Can Humans Understand Machines in Finance?

SKBI Newsletter | February 2025  

Can Humans Understand Machines in Finance?

Research Perspective

Industry Insights

Hong Zhang,
Professor of Finance
at SMU 

Danny Yong, 
Chief Investment Officer at Dymon Asia Capital

Machine learning has achieved remarkable success in predicting asset returns in the financial market. However, the underlying mechanisms behind this performance remain unclear. A recent SMU study demystifies this black box by linking machine learning trading to arbitrage, a fundamental concept in finance.

With AI and machine learning on the rise in financial markets, can we understand what they are doing and trust their recommended factors? Danny Yong (CIO, Dymon Asia Capital) and Ian Chung (Director, Dymon Asia Capital) shares their thoughts and experiences on this emerging topic with us.

Key Messages:

• In the US market, the best machine learning tools—neural networks—can generate above 20% yearly abnormal returns when using firm characteristics, such as momentum and value, as input factors.
• The study proposes dynamic arbitrage portfolios (DAPs) based on the famous Arbitrage Pricing Theory. A backtest is further adopted to rule out about 40% fake factors. Ultimately, DAPs can rank and select factors similarly to neural networks in the cross-section.
• This similarity enables DAPs to account for approximately half of the total annualized alpha of neural network returns (e.g., 10% out of the 20% annual alpha).
• When unpublished characteristics and small stocks are excluded from trading due to the biases they may introduce, DAPs can fully explain the performance of the neural networks.

Overall, the success of machine learning surprisingly aligns with the key concept of arbitrage in finance. This could pave the way for humans to better understand and benefit from machines in the financial market.

LU, Huahao; SPIEGEL, Matthew; and ZHANG, Hong. 2024. Machine Learning as Arbitrage: Can Economics Help Explain AI?. Working paper.

Key Challenges:

• How to identify reliable factors for trading is a million-dollar question in practice. Traditionally, the industry resorts to well-known and often reason-based factors, such as momentum. The new approach utilizes machine learning to identify factors. 
• However, machine learning may identify patterns that lack identifiable or coherent reasons.
• In practice, when machine learning identifies “good” factors, judgment is still needed to examine whether they are reasonable and tradable. In half of the cases, they are not.
• Liquidity and turnover are two relevant issues. Many factors are tradable only for large-cap stocks. However, many machine learning factors often load heavily on small stocks and imply unrealistic turnover.  

“The reality is that it is hard to put significant amount of capital in something you cannot explain.” says Danny. How to understand machine learning seems crucial for the next stage of development in capital market applications.

Meet the Author:
Prof Hong Zhang is the current director of SKBI and Keppel Professor in Financial Economics at SMU. His research interests include market efficiency and frictions, Fintech and AI, and social, Environmental and Cultural issues. He holds a Ph.D. from Yale University.

Meet the Expert:
Mr Danny Yong is the co-Founder of Dymon Asia Capital. He is the Co-CIO of the firm’s flagship Multi-Strategy Investment Fund (MSIF) and sits on the investment committee of Dymon Asia Private Equity. He has sat on the SKBI Advisory Board since 2019.

About SKBI:

The Sim Kee Boon Institute generates financial economic research through multidisciplinary collaborations involving not only the SMU community, but also research talent from around the world as well as industry and public-sector partners. The Institute will focus its efforts on the areas of (1) Market Innovations and FinTech, (2) Sustainability and Green Finance, and (3) Household Finance and Behaviour. To maintain relevance to finance practitioners and policy-makers, SKBI also adopts a view on Asian and global economic trends. View SKBI’s research. 

About the SKBI Newsletter:

This monthly newsletter provides a unique platform to connect academic researchers and industry experts. It aims to enhance the outreach of academic studies, while fostering dialogue on key insights and challenges and stimulating new ideas and collaborations.

© Copyright 2025 by Singapore Management University. All Rights Reserved.