Paper: arXiv 2504.02518
Authors: Simon Hirsch
Abstract
Probabilistic electricity price forecasting (PEPF) is vital for short-term electricity markets, yet the multivariate nature of day-ahead prices - spanning 24 consecutive hours - remains underexplored. At the same time, real-time decision-making requires methods that are both accurate and fast. We introduce an online algorithm for multivariate distributional regression models, allowing an efficient modelling of the conditional means, variances, and dependence structures of electricity prices. The approach combines multivariate distributional regression with online coordinate descent and LASSO-type regularization, enabling scalable estimation in high-dimensional covariate spaces. Additionally, we propose a regularized estimation path over increasingly complex dependence structures, allowing for early stopping and avoiding overfitting. In a case study of the German day-ahead market, our method outperforms a wide range of benchmarks, showing that modeling dependence improves both calibration and predictive accuracy. Furthermore, we analyse the trade-off between predictive accuracy and computational costs for batch and online estimation and provide an high-performing open-source Python implementation in the ondil package.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 7.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced multivariate distributional regression with LASSO regularization and online coordinate descent, indicating high mathematical sophistication; it includes a comprehensive case study with benchmark comparisons, open-source implementation, and detailed computational cost analysis, demonstrating strong empirical rigor.
Research Flowchart
flowchart TD
A["Research Goal:<br>High-dimensional Probabilistic<br>Electricity Price Forecasting"] --> B["Methodology:<br>Online Multivariate<br>Distributional Regression"]
B --> C{"Data & Inputs:<br>German Day-Ahead Market<br>24-hour price vectors + Covariates"}
C --> D["Computational Process:<br>Online Coordinate Descent<br>+ LASSO Regularization"]
D --> E["Regularization Path:<br>Iterative Complexity<br>Selection & Early Stopping"]
E --> F["Key Outcomes:<br>• Superior Accuracy<br>• Improved Calibration<br>• Efficient Computation<br>• Open-source Python package"]
F --> G["Conclusion:<br>Modeling dependence<br>enhances PEPF performance"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes an online algorithm for multivariate distributional regression models with regularization to efficiently model conditional means, variances, and dependence structures of electricity prices.
Source sentence
We introduce an online algorithm for multivariate distributional regression models, allowing efficient modeling of the conditional means, variances, and dependence structures of electricity prices.
Assumes 5
- The scale matrix must be positive definite. model
Source sentence
To ensure the positive definiteness of the scale matrix, we propose three unconstrained parameterizations.
- Online coordinate descent introduces an approximation error that can be controlled by a forget parameter γ. execution
Source sentence
This introduces an approximation error for the online case, which can be controlled by the forget parameter γ as shown in Hirsch et al. (2024).
- Prices are pre‑processed using incremental mean‑variance scaling. data
Source sentence
We use an incremental mean-variance scaling to pre-process the electricity prices.
- The dependence structure can be captured by a Gaussian copula assuming normality of the transformed variables. model
Source sentence
We employ a second model, where we sparsify the estimated dependence matrix of the Gaussian copula using the graphical LASSO.
Show 1 more
- Online learning updates the model using only the new observation. execution
Source sentence
in the online learning scheme, we only use the new observation for updating the model
Data
- German day-ahead market price dataset market-prices proprietary
Source sentence
We use the same data set as in Lipiecki et al. (2024), which consists of electricity prices for the German day-ahead market from 2015-01-01 to 2024-01-01 exclusive.
Costs
Not applicable
Finds 5
- Modeling the dependence structure significantly improves probabilistic forecasting performance. performance
Source sentence
Our results show that modeling the dependence structure in the day-ahead market improves probabilistic forecasting performance significantly.
- Calibration of 24‑dimensional joint prediction bands is markedly improved by modeling dependence. pricing-calibration
Source sentence
calibration for prediction bands of the 24-dimensional price path is improved significantly by modeling the dependence structure
- Online learning yields speed‑ups of 2‑3 orders of magnitude compared to repeated batch fitting. performance
Source sentence
online learning, compared to repeated batch fitting, yields better results for a given computation budget – with speed-ups of 2-3 orders of magnitude.
- Regularization is crucial to avoid overfitting in high‑dimensional settings. statistical
Source sentence
we like to highlight the importance of regularization to avoid overfitting in a high-dimensional setting
- The online multivariate distributional regression models achieve strong predictive accuracy across multiple multivariate scoring rules. predictive-accuracy
Source sentence
The online, multivariate distributional regression models deliver strong predictive accuracy across a range of multivariate scoring rules.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 29 related papers). Coverage is partial, so this is neither replication nor refutation.
1 paper in the archive
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (104)
- Covariance structures for high-dimensional energy forecasting (2022)
- Ante-dependence analysis of an ordered set of variables (1962)
- Scalable estimation for structured additive distributional regression (2025)
- Distribution of the estimators for autoregressive time series with a unit root (1979)
- Lag order and critical values of the augmented Dickey–Fuller test (1995)
- Probabilistic intraday electricity price forecasting using generative machine learning (2025) — link
- Asymmetric penalties underlie proper loss functions in probabilistic forecasting (2025) — link
- Note on a method for calculating corrected sums of squares and products (1962)
and 96 more
- bashtage/arch: Release 7.2 (2024) — link
- Day-ahead electricity price forecasting with high-dimensional structures: Univariate vs. multivariate modeling frameworks (2018)
- On the choice of initial guesses for the Newton-Raphson algorithm (2021)
- Choosing starting values for certain Newton–Raphson iterations (2006)
- Bootstrap prediction bands for forecast paths from vector autoregressive models (2011)
- Comparison of methods for constructing joint confidence bands for impulse response functions (2015)
- Trading on short-term path forecasts of intraday electricity prices (2022)
- Hedging renewable power purchase agreements (2024)
- The value of coordination in multimarket bidding of grid energy storage (2023)
- Sparse inverse covariance estimation with the graphical lasso (2008)
- State of the German short-term power market (2017)
- Forecasting day-ahead electricity prices: A review of state-of-the-art algorithms, best practices and an open-access benchmark (2021)
- Structured antedependence models for longitudinal data (1997)
- Computational aspects of likelihood-based estimation of first-order antedependence models (1998)
- LASSO-type penalization in the framework of generalized additive models for location, scale and shape (2019)
- Incorporating second-order functional knowledge for better option pricing (2000)
- Activation functions in deep learning: A comprehensive survey and benchmark (2022)
- Noncanonical links in generalized linear models–when is the effort justified? (2000)
- Using the softplus function to construct alternative link functions in generalized linear models and beyond (2024)
- Additive covariance matrix models: modeling regional electricity net-demand in Great Britain (2025)
- Scalable Fitting Methods for Multivariate Gaussian Additive Models with Covariate-dependent Covariance Matrices (2025) — link
- Online Distributional Regression (2024) — link
- Scikit-learn: Machine learning in Python (2011)
- Numba: A llvm-based python jit compiler (2015)
- gamlss. lasso: Extra Lasso-Type Additive Terms for GAMLSS (2021)
- Probabilistic forecasting in day-ahead electricity markets: Simulating peak and off-peak prices (2020)
- Covariance Estimation: The GLM and Regularization Perspectives (2011)
- Numerical analysis for statisticians (2010)
- Generalized additive models for location, scale and shape (2005)
- Probabilistic multivariate electricity price forecasting using implicit generative ensemble post-processing (2020)
- Multivariate probabilistic forecasting of electricity prices with trading applications (2025)
- Multiple split approach–multidimensional probabilistic forecasting of electricity markets — → 2407.07795
- Bidding and scheduling in energy markets: Which probabilistic forecast do we need? (2022)
- The value of probabilistic forecasts for electricity market bidding and scheduling under uncertainty (2024)
- Semi-structured distributional regression (2024)
- Deep distributional time series models and the probabilistic forecasting of intraday electricity prices (2023)
- Marginally calibrated deep distributional regression (2021)
- Forecasting electricity prices with expert, linear, and nonlinear models (2023)
- M5 competition uncertainty: Overdispersion, distributional forecasting, GAMLSS, and beyond (2022)
- Ensemble forecasting for intraday electricity prices: Simulating trajectories (2020)
- Online adaptive estimation of sparse signals: Where RLS meets the $\ell_1$-norm (2010)
- Iteratively reweighted adaptive lasso for conditional heteroscedastic time series with applications to ar–arch type processes (2016)
- Adaptive Covariance Matrix Estimation through block thresholding (2012)
- On-line estimation with the multivariate Gaussian distribution (2007)
- Online mixed missing value imputation using gaussian copula (2020)
- Probabilistic price forecasts for short-term trade optimization (2024) — link
- Bayesian copulae distributions, with application to operational risk management—some comments (2013)
- Regions of reliability in the evaluation of multivariate probabilistic forecasts (2023)
- Multivariate forecasting evaluation: On sensitive and strictly proper scoring rules (2019) — link
- Strictly proper scoring rules, prediction, and estimation (2007)
- Estimation of the continuous ranked probability score with limited information and applications to ensemble weather forecasts (2018)
- Generalized additive models for location, scale and shape: a distributional regression approach, with applications (2024)
- Distributional regression for data analysis (2024)
- Scoringrules: a python library for probabilistic forecast evaluation (2024) — link
- Comparing predictive accuracy, twenty years later: A personal perspective on the use and abuse of Diebold–Mariano tests (2015)
- Comparing predictive accuracy (2002)
- Discrimination ability of the energy score (2013)
- Variogram-based proper scoring rules for probabilistic forecasts of multivariate quantities (2015)
- Evaluating the discrimination ability of proper multi-variate scoring rules (2024)
- Coherent dispersion criteria for optimal experimental design (1999)
- Probabilistic forecasts, calibration and sharpness (2007)
- Online coordinate descent for adaptive estimation of sparse signals (2009)
- Adaptive generalized logit-normal distributions for wind power short-term forecasting (2021)
- Online adaptive lasso estimation in vector autoregressive models for high dimensional wind power forecasting (2019)
- Conformal prediction interval estimation and applications to day-ahead and intraday power markets (2021)
- Modeling multivariate intraday forecast update processes for wind power (2024)
- Recent developments in multivariate wind and solar power forecasting (2022) — link
- Distributional neural networks for electricity price forecasting (2023)
- NBMLSS: probabilistic forecasting of electricity prices via Neural Basis Models for Location Scale and Shape (2024) — link
- On-line conformalized neural networks ensembles for probabilistic forecasting of day-ahead electricity prices (2025)
- Adaptive conformal inference under distribution shift (2021)
- Conformal inference for online prediction with arbitrary distribution shifts (2024)
- Adaptive conformal predictions for time series (2022)
- Postprocessing of point predictions for probabilistic forecasting of day-ahead electricity prices: The benefits of using isotonic distributional regression — → 2404.02270
- An introduction to multivariate probabilistic forecast evaluation (2021)
- Assessing the performance of deep learning models for multivariate probabilistic energy forecasting (2021)
- Probabilistic Multivariate Time Series Forecasting and Robust Uncertainty Quantification with Applications in Electricity Price Prediction (2023)
- Rage against the mean–a review of distributional regression approaches (2023)
- Distributional modeling and short-term forecasting of electricity prices by generalized additive models for location, scale and shape (2011)
- From point forecasts to multivariate probabilistic forecasts: The Schaake shuffle for day-ahead electricity price forecasting (2023)
- Multivariate probabilistic crps learning with an application to day-ahead electricity prices — → 2303.10019
- Recent advances in electricity price forecasting: A review of probabilistic forecasting (2018)
- Iteratively reweighted least squares for maximum likelihood estimation, and some robust and resistant alternatives (1984)
- Smoothing reference centile curves: the LMS method and penalized likelihood (1992)
- Probabilistic Prediction of Oceanographic Velocities with Multivariate Gaussian Natural Gradient Boosting (2023)
- XGBoostLSS–An extension of XGBoost to probabilistic forecasting (2019) — link
- High-dimensional multivariate forecasting with low-rank gaussian copula processes (2019)
- Distributional random forests: Heterogeneity adjustment and multivariate distributional regression (2022)
- Multi-target xgboostlss regression (2022) — link
- Variable selection using a smooth information criterion for distributional regression models (2023)
- Robust fitting for generalized additive models for location, scale and shape (2021)
- Truly multivariate structured additive distributional regression (2025)
- Cholesky-based multivariate Gaussian regression (2022)
- Online missing value imputation and change point detection with the gaussian copula (2022)
- Online heterogeneous mixture modeling with marginal and copula selection (2011)
- A streaming algorithm for bivariate empirical copulas (2019)
3 of 104 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.