Using futures traded on the Chicago Board of Trade, Chicago Mercantile Exchange and New York Board of Trade, we test six alternative models of the return-generating process (RGP) in futures exchanges that adopt a price-limit regime. We rank the six models according to their return-prediction ability, based on the mean square error criterion, and we find that the near-limit model performed best for both the estimation period and the prediction period. A reliable prediction of the expected return can have important implications for both traders and policy makers, concerning related issues such as the employment of long or short strategy, margin requirements and the effectiveness of the price limit mechanism.