Deep Hedging of Green PPAs in Electricity Markets

TL;DR

The paper tackles hedging Green PPAs in highly incomplete electricity markets by integrating a stochastic model of weather-driven renewable infeeds with a forward-based price framework that captures cannibalisation. It then applies a neural-network-based Deep Hedging approach, optimizing hedges with respect to risk measures such as Expected Shortfall. Empirical results show our Deep Hedging strategies outperform static and simple dynamic hedges in terms of variance, skewness, and ES across scenarios, while providing interpretable deltas linked to price and weather signals. The work demonstrates the practical viability of ML-driven hedging for renewable-backed contracts and outlines future work on transaction costs, longer horizons, and multi-asset settings to further enhance hedging efficacy in energy markets.

Abstract

In power markets, Green Power Purchase Agreements have become an important contractual tool of the energy transition from fossil fuels to renewable sources such as wind or solar radiation. Trading Green PPAs exposes agents to price risks and weather risks. Also, developed electricity markets feature the so-called cannibalisation effect : large infeeds induce low prices and vice versa. As weather is a non-tradable entity the question arises how to hedge and risk-manage in this highly incom-plete setting. We propose a ''deep hedging'' framework utilising machine learning methods to construct hedging strategies. The resulting strategies outperform static and dynamic benchmark strategies with respect to different risk measures.

Figures (9)

• Figure 1: Interpolants and point-wise error approximating the sigmoid function with 10, 20 and 30 points on a uniform grid between -5 and 5.
• Figure 2: Paths of the stochastic model introduced in Section \ref{['section_model']}, infeed above (stylised wind onshore), power price below. Parameters used were: $t=0, T=48$, forecast $Q_1(0, T) = 0.5$, forward price $f(0, T) = 100$. Arrival of major weather forecasts at $t \in \{10, 14, 18, 34, 38, 42\}$. We have set $\varsigma(x) = \frac{1}{1+e^{-t}}$. Parameters of the stochastic processes are: $\kappa_1 = \kappa_2 = 0.1$, $\sigma_1 = \sigma_2 = 3.0$, $\kappa^P = 0.5$, $\sigma^P = 0.8$. The technology weights are $w_1 = 0.8, w_2 = 0.2$.
• Figure 3: Histogram and scatter-plots for power price and wind efficiencies at the final time $T=48h$.
• Figure 4: Resulting PnL distributions for dynamic volume hedging and deep hedging using the 5%-ES with 100.000 paths.
• Figure 5: Case at-the-money: Delta versus price one hour before expiry on all test paths. This is the at-the-money case where PPA price equals the forward price of 100.

Theorems & Definitions (1)

• proof