Application Of Support Vector Machine And Time Series Method In ETF Risk Measurement

Risk is everywhere,and the essence of risk management is dealing with uncertainty.For all walks of life,risk management is always one of the core issues,especially in financial markets,the importance of risk management is self-evident.One of the key aspects of risk management is risk measurement.Research on risk measurement is of great significance to all fields.The time series analysis method is widely used in various fields.The core idea is to model the time series to extend the trend of the sequence and make predictions.Among them,the generalized autoregressive conditional heteroscedasticity model(GARCH)is often used for time series volatility modeling.The copula function can be used to construct a joint distribution of random variables,what we need are only the edge distribution of a given random variable and the given copula class.Linear correlations are increasingly difficult to accurately describe the synergistic risks of portfolios,so Copula theory is introduced to better describe correlations.Support Vector Machine(SVM)is a method based on the principle of structural risk minimization in statistical learning theory.It has become a research hotspot because of its complete theoretical support and excellent learning performance.With the help of nuclear techniques,support vector machines can effectively deal with more complex nonlinear problems,which are widely used in classification,regression and probability density estimation problems.In this paper,the risk metric VaR is selected as the entry point,and the accuracy of the VaR calculation based on the support vector machine method and the time series analysis method is compared.Taking two ETF closing price data to construct asset portfolio as an empirical research object,combined with backtesting to compare the advantages and disadvantages of the two methods,the empirical results show that at relatively low confidence(95%,97.5%),the time series method is more conservative than the SVM method,the time series method tends to overestimate the real risk,and the SVM based probability density estimation method tends to underestimate the real Risk;at relatively high confidence(99%,99.5%,99.9%),the SVM method is more conservative when calculating VaR,the time series approach tends to underestimate the real risk,and the SVM approach tends to overestimate the real risk.