Handel na rynku dnia następnego z wykorzystaniem krótkoterminowych prognoz cen 
energii elektrycznej - generatywne metody uczenia maszynowego  
[Trading on short-term path forecasts of intraday electricity prices - generative 
machine learning methods]

Jieyu Chen (Institute of Statistics, Karlsruhe Institute of Technology, Germany)


We propose a novel multivariate generative model for electricity prices in intraday 
continuous trading markets. In this new class of nonparametric data-driven distributional 
regression models, path forecasts of intraday electricity prices are obtained directly as 
the output of a generative neural network. The generative model is trained by optimizing 
the energy score between the path forecasts and real prices, and is conditional on 
exogenous input variables including wind generation. In a case study of a selected German 
intraday electricity market, we demonstrate the benefits of our generative model in terms 
of both statistical error metrics and economic evaluation based on different trading 
strategies.