美联储-期权的定价核心（英）-2023.8-66页_2mb

The paper introduces advanced pricing kernels for option valuation, extending the economically appealing Rubinstein-Brennan kernels to a dynamic framework that allows both path-dependence and volatility-dependence. The authors specify pricing kernels as functions of latent variance paths and index levels, nest path-independent kernels (Ross, 2015) as special cases, and ensure consistency with affine dynamics in models like Heston (1993). Different kernel specifications, like completely affine versus affine prices of risk, can yield similar option fit but dramatically different economic implications, such as risk premiums and Sharpe ratios. The empirical analysis shows that affine pricing kernels can lead to implausible economic assumptions, whereas completely affine kernels provide more reasonable option fit and risk premium estimates when combined with appropriate parameter restrictions. The data (S&P 500 options and returns) provide insufficient power to distinguish these models statistically, confirming concerns similar to those raised by Merton (1980) about identifying long-term risk premia. The paper demonstrates that U-shaped or non-monotonic pricing kernels are not anomalies but naturally arise from the structure of the kernel when monotonicity restrictions are relaxed, providing a resolution to the pricing kernel puzzle. Furthermore, it highlights that even small modifications in the pricing kernel specification can lead to significantly different economic outcomes, challenging the common practice of relying solely on option fit to identify risk premia and implying the importance of careful economic restrictions in model estimation. The findings suggest that option pricing models with standard affine specifications may have implausible economic implications, emphasizing the need for rigorous econometric and economic assessment of dynamic asset pricing models.