Machine Learning NeEDS Mathematical Optimization with Prof Yael Grushka-Cockayne

Machine Learning NeEDS Mathematical Optimization
Branding the role of OR in AI with the Support of EURO

Title: Challenges of Combining Forecasts from Correlated Sources

Abstract: In this talk, we will explore some challenges with forming a consensus forecast when combining forecasts from multiple sources. We will propose the use of a common correlation heuristic for aggregating point forecasts. The forecast aggregation literature has a long history of accounting for correlation among forecast errors. Theoretically sound methods, however, such as covariance-based weights, have been outperformed empirically in many studies by a simple average or weights that account for forecast error variance but assume no correlation. We offer a heuristic that utilizes a common correlation between the forecasters, reducing the number of parameters to be estimated while still accounting for some level of correlation.