Research On Timing Strategy Of NARX Dynamic Neural Network

With the gradual improvement of China's securities market and the continuous development of the fund industry,investment institutions are no longer relying solely on traditional pure passive management index funds to manage index funds,but expect to utilize various strategies to optimize the investment portfolio and achieve active management,so as to obtain returns higher than the benchmark index.This will require the introduction of enhanced index fund.As the enhanced index fund aims to maintain the characteristics of the mark index as much as possible,buying corresponding proportion of securities based on the index proportion occupied by each different kind of securities,hence investment target of the index funds is not changeable.Therefore,to obtain excess return,it is essential for investment institutions to focus more on the optimal time to execute buy or sell decisions on funds.At present,there are many quantitative trading methods for trade timing in China's securities market.Examples include BP static network,SVM model and NARX dynamic network.However,we still have yet to discover any quantitative timing method suitable for China's A-share market CSI 300 index funds.Hence,this article hopes to find a suitable quantitative timing method for CSI 300 index,so that the constructed enhanced index fund can outperform the benchmark index and obtain excess returns.This paper makes an empirical analysis on the effectiveness of the quantitative timing method in China's A-share market and the selection of the optimal timing method.Firstly,compare the imitative effects of the three models by using BP static neural network,support vector machine model and NARX dynamic neural network to predict the price of csi 300 index respectively.The empirical results show that the NARX dynamic neural network has the best imitative effect.Secondly,on the basis of stock price fitting,the three models are further used to classify the pattern of stock index trend,which can be divided into three patterns:"stage top","stage bottom" and "continuous middle".The empirical results show that the NARX dynamic neural network has the highest pattern recognition accuracy.Finally,since the NARX dynamic neural network has the best performance in both fitting effect and pattern classification,the time-selected enhanced index fund is constructed based on the pattern classification judgment of the network.If it is at the "bottom of the stage",execute the buy;If it is in the "continuous middle",keep the original position.And will choose enhanced fund performance index(alpha and beta,Sharpe ratio,risk value,maximum retracement)combined with market index comparison,calculation results show that quantitative timing based on NARX dynamic neural network classification model constructing portfolio of active timing(when choosing enhanced fund)better than market index combination,higher rates of return less risky.Therefore,NARX network can be applied to China's securities market and act as a guide for timing investment.The results of this study and the corresponding model program can provide support for the securities companies and fund managers to construct the timing enhancement.