Faculty Sponsor
Dr. Di Shang
Faculty Sponsor College
Coggin College of Business
Faculty Sponsor Department
Accounting & Finance
Location
SOARS Virtual Conference
Presentation Website
https://unfsoars.domains.unf.edu/2021/posters/modeling-real-time-prices-for-energy-markets/
Keywords
SOARS (Conference) (2021 : University of North Florida) – Archives; SOARS (Conference) (2021 : University of North Florida) – Posters; University of North Florida -- Students -- Research – Posters; University of North Florida. Office of Undergraduate Research; University of North Florida. Graduate School; College students – Research -- Florida – Jacksonville – Posters; University of North Florida – Graduate students – Research – Posters; University of North Florida. College of Computing; Engineering & Construction -- Research -- Posters
Abstract
Wholesale power markets operate as both “day-ahead” and “real-time” markets, where transactions for power are placed a day before flow and an hour before flow, respectively. This project focuses on comparing models for predicting the price of the real-time market, using the PJM region as a case study. The real-time market is notoriously more volatile than its day-ahead counterpart, so the ability to predict the price of power on an hourly basis is extremely valuable for making decisions such as when and where to transact and when to hedge against volatility. Additional variables considered include temperature, power demand (load), generation by fuel type, lags of the day-ahead price, and the lags of the components of the total real time price. All variables are at an hourly granularity. The models developed to predict real-time price include a neural network regression, a boosted decision tree regression, two specifications of multiple linear regressions, and a seasonal ARIMA. The models were used to predict a week of pricing and were evaluated by comparing their Mean Absolute Errors (MAEs) and Root Mean Squared Errors (RMSEs). The second regression model had the lowest RMSE of 6.84 and the second-lowest MAE of 4.12. This model is recommended to predict real-time price because RMSE is a better indicator of a model’s ability to adapt to large swings in value. Additionally, the regression is based on fundamental factors of demand, temperature, and generation mix, while the seasonal ARIMA is based on the time series analysis of the real-time price.
Rights Statement
http://rightsstatements.org/vocab/InC/1.0/
Included in
Modeling Real Time Prices for Energy Markets
SOARS Virtual Conference
Wholesale power markets operate as both “day-ahead” and “real-time” markets, where transactions for power are placed a day before flow and an hour before flow, respectively. This project focuses on comparing models for predicting the price of the real-time market, using the PJM region as a case study. The real-time market is notoriously more volatile than its day-ahead counterpart, so the ability to predict the price of power on an hourly basis is extremely valuable for making decisions such as when and where to transact and when to hedge against volatility. Additional variables considered include temperature, power demand (load), generation by fuel type, lags of the day-ahead price, and the lags of the components of the total real time price. All variables are at an hourly granularity. The models developed to predict real-time price include a neural network regression, a boosted decision tree regression, two specifications of multiple linear regressions, and a seasonal ARIMA. The models were used to predict a week of pricing and were evaluated by comparing their Mean Absolute Errors (MAEs) and Root Mean Squared Errors (RMSEs). The second regression model had the lowest RMSE of 6.84 and the second-lowest MAE of 4.12. This model is recommended to predict real-time price because RMSE is a better indicator of a model’s ability to adapt to large swings in value. Additionally, the regression is based on fundamental factors of demand, temperature, and generation mix, while the seasonal ARIMA is based on the time series analysis of the real-time price.
https://digitalcommons.unf.edu/soars/2021/spring_2021/78