Abstract
This thesis examines the forecasting proficiency of several linear and non-linear time series models on the day-ahead hourly electricity spot prices of the New York state market, the New York Independent System Operator (NYISO). The study considers Markov Chains, Linear Regressions, VAR, AR, ARMA, and ARIMA models. The analysis shows that the ARIMA models consistently forecast with the greatest accuracy across each of the NYISO’s geographically diverse zones and in each season studied. ARIMA models handle large price spikes during peak periods with the greatest accuracy, consistently forecast evening hours with the highest demand and volatility, and maintain accuracy during mid-week days with historically volatile prices. The analysis also shows that exogenous explanatory variables increase forecast accuracy substantially. The strongest models for the NYISO, and markets that function similarly to the NYISO, should incorporate forecasted values of future price and load. ARIMA models with these exogenous variables consistently have MAPEs of 4% compared to past studies with MAPEs of 6-10%. Translated over a two week period, this results in roughly $3.46 million in savings.