A new paper on arXiv, "The Hidden Look-Ahead: Point-in-Time Leakage in Electricity Market Data," examines a subtle but important flaw in empirical studies of electricity prices. Such studies routinely state that their models rely only on information available at the time of the trading decision. However, the paper points out that when weather-driven forecasts are drawn from commercial archives, this condition is harder to satisfy than it appears.
The issue is that commercial weather archives may contain data that was not actually available in real time, either because of revisions, backfill, or other forms of post-hoc adjustment. If a model uses these data as inputs, it can inadvertently incorporate future information, leading to overly optimistic estimates of predictive performance. This is a classic look-ahead bias, but one that is easy to miss because the data appear to be historical.
The paper does not provide specific examples or quantify the impact in the abstract, but it underscores a methodological concern for anyone building electricity price models. The authors argue that researchers must verify the point-in-time availability of weather data, not just assume that a commercial archive is free of leakage. This is a cautionary note for the growing field of machine learning in energy markets, where data quality is as important as model complexity.