Resumen
The present work aims to verify the efficiency of factors of return in predicting stocks? returns traded in BM&FBovespa. Have been tested 39 models, grouping 16 variables in their totality, in families and in isolation. All the models have been tested using as factors? payoff the last 12 months OLS coefficient, and the isolated variable models were also tested using a static coefficient. The variables earnings/price and revenue/price, with static payoff, presented higher average predictability efficiency than the model using the totality of variables, confronting with the multifactoriality marginal gain. Significant predictability was verify for the i) multifatorial models: totality of variables and ?cheapness? family; ii) isolated variable models: dividend yield, earnings/price, revenue/price; ROA and excess 12 months return with static payoff, and earnings/price and market value with moving payoff. The existence of predictive capacity is opposed to the market efficiency.