Paper: arXiv 2412.00062
Authors: Xuesong Wang, Sharaf K. Magableh, Oraib Dawaghreh, Caisheng Wang, Jiaxuan Gong, Zhongyang Zhao, Michael H. Liao
Abstract
Virtual bidding plays an important role in two-settlement electric power markets, as it can reduce discrepancies between day-ahead and real-time markets. Renewable energy penetration increases volatility in electricity prices, making accurate forecasting critical for virtual bidders, reducing uncertainty and maximizing profits. This study presents a Transformer-based deep learning model to forecast the price spread between real-time and day-ahead electricity prices in the ERCOT (Electric Reliability Council of Texas) market. The proposed model leverages various time-series features, including load forecasts, solar and wind generation forecasts, and temporal attributes. The model is trained under realistic constraints and validated using a walk-forward approach by updating the model every week. Based on the price spread prediction results, several trading strategies are proposed and the most effective strategy for maximizing cumulative profit under realistic market conditions is identified through backtesting. The results show that the strategy of trading only at the peak hour with a precision score of over 50% produces nearly consistent profit over the test period. The proposed method underscores the importance of an accurate electricity price forecasting model and introduces a new method of evaluating the price forecast model from a virtual bidder’s perspective, providing valuable insights for future research.
Complexity vs Empirical Score
- Math Complexity: 6.5/10
- Empirical Rigor: 7.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs an advanced Transformer-based deep learning architecture with attention mechanisms, which is mathematically dense. It demonstrates high empirical rigor by using real ERCOT market data, implementing a walk-forward backtest, and reporting practical trading performance metrics.
Research Flowchart
flowchart TD
A["Research Goal: Forecast Electricity Price Spread for Virtual Bidding"] --> B["Input Data & Features"]
B --> C["Transformer Model Architecture"]
C --> D["Walk-Forward Training & Validation"]
D --> E["Price Spread Prediction"]
E --> F["Backtesting Trading Strategies"]
F --> G["Key Findings"]
subgraph B ["Inputs"]
B1["ERCOT Market Data"]
B2["Load & Renewable Forecasts"]
B3["Temporal Attributes"]
end
subgraph F ["Strategies"]
F1["Peak Hour Trading"]
F2["Threshold-Based Trading"]
end
subgraph G ["Outcomes"]
G1["Peak Strategy: >50% Precision & Consistent Profit"]
G2["Forecast Accuracy Critical for Virtual Bidding Success"]
end
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 a Transformer-based deep learning model to forecast the price spread between real-time and day-ahead electricity prices in the ERCOT market and develops trading strategies based on its predictions.
Source sentence
A Transformer-based deep learning model is proposed to forecast the spread between SCED and DAM prices in the highly volatile ERCOT market.
Assumes 5
- The uplift cost of trading 1 MWh energy is $5.00. market
Source sentence
The uplift cost of trading 1 MWh energy is $5.00.
- Bids and offers always get cleared by placing the price at price cap or price floor. execution
Source sentence
The bids and offers always get cleared by placing the price at price cap or price floor, details please refer to [9].
- Only 1 MWh can be traded per day, possibly split across multiple hours. execution
Source sentence
There is a budget limit each day, i.e., only 1 MWh is traded per day, but it may be split in multiple hours evenly if multiple hours are selected to trade in a single day.
- Bids and offers will not change the system lambda. market
Source sentence
The bids and offers will not change the system lambda.
Show 1 more
- Price spread is quantized into five ranges with thresholds at $5 and $12. model
Source sentence
The price spread is quantized into 5 ranges: (−inf, −12), [−12, −5), [−5, 5), [5, 12), [12, inf). The threshold of $5.00 is chosen as the break even point of the virtual bidding considering the uplist cost. The threshold of $12.00 is about the median of the price spreads that is outside $5.00.
Data
- ERCOT market data (load forecast, solar and wind generation forecasts, SCED and DAM price market-prices public
Source sentence
Price spread: the difference between SCED and DAM system lambda prices.
Costs
Modeled
Uplift cost of $5.00 per 1 MWh traded is incorporated as the break‑even threshold in the trading strategy.
Source sentence
The uplift cost of trading 1 MWh energy is $5.00.
Finds 3
- Trading only at the peak hour with a precision score over 50% yields nearly consistent profit over the test period. performance
Source sentence
The results show that the strategy of trading only at the peak hour with a precision score of over 50% produces nearly consistent profit over the test period.
- Model predictions for hour 19 generate the highest cumulative profit among all hours. performance
Source sentence
The results demonstrated that the model can provide valuable predictions, particularly for hour 19, where trading strategies based on the model’s output at this hour yielded the highest cumulative profit.
- Performance differences between training settings are minimal; model C has slight precision/F1 advantage, model F excels in accuracy/recall. performance
Source sentence
Overall, the performance differences between training settings are minimal, with model C showing slight advantages in precision and F1 score, while model F excels in accuracy and recall.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 12 related papers). Coverage is partial, so this is neither replication nor refutation.
1 paper in the archive
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.