The returns to scale for nineteen South Asian countries are estimated using window and cumulative rolling stochastic frontier regression analysis. The stochastic frontier analysis accounts for technical inefficiency of Hicks non-neutral technology production function in the estimation of the returns to scale. The window rolling regression and cumulative rolling regression allows the estimation of short and long run time-varying returns to scale, respectively. Empirical application to Asian agriculture sector using Food and Agricultural Organization data from 1961-2008 indicates returns to scale are under (over) estimated by the traditional panel models in the short (long) run time-varying estimation. The time-varying estimates of returns to scale indicate decreasing trend in the short run compared to long run analysis. Keywords: Asian agriculture sector, stochastic frontier analysis, window and cumulative time-varying input elasticities and returns to scale, one-way fixed effect, 1961-2008.