REITs是区别于股票和债券之外，兼具股性及债性的金融资产，投资范围以不动产为主。美国自1960年以来即有REITs产品问世，新加坡、澳大利亚、英国等国家亦有较为成熟的REITs市场。我国于2021年6月21日开始进行REITs试点，首批共计9只，并具备底层资产优质等特点。截至2023年2月23日，我国共有25只REITs上市，种类包括产业园类、高速公路类、仓储物流类、生态环保类及保障性租赁住房类。由于我国REITs起步较晚，相关研究不如国外充分且定量研究较为缺乏。因此笔者期望通过定量研究方法探究影响我国REITs二级市场表现的因素，在为国内后续研究提供参考的同时，为C-REITs投资提供可行方案。本文中，首先收集了49个关于债券市场、股票市场及宏观因素的日频数据，接着使用多元线性回归方法分别对我国公募REITs产品按市值加权的日收益率及换手率进行研究，识别了包括中期票据收益率，城投债收益率，股票指数收益率在内的影响因素。随后进一步进行Granger因果检验，考察变量对二级市场表现在时间序列上的预测能力，发现一阶差分后的5年期AAA+中债中期票据到期收益率及上证50指数日收益率对日收益率存在预测能力；一阶差分后的WIND房地产指数对日换手率具备预测能力。最后，使用多种机器学习方法进一步识别具有预测能力的变量，并比较包含不同变量的模型预测能力。发现使用多元线性回归法识别因素的机器学习方法在预测中表现最好，基于此构建的纯多头策略在考虑成本的情况下能产生超额收益，是C-REITs投资的可行方案。
REITs are financial assets whose return patterns have both equity and debt properties. Their investment scope is mainly real estate. REITs products have been available in the United States since 1960. Countries such as Singapore, Australia and the United Kingdom also have relatively mature REITs markets. China began to carry out the REITs pilot project on June 21, 2021, with a total of 9 in the first batch, which has the characteristics of high-quality underlying assets. As of February 23, 2023, there were 25 REITs listed in China, including industrial parks, highways, warehousing and logistics, ecological and environmental protection and affordable rental housing.Since China‘s REITs started relatively late, the relevant researches, especially quantitative ones, are lacking. Therefore, the author hopes to explore factors affecting the secondary market performance of C-REITs through quantitative research methods, thus providing reference for the follow-up researches and a solution for C-REITs investment.In this paper, 49 daily frequency data on bond market, stock market and macro factors are collected. Then multiple linear regression is applied to study the secondary market performance of C-REITs weighted by market value, and identify influencing factors, including the yield of medium-term notes, the yield of chentou bonds, and the return of stock index. Granger causality test is further carried out to examine the predictive ability of variables on performance of C-REITs in time series. It is found that the yield to maturity of five-year AAA+ medium-term notes after the first order difference and daily yield of SSE 50 index have the predictive ability on daily return; The WIND real estate index after the first-order difference has the ability to predict daily turnover rate. Finally, variables are further filtered using multiple machine learning methods, and the performance of models containing different variables is compared. It is found that the machine learning methods using factors identified by multiple linear regression are the best in forecasting. The long only strategy based on this can generate excess returns under the consideration of cost, which is a feasible solution for C-REITs investment.
