Stealing profits: Spread-based temporal hierarchy forecasting for day-ahead electricity markets

Battle Count: 0

Highly relevant for quantitative trading in energy markets. It demonstrates that statistical accuracy (RMSE/MAE) does not always correlate with economic profit, and provides a specific method (Spread THieF) to improve trading decisions by focusing on relative price movements (spreads) which are critical for arbitrage.

Moderate to High. Requires implementing hierarchical forecast reconciliation (minimum-trace method), constructing large covariance matrices (300x300 for spreads), and managing multiple base models (ARX, NARX, TabPFN). The use of shrinkage estimators for covariance adds complexity.

The paper uses publicly available data from ENTSO-E and Investing.com. The forecasting models (ARX, NARX) are standard, and TabPFN is a released foundation model. The reconciliation method is mathematically defined. However, specific code repositories are not explicitly linked in the text provided, though the methodology is detailed enough for re-implementation.