Improving load forecasts in Italian bidding zones: A coherent combination approach
Forecast reconciliation is a well-established post-forecasting process that adjusts base forecasts produced by a single expert or model to ensure coherence within a constrained forecasting framework. However, in many practical applications, multiple forecasts are available for each variable, originating from different models or experts. Coherent forecast combination extends traditional reconciliation by integrating forecast combination and reconciliation into a unified framework to improve accuracy and satisfy the coherence property. In addition, we explore the theoretical properties of coherent forecast combination, its advantages over single-task combination and single-expert reconciliation approaches, and its practical implementation with the R package FoCo2 (available on CRAN). Using the 15-minutes Italian load (disaggregated by 7 bidding zones) dataset, our method shows superior accuracy compared to single-task base and combined forecasts, and a state-of-the-art single-expert reconciliation technique, demonstrating to be an effective approach to forecasting linearly constrained multiple time series.
142/200 words
Poster
Download pdf