acceptodds
Withdrawn from ICLR 2027

Regime-Aware Cross-Market Electricity Price Forecasting with TimesFM and Large Language Models

Withdrawn from ICLR 2027, so left out of the paper lists and the map.

Abstract

Electricity price forecasting is challenging because electricity markets are influenced by complex and rapidly changing interactions among demand, renewable generation, weather conditions, transmission constraints, market rules, and operational events. Time-series foundation models such as TimesFM provide a promising backbone for modeling numerical temporal dynamics, but they primarily rely on historical observations and may struggle when market regimes change. Large language models offer complementary capabilities for interpreting unstructured information, including regulatory updates, market announcements, operational reports, and extreme-event descriptions, but they are not designed to directly produce calibrated numerical forecasts. We propose a regime-aware cross-market electricity price forecasting framework that combines TimesFM with large language models. TimesFM models the temporal dynamics of electricity prices and related numerical variables, while a large language model converts time-stamped textual information into structured market-regime representations. These representations are integrated through a lightweight market-adaptation module, allowing a shared forecasting backbone to capture common patterns across markets while adapting to market-specific characteristics. The framework further supports probabilistic forecasting and explicitly models price spikes and uncertainty associated with regime transitions. We design a leakage-controlled evaluation protocol and conduct cross-market transfer experiments to assess the model’s robustness under distribution shifts and limited data availability. The study evaluates forecasting accuracy, probabilistic calibration, extreme-price prediction, transferability, and computational efficiency, providing a systematic investigation of how language-derived market knowledge can enhance time-series foundation models for electricity price forecasting.

Primary area:
transfer learning, meta learning, and lifelong learning
Keywords:
Electricity Price Forecasting, Time-Series Foundation Models, TimesFM, Large Language Models, Cross-Market Transfer Learning

Reviews

Among:
Shape:
1234this paper, Bottom 2%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.33 over 3 reviews, above 2% of the others.
Reviewer_uNaCOpenReview ↗
Rating
1
Confidence
4
Informativeness
2
Presentation
2
Soundness
1
Read review
Paper summary

This paper introduces REGIME-TIMESFM, a method that combines TimesFM’s numerical forecasts with features—such as market events and policy states—extracted by an LLM from official documents, refining the predictions via a Ridge residual adapter incorporating interaction terms. The approach also utilizes residual quantiles to generate prediction intervals and employs an auxiliary classifier to estimate extreme price risk. Experiments primarily cover 1-, 6-, and 24-hour forecasts for the Guangdong and Jiangsu markets, comparing the original TimesFM against numerical adapters and language-conditioned adapters, while analyzing various market states and text alignment methods. The authors report that the benefits of language features are conditional and note the withdrawal of external test results for Sichuan due to timestamp-related issues.

Critical strengths

The authors clearly distinguish between numerical prediction and text information extraction, retaining extraction failures as missing information rather than defaulting to "no event." The decision to retract scores after discovering timestamp issues in the Sichuan test also demonstrates a prudent approach to handling the boundaries of evidence. While these design choices are sound, they do not yet fully compensate for the shortcomings in the current core validation.

Critical weaknesses

In Table 3, compared to the numerical adapter, the language-conditioned model shows a performance degradation of 0.23 and parity at the three Guangdong forecast horizons; for Jiangsu, it degrades by 0.58 at the 1-hour horizon and improves by only 0.08 at the 6-hour horizon. The most notable improvement occurs at the Jiangsu 24-hour horizon, where the MAE drops from 42.04 to 34.33; however, the original TimesFM achieves 33.18, outperforming it. Thus, these results primarily demonstrate that language features mitigate the issue of a degraded adapter, rather than proving superiority over currently available numerical forecasting methods. More critically, the "strict_observed_future_text_prohibited" control in Figure 5 yields MAEs of 42.04 (Jiangsu 24-hour) and 31.71 (Jiangsu 6-hour), neither of which outperforms the corresponding numerical adapters in Table 3. Conversely, the "correct_aligned_text" setting in Figure 5 aligns with the main table's language-based results, showing MAEs of 34.33 and 31.38, respectively. The paper fails to clearly explain the relationship between these two alignment methods or to establish evidence that the primary gains persist under the strict protocol. Given that preventing temporal leakage is a key contribution, this issue directly impacts the paper's conclusions rather than serving merely as a secondary sensitivity analysis. Furthermore, the text controls do not convincingly demonstrate the superiority of "structured linguistic knowledge" over simple information; the document count control for Jiangsu 6-hour (31.23) outperforms the full language model (31.38), and the vocabulary control for Jiangsu 24-hour (33.72) also outperforms the full language model (34.33). The paper highlights the full model's superiority over certain controls while failing to adequately explain why it underperforms against other, simpler controls. Consequently, the current results make it difficult to disentangle the value of semantic content from the effects of document frequency, lexical characteristics, or adapter design. Time-MMD has already aligned external reports and event texts with numerical sequences, employing LLMs for filtering, fact extraction, and summarization, while providing an evaluation framework that integrates language models with numerical forecasting models. Consequently, there are already direct precedents for "augmenting numerical forecasting with external text" and "text-numerical temporal alignment." The added value of the present paper must therefore stem from a reliable representation of market states or more rigorous evidence of validity—aspects that the current experiments have not yet sufficiently established.

Both Table 5 and Figure 4 report the conditional MAE for the language model minus the numerical adapter, yet the values ​​corresponding to the same market, time horizon, and stratification differ. These discrepancies far exceed the limits of display precision, and the paper offers no explanation attributing them to different data splits, memory windows, or experimental configurations. Conditional stratification underpins the core conclusion regarding "when text is useful"; until these values ​​are reconciled, that conclusion cannot be reliably verified. The "prior-donor" control values ​​reported in Section 6.4 also fail to match the corresponding controls shown in Figure 5, further raising concerns about the inconsistent use of result versions. A line-by-line reconciliation of experimental records is required here, rather than merely revising figure captions. While the title and abstract emphasize cross-market prediction—and Section 5.2 claims to evaluate adaptation to target markets over horizons of 0, 1, 7, and 30 days—the current manuscript presents no results, adaptation curves, or comparative analyses regarding transfer from source to target markets. Local prediction results for the two markets cannot substitute for transfer experiments.

Questions for authors

see weaknesses

Reviewer_xHNoOpenReview ↗
Rating
2
Confidence
3
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

This paper proposes Regime-TimesFM, a framework for electricity price forecasting that combines TimesFM with LLM-extracted market information. The LLM extracts structured features from official documents, and a Ridge regression adapter uses these features to correct the TimesFM predictions. The authors evaluate the method on electricity markets in Guangdong and Jiangsu, with additional analysis of different market regimes and text availability.

Critical strengths
  1. The paper studies an interesting problem: whether external text information can improve electricity price forecasting. The use of document timestamps to avoid information leakage is important for this task.

  2. The authors provide several analyses of when text information is useful. The results also show that text does not always help, which is a useful observation.

Critical weaknesses
  1. Limited novelty. The proposed method mainly combines TimesFM, LLM-extracted text features, and a Ridge regression adapter. Similar approaches using LLM-derived news information for electricity price forecasting have already been studied. The paper also does not explicitly model regime transitions or regime-dependent dynamics. I think the technical contribution is limited. The related work on text-enhanced time series forecasting is also not sufficient.

  2. Limited experimental evidence. The experiments mainly compare different variants of TimesFM on two electricity markets. There are no strong electricity-specific or existing text-enhanced forecasting baselines. More importantly, the LLM-conditioned model does not consistently outperform the original TimesFM. The external evaluation is also incomplete due to timestamp issues. These results make it difficult to judge the effectiveness and generalization of the proposed method.

Questions for authors
  1. What is the main technical difference between this method and existing approaches that use LLM-extracted news features for electricity price forecasting?

  2. Can the authors compare with electricity-specific forecasting models or existing text-enhanced methods? Also, why does the LLM-conditioned model fail to improve over the original TimesFM in most settings?

Reviewer_Hm4UOpenReview ↗
Rating
1
Confidence
3
Informativeness
1
Presentation
2
Soundness
1
Read review
Paper summary

This paper combines TimesFM with LLM-extracted market information through a lightweight residual adapter for electricity price forecasting. The division of responsibilities is sensible, and the attention to document timing, failed extractions, and transparent reporting of an invalid external evaluation is valuable. The authors also appropriately acknowledge that language features provide mixed benefits.

Critical strengths
  1. The paper uses LLM to extract contextual information while TimesFM handles numerical forecasting, giving each component a clear role.

  2. The evaluation uses realistic data and considers document availability, delayed physical covariates, chronological splits, and separate calibration data. Comparisons across horizons, regimes, and text representations help investigate when language information contributes.

Critical weaknesses
  1. The key weakness is the novelty. The paper is more about using the LLM in the electricity market. The technical contribution is limited.

  2. The results are not promising. In Table 3, the language-conditioned model never beats the strongest numerical alternative. For Jiangsu at 24 hours, its MAE improves over the numerical adapter from 42.04 to 34.33, but plain TimesFM achieves 33.18. Figure 5 reports Jiangsu’s 24-hour MAE returning to 42.04 under strict observed-time alignment, leaving the practical availability of the useful text signal uncertain.

  3. Table 5 and Figure 4 give different values for matching regime comparisons. Please explain.

Questions for authors

Please address the weaknesses.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Discussion (0)

Sign in to comment.