Research On Portfolio Strategy Based On Stock Anomalies

In this paper,based on the two machine learning algorithms of support vector machine and decision tree,the stock return rate is predicted,screened and reclassified,and finally a portfolio strategy that can stably obtain excess returns is constructed.This article builds a portfolio strategy that will be divided into.First,select the sample interval from 2004 to 2017,and use the data in the interval as the sample data to construct the required classification factor.Second,the data is pre-processed,outliers are eliminated,missing values are added,and the data set for machine learning training is as complete and accurate as possible.Third,select the kernel function of the classification and the parameter optimization of the classifier.Fourth,the classifier is subjected to branching processing to reduce unnecessary classification nodes to prevent the risk of overfitting.Fifth,evaluate the model.The classification prediction effect of the model is measured by comparing with the yield of the HS300 Index.In this paper,after preprocessing and and data normalization operations,66 statistically significant anomalies were selected by factor validity test from 85 initial factors.Then the effective factors are used as single-factor portfolio strategy respectively.The empirical results show that compared with the multi-factor portfolio strategy,the single-factor strategy is insufficient,so a multi-factor portfolio strategy is proposed.The anomalies are input into the support vector machine prediction model and the decision tree classification model as classification factors.First,the data set is input in the support vector machine prediction model and divided into a training set,a verification set,and a test set.Each training prediction model selects all the historical data that can be selected at that time point for learning,and uses the comparison with the next rate of return to judge whether the prediction is correct or not,and adjusts the algorithm model to calculate the next cycle of the stock sample.The probability.Then,in the decision tree classification model,the attributes uesd in classification are calculated and compared,and 10 groups of classification factors are obtained one by one.Each group of classification factors contains 20 stock samples.Take the best group of classification attributes,construct a portfolio by the same weight method,and calculate the income indicator and risk indicator of this portfolio.The results show that this portfolio can stabilize more than the Shanghai and Shenzhen 300 Index.Finally,through statistical tests,the benefits of this strategy are not accidental.