In this paper, we estimate the time-varying betas or input elasticities and returns to scale in a stochastic frontier cumulative rolling regression analysis framework using FAO data of Southeast Asian countries from 1960-2002. Empirical results indicated returns to scale are overestimated by pooled and panel models relative to stochastic frontier models without accounting for technical inefficiencies. Second, the time-varying estimates of betas or input elasticities and returns to scale indicate variations not only between the models but also over time.