Research On Value At Risk Of Stocks Based On ICA And GARCH Model

The measurement of stocks risk in the financial market is very important,which not only affects operational efficiency of the financial market,but also has a key reference value for investors and securities managers to make investment planning.Therefore,it is very necessary to study the volatility of the stocks market and measure its risk.The value at risk of stocks based on the improved Fast ICA algorithm and GARCH model is studied in this paper.The main research contents are as follows.(1)In order to improve its separation performance,the Fast ICA algorithm is modified from two aspects,and then the TSICA algorithm is proposed in this paper.On the one hand,a piecewise function is constructed based on Tukey function and Softsign function to enhance the separation accuracy of the algorithm.On the other hand,the Newton’s third-order iteration format is introduced to raise the convergence speed.The performance of the TSICA algorithm is evaluated by performing 100 separation experiments on mixed signals.The experimental results show that the TSICA algorithm has higher separation accuracy.Meanwhile,the number of iterations is significantly reduced,which is only 6.8%of the Fast ICA algorithm and 25.4%of the MTICA algorithm.(2)In order two increase the prediction accuracy of GARCH model on volatility,and further improve the measurement effect of stocks VaR,the TSICA algorithm and TnA method are applied for denoising pretreatment of data before GARCH modeling,and the TSICA-GARCH model for volatility prediction is constructed subsequently in this paper.Firstly,the TSICA algorithm is used to decompose the selected stocks data into independent components.Moreover,TnA algorithm is introduced for sorting to identify noise and get rid of it.From that,the return series is reconstructed.Then the GARCH mode is used to predict the volatility of the reconstructed sequence.Therefore,the TSICA-GARCH model combining those two models is formed,which measures VaR in the stocks market and back-tested.According to statistical tests on 74 stocks indexes in China stocks market,the indexes conforming to the GARCH model are chosen and then are put into the experimental analysis.The results show that the measurement of VaR based on TSICA-GARCH model is more efficient than that based on GARCH model.(3)In view of the fact that the price range can reflect the stock price changes,and can embody the law of price fluctuation,the TSICA-Range-GARCH model is constructed by introducing the price range of stocks into the TSICA-GARCH model to enhance the model’s characterization of volatility.Firstly,the TSICA algorithm is used to de-noise the selected stocks data.Then the price range and GARCH model are combined to form a volatility prediction model which is called the Range-GARCH model.Furth more,VaR is measured and back-tested.From the empirical analysis of Range-GARCH modeling for 52 indexes with ARCH effect,it can be seen that the measurement effect of VaR are TSICA-Range-GARCH model,TSICA-GARCH model,and GARCH model in descending order,(4)Compared with VaR,CVaR is a consistent risk measure method.CVaR can reflect more effectively the real situation of risks under extreme market conditions.Therefore,we apply the TSICA-Range-GARCH model to predict volatility,and on this basis,to measure CVaR and backtested.From the empirical analysis of 12 stocks in life-related industries during the COVID-19,it can be seen that CVaR is more accurate than VaR in measuring risk under the extreme situation of COVID-19 epidemic.