近年来，量化投资从幕后走向台前，成为个人投资者获取超额收益的重要手段，大数据、机器学习等计算机技术的革新，也不断地推动着量化投资向着更完善的方向发展。遗憾的是，尽管国内量化投资的潮流日渐兴盛，但与国外相比中国的量化投资仍然处于一个初级阶段。目前国内关于量化投资的研究大多聚焦在FF-3、FF-5等因子模型上，但实践证明上述模型对于A股的适用性较弱，不太符合A股的市场结构。此外，包含了流动性、波动性、投资者情绪和错误定价等诸多信息在内的交易摩擦类因子在外国股票市场上已经验证了有效性，但在中国股市上的应用仍相对较少。为了构建更符合A股市场的多因子定价模型，提高模型的解释能力，本文以更加符合A股市场结构的CH-3因子模型为基准，通过等权重法、综合Z打分法和IC均值加权法分别提取交易摩擦信息，构建三种交易摩擦因子，并将上述三种交易摩擦因子与CH-3因子模型相结合，形成交易摩擦四因子模型，以期获得更具解释能力和有效性的多因子模型。
本文的实证共分为三部分，第一部分是构建交易摩擦因子，本文将九个交易摩擦类原始因子作为候选因子，用IC检验和Fama-Macbeth回归两种检验方法进行因子筛选，并剔除效果较差的三个因子，对剩下的六个因子通过等权重法、综合Z打分法和IC均值加权法构建三种交易摩擦因子，描述性统计结果显示三种交易摩擦因子与收益率均为负相关；第二部分是检验三种交易摩擦因子模型的解释能力，本文首先使用5×5分组检验对FF-3、CH-3和三种交易摩擦因子模型在两种维度上分25组回归，发现交易摩擦因子模型在A股市场中具备较强的解释能力，且采用综合Z打分法和IC均值加权法构建的交易摩擦因子PTF和PTF-IC要优于等权重法构建的交易摩擦因子PTF-IM；之后分别对上述模型使用GRS检验，并发现引入了PTF因子和PTF-IC因子的多因子模型的解释能力相对较强；第三部分是五种模型的资本市场回测检验，本文在定义了交易参数和变量的基础上，运用聚宽量化平台进行了回测，结果发现PTF-IC4因子模型获得了最高的超额收益，具备良好的市场表现。上述实证结果表明，引入了交易摩擦因子的多因子模型具备优秀的适用性和有效性，采用综合Z打分法和IC均值加权法均能有效提取交易摩擦类因子所含的信息且后者在实际资本市场上的表现更优。

In recent years, quantitative investment has become an important means for individual investors to obtain excess returns from behind the scenes to the front. The innovation of computer technology such as big data and machine learning also continuously promotes the development of quantitative investment in a more perfect direction. Unfortunately, despite the growing trend of quantitative investment in China, it is still at an early stage compared with foreign countries. At present, most domestic researches on quantitative investment focus on FF-3, FF-5 and other factor models, but the practice has proved that the above models are weak in applicability to A-shares and do not conform to the market structure of A-shares. In addition, trade friction factors, which include a wide range of information such as liquidity, volatility, investor sentiment and mispricing, have been proven effective in foreign stock markets, but are still relatively rarely used in China's stock market. In order to build A multi-factor pricing model more in line with the A-share market and improve the explanatory ability of the model, this paper takes the CH-3 factor model more in line with the A-share market structure as the benchmark, constructs three trading friction factors by equal weight method, comprehensive Z-scoring method and IC mean weight method, and combines the above three trading friction factors with the CH-3 factor model. A four-factor model of transaction friction is constructed in order to obtain a more explanatory and effective multi-factor model.
The empirical evidence in this paper is divided into three parts. The first part is the construction of trading friction factors. Nine original trading friction factors are taken as candidate factors in this paper, and the three factors with poor effect are eliminated by IC test and Fama-Macbeth regression. Three trade friction factors were constructed for the remaining six factors by equal weight method, comprehensive Z-scoring method and IC mean weight method. Descriptive statistical results showed that the three trade friction factors were negatively correlated with the rate of return. The second part is to test the explanatory ability of the three trading friction factor models. In this paper, the 5×5 grouping test is used to divide the FF-3, CH-3 and three trading friction factor models into 25 groups of regression, and it is found that the trading friction factor model has A strong explanatory ability in the A-share market. Moreover, the trade friction factor constructed by comprehensive Z-scoring method and IC mean weight method is better than that constructed by equal weight method. Then GRS test was used to test the above models, and it was found that the multifactor model with the introduction of PTF factor and PTF-IC factor had the strongest explanatory ability. The third part is the back test of the five models. On the basis of defining the trading parameters and variables, this paper uses the broadening quantization platform to carry out the back test. The results show that the PTF-IC4 factor model obtains the highest excess return and has a good market performance. The above empirical results show that the multi-factor model with the transaction friction factor has excellent applicability and effectiveness. The comprehensive Z-scoring method and IC mean weighting method can effectively extract the information contained in the transaction friction factor, and the latter has better performance in the actual capital market.