基于均值回归的超额价差套利策略研究

Research on Excess Spread Arbitrage Strategy Based on Mean Regression

随着机器学习量化策略的普及,市场反而对简单有效的低维策略需求上升。本文基于均值回归理论的基本逻辑,构建带有前期价格信息的移动权重模型,该模型打破了原先对均值回归主次资产组合的限制,具有更强的统计意义,同时存在操作难度较大的问题。在策略构建中,围绕避免过拟合的思路,对模型的复杂部分进行简化,并设计出可行的基于均值回归的超额价差套利策略,该策略在2013—2020年螺纹钢期货的历史回测中取得220.0%的总收益率和27.5%的年化收益率,并在2021年的真实回测中取得了几何年化收益率245.81%和夏普比率9.39的优异成绩。此外,交易策略还具有净值曲线平滑、回撤小的优势,策略在期货市场普遍适用,在其他市场也有较广的应用空间。

With the popularization of machine learning quantitative strategies, the market has increased demand for simple and effective low dimensional strategies. Based on the basic logic of mean regression theory, this paper constructs a mobile weight model with early price information. The model breaks the original restriction on the primary and secondary portfolio of mean regression, and it has stronger statistical significance, meanwhile, it is hard to operate. In the strategy construction, focusing on the idea of avoiding over fitting, this paper simplifies the complex part of the model, and designs a feasible excess spread arbitrage strategy based on mean regression. The strategy achieved a total return of 220.0% and an annualized return of 27.5% in the historical back test of rebar futures from 2013 to 2020. In the real back test in 2021,the geometric annualized rate of return is 245.81% and the sharp ratio is 9.39. In addition, the trading strategy also has the advantages of smooth net worth curve and small pullback. The strategy is generally applicable in the futures market and has a wide application space in other markets.