Research On Quantitative Investment Strategy Based On Machine Learning Technology

In recent years,machine learning models represented by neural networks have made great achievements in the fields of natural language processing and image recognition.As the most efficient industry in the economic society,the financial investment industry has always attached importance to the integration of the latest science and technology and financial practice.Therefore,the application of related model methods in the field of finance has also become a research hotspot in academia and industry.This article summarizes the current research situation of quantitative investment based on machine learning algorithms.It mainly conducts research from two aspects: quantitative stock selection based on cross-section data and quantitative trading timing based on time series data.details as follows:(1)Research on quantitative stock selection based on Randomly Distributed Embedding(RDE).This study uses the stock price data of the target stocks in the same industry as the data source,based on a random distribution embedded frame,to predict the stock's rise and fall.Specifically,the historical data of stocks in the industry is used as input to construct a projection of the industry stock attractors on a low-dimensional space to achieve a prediction of the target stock.The predictions of multiple lowdimensional spaces are integrated to realize the prediction of the future rise and fall of the target stock.(2)Quantitative timing research based on convolutional neural network.This study takes historical trading data of a single stock as input,and uses convolutional neural networks and gradient boosting trees to predict stock buying and selling points.Based on the lack of daily frequency data of stocks,artificial complexity is adopted to reduce the model complexity.Aiming at the characteristics of imbalanced training data category,try to improve the model effect by using methods such as adjusting threshold and model fusion.Applying machine learning-based algorithms to the research of quantitative strategies can achieve better results.Based on a large number of experimental comparisons,this paper believes that based on the understanding of the characteristics of the stock market data,the reasonable transformation of new machine learning algorithms into quantitative strategy research can achieve more predictive effects than traditional models and provide excess returns in the securities market stand by.