This study examines whether the short-term variation in the japanese size and value premium is sufficiently predictable to be exploited by a timing strategy. In the spirit of pesaran and timmermann [j. Finance 50 (1995) 1201], we employ a dynamic modeling approach in which we explicitly allow for permutations among the determinants in order to mitigate typical data-snooping biases. Using a base set of candidate regressors, we perform an in-sample estimation of all economically sensible models. Subsequently, a “best” model is determined according to a selection criterion. However, whereas most studies use in-sample model selection criteria, we introduce an out-of-sample training period to select our models. We then implement our strategy in a second-stage out-of-sample period: the trading period. All stages re-occur on a monthly basis via a rolling window framework. The results confirm sufficient predictability under lower transaction cost levels. Under high transaction costs scenarios it is more difficult to obtain incremental benefits.