Selecting weighting schemes under structural change in volatility forecasting

Identify which estimation data-weighting schemes—rolling window truncation versus exponential down-weighting—are most appropriate for handling structural change when training daily realized variance forecasting models for equities in walk-forward out-of-sample setups.

Background

Financial time series often exhibit structural changes due to evolving market conditions and participant behavior. The authors discuss using rolling windows (discarding old data) or exponential weighting (gradually down-weighting old data) for estimation but note uncertainty about which is preferable in financial applications. Choosing the proper scheme impacts bias-variance trade-offs and forecast robustness.