一个基于FLS算法的高频统计套利交易系统

发布时间：2018-02-16 16:59

本文关键词： 算法交易 FLS 卡尔曼滤波时变系数在线估计　出处：《西南财经大学》2014年硕士论文　论文类型：学位论文

【摘要】：算法交易也称为程序化交易,是一种将投资策略逻辑程式化,然后借助计算机进行买卖决策的投资方式。在实际运用中,一个高效的交易系统需要实时的分析和处理海量数据,然后给出交易信号。其中最为重要的一个任务便是在不施加任何概率先验假设的条件下挖掘出各个数据流的相关关系以及其随时间的演变情况。本文首先介绍了一种估计时变参数的有效算法——灵活最小二乘算法(flexible least squares,简称FLS),该算法是普通最小二乘的一种变形,它能够在没有概率假设的情况下快速的估计出随时间演变的回归系数。其次,为了能更好的理解该算法,本文还讨论了FLS算法与卡尔曼滤波算法之间的内在联系以及给出了一些改进。最后,我们将FLS算法运用到了高频统计套利之中,并且在此基础上构建了一套完整的程序化交易系统,并取得了良好的业绩表现。
[Abstract]:Algorithmic transaction, also known as programmed transaction, is an investment method that makes investment strategy logical and then makes trading decisions with the help of computer. In practice, an efficient trading system needs to analyze and process massive amounts of data in real time. Then the transaction signal is given. One of the most important tasks is to excavate the correlation of each data stream and its evolution over time without applying any probabilistic priori hypothesis. Flexible least squares, an efficient algorithm for estimating time-varying parameters, is a variant of ordinary least squares. It can estimate the regression coefficients over time quickly without the assumption of probability. Secondly, in order to better understand the algorithm, This paper also discusses the relationship between the FLS algorithm and the Kalman filter algorithm and gives some improvements. Finally, we apply the FLS algorithm to the high frequency statistical arbitrage. And on this basis, a complete set of program trading system, and achieved good performance.
【学位授予单位】：西南财经大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F830.59

【参考文献】