A1326
Title: An angular combination of forecasting distributions in the energy market
Authors: Laura Garcia-Jorcano - Universidad de Castilla-La Mancha (Spain) [presenting]
Lidia Sanchis-Marco - University of Castilla-La Mancha (Spain)
Abstract: A dynamic angular combination of forecasting probability distributions integrating conditional autoregressive value-at-risk (CAViaR)-based quantiles fitted using a metalog distribution is proposed for the energy market. This approach improves the accuracy of predicting changes in distributions during periods of market stress and provides greater flexibility in adapting to changing market conditions while more effectively capturing the nonlinear and volatile behavior of financial data compared with traditional combination methods of distributions (e.g., vertical and horizontal averaging) and individual distributions. Multinomial quantile backtesting confirms that angular and horizontal combinations improve the forecasting of extreme losses but not extreme gains compared with individual models. Angular combination represents a flexible and robust method for forecasting probability distributions without restrictive assumptions in environments with structural breaks, asymmetry, and severe tail risk as observed in the energy market, with valuable implications for risk management and decision-making processes in the energy market.
