Research On Electricity Price Volatility Based On Neural Network

It is very important for market participants to predict electricity price volatility accurately.Unlike the stock market,the electricity market is affected by weather,social activities,emergencies and other factors.The volatility of electricity price has the characteristics of seasonal pattern,reverse leverage effect and high volatility.In terms of the reverse leverage effect of the power market,this paper adopts various leverage characterization methods and establishes a variety of high-frequency HAR models with leverage to predict price volatility in the power market.The study found that the widely used HAR model did not perform well in the volatile period,and the performance of each model is different from the existing research conclusions.The prediction performance of HAR models is sensitive to sample data.Considering that this may be due to the lack of market effectiveness during the period of shocks,the applicability of different models will be different.Since the artificial neural network does not need to specify the model and only extracts information from the data,it can effectively avoid the problem of model misdesignation.For the first time,an artificial neural network model for forecasting electricity price volatility is established.In addition,the realized estimator based on high-frequency price data is innovatively considered as network input data.Based on the electricity market price data of New South Wales,Australia,the robustness of the forecasting results of various HAR models and high frequency neural networks is tested.The results show that the prediction error of the artificial neural network based on high frequency data is smaller than that of the traditional HAR model in both stationary and oscillatory periods.In the robustness test,the neural network model ranks first;in the sectional test,the prediction ability of the neural network model is obviously better than that of the HAR model.Therefore,the artificial neural network based on high frequency data can effectively solve the problem that the traditional model is not applicable in the period of shocks,reduce the prediction error and obtain more robust prediction performance.