Research On Implied Volatility Surface Prediction And Arbitrage Strategy Based On Convolution Neural Network

Option is one of the most widely used derivatives,and investors are more and more inclined to use the hedging of option portfolio for risk management and asset allocation.However,any investment needs to pay close attention to market fluctuations,and implied volatility reflects investors’ expectations of future price fluctuations in the option market.In addition,there are various arbitrage opportunities in the real option market.As long as investors’ expectations of implied volatility are consistent with the changes in the real market,they are more likely to make profits.Therefore,accurate prediction of implied volatility is the core for investors to establish arbitrage strategies to obtain higher returns.Accurate prediction of implied volatility can’t ignore the remaining period and the execution price,and the implied volatility surface just reflects the relationship between them and the change form of implied volatility,so it is of great significance for investors to accurately fit the implied volatility surface to grasp more market information and obtain stable income through arbitrage strategy.Under this background,this paper chooses SSE 50 ETF options as the research object,and studies the implied volatility in two stages.In the first stage,this paper uses VGG convolutional neural network as the benchmark model for forecasting the implied volatility surface,and innovatively introduces VIX index to better capture the market fluctuation of the option market at the initial stage of the epidemic based on the consideration of extreme events such as the epidemic.VIX and other related factors of the implied volatility form the two-dimensional input tensor of the model to realize the prediction of the implied volatility surface.The results show that the model can fit the fluctuation trend of the implied volatility with low error.In addition,the financial constraint of implied volatility without arbitrage is also incorporated into the above model as the penalty term of loss function,which makes the "black box" model of convolutional neural network financially interpretable,and through horizontal comparison with other deep learning models,it is proved that the model constructed in this paper has good forecasting effect and generalization ability;Finally,cubic Bessel spline interpolation is used to fit the implied volatility surface predicted by the real market and the benchmark model,and both of them are consistent in term structure and smile structure,which further proves that the VGG convolutional neural network model has certain research and application value in implied volatility prediction.In the second stage,on the basis of the forecast results of the first stage,taking "high throwing and low sucking" as the core trading principle,the classical cross-arbitrage strategy is selected to construct the option arbitrage strategy.Brin track is introduced to judge the timing of volatility trading.When the middle track breaks through the upper and lower limits of Brin track,it is considered that the option market fluctuates abnormally,and volatility trading can be done at this time.The results of arbitrage strategy show that the trading behavior based on this strategy is successful and effective.To sum up,there are two main contributions and innovations in this paper.On the one hand,new factors are introduced.Compared with other factors extracted from implied volatility,VIX index is an exogenous variable,which can expand the model information and capture the turbulent market conditions during the epidemic.On the other hand,it is a new application.This paper is not limited to using machine learning method to predict the implied volatility surface,but tries to give an arbitrage strategy based on the predicted surface results,which provides investors with arbitrage ideas and has certain theoretical and practical significance.