Study And Application Of Stock Index Prediction And Investment Strategy In Quantitative Trading

In recent years,with the rapid development of computer technology,the field of computer finance has received extensive attention and research,including quantitative investment,stock forecasting,and stock cluster analysis.Due to the potential economic benefits,the research of stock financial market has always been a hot issue in economics,mathematics,engineering and other disciplines.At the same time,computational finance research is also full of challenges.Financial data has the characteristics of nonlinearity,high noise,and random drift,which makes the problems of stock forecasting and investment decision-making full of uncertainty.Quantitative investment as a representative of computer finance,its core issues are stock price forecasting and investment decisions.There are already many machine learning methods that are used in stock forecasting and investment decision tasks,such as deep learning and reinforcement learning.In the tasks of stock forecasting,data preprocessing is particularly important.Different features and different data processing methods will have great impact on the forecasting results,leading to unsatisfactory trading results,but previous data engineering is often ignored by researchers.In addition,investment decisions are usually made by investors subjectively,lacking the ability to adapt to the financial market.Considering the above problems,we propose two quantitative investment frameworks,which are studied and analyzed from two aspects: stock price forecasting and investment decision-making.The research work completed in this paper mainly includes:(1)In stock forecasting tasks,market data and some technical indicators are usually used as input features of the forecasting model.However,there are many kinds of technical indicators in the world.Scholars and investors often choose technical indicators according to their personal preferences and investment experience.In this paper,we use the maximum information coefficient feature selection method to measure and evaluate different candidate technical indicators,and select the most favorable features for the prediction results as the input of the prediction model.At the same time,the convolutional neural network is used to extract the features of each technical indicator separately to prevent the interference of different indicator information.(2)When using machine learning to predict stock price,the stock data is usually normalized by global normalization.However,when the stock price of the test set breaks the maximum value or falls below the minimum value,the model will not be able to effectively fit the stock price.In this paper,the historical data is processed by the local normalization method in the sliding window.And the fluctuation ranges in the future are predicted.By this way,the stock data cross-border problem is effectively solved.This question also confirms that the local normalization method is more conducive to LSTM prediction of stock time series data.(3)Improve the traditional box investment strategy.Traditional box investment strategy is based on the forecasting results of the upper and lower bounds of the box.Then the investment strategy is manually formulated.Based on the forecasting methods mentioned above,on the premise of effectively forecasting the rise and fall of stock price,this paper adopts the reinforcement learning method to make the model learn the investment strategy independently in the market by trial and error.So as to make the ultimate learning investment strategy more suitable to the stock market and reduce the workload of manual strategy formulation.