Rolling Forecast Origin

Algorithm

A rolling forecast origin represents a dynamically updated input point for time series forecasting models, particularly relevant in cryptocurrency and derivatives markets where stationarity is often absent. This methodology contrasts with fixed-origin forecasts by continually incorporating the most recent observed data, effectively ‘rolling’ the starting point of the prediction window forward in time. Consequently, the model adapts to evolving market conditions, mitigating the impact of structural breaks and non-linear dependencies inherent in financial data. Its implementation requires careful consideration of computational cost and potential overfitting, demanding robust backtesting and validation procedures.