基于数据挖掘方法的量化交易系统设计与研究

发布时间：2018-06-24 01:17

本文选题：数据挖掘 + 量化选股　；参考：《安徽工业大学》2017年硕士论文

【摘要】：截止到2013年底,全球所有对冲基金管理的资金规模已经超过27千亿美元。使用量化投资方式管理的各类基金和资管产品的资金量大约占到全球总交易量的3成左右,在全球各种大型的证券交易所中,各类量化投资方式贡献了将近50%的成交量。量化交易策略的构建,首先要对证券期货市场上的信息进行统计分析,然后对量化模型进行历史数据的回测,回测效果好且稳定的模型才会投入到实盘操作中。本文针对量化交易的实际应用,设计了一种基于数据挖掘方法的量化交易系统,所用的主要开发工具为数值计算软件MATLAB,设计了量化选股、策略回测、时序分析和组合管理4大核心模块,以支持简单的交易决策。本文主要内容和结果如下:(1)利用多因子模型进行选股并以此构建投资组合,组合能在较长周期范围内大幅跑赢业绩比较基准,具有稳步上升的超额收益。使用SVM算法构建股票涨跌分类器,并对所有股票的未来涨跌进行预测。更重要的是提出了将层次聚类与K-均值聚类叠加,构造两阶段聚类模型,最终选出了盈利能力最强的股票类别。(2)利用沪深300股指期货合约的5分钟收盘数据对经典的双均线趋势策略进行回测,通过参数扫描法优化均线参数,在样本内测试集得到了较高的年化夏普比率。(3)利用ARIMA模型和灰色预测模型预测个股股价走势,马氏预测模型和SVM回归模型用来预测大盘指数走势。结果显示对于个股的预测结果误差较大,但对于大盘指数的预测精度较高。(4)运用均值-方差模型,对过去一段时间期望收益率最高的20只股票进行资产配置,设置个股、行业大类配置比例,确定当组合风险最小时,各股票的权重占比。运用绩效评估指标,对3只具有代表性的公募基金进行分析比较,并评估其整体表现。利用VAR模型的3种计算方法,计算了组合的风险价值。
[Abstract]:By the end of 2013, all hedge funds managed more than $2 trillion and 700 billion in the world. The amount of funds and management products managed by quantitative investment accounts for about 3 of the global trading volume. In all the large stock exchanges around the world, a variety of quantitative investment methods have contributed nearly 50%. The construction of quantitative transaction strategy is to make a statistical analysis of the information on the stock and futures market, and then to test the historical data of the quantitative model, and to return the good and stable model to the actual operation. In this paper, a quantized method based on data mining is designed to quantify the actual application of the transaction. Trading system, the main development tool used for the numerical computing software MATLAB, designed the quantitative selection of stock selection, strategy back test, time series analysis and combination management 4 core modules to support simple transaction decision-making. The main content and results are as follows: (1) using multi factor model into stock selection and building investment portfolio, combination can be in a long week. The SVM algorithm is used to build the stock fluctuation classifier and predict the future rise and fall of all stocks. The more important thing is to put forward a hierarchical clustering and K- mean clustering to construct the two stage clustering model, and finally select the most profitable stock. Category. (2) using the 5 minute closing data of the Shanghai and Shenzhen stock index futures contract in 5 minutes, the classic dual trend strategy is measured, the parameters are optimized by the parameter scanning method, and the higher annual SHARP ratio is obtained in the sample test set. (3) the ARIMA model and the grey pretest model are used to predict the stock price trend, martensitic prediction model and SVM The regression model is used to predict the trend of the big plate index. The results show that the prediction results for the stock are more accurate, but the prediction accuracy is higher for the big plate index. (4) the average variance model is used to configure the assets of the 20 stocks with the highest expected return on the past period of time, set up a stock, the proportion of the industry large category, and determine the combination wind. The risk is the hour, the weight of each stock is accounted for. By using performance evaluation index, 3 representative public funds are analyzed and compared, and their overall performance is evaluated. By using 3 methods of VAR model, the risk value of the combination is calculated.
【学位授予单位】：安徽工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP311.13;F724.5

【参考文献】