Power futures return forecasting

This project studies the daily variation of a short-term (24-hour) electricity future, interpreted as a daily revision of market expectations about the future day-ahead electricity price.

The modeling framework follows the merit-order logic described in Fundamental Price Drivers on Continental European Day-Ahead Power Markets by Geissmann & Obrist (2018).

In electricity markets, the price is determined by the marginal generating technology required to meet demand. Consequently, shocks to fuel costs, renewable production, demand, and cross-border flows shift the merit order and generate revisions in electricity price expectations. These shocks therefore explain the daily variations of electricity futures prices.

The explanatory variables are grouped into three fundamental categories.

Marginal cost drivers

• GAS_RET – natural gas price variation
• COAL_RET – coal price variation
• CARBON_RET – CO₂ allowance (EUA) price variation

These variables capture shocks to the marginal cost of thermal generation, which affect electricity prices when the corresponding technology is marginal.

Merit-order (quantity) drivers

• x_SOLAR, x_WINDPOW – renewable generation
• x_HYDRO – hydro generation
• x_NUCLEAR – nuclear generation
• x_CONSUMPTION – electricity demand
• x_RESIDUAL_LOAD – demand net of renewables

These variables represent shifts in the supply curve of low-marginal-cost generation, which directly move the marginal technology in the merit order.

Cross-border trade drivers

• x_NET_IMPORT, x_NET_EXPORT
• DE_FR_EXCHANGE, FR_DE_EXCHANGE

These variables capture market coupling effects between France and Germany, reflecting how electricity flows modify effective supply and demand.

Finally, the modeling explicitly accounts for heterogeneity between markets. France and Germany are modeled separately because their electricity systems differ structurally (generation mix, marginal technologies, and trade patterns), implying different underlying statistical distributions. In addition, a third regime captures days where German data are absent and only French fundamentals are observed, which follow a distinct distribution.