Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Orderbook Feature Learning and Asymmetric Generalization in Intraday Electricity Markets

Runyao Yu, Ruochen Wu, Yongsheng Han, Jochen L. Cremer

TL;DR

This paper extracts 384 features from the orderbook and identifies a set of powerful features via feature selection and performs a systematic generalization study across countries and product types, from which an asymmetric generalization phenomenon is revealed.

Abstract

Accurate probabilistic forecasting of intraday electricity prices is critical for market participants to inform trading decisions. Existing studies rely on specific domain features, such as Volume-Weighted Average Price (VWAP) and the last price. However, the rich information in the orderbook remains underexplored. Furthermore, these approaches are often developed within a single country and product type, making it unclear whether the approaches are generalizable. In this paper, we extract 384 features from the orderbook and identify a set of powerful features via feature selection. Based on selected features, we present a comprehensive benchmark using classical statistical models, tree-based ensembles, and deep learning models across two countries (Germany and Austria) and two product types (60-min and 15-min). We further perform a systematic generalization study across countries and product types, from which we reveal an asymmetric generalization phenomenon: models trained on more liquid markets or products transfer well to less liquid ones, whereas the reverse transfer leads to substantial performance degradation.

Orderbook Feature Learning and Asymmetric Generalization in Intraday Electricity Markets

TL;DR

This paper extracts 384 features from the orderbook and identifies a set of powerful features via feature selection and performs a systematic generalization study across countries and product types, from which an asymmetric generalization phenomenon is revealed.

Abstract

Accurate probabilistic forecasting of intraday electricity prices is critical for market participants to inform trading decisions. Existing studies rely on specific domain features, such as Volume-Weighted Average Price (VWAP) and the last price. However, the rich information in the orderbook remains underexplored. Furthermore, these approaches are often developed within a single country and product type, making it unclear whether the approaches are generalizable. In this paper, we extract 384 features from the orderbook and identify a set of powerful features via feature selection. Based on selected features, we present a comprehensive benchmark using classical statistical models, tree-based ensembles, and deep learning models across two countries (Germany and Austria) and two product types (60-min and 15-min). We further perform a systematic generalization study across countries and product types, from which we reveal an asymmetric generalization phenomenon: models trained on more liquid markets or products transfer well to less liquid ones, whereas the reverse transfer leads to substantial performance degradation.
Paper Structure (19 sections, 5 equations, 2 figures, 4 tables)

This paper contains 19 sections, 5 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Preliminary
  3. Feature Extraction and Selection
  4. Feature Extraction
  5. Feature Selection
  6. Model Comparison
  7. Classical Statistical Models
  8. Tree-Based Ensemble Learning Models
  9. Deep Learning Models
  10. Evaluation Metrics
  11. Generalization Assessment
  12. Cross-Country Generalization
  13. Cross-Product-Type Generalization
  14. Experiment
  15. Feature Extraction and Selection
  16. ...and 4 more sections

Figures (2)

  • Figure 1: (a) Overview of the workflow: extracting 384 orderbook features, performing sparse feature selection, benchmarking classical, tree-based, and deep models, producing probabilistic forecasts of the ID$_3$ index, and assessing cross-market and cross-product generalization. (b)-(e) Histograms of 60-min and 15-min ID$_3$ from Germany and Austria. The price indices exhibit high skewness and dispersion during the energy crisis in 2022, gradually reverting to a more stable distribution in 2023 and 2024. (f)-(g) ID$_3$ trajectories in 2024 (range limited to [–500, 1000] €/MWh for better visual comparison). Volatility increases in the order: AT, 60-min $<$ DE, 60-min $<$ AT, 15-min $<$ DE, 15-min. (h)-(j) Distribution of absolute feature importance by feature type, look-back window size, and market side, respectively. (k) Testing loss (AQL) for different feature sets across countries and resolutions.
  • Figure 2: Analysis of model performance against market liquidity. (a) Comparison of market liquidity. (b) Loss ratio versus trade-count ratio.