A1362
Title: Weighted inference from long historical samples under structural change
Authors: Etsusaku Shimada - Iwate Prefectural University (Japan) [presenting]
Abstract: Empirical work often relies on long historical samples even when the relevance of older observations changes over time. Weighted inference from historical data under structural change is studied, focusing on how alternative weighting schemes affect current-oriented estimation. The framework treats weights as a way to control the contribution of observations generated under different structural environments, rather than simply as a device for variance reduction. Estimation and asymptotic inference are developed under weak dependence, with particular attention to the effective sample size induced by downweighting older observations. Simulation experiments compare full-sample, rolling-window, and smoothly weighted procedures, illustrating how recency emphasis changes bias, variance, and estimation accuracy when structural relationships evolve gradually. The results clarify how long historical samples can be used more transparently when empirical environments are not stable over time.
