消费的资本资产定价模型（CCAPM）是将微观资产收益与宏观经济波动相联系的可估计理论模型，它既可以解释居民消费行为特征，也可以解释居民的储蓄（投资）行为特征。本文以传统CCAPM模型为基础，将货币谨慎动机持有与住房服务消费加入到时间可分及不可分的跨期效用模型中建立了多个修正模型，以此研究中国居民部门的消费和储蓄（投资）行为特征。这些模型旨在：1）考察居民的一般消费、住房服务消费和货币持有等因素对中国股票资产和住房资产的影响；2）探讨何种模型可以更准确地同时解释中国居民的消费行为和资产持有行为；以及，3）通过效用函数的具体形式及其参数估计，探讨中国居民如何在各期之间、消费品与资产组合之间进行资源配置，以及如何应对不确定性。在分析中，本文将上述所有模型分别应用于分析中国居民持有股票资产与住房资产收益率的实证检验中。本文同时采用了广义矩估计法（GMM）和最小距离筛分法（SMD），利用中国1991年第一季度到2019年第四季度股票资产市场与住房资产市场相关数据，以及消费增长率等宏观变量的季度数据对各不同模型进行了实证估计，并将各模型拟合效果运用恰当的统计指标进行比较。通过比较参数估计结果与模型拟合表现，可以得到如下基本结论：1）采用传统的CCAPM模型无法较好地拟合股票资产与住房资产收益率实际数据，只有将居民的货币持有与住房因素加入到模型中，才可以对股票和住房资产的定价给予较好的解释；2）加入货币持有和住房的递归效用模型是对中国股票资产和住房资产定价解释力最强的模型，该模型构成的定价核可以对两种资产的收益率变化进行较好的拟合；3）半参SMD的估计方法放松了模型假设，与原始模型希望表达的含义更为接近，可以显著提升模型拟合效果；4）中国居民在消费和投资决策中，表现出较强的风险厌恶，这具体体现在两方面：一是中国居民风险厌恶系数较高并且倾向于提前解决不确定性；二是为应对各种不确定性和冲击，中国居民愿意持有较大比率的货币资产，对住房资产所带来的效用也十分看重。
Consumption-based asset pricing model (CCAPM) is an empirically testable framework that links micro-level asset returns and macro-level fluctuations. It is a useful tool to explain both consumption and saving (investment) behaviors. This thesis provides various refinements to the classical model by incorporating currency holding and/or housing service into time-separable and nonseparable preferences. Through these new variants of CCAPM, this thesis a) investigates how household consumption, housing service consumption and currency holding affect China stock and real estate asset pricing, b) tests which model has the best explanatory power for Chinese household consumption and investment, and c) by estimating the structural parameters, discusses how Chinese households allocate resources intratemporally among consumption and intertemporally over time, and how they treat uncertainty. To achieve the goals, we estimate and statistically compare the performance of each model using China stock market and consumption data. We use both generalized method of moments (GMM) for the parametric models and sieve minimum distance estimator (SMD) for the semiparametric models, and apply the estimators to quarterly data of stock, real estate and consumption from 1991 to 2019. Based on the results, we find that a) including currency holding and housing service consumption in the utility function pushes up the goodness to fit to a level that cannot be attained using the classical model, b) under recursive preferences, these two elements lead to the best CCAPM among the variants we consider, which has the greatest explanatory power and the best performance in in-sample prediction of the asset returns, c) SMD estimator achieves the best performance by relaxing assumptions and preserving a flexible functional form of the recursive preferences, and d) Chinese households have strong risk aversion as i) they have a large coefficient of relative risk aversion and tend to resolute uncertainties early and ii) they value money and housing as precautions against uncertainties and shocks.
