Essays in macro asset pricing models with news sentiment

Abstract

This thesis investigates the role of investor sentiment in asset pricing through the lens of Shefrin's behavioral pricing kernel, quantile preferences, and probability weighting functions from non-expected utility theory. Each chapter examines a distinct dimention of how sentiment influences asset prices, offering both theoretical insights and empirical validation across different model frameworks.

Chapter 2 conducts an empirical test of the consumption-based CAPM by approximating the stochastic discount factor (SDF) as a linear function of consumption growth and news sentiment. This approach aims to reconcile the tension between the high cross-sectional explanatory power of conditional asset pricing models and the underlying assumptions. By explicitly incorporating sentiment into the pricing kernel, the model offers a more behaviorally consistent representation of investor preferences. Notably, the inclusion of sentiment results in a less negative SDF and a more moderate estimate of risk aversion, thereby aligning empirical estimates more closely with economically plausible values.

Chapter 3 introduces a quantile-preferences asset pricing model that incorporates a sentiment index constructed using Principal Component Analysis (PCA) on a comprehensive set of survey-based, market-based, and news-based sentiment indicators. The framework leverages a Panel Quantile Regression (PQR) model to investigate how sentiment influences the cross-sectional distribution of asset returns, with particular emphasis on extreme tail events. By controlling for unobserved heterogeneity across financial assets, the PQR approach enables the identification of systematic patterns in the tail behavior of returns. A comparison between PQR estimates and those from individual Univariate Quantile Regressions (UQR) reveals that, once heterogeneity is accounted for, the first principal component (PC1)—interpreted as a proxy for aggregate sentiment—exerts a more pronounced negative effect on the upper quantiles of returns than indicated by most UQR estimates.

Chapter 4 investigates the relationship between tail-overweighting parameters embedded in the probability weighting functions of non-expected utility models and an external sentiment measure, proxied by the first principal component (PC1) extracted via PCA in Chapter 3. Among the parametric weighting functions analyzed, the Prelec probability weighting function exhibits the strongest empirical alignment, with its curvature parameter $\alpha$ showing a notable $80 \%$ correlation with PC1. The findings offer empirical evidence of a significant correlation between probability weighting parameters and an external sentiment measure in real-world, non-experimental data. This contributes to the broader behavioral asset pricing literature by shedding new light on the dynamic nature of probability weighting and its implications for resolving the pricing kernel puzzle.

More information

Identifiers

ORCID for Xuefeng Yang: orcid.org/0009-0005-7902-9854

ORCID for Michael Hatcher: orcid.org/0000-0001-8506-1950

ORCID for Jose Olmo: orcid.org/0000-0002-0437-7812

Catalogue record

Export record