Day-ahead Of Electricity Price Forecasting Using Nonparametric GARCH Model

In March 2015, the State Council put forward a document of several opinions on further deepening the reform of electricity power system that is designed to improve the electricity power market completion mechanism. With the degree of marketization of our electricity power deeper, electricity price are more easily affected by the market environment and show the features such as volatile extreme jump, to give participants a high risk of electricity market. So that forecasting electricity received all market participants’ attention. Generation companies use electricity prices forecasting to Optimize bidding strategies. Consumers use electricity prices forecasting to integrate of electricity purchasing portfolio. Market regulators use electricity prices forecasting to promote the stable and order development of electricity market. It can be seen that forecasting electricity accurately is very important.This paper analyzes the fluctuation characters of electricity price in PJM electricity market and Nord Pool electricity market(include Sudan and Estonia) and build model nonparametric GARCH model to forecast electricity price. First we build ARMA models. And then, we get the residuals and find that the residuals have heteroskedastciticty. So we need to further process the residuals to forecast electricity price. First we apply parametric GARCH models that include GARCH model, TGARCH model,EGARCH model, APGARCH model and ACGARCH model to forecast day-ahead electricity prices. With the assumption that residuals obey normal and student’s t distributions, we compare the prediction accuracy of different GARCH models. We find that in most cases the asymmetric GARCH models perform well.Through the study we found that parameters established GARCH model assumption is not always appropriate. On this basis, we put forward nonparametric GARCH(NPGARCH) model to electricity price forecasting. After study the theoretical basis and estimation principle of NPGARCH model, without setting the residual distribution, this pepper use the kernel function regression method to estimate the model, then to forecast the selected electricity price of the four electricity market. Through comparing prediction precision of the parametric GARCH models and nonparametric GARCH model, we get that nonparametric GARCH model can fit price fluctuation characteristics better, and improve the prediction accuracy.