A Construction And An Applied Study Of HAR Model Based On High-Frequency Data

Affected by various factors at home and abroad,Chinese non-ferrous metal industry has experienced problems such as market downturn and overcapacity in recent years.Exchange rate,interest rate as well as import and export policies have increased the uncertainty of non-ferrous metals.As one of the most important non-ferrous metals and six non-ferrous metal futures,industry output and price of aluminum have been seriously affected.The futures market has the main function of price discovery and avoiding price risk.Therefore,the fluctuation rule of Shanghai Aluminum Futures has become the research focus of investors and regulators,and volatility modeling and forecasting is an effective way to reveal its fluctuation rule and market risk.Earlier research on financial volatility prediction models usually used daily data or lower frequency data,so this paper establishes a classic model of GARCH(1,1)based on low-frequency data at first,because of its the characteristics of simple form and good prediction effect.The empirical results show that the GARCH(1,1)model under the t distribution is optimal in the sample fitting.In recent years,the availability of intra-day high-frequency trading data has become higher,providing a new means for the study of financial volatility.Chinese financial market has obvious characteristics of heterogeneity,so this article uses the HAR family model to study the impact of heterogeneity on financial market fluctuations.Through empirical research,this paper finds that there is a significant asymmetry in the volatility of Shanghai Aluminum futures and high-order moment risk.Therefore,symbol jumps and realized high-order moments are introduced into the HAR family model improve the ability of the HAR model to explain the market and predict the volatility of Shanghai Aluminum futures.The empirical results show that the model combined continuous and jumping wave decomposition with positive and negative sign jumping variation,and added realized high-order moment,that is,the HAR-RV-CJ-DSJV-SK model has the strongest in-sample fitting ability.To further evaluate the pros and cons of the model,this paper evaluates the outof-sample prediction capabilities of the above 11 models.This paper uses six loss functions and the merits of the external prediction ability of each model sample are evaluated by the MCS test based on the rolling time window.The empirical study of the out-of-sample prediction shows that the HAR-RV-CJ-D-SJV-SK model has the most outstanding out-of-sample prediction ability.