Research On Market Effect Analysis Method Of Policies

Stock forecasting is one of the hot studies in academia in recent years,but stock price fluctuations are affected by a variety of factors,which makes forecasting difficult.Policy as a long-term national development plan to ensure the smooth operation of the securities market.The inherent advance anticipation characteristic of the stock market makes the changes in national policies inevitably reflected in the stock market activities,therefore,mining the influence of policy factors can provide effective reference information for investors.However,there are still two shortcomings: firstly,there are fewer classification criteria for policy content,and researchers mostly use manual annotation methods when classifying policies,which is costly to classify;secondly,most of them use machine learning methods for stock price prediction,and the model interpretability is limited.This paper aims to explore a market effect analysis method for policies,using random forests and LIME locally interpretable models to explore the factors influencing stock market volatility,i.e.,the importance of the characteristics of policy categories.The main contributions of the paper are as follows.1.To address the problem of lack of classification criteria and timeconsuming classification methods for policy content,the paper proposes three dimensions for classifying dairy industry policies over a long period of time,and combines the idea of semi-supervised learning with the singleview multi-classifier approach Tri-training to construct a collaborative training classifier based on Support Vector Machine(SVM),Naive Bayes(NB),and Back Propagation Neural Network(BPNN)to achieve policy text classification.2.Collect stock data of listed companies in the dairy industry and extract market signals from them.Use Discrete Wavelet Transform(DWT)technique to process the stock time-series signals,decompose the stock features into multi-scale components,and reconstruct them into highfrequency and low-frequency signals to achieve the separation of stock signal fluctuation trends and abrupt noise.3.To address the problem of low interpretability of the model,after using random forests for regression analysis of policy categories with highfrequency and low-frequency signals of stock characteristics,the paper further introduces LIME locally interpretable model to interpret the individual output results of random forests and the overall decision behavior of the model to derive the direction and degree of influence of each type of policy events on the short-term and long-term fluctuations of stock market characteristics,so as to provide a reference basis for policy formulation and the decision-making behavior of stock market participants.