【摘要】 风险与收益是投资者最关心的两个核心变量，因此，收益率的预测成为国内外学者的研究焦点。在经典的资产定价框架下，同期的预期收益率和方差之间存在理论关系。由于方差是时变的，在当前时刻无法知道未来方差。不少学者利用期权隐含方差、隐含高阶矩、方差风险溢酬预测未来收益率。实际上，隐含方差是风险中性世界中的预期方差，它等于现实世界的预期方差加方差风险溢酬。因此，用隐含方差或方差风险溢酬预测未来收益率存在遗漏预测变量的问题。为了解决这个问题，采用当前时刻的现实世界的预期方差预测未来的收益率。本研究从方差分解和风险与收益率非线性的角度，考察是否可以提高方差预期值对中国股票市场收益率的预测效果，并考察方差分解和非线性两者的叠加效果对股票市场收益率的预测效果，还从资产配置的角度评估方差非线性项预期值以及考虑方差分解对股票市场收益率预测的经济意义。研究结果表明，在2003年至2022年样本期内，方差预期值对未来股票市场收益率有负向的预测能力；综合考虑方差分解和风险与收益率非线性，可以最大程度上提高方差预期值指标对未来收益率的预测能力，取得经济上和统计上都显著的预测效果；说明方差预期值系列指标在中国股市可作为月度收益率的预测指标。在近十年样本内，无论是已实现方差预期值、下行已实现方差预期值、上行已实现方差预期值，还是它们的二次方和三次方，对未来收益率的预测力都更强。基于方差分解和非线性的研究视角，拓展了方差对收益率预测能力的研究，深化了对股票整体收益率可预测性的理解。研究结论对投资中国股票市场的理性投资者具有实践指导意义。理性投资者可以利用这种稳健的样本外预测能力，有效提高资产配置的效用水平，特别是短期指数投资者应多关注下行已实现方差的高次项风险。更多还原

【Abstract】 Risk and return are the two core variables that investors are most concerned, and the prediction of return has therefore become the focus of research by domestic and foreign scholars.In the classical asset pricing framework, there is a theoretical relationship between the expected return and variance in the same period. Since the variance is time-varying, the future variance is unknown at the current moment. Many scholars use implied variance, implied higher-order moments, and variance risk premium extracted from options to predict future returns. In fact, the implied variance is the expected variance in the risk-neutral world, which is equal to the expected variance in the physical world plus the variance risk premium. Therefore, there is a problem of missing predictive variables when using implied variance or variance risk premium to predict future returns. In order to solve this problem, this study uses the expected variance of the physical world at the current moment to predict future returns.This study examines whether the prediction effect of the expected variance on China′s stock market returns can be improved from the perspective of variance decomposition and nonlinearity of risk and return. At the same time, explore the predictive effect of the intersection of decomposition and nonlinearity. We also assess their economic significance from an asset allocation perspective. Finally, the robustness of the results was tested by replacing different indices and estimating methods of the expected variance.The results show that the expected variance has negative predictive power for the stock market returns during the sample period from 2003 to 2022. And the intersection of nonlinearity and decomposition can greatly improve the ability of the expected variance for the next month, with the predicted effect being both economically and statistically significant. It shows that the expected variance series indicators can be used as a predictor of monthly returns on China′s stock market returns. Moreover, in the sample of the past ten years, whether it is the total expected realized variance, expected downside realized variance, expected upside realized variance, or their quadratic and cubic, they are more predictive of stock returns.Based on the perspective of variance decomposition and nonlinearity, it expands the study of variance′s ability to predict returns, and deepens the understanding of the predictability of the stock market returns. Relevant conclusions have practical guiding significance for rational investors investing in China′s stock market. And rational investors can use this robust out-ofsample predictive ability to improve the utility level of asset allocation. In particular, short-term index investors should pay more attention to the higher order risk of downside realized variance.更多还原