What is the data analysis method for stretching and contracting?

Stretch/contract is a data analysis method for time series forecasting that combines forecasts from models fitted to different historical windows. It specifically addresses varying optimal history lengths for different forecasting horizons. This approach enhances forecast reliability through statistical ensemble principles. The method requires generating two forecast sets: stretched forecasts using longer historical data and contracted forecasts using shorter data. Forecast weights are optimized separately for each future horizon to minimize variance or error metrics. The optimal history window length typically decreases as the forecast horizon increases. This technique applies when historical patterns show non-stationarity or varying relevance over time. Model selection and computational resources are key implementation considerations. Its primary application lies in improving forecast accuracy for operational planning where horizon-specific data relevance varies. Stretch/contract enables adaptive learning, delivering superior performance in scenarios like demand forecasting, economic trend projection, and resource planning compared to fixed-window methods. This flexibility offers significant value in dynamic business environments requiring robust forward estimates.