuals to take into account recent changes in volatility. The precision of predictability is measured as the deviation between forecasts and actual values of the level and volatility of returns. The main finding is that higher (lower) predictability is associated with lower (higher) portfolio risk.