Improving forecasting by subsampling seasonal time series

Time series forecasting plays an increasingly important role in modern business decisions. In today's data-rich environment, people often aim to choose the optimal model for their data. However, identifying requires professional knowledge and experience, making accurate a challenging task. To mitigate importance of selection, we propose simple reliable algorithm improve performance. Specifically, construct multiple time with different sub-seasons from original series. These derived highlight sub-seasonal patterns series, it possible methods capture diverse components Subsequently, produce forecasts these separately classical statistical models (ETS or ARIMA). Finally, are combined. We evaluate our approach on widely-used competition data sets (M1, M3, M4) terms both point prediction intervals. observe performance improvements compared benchmarks. Our is particularly suitable robust higher frequency. demonstrate practical value proposition, showcase hourly load that exhibit seasonal patterns.