IF1407合约的长记忆回归模型与投资策略研究

A long memory regression model of IF1407 contract and investment strategy

在高频交易中,金融资产价格变化的时间持续期是交易者关注的重点.在以往的文献中,人们一般是用ARFIMA(p,d,q)模型或者ACD模型研究时间持续期.然而,我们在分析股指期货的高频数据时发现,时间持续期不仅与其自身滞后期相关,还与这期间内没有发生价格变化的交易次数和价格变化的具体值有关.因此,本文将ARFIMA(0,d,0)模型与回归模型结合起来,提出了时间持续期与其他因素存在线性关系的新模型,并且时间持续期还具有长记忆特征.我们试图用此模型研究高频交易中价格变化的时间持续期、持续期内没有发生价格变化的交易次数以及价格变化的具体值三者之间的关系.模拟结果表明我们提出的profile-最小二乘法能够较好地估计新模型中的参数.实证部分用我国沪深300股指期货的高频交易数据来说明新模型的应用价值.

The time duration for price changes of financial asset is the focus of attention of trades in high frequency data analysis. In the past, people generally use the ARFIMA （p, d, q） model or the ACD model to study time duration. When high frequency data of stock index futures are analyzed, the time duration is found to be related not only to its lag terms, but also to the number of trades in the period with no price change, and to the size of the price change in Renminbi yuan. The ARFIMA （0, d, 0） and the regression models are combined as a new model that allows the study of the linear relationship between time duration, which has a long memory, and other factors. This model is used to study the relationships among three parameters. （1） time duration between two price changes in high frequency trading, （2） number of trades during period of no price change, and （3） value of price change. Simulation results show that it is appropriate to use the profile least-squares method to estimate parameters in the new model. We used high frequency trading data of Shanghai and Shenzhen 300 stock index futures to illustrate application value of the new model.