股票推荐系统的算法

THE ALGORITHM OF SROCK RECOMMENDATION SYSTEM

孔浩森

11749016

硕士

金融

刘鲁川

2019-05-25

2019-07-05

哈尔滨工业大学

深圳

随着我国资本市场的不断扩容，越来越多的公司在市场上公开发行股票，募集资金。同样，伴随着国民经济的不断发展，居民收入水平不断提高，越来越多的居民将购买股票作为一种重要的投资理财方式。而我国的股票投资市场仍以散户为主，从众多股票里进行股票挑选，并推荐给个人投资者，尤其是缺少投资经验和相应专业知识的投资者，为其提供决策支持，具有很大的实践价值。本文对股票的推荐方法进行了分类，明确提出了个性化推荐方法和非个性化推荐方法。个性化推荐方法是借鉴商品领域的个性化推荐算法，将其移植到股票个性化推荐领域，面向某个投资者。而个性化推荐存在个人的交易持仓等隐私数据难以获得和根据个人投资的历史数据及个人风险偏好特征得出的推荐股票，未必能取得良好的市场表现等问题，因而本文推荐算法关注的重点在于非个性化推荐领域，面向所有投资者。本文重点针对非个性化推荐领域的传统方法进行了梳理，并将其分为传统非个性化推荐方法和量化非个性化推荐方法两类。在非个性化推荐领域，择股与择时，始终都是股票投资过程中绕不开的重要问题。在构建股票择股推荐方法的过程，本文采用了GARP策略的基本思想，选取中证800指数这一具有代表性的基准组合，并从众多指标中筛选出了有效的单因子，然后构建了一个多因子组合，并对该组合的择股推荐效果进行了敏感性分析，进一步优化了投资组合的选取过程。在对数据挖掘分类算法进行了梳理，并对支持向量机算法的原理进行了详细分析之后，本文提出了基于支持向量机算法的择时模型。具体是选取中证800指数的日频数据，并进行处理，设置了自变量，对中证800指数的涨跌情况的因变量进行分类预测处理。经过实证检验，证明这种择股加择时的推荐算法模型在股票非个性化推荐领域具有良好的效果。

With the continuous expansion of China's capital market, more and more companies are publicly issuing stocks in the market and raising funds. Similarly, along with the continuous development of the national economy, the income level of residents continues to increase, and more and more residents will purchase stocks as an important means of investment and financing. However, China's stock investment market is still dominated by individual investors. Stocks are selected from a large number of stocks and recommended to individual investors, especially investors who lack investment experience and corresponding professional knowledge to provide decision support. This paper classifies the recommendation methods of stocks, and clearly proposes personalized recommendation methods and non-personalized recommendation methods. The personalized recommendation method is to learn from the personalized recommendation algorithm in the commodity field and transplant it into the stock personalized recommendation field for an investor. It is difficult to obtain private data and recommended stocks based on personal investment historical data and personal risk preference characteristics, may not be able to achieve good market performance , so the focus of this paper is Non-personalized recommendation for all investors. This paper focuses on the traditional methods of non-personalized recommendation, and divides it into two types: traditional non-personalized recommendation methods and quantitative non-personalized recommendation methods. In the field of non-personalized recommendation, stock selection and timing are always important issues that cannot be avoided in the stock investment process.In the process of constructing stock recommendation recommendation method, this paper adopts the basic idea of GARP strategy, selects the representative benchmark combination of CSI 800 Index, and selects effective single factor from many indicators, and then constructs a Multi-factor combination, and I used sensitivity analysis of the portfolio recommendation effect of the combination to further optimized the selection process of the portfolio. After sorting out the data mining classification algorithm and analyzing the principle of SVM algorithm in detail, this paper proposes a timing model based on support vector machine algorithm. Specifically, the daily frequency data of the CSI 800 Index is selected and processed, and the independent variables are set to classify and predict the dependent variable of the CSI 800 index. After empirical test, it is proved that the recommendation algorithm model of this stock selection has a good effect in the field of stock non-personalized recommendation.

股票推荐系统 择股 择时 GARP策略 支持向量机算法 实证分析

stock recommendation system stock selection timing GARP strategy support vector machine algorithm empirical analysis

中文

联合培养

南方科技大学

孔浩森. 股票推荐系统的算法[D]. 深圳. 哈尔滨工业大学,2019.