Introduced by Negi Fazeli, Ahmed Issaouai, Patrick Lerchi, Marie Parent, and Shaojin Wei
Investors in energy infrastructure are using outdated tools to bet on future
electricity markets. By current market standards, prices are predicted with few features and over-simplified models. Furthermore, these old models are unable to adapt to forces that disrupt the usual trend of energy prices, like new wind technology or unusually high hydroelectricity generation from a year with a heavy snowpack.
Our team wanted to change this. Using work from a previous semester, we narrowed a list of 74 price-predictive features to 11 using Pearson correlation, LASSO selection, and VIF measures. These 11 features were then forecast into the future and classified with a regression model to obtain a ’baseline’ prediction of future energy prices. Finally, we implemented a segmented ARX optimization tool developed at Stanford to assess how feature impact on electricity price changed over time. By prioritizing user flexibility and intuitive output format, we were able to build a platform that can pair the industry experience of financial investors with the predictive capability of data science. Our work can definitely be improved by future Data-X students, and we look forward to seeing how this project evolves in following semesters.