Electricity price forecasting based on nonparametric GARCH

Based on nonparametric theory for conditional heteroskedasticity function, an improved method of electricity price forecasting is proposed. On the basis of real electricity price time series, conditional variance function is modeled for stochastic volatility, and the model is determined by means of non-parametric estimation. In the nonparametric estimation process, an iterate algorithm is introduced to overcome the problem that volatility is unobserved latent variable so that the confidence of estimated conditional variance function is weak. On the study of stochastic volatility of day-ahead electricity price in Humb spot in California, the forecasting is made. And the results of test show that the proposed method has the capability of forecasting electricity prices characteristic of volatility clustering, and improves the accuracy of price spikes forecasting.