High Frequency Prediction Volatility With Applications In Finance Based On Deep Learning LSTM Model

The prediction of volatility of financial assets has been a hot topic in the academic research.Traditional low-frequency and high-frequency volatility prediction models are usually constructed based on time series data.With the great achievements of machine learning and deep learning in many fields,how to improve the prediction accuracy of volatility by using the research results in these fields for reference and taking the advantages of traditional time series model needs to be further studied.At present,the macro economy is influenced by the impact of novel coronavirus outbreak and of international crude oil prices,and the financial risk in the whole world continues to accumulate,so it is particularly important to find appropriate tools for risk management.With the continuous development of the domestic options market,the types and scale of financial options and commodity options are growing steadily.Rational use of options which are characterized by leverage and asymmetric returns can reduce portfolio risks and increase portfolio returns.Volatility is an important factor not only in option pricing but also in option arbitrage investment,therefore it is worthwhile for investors to make better use of volatility information to construct option investment strategy.Firstly,this paper uses the financial high frequency data to calculate the realized volatility which will be implemented as a financial asset volatility measurement.Fusing three kinds of information: High frequency realized volatility,technical Indexes,and Time series model feature information,we employ the deep learning LSTM model to construct the LSTM-HIT volatility prediction model Secondly,using the semiparametrical Extreme Value Theory(EVT)to estimate the quantile of standardized return,we construct LSTM-HIT-EVT risk management VAR model.Moreover,the LSTM-HIT model prediction results were used to generate the volatility strategy signal,and then the option volatility arbitrage strategy will be constructed for option investment.Finally,the empirical analysis of the above model is carried out by using the 5-minute high-frequency data of Shanghai Composite Index and 50 ExchangeTraded-Fund with its daily options data,and the comparative analysis is made with eight common machine learning models and six traditional time series models.The empirical results show that,based on multidimensional feature information,LSTM-HIT volatility prediction model is more accurate than the LSTM-H model,machine learning models and traditional time series model.Combining LSTM-HIT model with the semi-parametrical Extreme Value Theory(EVT),we construct the LSTM-HIT-EVT risk measurement model,and the empirical results also show that LSTM-HIT-EVT model is more accurate than the traditional VAR models under the evaluation criteria of condition coverage test and MCS test based on three kinds of VaR loss function.The LSTM-HIT volatility arbitrage strategy constructed in this paper achieves the best performance both on the return and risk evaluation metrics which shows that volatility arbitrage strategy constructed based on LSTM-HIT has its advantages.