arxiv: 2605.13446 · v1 · submitted 2026-05-13 · 📊 stat.AP

Recognition: 2 theorem links

· Lean Theorem

Scenario generation of intraday electricity price paths for optimal trading in continuous markets

Andrzej Pu\'c, Joanna Janczura

Pith reviewed 2026-05-14 18:31 UTC · model grok-4.3

classification 📊 stat.AP

keywords scenariotradingcontinuouselectricityintradaymarketspriceadaptive

0 comments

The pith

A kernel-based regression model plus scenario generation from forecast errors and a new Support Vector Sorting step produces ensemble price trajectories that improve both statistical accuracy and trading profits over benchmarks on German intraday continuous market data.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Electricity prices in continuous intraday markets move quickly as new trades arrive. The authors start with a support vector regression model that first corrects for known biases in price forecasts. They then create many possible future price paths by adding realistic error patterns drawn from how the model has missed on fundamental drivers such as load and renewable generation. A new sorting procedure called Support Vector Sorting picks a small number of these paths that still capture the main range of possible outcomes. These selected scenarios feed into trading strategies that can update positions as new information arrives. On transaction-level data from the German market the ensemble forecasts beat simple benchmarks in both how well they match actual prices and how much money the resulting trades make. Reweighting the scenarios as time passes further lifts profits while keeping downside risk roughly the same. The approach therefore links a flexible statistical learner to practical decision rules under uncertainty.

Core claim

Combining kernel-based learning with scenario driven uncertainty and adaptive updating provides a flexible and effective approach for forecasting and trading in continuous electricity markets.

Load-bearing premise

That forecast errors of fundamental variables can be used directly to generate scenarios whose statistical properties remain representative of future price uncertainty without additional calibration or regime detection.

Figures

Figures reproduced from arXiv: 2605.13446 by Andrzej Pu\'c, Joanna Janczura.

**Figure 1.** Figure 1: Synthetic example of actual and forecast trajectories for a fundamental variable in one training window sample. The [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗

**Figure 2.** Figure 2: Timeline of data availability for all exogenous variables used in the study. [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗

**Figure 3.** Figure 3: Example of the evolution of weighted median paths and bands under [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗

**Figure 4.** Figure 4: Pinball score (PB) averaged over days in test window, deliveries and path steps for different forecasting approaches. [PITH_FULL_IMAGE:figures/full_fig_p018_4.png] view at source ↗

read the original abstract

Continuous intraday electricity markets play an increasingly important role in short-term trading and balancing, yet decision-making under rapidly evolving price dynamics remains challenging. This paper proposes a comprehensive framework for ensemble forecasting of intraday electricity price trajectories and their translation into adaptive trading decisions. Building on a corrected Support Vector Regression model, the approach extends point predictions to probabilistic trajectory forecasts by introducing scenario generation based on forecast errors of fundamental variables and proposing a novel Support Vector Sorting procedure for the efficient selection of representative scenarios. The framework is evaluated using transaction level data from the German intraday continuous market. Empirical results show improvements over benchmark methods in both statistical and economic terms. Fundamental scenarios enhance median trajectory accuracy but produce more concentrated predictive distributions, while historical simulation with scenario selection better captures tail risk. From an economic perspective, ensemble-based forecasts outperform naive benchmarks across most of the trading strategies. Dynamic updating through scenario reweighting further improves profitability with limited impact on downside risk. Overall, the results demonstrate that combining kernel-based learning with scenario driven uncertainty and adaptive updating provides a flexible and effective approach for forecasting and trading in continuous electricity markets.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Paper introduces a practical SVR-based scenario method for intraday electricity trading with a new sorting step, but the robustness to regime shifts remains unproven.

read the letter

The core of this paper is a framework that takes corrected Support Vector Regression point forecasts for intraday prices, generates trajectory scenarios from the forecast errors of chosen fundamental variables, and then applies a new Support Vector Sorting procedure to pick a manageable set of representative paths. They test the whole thing on transaction-level data from the German continuous intraday market and report better median accuracy plus higher trading profits than simple benchmarks, with some extra lift from dynamic scenario reweighting. The linkage of fundamental errors to price uncertainty and the sorting routine for selection look like the genuinely new pieces relative to standard historical simulation or plain SVR work in this area. The end-to-end economic evaluation with actual trading strategies is also a plus, as it shows how the forecasts could be used in practice rather than stopping at statistical scores. The main soft spot is the untested assumption that the fitted error distributions stay representative across market regimes. Electricity prices often show abrupt spikes and shifts that are only loosely tied to the fundamentals they use, so reusing the same error stats risks understating tails, which matches the abstract's note that fundamental scenarios end up more concentrated than historical ones. The reported results also give no error bars, no explicit out-of-sample calibration checks, and limited detail on how post-selection affects the ensemble properties, so the claimed gains are hard to judge for stability. This is for people already working on short-term energy market models or real-time trading algorithms. A reader who needs concrete scenario sets for optimization or who wants to adapt kernel methods to power prices would pick up usable techniques and see how they perform in a live market setting. It is worth sending for peer review so referees can check the sorting algorithm and any hidden sensitivities in the scenario construction.

Referee Report

3 major / 2 minor

Summary. The paper proposes a framework for ensemble forecasting of intraday electricity price trajectories in continuous markets. It starts from a corrected Support Vector Regression point forecast, extends it to probabilistic paths via scenario generation driven by forecast errors of fundamental variables, and introduces a Support Vector Sorting procedure to select a compact set of representative scenarios. The approach is tested on transaction-level data from the German intraday continuous market, with claims of statistical gains in median accuracy and economic gains in trading profitability relative to benchmarks; dynamic reweighting of scenarios is reported to further improve results with limited downside impact.

Significance. If the empirical claims hold after addressing the validation gaps, the work supplies a practical, kernel-based route to scenario-driven trading decisions in fast-moving electricity markets. The combination of SVR point forecasts with fundamental-error scenarios and an explicit selection step offers a reproducible template that could be adopted by market participants; the real-data evaluation and the reported economic outperformance are the primary sources of value.

major comments (3)

[Abstract and §3] Abstract and §3 (scenario generation): the central premise that raw forecast errors of the chosen fundamentals remain distributionally representative of future price uncertainty is load-bearing yet untested for regime shifts; electricity prices exhibit spikes and mean-reversion changes only weakly linked to the fundamentals, so the generated ensembles may systematically understate tail risk without regime detection or recalibration.
[§4] §4 (empirical evaluation): no quantitative error bars, bootstrap intervals, or out-of-sample calibration diagnostics are supplied for the reported improvements in median trajectory accuracy or trading profitability; without these, it is impossible to judge whether the gains over benchmarks are statistically distinguishable from sampling variation.
[§3.2] §3.2 (Support Vector Sorting): the procedure re-uses the same training-period forecast-error distribution for scenario construction on held-out data, creating a modest circularity that is not quantified; the paper must demonstrate that the selected scenarios preserve tail properties out-of-sample rather than merely asserting economic value preservation.

minor comments (2)

[§3.2] Notation for the Support Vector Sorting algorithm (Algorithm 1) should be aligned with standard SVR notation to avoid confusion between the sorting kernel and the original regression kernel.
[Figures] Figure 4 (or equivalent) comparing scenario distributions would benefit from explicit overlay of empirical quantiles to make the concentration of fundamental scenarios versus historical simulation visually clearer.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below with clarifications and indicate revisions where the concerns can be directly addressed through additional analysis or discussion.

read point-by-point responses

Referee: [Abstract and §3] Abstract and §3 (scenario generation): the central premise that raw forecast errors of the chosen fundamentals remain distributionally representative of future price uncertainty is load-bearing yet untested for regime shifts; electricity prices exhibit spikes and mean-reversion changes only weakly linked to the fundamentals, so the generated ensembles may systematically understate tail risk without regime detection or recalibration.

Authors: We acknowledge that the distributional representativeness assumption is central to the scenario generation step. Our evaluation uses a multi-year German intraday dataset that includes multiple volatility regimes and price spikes, providing some empirical coverage of varying conditions. To directly address the concern, we will add a new subsection in the revised manuscript that splits the out-of-sample period by volatility quartiles and reports quantile coverage (including tails) separately for each sub-period. This will quantify any degradation in tail representation across regimes. revision: yes
Referee: [§4] §4 (empirical evaluation): no quantitative error bars, bootstrap intervals, or out-of-sample calibration diagnostics are supplied for the reported improvements in median trajectory accuracy or trading profitability; without these, it is impossible to judge whether the gains over benchmarks are statistically distinguishable from sampling variation.

Authors: We agree that statistical uncertainty measures are needed to substantiate the reported gains. In the revision we will add bootstrap confidence intervals (e.g., 1000 resamples) for all median accuracy and profitability metrics, together with out-of-sample calibration diagnostics such as reliability diagrams and PIT histograms for the ensemble forecasts. These additions will allow readers to assess whether improvements exceed sampling variation. revision: yes
Referee: [§3.2] §3.2 (Support Vector Sorting): the procedure re-uses the same training-period forecast-error distribution for scenario construction on held-out data, creating a modest circularity that is not quantified; the paper must demonstrate that the selected scenarios preserve tail properties out-of-sample rather than merely asserting economic value preservation.

Authors: The training-period error distribution is deliberately reused to avoid look-ahead bias when constructing scenarios for the test set; this is standard in scenario-generation literature. We will strengthen the presentation by adding explicit out-of-sample diagnostics: comparison of 95 % and 99 % quantile coverage and extreme-value statistics between the selected scenarios and realized prices on the held-out data. These results will be reported alongside the existing economic metrics. revision: yes

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 1 invented entities

The central claim rests on the assumption that forecast errors observed on fundamental variables remain statistically exchangeable with future errors and that a small selected subset of scenarios preserves the relevant risk measures for trading decisions.

free parameters (2)

SVR hyperparameters (C, epsilon, kernel parameters)
Standard SVR tuning parameters that must be chosen or cross-validated on the training data.
Number of scenarios retained after Support Vector Sorting
User-chosen integer that controls the trade-off between computational cost and coverage of the predictive distribution.

axioms (2)

domain assumption Forecast errors of fundamental variables are stationary and can be sampled to represent price-path uncertainty
Invoked when the paper states that scenarios are generated from forecast errors of fundamental variables.
ad hoc to paper The selected representative scenarios preserve the economic value of the full ensemble for trading decisions
Implicit in the claim that the novel sorting procedure yields efficient yet effective scenario sets.

invented entities (1)

Support Vector Sorting procedure no independent evidence
purpose: Efficient selection of representative scenarios from a larger ensemble
New algorithmic step introduced to reduce the number of scenarios while retaining key statistical properties.

pith-pipeline@v0.9.0 · 5489 in / 1568 out tokens · 25843 ms · 2026-05-14T18:31:22.732536+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
corrected kernel K(xi,xj)=exp(−l∥xi−xj∥)exp(−g∥ŷi−ŷj∥²) … scenario generation based on forecast errors of fundamental variables … Support Vector Sorting … Wasserstein distance
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear
ensemble path forecasting … fundamental scenarios … historical simulation

Reference graph

Works this paper leans on

26 extracted references · 24 canonical work pages · 1 internal anchor

[1]

doi: https://doi.org/10.1016/j.ijforecast.2014.08.013

ISSN 0169-2070. doi: https://doi.org/10.1016/j.ijforecast.2014.08.013. URLhttps: //www.sciencedirect.com/science/article/pii/S0169207014001423. Monica Borunda, Luis Conde-López, Gerardo Ruiz-Chavarría, Guadalupe Lopez Lopez, Victor M. Alvarado, and Edgardo de Jesús Carrera Avendaño. Advances in similar day methods for short-term load forecasting for power...

work page doi:10.1016/j.ijforecast.2014.08.013 2070
[2]

doi: 10.3390/forecast8020032

ISSN 2571-9394. doi: 10.3390/forecast8020032. URLhttps: //www.mdpi.com/2571-9394/8/2/32. Chih-Chung Chang and Chih-Jen Lin. Libsvm: A library for support vector machines.ACM Transactions on Intelligent Systems and Technology (TIST), 2(3),

work page doi:10.3390/forecast8020032
[3]

doi: 10.1145/1961189.1961199

ISSN 2157-6904. doi: 10.1145/1961189.1961199. URLhttps://doi.org/10.1145/1961189.1961199. Jieyu Chen, Sebastian Lerch, Melanie Schienle, Tomasz Serafin, and Rafał Weron. Probabilistic intraday electricity price forecasting using generative machine learning,

work page doi:10.1145/1961189.1961199
[4]

Harris Drucker, Christopher J

URLhttps://arxiv.org/abs/ 2506.00044. Harris Drucker, Christopher J. C. Burges, Linda Kaufman, Alex Smola, and Vladimir Vapnik. Support vector regression machines. In M.C. Mozer, M. Jordan, and T. Petsche, editors,Advances in Neural Information Processing Systems, volume

work page arXiv
[5]

URLhttps://proceedings.neurips.cc/ paper_files/paper/1996/file/d38901788c533e8286cb6400b40b386d-Paper.pdf. ENTSO-E. Transparency platform knowledge base, 2026a. URLhttps://transparencyplatform. zendesk.com/hc/en-us/categories/12818231533716-Knowledge-base. Accessed: 2026-04-06. ENTSO-E. Transparency platform, 2026b. URLhttps://transparency.entsoe.eu/. Acc...

work page arXiv 1996
[6]

European Energy Exchange.https://www.epexspot.com

EPEX SPOT. European Energy Exchange.https://www.epexspot.com. Accessed: 2021-10-25. EPEX SPOT SE. Trading at EPEX SPOT – trading brochure.https://www.epexspot.com/sites/ default/files/download_center_files/20-01-24_TradingBrochure.pdf,

2021
[7]

Gneiting and A

doi: 10.1198/016214506000001437. URL https://doi.org/10.1198/016214506000001437. Holger Heitsch and Werner Römisch. Scenario reduction algorithms in stochastic programming.Com- putational Optimization and Applications, 24(2):187–206,

work page doi:10.1198/016214506000001437
[8]

Random forests.Machine Learning, 45(1):5–32, 2001

ISSN 1573-2894. doi: 10.1023/A: 1021805924152. URLhttps://doi.org/10.1023/A:1021805924152. Simon Hirsch and Florian Ziel. Multivariate simulation-based forecasting for intraday power markets: Mod- eling cross-product price effects.Applied Stochastic Models in Business and Industry, 40(6):1571–1595, 2024a. doi: https://doi.org/10.1002/asmb.2837. URLhttps:/...

work page doi:10.1023/a:
[9]

Tim Janke and Florian Steinke

Accessed: 2025-12-27. Tim Janke and Florian Steinke. Forecasting the price distribution of continuous intraday electricity trad- ing.Energies, 12(22),

2025
[10]

doi: 10.3390/en12224262

ISSN 1996-1073. doi: 10.3390/en12224262. URLhttps://www.mdpi.com/ 1996-1073/12/22/4262. Christopher Kath and Florian Ziel. The value of forecasts: Quantifying the economic gains of accurate quarter-hourly electricity price forecasts.Energy Economics, 76:411–423,

work page doi:10.3390/en12224262 1996
[11]

doi: 10.1016/j.eneco.2018.10.005

ISSN 0140-9883. doi: 10.1016/j.eneco.2018.10.005. URLhttp://dx.doi.org/10.1016/j.eneco.2018.10.005. Grzegorz Marcjasz, Tomasz Serafin, and Rafał Weron. Selection of calibration windows for day-ahead electricity price forecasting.Energies, 11(9),

work page doi:10.1016/j.eneco.2018.10.005 2018
[12]

doi: 10.3390/en11092364

ISSN 1996-1073. doi: 10.3390/en11092364. URL https://www.mdpi.com/1996-1073/11/9/2364. Grzegorz Marcjasz, Bartosz Uniejewski, and Rafał Weron. Beating the naïve—combining LASSO with naïve intraday electricity price forecasts.Energies, 13(7),

work page doi:10.3390/en11092364 1996
[13]

doi: 10.3390/en13071667

ISSN 1996-1073. doi: 10.3390/en13071667. URL https://www.mdpi.com/1996-1073/13/7/1667. Vivek Mohan, Jai Govind Singh, and Weerakorn Ongsakul. Sortino ratio based portfolio optimization considering EVs and renewable energy in microgrid power market.IEEE Transactions on Sustainable Energy, 8(1):219–229,

work page doi:10.3390/en13071667 1996
[14]

Rudy Morel, Stéphane Mallat, and Jean-Philippe Bouchaud

doi: 10.1109/TSTE.2016.2593713. Rudy Morel, Stéphane Mallat, and Jean-Philippe Bouchaud. Path shadowing Monte Carlo.Quantitative Finance, 24(9):1199–1225,

work page doi:10.1109/tste.2016.2593713 2016
[15]

URLhttps://doi.org/10.1080/ 14697688.2024.2399285

doi: 10.1080/14697688.2024.2399285. URLhttps://doi.org/10.1080/ 14697688.2024.2399285. Michał Narajewski and Florian Ziel. Econometric modelling and forecasting of intraday electricity prices. Journal of Commodity Markets, 19:100107, 2020a. ISSN 2405-8513. doi: https://doi.org/10.1016/j.jcomm. 2019.100107. URLhttps://www.sciencedirect.com/science/article/...

work page doi:10.1080/14697688.2024.2399285 2024
[16]

JakubNowotarskiandRafałWeron

doi: https://doi.org/10.1016/j.apenergy.2024.124975. JakubNowotarskiandRafałWeron. Recentadvancesinelectricitypriceforecasting: Areviewofprobabilistic forecasting.Renewable and Sustainable Energy Reviews, 81:1548–1568,

work page doi:10.1016/j.apenergy.2024.124975 2024
[17]

doi: https://doi.org/10.1016/j.rser.2017.05.234

ISSN 1364-0321. doi: https://doi.org/10.1016/j.rser.2017.05.234. URLhttps://www.sciencedirect.com/science/article/ pii/S1364032117308808. F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duch- esnay. Scikit-le...

work page doi:10.1016/j.rser.2017.05.234 2017
[18]

doi: https://doi.org/10.1016/j.ijforecast.2025.11.007

ISSN 0169-2070. doi: https://doi.org/10.1016/j.ijforecast.2025.11.007. URLhttps://www.sciencedirect.com/science/ article/pii/S0169207025001098. 25 Jesse Read, Bernhard Pfahringer, Geoff Holmes, and Eibe Frank. Classifier chains for multi-label classifi- cation. In Wray Buntine, Marko Grobelnik, Dunja Mladenić, and John Shawe-Taylor, editors,Machine Learni...

work page doi:10.1016/j.ijforecast.2025.11.007 2070
[19]

arXiv preprint

URLhttps://arxiv.org/ abs/2510.15011. arXiv preprint. Tomasz Serafin, Grzegorz Marcjasz, and Rafał Weron. Trading on short-term path forecasts of intraday elec- tricity prices.Energy Economics, 112:106125,

work page arXiv
[20]

doi: https://doi.org/10.1016/j.eneco

ISSN 0140-9883. doi: https://doi.org/10.1016/j.eneco. 2022.106125. URLhttps://www.sciencedirect.com/science/article/pii/S014098832200281X. Anna Staszewska. Representing uncertainty about response paths: The use of heuristic optimisation methods.Computational Statistics&Data Analysis, 52(1):121–132,

work page doi:10.1016/j.eneco 2022
[21]

doi: https://doi.org/10.1016/j.csda.2006.12.023

ISSN 0167-9473. doi: https://doi.org/10.1016/j.csda.2006.12.023. URLhttps://www.sciencedirect.com/science/article/ pii/S0167947306005007. Mucun Sun, Cong Feng, and Jie Zhang. Probabilistic solar power forecasting based on weather scenario generation.Applied Energy, 266:114823,

work page doi:10.1016/j.csda.2006.12.023 2006
[22]

doi: https://doi.org/10.1016/j.apenergy

ISSN 0306-2619. doi: https://doi.org/10.1016/j.apenergy. 2020.114823. URLhttps://www.sciencedirect.com/science/article/pii/S0306261920303354. Léonard Tschora, Erwan Pierre, Marc Plantevit, and Céline Robardet. Electricity price forecasting on the day-ahead market using machine learning.Applied Energy, 313:118752,

work page doi:10.1016/j.apenergy 2020
[23]

doi: https: //doi.org/10.1016/j.apenergy.2022.118752

ISSN 0306-2619. doi: https: //doi.org/10.1016/j.apenergy.2022.118752. URLhttps://www.sciencedirect.com/science/article/ pii/S0306261922002057. Bartosz Uniejewski and Florian Ziel. Probabilistic forecasts of load, solar and wind for electricity price forecasting. arXiv preprint,

work page doi:10.1016/j.apenergy.2022.118752 2022
[24]

Bartosz Uniejewski, Grzegorz Marcjasz, and Rafał Weron

URLhttps://arxiv.org/abs/2501.06180. Bartosz Uniejewski, Grzegorz Marcjasz, and Rafał Weron. Understanding intraday electricity markets: Variable selection and very short-term price forecasting using LASSO.International Journal of Forecast- ing, 35(4):1533–1547,

work page arXiv
[25]

doi: https://doi.org/10.1016/j.ijforecast.2019.02.001

ISSN 0169-2070. doi: https://doi.org/10.1016/j.ijforecast.2019.02.001. URL https://www.sciencedirect.com/science/article/pii/S0169207019300123. Cédric Villani.Optimal Transport: Old and New, volume 338 ofGrundlehren der Mathematischen Wissenschaften. Springer-Verlag, Berlin, Heidelberg,

work page doi:10.1016/j.ijforecast.2019.02.001 2070
[26]

OrderFusion: Encoding Orderbook for End-to-End Probabilistic Intraday Electricity Price Forecasting

URLhttps://arxiv.org/abs/2502.06830. Appendix A. Additional figures 26 −10 −5 0 5 10 load [GW] −5.0 −2.5 0.0 2.5 5.0 solar [GW] 1 2 3 4 5 6 7 8 9 10 11 h −10 −5 0 5 10 wind [GW] 100 101 102 103 104 # (log scale) Figure A.5: Colormap representing the histograms of fundamental scenarios∆(i) fs for each quarter-hourly step in a path. Note that, due to differ...

work page internal anchor Pith review Pith/arXiv arXiv 2000