Stock Trading Decision Making: A Hybrid Approach Based On Stock Selection,Timing And Portfolio Selection

Stock forecasting,timing and investment portfolios have always been hot issues in the field of quantitative finance.Investors have been paying attention to these three issues,which stocks investors choose when investing,when they hold stocks for trading,and how to avoid investment risks by diversifying stock investment.This article builds an investment strategy model for predicting stock selection,timing,and portfolio combination around the above three issues.The research content is as follows:(1)Support vector regression with modified firefly algorithm for stock price forecastingIn this section,a stock return prediction model combining the improved Firefly algorithm and support vector regression(SVR)is constructed,and the model is used to predict the stock return.The model has two stages.In the first stage,a sufficient condition for the Firefly algorithm(FA)to converge is proposed and proved.According to this condition,a FA with a dynamic adjustment strategy and a chaotic strategy based on reverse movement is proposed.In the second stage,the modified Firefly algorithm(MFA)and the support vector regression model are combined to predict the stock return,and the,MFA is used to optimize the parameters of the support vector regression.Finally,the performance of the algorithm is verified by empirical results.The empirical results show that: Compared with other improved Firefly algorithm,it has superior performance.The proposed MFA and SVR model combine the rate of return prediction model.An effective tool for predicting stock returns.(2)Cascading Naive Bayes onto extreme gradient boosting for stock timingIn this study,chooses the timing of the higher-yielding stocks selected in the previous study.we first use the piecewise linear representation method to identify the stock trading signal,and then build a classification model combining naive Bayes and extreme gradient lifting algorithm to select the stock time.The construction of the model is mainly divided into two steps: using naive Bayes method to select the time according to the technical indicators,and then taking the result as an indicator,together with the original technical indicators as the input indicators of extreme gradient lifting algorithm for classification.Through the use of UCI open data set,the model is verified to have higher classification accuracy than other comparison methods.Finally,the method is used to select the time,and 13 technical indicators commonly used in the research are used as the input.The stock price change direction of the following trading day and the trading signal identified by the piecewise linear representation method are used as the output.The empirical results show that the proposed method has higher classification accuracy than other classification methods,and trades under different trading strategies according to the trading signals predicted by each classification model.The results show that trading according to the trading signals predicted by this method can obtain better results in each trading strategy.(3)Asset allocation based on stock forecast and timingIn this study,allocates stock assets on the basis of the previous two parts of research.,three investment portfolio models are used to allocate assets.Select the stocks to be allocated,according to the forecast method of predicting stock returns mentioned in the previous article,the stocks with the highest predicted returns will be allocated according to the ratio of each stock after allocation and the timing strategy.Two realistic constraints are considered,one is the investment constraint that does not allow short selling,and the other is the constraint of transaction costs,and the configuration is performed under these two constraints,and the return is calculated.The empirical results show that when using a mixed strategy based on predictive stock selection,timing and portfolio investment,higher returns can be obtained.