Situation: Participants in wholesale electricity markets can improve their financial performance by taking advantage of the price spread between day ahead and real-time markets. Trade / bid decisions can be improved by more accurately predicting the locational marginal prices (LMP) to be set by the Independent System Operators (ISO). However, prices and fees are hard to predict since they are calculated using complex market algorithms, and they are only settled by the ISOs after bids are made – sometimes weeks later.
Solution: nDimensional built a GPT-based bid optimizer for a New York quant fund that enables them to benefit from better MISO LMP predictions. A GPT model was trained on vast quantities of MISO data to learn the complex dynamics and nuances about the market as a whole, in order to generate intelligent responses to bidding questions. An RL agent was created to direct the GPT models to optimize daily arbitrage trades between day ahead and real time prices. Although the fund had been successfully using AI for more than a decade, the new GPT-based approach has delivered substantially more lucrative results.