Trading Electricity Markets Using Neural Networks

We tackle the problem of developing an automated trading strategy to profit in the British intraday continuous electricity markets. We must train a feedforward neural network to predict one-hour-ahead total electricity transmission system demand. In live testing to ensure no look-ahead bias, we present results with accuracy better than National Grid’s own demand forecasts. We then train a second feedforward neural network, using our demand forecast as an input to the network, to predict one-hour-ahead net imbalance volume (NIV), and use this predicted NIV as a trading signal to buy and sell 30-minute electricity contracts. In live testing, between 09 March and 22 March 2020, the trading algorithm made 599 simulated trades, with 431 trades returning a profit (an accuracy of 72%). These results demonstrate the potential of neural network driven automated trading strategies to make significant risk-adjusted excess returns (i.e., profits) in the intraday electricity markets.