Volatility Forecasting Based On High-frequency Data And Microstructure Noise:Empirical Study

Volatility forecasting is significant to risk management and asset pricing.In recent years,many scholars have conducted researches on volatility forecasting based on high-frequency data.However,the noise contained in high-frequency data has a great impact on the prediction effect.This paper aims to extract noise and add it to the volatility forecasting models to explore whether noise can improve forecasting accuracy,thereby bringing certain economic returns to investors.The research ideas and methods of this paper are as follows: we use high-frequency trading data of the US S＆P 500 index ETF from 2014 to 2020,add noise variance into HAR family volatility forecasting models to perform in-sample fitting and out-of-sample prediction.By constructing indicators of liquidity and information asymmetry,and decomposing noise variance,we explain how noise affects volatility in ternms of market microstructure.We also use noise to forecast continuous and jump fluctuation,which are two components of volatility.By changing the sample data,using the expansion window,using other noise variables and using Monte Carlo simulation,we prove the robustness of the conclusion.Finally,the results of volatility forecasting are applied to formulate empirical asset allocation strategies to reflect the economic value.The conclusions are: First,medium and short-term noise has a positive impact on volatility,and the additional information contained in noise can significantly improve the volatility forecast effect.Second,turnover rate,information asymmetry and daily trading volume have a significant impact on noise changes,which may be the main source of additional information contained in noise,and the liquidity part of noise has a greater contribution to improving volatility forecasting.Third,the prediction accuracy of volatility components can also be improved by considering noise especially the jump,and the continuous and jump fluctuations predicted separately can be added together to obtain more accurate volatility prediction value.Fourth,the empirical asset allocation strategy results indicate that investors would like to pay an annual fee of 8 basis points to obtain a more accurate forecast value and a higher return on investment.The innovations of this paper are as follows: First,we combine the formation mechanism of market microstructure noise with volatility forecasting and explore whether the liquidity and information asymmetry components in the noise can improve the effect of volatility forecasting.Second,we propose a method of using noise to predict the continuous fluctuation and jump fluctuation and then combine them to get the volatility forecast.Finally,based on the forecast results,we formulate the empirical asset allocation strategy and give the actual economic value of the research.