A Bagging Long Short-term Memory Network for Financial Transmission Rights Forecasting

A financial transmission right (FTR) is a financial derivative based on the differences between locational marginal prices (LMPs) of different nodes. FTRs are introduced into the electricity market to help market participants hedge against the risks of transmission congestion. Also, virtual bidders with advanced computational power and forecasting capabilities are encouraged to participate in the FTRs market. This enables FTRs a good crowd-sourcing indicator of transmission congestion. Hence, FTRs prediction can be very valuable to various stakeholders in the electricity sector. Noticing the value of FTRs, we propose Univariate Long Short-term Memory (ULSTM) and Multivariate Long Short-term Memory (MLSTM) networks for FTRs forecasting. To find out applicable predictors, we conduct ablation studies that compare the prediction accuracy of the Long Short-term Memory (LSTM) networks by varying features and input lag lengths. Although the incorporation of associated features generally improves the forecast accuracy, the periodic noises of these features may ruin the forecasting ability of our proposed network. To smooth out the noises, we further propose a bagging method to enhance our prediction networks.