Two Essays in Asset Pricing and FinTech: Out-of-Sample Equity Premium Predictability and Stock Co-jump Networks with Mixed Membership

Abstract

In Chapter 1, we introduce a new method for forecasting stock returns. Despite exhibiting strong in-sample predictive power, a wide range of predictors proposed in the literature are shown by Goyal and Welch [2008] to underperform the historical average in out-of-sample forecasts of the equity premium. We propose an unconventional approach for out-of-sample equity premium prediction that avoids parameter estimation, adopting a conservative constant as the predictive coefficient. Our methodology retains the same zero-variance advantage as the historical average, while achieving lower bias, thereby outperforming the benchmark. We show that our forecast first-order stochastically dominates the historical average. Using our methodology, we reveal that many predictors exhibit statistically and economically significant out-of-sample gains for the market return. We also compare our method with machine learning models and apply our framework to out-of-sample bond return predictability.

In Chapter 2, we study stock dependence across many firms using pairwise co-jump networks with high-frequency data. Stock cojumps contain important information about the risk linkage among stocks. Ding et al. [2024] discovered a block structure among stocks based on their co-jumps and proposed a model DCBM-DMP. In this paper, we exhibit that this structure can greatly benefit from an essentially additional component: mixed membership. Specifically, we propose a Degree Corrected Mixed Membership network model with Dependent Multivariate Poisson edges (DCMM-DMP) and develop a Mixed Spectral Clustering On Ratios-of-Eigenvectors for networks with Dependent Multivariate Poisson edges (Mixed-SCORE-DMP) algorithm. We show that Mixed-SCORE-DMP is asymptotically consistent in estimating the mixed membership structures. Empirically, we show that (1) the mixed membership network DCMM-DMP enhances DCBM-DMP by providing a more precise risk structure; (2) “peers” defined by our mixed membership model offer significant advantages in return prediction over benchmarks, such as GICS, self-grouping, or counting analyst coverage linkage; (3) a “purity” measure based on our model provides insightful perspectives about stocks’ risk profile and investment opportunities. We also construct a lead-lag jump network to study the leading and lagging group effect.

Date of Award	2025
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology
Supervisor	Yingying LI (Supervisor) & Jialin YU (Supervisor)