Machine learning in asset pricing and portfolio optimization

Liang, Jiawen (2025) Machine learning in asset pricing and portfolio optimization. PhD thesis, University of Glasgow.

Abstract

In the rapidly evolving field of finance, asset pricing and portfolio optimization are facing challenges due to technological advancements, shifting economic landscapes, regulatory changes, and increased complexity in financial markets. This thesis contains three essays that explore the use of advanced machine learning approaches in financial advising and asset pricing. The first essay improves robo-advisors’ performance by combining reinforcement learning (RL) with importance sampling that focuses on rare events, leading to better investment outcomes. The second essay employs inverse optimization to estimate investors’ risk aversion under normal and disaster conditions, and then optimizes portfolios based on the learnt risk aversion by deep RL. The third essay proposes a framework for asset pricing that uses neural networks to model nonlinear pricing kernels and includes considerations of environmental, social, and governance (ESG) factors in explaining cross-sectional asset prices. Details of the three essays are summarized below:

Robo-advising under rare disasters Robo-advisors provide automated portfolio management services to investors, and their growth has been unprecedented in the past few years. However, empirical evidence shows that roboadvisors underperformed during the recent COVID-19 pandemic. This may be because rare disasters are highly unlikely to occur and yet have a huge impact on financial markets. Our study develops a novel computational framework to improve the performance and robustness of robo-advising in the presence of rare disasters. It integrates RL with importance sampling. Instead of sampling the transition probability from a ground-truth probability distribution, we sample it from a proposal distribution, where the event of interest occurs more frequently. The proposed algorithm is validated by data covering the 2008 financial crisis and the COVID-19 pandemic, showing superior performance over benchmarked methods. The estimated quarterly return of the robo-advising portfolio using the optimal policy of the proposed algorithm is 0.512%, significantly higher than both the benchmarked policy and the average quarterly return, which are-0.639% and-14.55%, respectively. This improvement is attributed to targeted learning about rare disasters, enabling robo-advisors to reduce exposure to risky assets. The proposed algorithm is model-free and reduces the variance of value estimates through importance sampling.

Risk aversion and portfolio optimization for robo-advising We develop a novel framework for learning investors’ risk aversion using low-resolution data, a common issue arising from short trajectories recording investors’ portfolio choices, particularly during disaster events. Furthermore, the observed portfolio choice is often affected by behavioural biases. Our approach combines online inverse optimization with deep RLto simultaneously estimate risk aversion and determine optimal investment strategies under both normal and disaster states. Utilizing real mutual fund data, we demonstrate that our algorithm’s risk aversion estimation converges asymptotically to the optimal risk aversion during the learning process. Critically, based on the learned risk aversion and trained deep RL model, we show that robo-advisors adopting our approach can effectively tailor investment strategies to suit investor risk aversion under varying market conditions, outperforming traditional funds. This highlights the potential for our framework to enhance investment decision-making and better represent investor interests in both stable and volatile market environments.

Nonlinear pricing kernels via neural networks This study proposes a nonlinear pricing kernel approximated through neural networks, addressing limitations of traditional linear models, which capture linear relationships and are prone to overfitting when applied to the factor zoo. The proposed model specification test examines the validity of the nonlinearity assumption of the pricing kernel. Through optimal neural network selection, our findings reveal that a one-layer neural network significantly reduces quadratic pricing errors, indicating its superior pricing performance compared to deep neural networks. Moreover, the role of ESG variables in asset pricing, particularly within the extensive range of factors, remains underexplored. The significance test designed for neural networks shows that ESG variables are significant in asset pricing.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Subjects: H Social Sciences > HG Finance
• Colleges/Schools: College of Social Sciences > Adam Smith Business School > Accounting and Finance
• Funder's Name: China Scholarship Council (CSC)
• Supervisor's Name: Chen, Professor Cathy Yi-Hsuan and Chen, Dr. Bowei
• Date of Award: 2025
• Depositing User: Theses Team
• Unique ID: glathesis:2025-84858
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 04 Feb 2025 16:23
• Last Modified: 04 Feb 2025 16:28
• Thesis DOI: 10.5525/gla.thesis.84858
• URI: https://theses.gla.ac.uk/id/eprint/84858