Electricity Market Price Forecast via LLM-based Sentiment Analysis and TimeXer

Accurate forecasting of electricity market prices is essential for market participants to make profits and manage risks. However, existing forecasting models ignore the impact of news content, and often overlook the dependency between market sentiment and price time series. To address these issues, this paper proposes a time series forecasting approach that incorporates sentiment analysis via large language models (LLMs). By analyzing news texts with an LLM, sentiment scores can be extracted and subsequently integrated into the forecasting process. The TimeXer model considers these LLM-derived sentiment scores as exogenous variables, improving the accuracy of electricity market price forecasting while effectively capturing the dependency between market sentiment and prices. Our method demonstrates superior performance on an electricity market price dataset from the California Independent System Operator (CAISO).