基于高频数据下商品期货统计套利分析

发布时间：2018-04-15 20:06

本文选题：统计套利 + 协整　；参考：《浙江工商大学》2014年硕士论文

【摘要】：量化投资在海外的发展已有30多年的历史,其投资业绩稳定,市场规模和份额不断扩大,时至今日,量化投资已经成为美国市场上一种重要的投资方法。国内股指期货上市,融资融券业务、国债期货的推出以及期权的筹备,使得量化投资策略的实现有了一定的基础。而且,2012年11月,期货资产管理业务正式开闸,将给期货行业带来结构性变化。由于期货的杠杆性,风险比股票市场高很多,因此以程序化和量化投资为核心策略的交易将成为资产管理业务主要的投资方式,以追求稳定的收益。统计套利是量化投资中一种非常重要的方法,它是一种市场中性策略,即收益不受“牛熊”市的影响。本文主要研究了统计套利在商品期货市场上的应用。通过流动性检验、相关性检验以及协整检验,最终我们选择RB1309和RB1310为实证对象。首先,我们提出了基于协整理论的统计套利框架,主要介绍了套利合约的选择、套利成本费用的计算、套利流程的设计以及策略的绩效评价；其次,考虑到金融市场具有时变的特性,因此,我们将滑动窗口理论引入到统计套利里来,使参数具有时变的特征。滑动窗口下的统计套利策略主要有以下几个研究：第一：对开仓阂值和止损阈值进行了优化。传统的开仓阈值采用固定值的形式。众所周知,当开仓阈值较大时套利机会减少,而阈值较小时则可能出现频繁开仓导致严重亏损的情况。因此,基于动态优化的思想,我们采用窗口内收益最大化来确定开仓阈值,以实现自适应的动态调整参数,避免阂值过高或者过低而影响收益。对于止损,我们基于VaR的思想来确定阈值,并通过GARCH模型来估计价差的波动率。第二：冲击成本的估算。冲击成本是交易成本中非常重要的组成部分。很多策略实盘运行时总是离预期相差较大,这主要是忽略了对冲击成本的考虑。很多相关文献同样也考虑了冲击成本,但都是直接给定,没有通过数学模型进行比较客观的估算。本文利用商品期货的分笔数据,通过相应的数学模型估算出了套利合约的冲击成本。第三：协整系数的估计以及策略回测。不同的协整系数会导致不同的价差,从而产生不同的套利过程。本文采用OLS估计和时变系数估计两种方法来获得协整系数,并将两策略分别作用于测试样本的回测,效果较好。同时,通过敏感性分析给出了参数的稳健性。考虑到两策略相对独立,因此,我们对OLS策略和时变系数策略进行了有效组合,并进行了3个月的历史回测,通过收益分析和风险分析得出,组合模型是有效的。最后,本文给出了研究的不足以及展望。
[Abstract]:Quantitative investment has been developed overseas for more than 30 years. Its investment performance is stable and its market size and market share is expanding. Up to now, quantitative investment has become an important investment method in American market.The listing of domestic stock index futures, the financing and margin trading, the introduction of treasury bond futures and the preparation of options make the realization of quantitative investment strategy have a certain basis.Moreover, in November 2012, the futures asset management business officially opens the gate, will bring the structural change to the futures industry.Because of the leverage of futures, the risk of futures is much higher than that of stock market, so the transaction with program and quantitative investment as the core strategy will become the main investment way of asset management business, in order to pursue stable income.Statistical arbitrage is a very important method in quantitative investment. It is a market neutral strategy, that is, the income is not affected by the "bull bear" market.This paper mainly studies the application of statistical arbitrage in commodity futures market.Through liquidity test, correlation test and cointegration test, we choose RB1309 and RB1310 as the empirical object.Firstly, we propose a statistical arbitrage framework based on cointegration theory, which mainly introduces the selection of arbitrage contract, the calculation of arbitrage cost, the design of arbitrage process and the performance evaluation of strategy.Considering that the financial market is time-varying, we introduce the sliding window theory into statistical arbitrage to make the parameters time-varying.The statistical arbitrage strategy under sliding window is mainly studied as follows:First, the opening threshold and stop loss threshold are optimized.The traditional opening threshold is in the form of fixed value.As we all know, when the opening threshold is large, the arbitrage opportunity decreases, and the low threshold may lead to serious losses caused by frequent opening.Therefore, based on the idea of dynamic optimization, we use the maximization of income in the window to determine the opening threshold, so as to realize adaptive dynamic adjustment parameters to avoid the effect of too high or too low threshold value on revenue.For stop loss, we determine the threshold based on the idea of VaR, and estimate the volatility of price difference by GARCH model.The second is the estimation of impact cost.Impact cost is a very important part of transaction cost.Many strategies are always running at a much different time than expected, which largely ignores the cost of impact.Many related literatures also consider the impact cost, but they are all given directly and not estimated objectively by mathematical model.In this paper, the impact cost of arbitrage contract is estimated by the corresponding mathematical model.The third is the estimation of cointegration coefficient and the strategy back measurement.Different cointegration coefficients will lead to different spread, which will result in different arbitrage process.In this paper, two methods, OLS estimation and time-varying coefficient estimation, are used to obtain cointegration coefficients.At the same time, the robustness of the parameters is given by sensitivity analysis.Considering that the two strategies are relatively independent, we combine the OLS strategy and the time-varying coefficient strategy effectively, and carry out a 3-month historical review. Through the income analysis and the risk analysis, we conclude that the combination model is effective.Finally, this paper gives the deficiency and prospect of the research.
【学位授予单位】：浙江工商大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F724.5

【相似文献】