Paper: arXiv 2511.13616

Authors: Katarzyna Maciejowska, Arkadiusz Lipiecki, Bartosz Uniejewski

Abstract

In recent years, a rapid development of forecasting methods has led to an increase in the accuracy of predictions. In the literature, forecasts are typically evaluated using metrics such as Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). While appropriate for statistical assessment, these measures do not adequately reflect the economic value of forecasts. This study addresses the decision-making problem faced by a battery energy storage system, which must determine optimal charging and discharging times based on day-ahead electricity price forecasts. To explore the relationship between forecast accuracy and economic value, we generate a pool of 192 forecasts. These are evaluated using seven statistical metrics that go beyond RMSE and MAE, capturing various characteristics of the predictions and associated errors. We calculate the dynamic correlation between the statistical measures and gained profits to reveal that both RMSE and MAE are only weakly correlated with revenue. In contrast, measures that assess the alignment between predicted and actual daily price curves have a stronger relationship with profitability and are thus more effective for selecting optimal forecasts.

Complexity vs Empirical Score

  • Math Complexity: 5.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced statistical and econometric methods, including complex forecasting models (ARX, NARX, LEAR) and a custom battery trading simulation, requiring significant data processing and implementation. However, it lacks explicit code or algorithmic details in the excerpt, focusing on empirical results and methodology rather than mathematical derivations.

Research Flowchart

  flowchart TD
  A["Research Goal: Assess the economic value<br>of electricity price forecasts<br>for battery storage operation"] --> B{"Methodology"}
  
  B --> C["Input: 192 diverse day-ahead<br>electricity price forecasts"]
  B --> D["Statistical Evaluation:<br>7 metrics beyond RMSE/MAE<br>e.g., MAPE, Correlation"]
  
  C --> E["Simulation: Battery Storage<br>Optimization based on forecasts"]
  D --> F["Comparison: Dynamic correlation<br>between statistical metrics<br>and actual profits"]
  
  E --> G["Outcome: Calculated revenue<br>for each forecast"]
  G --> F
  
  F --> H{"Key Findings"}
  H --> I["RMSE & MAE show<br>weak correlation with revenue"]
  H --> J["Measures of curve alignment<br>strongly predict profitability"]
  H --> K["Best practice: Select forecasts<br>based on economic utility, not<br>just statistical error"]

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 two new measures describing dispersion of errors and association of forecasts to capture economic value for BESS arbitrage.

Source sentence
We propose and examine two measures describing dispersion of errors and association of forecasts, which reflect statistical properties of the electricity price forecasts that are particularly relevant for the BESS decision-making process.

Assumes 6

  • The battery completes one full charge–discharge cycle per day and does not transfer energy between consecutive days. execution
    Source sentence
    Specifically, we assume that the battery completes one full charge–discharge cycle per day and does not transfer energy between consecutive days.
  • No additional constraints on the minimum or maximum state of charge are imposed. execution
    Source sentence
    Furthermore, no additional constraints on the minimum or maximum state of charge are imposed, although such limits are typically required in real‑world applications to extend battery lifetime.
  • Battery degradation is not modeled explicitly; it is indirectly reflected through reduced operating costs. execution
    Source sentence
    Finally, given the relatively short evaluation period considered in this study, battery degradation is not modeled explicitly within the decision problem; instead, it is indirectly reflected through reduc…
  • BESS earns income solely from day‑ahead market arbitrage, buying low and selling high. market
    Source sentence
    It is assumed that BESS earns income from the arbitrage in DA market. BESS aims at buying energy in periods of low prices and selling at hours of high prices.
Show 2 more
  • Operating costs are a levelled cost per unit of energy throughput, incorporating investment, O&M, and degradation. model
    Source sentence
    The operating costs and battery efficiency are taken from [6] and represent the levelled cost per unit of energy throughput. This takes into account investment and O&M costs, as well as implicitly incorporating battery degradation through the assumed number of charge–discharge cycles over the lifetime of the battery.
  • The BESS specifications (energy capacity 3 MWh, power ratings 3 MW and 1 MW) are representative of grid‑scale lithium‑ion batteries. model
    Source sentence
    Both batteries are assumed to have the same energy capacity of 3 MWh. They differ in terms of the power ratings that are 3 MW and 1 MW, for BESS‑a and BESS‑b respectively.

Data

Backtest sample: 2016-01 – 2024-12 · Germany–Luxembourg day-ahead electricity market · other · intraday

  • German‑Luxembourg day‑ahead electricity price · ENTSO‑E transparency platform market-prices public
    2016-01-02 to 2024-12-31 · Germany–Luxembourg bidding zone (Austria included until Oct 2018)
    Source sentence
    The core series is the hourly day‑ahead price for the Germany–Luxembourg bidding zone (Austria included until October 2018), sourced from the ENTSO‑E transparency platform.
  • System load, solar and wind forecasts · ENTSO‑E fundamentals public
    2016-01-02 to 2024-12-31
    Source sentence
    We add ENTSO‑E day‑ahead forecasts of system load, solar generation, and aggregated wind generation (onshore and offshore), with solar and wind combined into a single renewable generation series.
  • Commodity futures (coal, gas, oil, EUA) · Investing.com macro public
    2016-01-02 to 2024-12-31
    Source sentence
    Broader market influences are represented by commodity futures prices for coal (API2), natural gas (TTF), crude oil (Brent), and EU emission allowances (EUA), obtained from Investing.com.

Costs

Modeled Costs are part of the reported headline results.

Operating costs are set at 11.63 EUR/MWh, representing levelled cost per unit of energy throughput.

Source sentence
... are taken from [6] and represent the levelled cost per unit of energy throughput.

Finds 5

  • Traditional accuracy metrics (RMSE, MAE) are only weakly correlated with BESS profits. performance
    Source sentence
    We provide evidence that classical forecast accuracy measures, such as the RMSE or the MAE, are only weakly correlated with BESS profits.
  • Dispersion‑ and association‑based measures (Cov‑e, Corr‑f) show a strong correlation with BESS profitability. performance
    Source sentence
    In particular, measures capturing intraday error dispersion (Cov‑e) and the association between predicted and realized prices (Corr‑f) exhibit a strong relationship with BESS profitability.
  • Choosing the best‑performing forecasting model can increase profits by 16–58% over 2021–2024 compared to the poorest model. performance
    Source sentence
    Selecting the best‑performing model can increase profits by 16–58% during 2021–2024 compared to the poorest‑performing model.
  • NARX delivers the highest statistical accuracy, while LEAR‑based forecasts generate the highest profits. predictive-accuracy
    Source sentence
    Among the analyzed model classes, NARX delivers the highest statistical accuracy, while LEAR‑based forecasts generate the highest profits.
  • The pooled estimator is an attractive alternative to the commonly used heterogeneous approach. statistical
    Source sentence
    With respect to model specification, the pooled estimator emerges as an attractive alternative to the commonly used heterogeneous approach.

Checked by others

Contradicted by

  • Improving Algorithmic Trading for BESS on Intraday Electricity Markets through Distributional Price Forecasts (2026)
    On the question: How does forecast accuracy impact BESS profitability in intraday electricity markets?
    Compare the findings

    This paper: Traditional accuracy metrics (RMSE, MAE) are only weakly correlated with BESS profits.

    That paper: Forecast errors and instability are strongly negatively correlated with daily profits, indicating predictive accuracy drives returns.

    A finds a weak correlation between RMSE/MAE and profits, while B finds a strong negative correlation between forecast errors/instability and daily profits.

Cited by 5 papers in the archive (4 by independent authors): Improving Algorithmic Trading for BESS on Intraday … (discusses its findings); Probabilistic Forecasting for Day-ahead Electricity Prices, … (extends it); Deep Learning for Electricity Price Forecasting: A Review of … (uses its method or data)

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (47)

  1. Electricity price forecasting on the day-ahead market using machine learning (2022)
  2. Electricity Price Forecasting: The Dawn of Machine Learning (2022)
  3. Multi-service based economic valuation of grid-connected battery energy storage systems (2022)
  4. On the participation of energy storage systems in reserve markets using Decision Focused Learning (2025)
  5. Decision-Focused Learning Enhanced by Automated Feature Engineering for Energy Storage Optimisation (2026)
  6. Monetizing energy storage (2023)
  7. Prescriptive Trees for Integrated Forecasting and Optimization Applied in Trading of Renewable Energy (2022)
  8. Economic Impact of Electricity Market Price Forecasting Errors: A Demand-Side Analysis (2010)
and 39 more
  1. Quantifying the value of probabilistic forecasts when trading renewable hybrid power parks in day-ahead markets: A Nordic case study (2024)
  2. An Integrated Approach for Value-Oriented Energy Forecasting and Data-Driven Decision-Making Application to Renewable Energy Trading (2019)
  3. Short-Term Load Forecasting With a New Nonsymmetric Penalty Function (2011)
  4. Toward Cost-Oriented Forecasting of Wind Power Generation (2018)
  5. Cost-oriented load forecasting (2022)
  6. Loss functions in regression models: Impact on profits and risk in day-ahead electricity trading (2025)
  7. Toward Value-Oriented Renewable Energy Forecasting: An Iterative Learning Approach (2023)
  8. Decision-focused learning: foundations, state of the art, benchmark and future opportunities (2024)
  9. What Is a Good Forecast? An Essay on the Nature of Goodness in Weather Forecasting (1993)
  10. Day-ahead electricity price forecasting with high-dimensional structures: U nivariate vs. multivariate modeling frameworks (2018)
  11. Forecasting day-ahead electricity prices: A review of state-of-the-art algorithms, best practices and an open-access benchmark (2021)
  12. Forecasting Electricity Prices (2023)
  13. Distributional neural networks for electricity price forecasting (2023)
  14. On the importance of the long-term seasonal component in day-ahead electricity price forecasting with NARX neural networks (2019)
  15. A note on averaging day-ahead electricity price forecasts across calibration windows (2019)
  16. Probabilistic electricity price forecasting with NARX networks: Combine point or probabilistic forecasts? (2020)
  17. Training feedforward networks with the Marquardt algorithm (1994)
  18. Regression shrinkage and selection via the lasso (1996)
  19. Regularization for electricity price forecasting — → 2404.03968
  20. Selection of Calibration Windows for Day-Ahead Electricity Price Forecasting (2018)
  21. Averaging predictive distributions across calibration windows for day-ahead electricity price forecasting (2019)
  22. Forecasting I talian electricity zonal prices with exogenous variables (2012)
  23. Do jumps and cojumps matter for electricity price forecasting? Evidence from the German-Austrian day-ahead market (2022)
  24. Identifying spikes and seasonal components in electricity spot price data: A guide to robust modeling (2013)
  25. Variance stabilizing transformations for electricity spot price forecasting (2018)
  26. Principles of Forecasting: A handbook for researchers and practitioners (2001)
  27. Forecast combinations (2006)
  28. Why does forecast combination work so well? (2020)
  29. Probabilistic forecasting of electricity spot prices using F actor Q uantile R egression A veraging (2016)
  30. Postprocessing of point predictions for probabilistic forecasting of day-ahead electricity prices: The benefits of using isotonic distributional regression — → 2404.02270
  31. Daily peak electrical load forecasting with a multi-resolution approach (2023)
  32. Short-term electricity prices forecasting in a competitive market by a hybrid intelligent approach (2011)
  33. Electricity prices forecasting by a hybrid evolutionary-adaptive methodology (2014)
  34. Price forecasting of day-ahead electricity markets using a hybrid forecast method (2011)
  35. Deterministic and probabilistic interval prediction for short-term wind power generation based on variational mode decomposition and machine learning methods (2016)
  36. Short-term scheduling of gas-fired CHP plant with thermal storage using optimization algorithm and forecasting models (2021)
  37. AI agents envisioning the future: Forecast-based operation of renewable energy storage systems using hydrogen with Deep Reinforcement Learning (2022)
  38. Adaptive optimal operation of grid-connected battery systems under varying electricity market volatility (2026)
  39. The value of electricity storage arbitrage on day-ahead markets across E urope (2023)

2 of 47 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.