Paper Type
Complete
Paper Number
PACIS2025-1917
Description
Price jumps on the continuous intraday electricity market are a considerable challenge for electricity traders, because surplus or demand has to be balanced out at almost any price. Forecasting these jumps would therefore be of considerable economic benefit. With market data of EPEX Spot from 2021 to 2023, we are facing the challenge to classify relevant price patterns into three classes ‘no_jump’, ‘negative_jump’, and ‘positive_jump’. The data set contains more than 100,000 observations in 15-minutes steps, containing trade data like prices, volume or actual and estimated renewable generation. For classification, decision trees, random forests and neural networks were used. The balance between alerts and false alerts was set considering its economic implications. False alerts are only critical and cause economic damage if a jump is predicted in wrong direction. Thus, our system is designed to detect a significant proportion of price jumps and opens up reaction potential for traders.
Recommended Citation
Lackes, Richard; Sengewald, Julian; and Wilz, Mathis, "Price jumps on the intraday energy market - design and implementation of an alarm system with machine learning methods" (2025). PACIS 2025 Proceedings. 24.
https://aisel.aisnet.org/pacis2025/aiandml/aiandml/24
Price jumps on the intraday energy market - design and implementation of an alarm system with machine learning methods
Price jumps on the continuous intraday electricity market are a considerable challenge for electricity traders, because surplus or demand has to be balanced out at almost any price. Forecasting these jumps would therefore be of considerable economic benefit. With market data of EPEX Spot from 2021 to 2023, we are facing the challenge to classify relevant price patterns into three classes ‘no_jump’, ‘negative_jump’, and ‘positive_jump’. The data set contains more than 100,000 observations in 15-minutes steps, containing trade data like prices, volume or actual and estimated renewable generation. For classification, decision trees, random forests and neural networks were used. The balance between alerts and false alerts was set considering its economic implications. False alerts are only critical and cause economic damage if a jump is predicted in wrong direction. Thus, our system is designed to detect a significant proportion of price jumps and opens up reaction potential for traders.
Comments
AI ML