The combination of forecasts using rank based techniques

Más información sobre el libro

The increasing availability of multiple forecasts for the same event, driven by advancements in computing and competitive forecasters, highlights the importance of combining forecasts. Decision-makers, such as politicians, executives, or farmers, often face the challenge of choosing between various forecasts, whether it’s for GDP growth, sales predictions, or daily temperatures. In such scenarios, combining forecasts becomes crucial. A decision-maker might opt for a forecast that performed well in the past, thereby employing a basic combining technique. Alternatively, they might choose to average the forecasts, a simpler method that disregards the forecasters' prior performance. This work introduces innovative non-parametric forecast combining methods based on the ranks of past forecast errors. These new techniques are evaluated against existing methods through extensive data analysis, simulations, and a selection procedure. The primary objective is to determine when and why certain combining techniques are most effective, ultimately aiming to develop a universal algorithm for forecast selection.