Research On Optimizing Quantitative Stock Selection Of BP Neural Network Based On Boosting Algorithm

In recent years,the domestic quantitative investment industry is developing rapidly,and more and more investors have set their sights on the field of quantitative investment,such as quantitative timing and quantitative stock selection,and high-frequency arbitrage transactions.Quantitative investment essentially believes that portfolio income can be decomposed into various factors and can be expressed quantitatively.Investors expect to obtain excess returns through quantitative,systematic and precise control of portfolio allocation.In the face of high-dimensional and complex financial data,neural networks are widely used in the field of quantitative investment due to their powerful nonlinear mapping capabilities and self-learning capabilities.However,with the deepening of the neural network algorithm in the field of quantitative investment,more and more problems have also been exposed in the research and modeling process.For example,the model optimization method tends to be simple,and the neural network algorithm is sensitive to data noise and Problems such as easy overfitting,and failure to repair the defects of these neural network stock selection models will cause serious regrets for investors 'failures and decision-making mistakes.Therefore,how to improve the shortcomings of the existing neural network model and optimize the model,construct a scientific and effective quantitative stock selection model,has far-reaching implications for investors.Based on the above analysis and cognition,the research object of this thesis is 145 related listed companies of the main board general manufacturing industry.The total number of samples in the data set is 181432.The sample interval is from January 2014 to December 2019.A total of 38 relevant financial indicators are selected.And technical indicators,according to the correlation analysis technique,10 characteristic indicators that are weakly correlated with stock returns are eliminated,and then the training data set and the test set are divided into the sample data set.When constructing the model,the quantitative stock selection problem is transformed into the model classification problem in machine learning,and the most widely used BP neural network in the neural network is selected as the main body of the model.After the BP model is built,its own limitations are addressed,such as slow convergence speed,Sensitive to data noise,easy to fall into local optimal solution,etc.,combined with the characteristics of iterative training of integrated learning Boosting algorithm iterative training misjudgment samples,the model is optimized,and the Boosting-BP model is constructed for empirical analysis and performance back testing.(1)The BP model has limitations such as slow convergence,easy to fall into local minimums,and sensitivity to noise.However,financial data is bound to be mixed with many noises,which will affect the classification effect of the model.Considering that the Boosting algorithm can build multiple bases.The classifier decomposes the data noise,and the classification accuracy of the Boosting-BP model has been greatly improved after the introduction of the integrated learning algorithm.It can be seen that the Boosting algorithm has played a certain role in optimizing the defects of the BP model.After comparing the total return and excess return,it is found that the Boosting-BP model performs better than the BP model.In addition,in terms of model stability and risk control,the Boosting-BP model has also achieved good results.Boosting-BP model's stock selection performance is better than BP model.(2)This thesis provides a set of effective model construction methods and details the construction process of stock selection models,including sample selection and factor selection,data processing,model construction,parameter optimization,and stock selection strategy.Provide investors with an effective decision-making reference.(3)Boosting algorithm,as one of the representative algorithms of integrated learning,combined with BP neural network enriches the existing model combination methods,and provides new research ideas for the optimization method of machine learning algorithms.Compared with the BP model,the stock picking income of the Boosting-BP model has been greatly improved,which shows that the application of integrated learning algorithms in quantitative stock picking has certain feasibility.Based on the above conclusions,this thesis believes that the related research of Boosting-BP model has the following important significance,First of all,the integrated learning algorithm is rarely used in quantitative research.The empirical research part of this article can verify the effectiveness of the integrated learning algorithm in quantitative stock selection,and provide practice for the application of machine learning,integrated learning,neural network and other technologies to quantitative stock selection research significance.Secondly,this thesis proposes an effective model building method and introduces the model building process in detail,which provides investors with an effective decision-making reference.Finally,this thesis broadens the existing model combination and algorithm optimization methods,and provides innovative ideas for model optimization methods in the field of quantitative investment.