Quantamental Method Based On Top Rank And Deep Network

In the stock market,quantitative investment which uses computer programs to replace people's subjective judgments can prevent investors from making irrational in-vestment decisions due to mood fluctuations.Through the text mining and extraction of the public company's financial statements and public data,the quantitative invest-ment strategy can achieve the analysis of the fundamentals of listed companies,and look for high-quality companies with better profitability for asset allocation.In order to further enhance the effect of quantitative investment strategy,this paper studies the quantitative stock selection model and the announcement text information extraction in the fundamental quantitative strategy.The main innovations are as follows:This paper proposes a quantitative stock selection algorithm based on top rank model.In the past,the quantitative stock selection algorithm was based on the classi-fication model,and it was difficult to reflect the comparison between the advantages and disadvantages of multiple stocks.From the perspective of optimizing ranking,this paper proposes a top rank model under two windows with different sizes,focusing on predicting the minority stocks with the best performance and their order,which is more conducive to asset allocation.This method constructs a large window to follow the evolution of the stock price's pattern in a long period of time,and uses small windows to capture the short-term trend of the stock price,so as to achieve accurate ranking.The method has the advantages of low time complexity and strong interpretability,and the obtained linear weight of the factor can also be combined with the traditional multi-factor strategy for stock selection.The empirical test results show that the proposed method is significantly better than the comparison method,and it can stably obtain excess returns.This paper proposes a progressive annotation information extraction method based on deep model,which helps the quantitative algorithm to quickly and accurately reflect the market changes by timely and accurately obtaining the latest public information of listed companies.This method splits the text information extraction problem into sentence-word two-layer structure and performs progressive decomposition.At the statement level,the model uses a convolutional neural network model of improved word vectors,and the rolling update words are embedded in the vector space to achieve the positioning of a small number of key statements in hundreds of statements.After the relevant sentence is obtained,the final target data is extracted by combining the artificial rule+TF-IDF+regular expression.This paper takes the announcement of the listed company guarantee as the entry point.The experimental results show that the performance of this method is significantly better than the comparison method in the announcement extraction task,so it can effectively help the quantitative stock selection.