Electricity Price Forecasting with Deep Learning and Stochastic Time Change

The goal of this project was to assess how accurate one can predict electricity prices in a day ahead market using deep learning models with stochastic time change as part of the training data. Inspired by these two articles: 1 and 2, I experimented with forecasting electricity prices using different deep learning models. The training set was time-date data and a stochastic time change element. The results were not as competitive as current methodologies, but I believe with further adjustments it can be a viable tool. The main issue I faced was during some period of the data, volatility of prices exploded and the model couldn’t recognize this shift in regime or account for the fat tails.