Trading Electrons: Predicting DART Spread Spikes in ISO Electricity Markets

Battle Count: 65

Highly relevant to quantitative trading in electricity markets. The paper directly addresses the joint problem of signal generation (spike forecasting) and execution (optimal position sizing with price impact), which is central to algorithmic trading. The structural price impact model and closed-form optimal quantities are directly applicable to virtual bidding strategies. The multi-zone optimization framework with cross-zone congestion effects provides a rigorous approach to portfolio construction in electricity markets. The empirical validation across three major US ISOs with out-of-sample results strengthens practical relevance.

Moderate to high complexity. Requires: (1) access to and processing of hourly bid stack data from ISOs, (2) zone-specific logistic regression training with careful feature engineering (50-dimensional features for NYISO), (3) numerical estimation of supply/demand curve slopes via finite differences, (4) multi-zone quadratic optimization with three candidate solutions and regime selection, (5) seasonal and load-regime stratification for impact calibration. The closed-form solutions simplify the optimization step, but the overall pipeline involves multiple calibration stages and market-specific parameter tuning.

The paper uses publicly available data from NYISO, ISO-NE, and ERCOT portals via the GridStatus platform. Detailed feature construction, model specifications, calibration procedures, and threshold selections are described. However, no code repository is provided, and some calibration choices (e.g., Top-10 spike hours, specific finite-difference procedures) require careful implementation. The chronological train/validation/test splits are clearly defined.