Big Ideas

What is the state of machine learning in finance? In their July 2023 paper entitled “Financial Machine Learning”, Bryan Kelly and Dacheng Xiu survey studies on the use of machine learning in finance to further its reputation as an indispensable tool for understanding financial markets. They focus on the use of machine learning for statistical forecasting, covering regularization methods that mitigate overfitting and efficient algorithms for screening a vast number of potential model specifications. They emphasize areas that have received the most attention to date, including return prediction, factor models of risk and return, stochastic discount factors and portfolio choice. Based on the body of machine learning research in finance, they conclude that: Keep Reading

What happens as more and more web-scraped training data for Large Language Models (LLM), such as ChatGPT, derives from outputs of predecessor LLMs? In their May 2023 paper entitled “The Curse of Recursion: Training on Generated Data Makes Models Forget”, Ilia Shumailov, Zakhar Shumaylov, Yiren Zhao, Yarin Gal, Nicolas Papernot and Ross Anderson investigate changes in LLM outputs as training data becomes increasingly LLM-generated. Based on simulations of this potential trend, they find that: Keep Reading

In his 2023 book, The Uncertainty Solution: How to Invest with Confidence in the Face of the Unknown, author John Jennings seeks “to provide individual investors with mental models that will help them make better investment decisions, practice better investment behavior, and be better consumers of investment advice… This book is not about how to invest but rather how to think about investing. It is the culmination of my thirteen-year quest for investment wisdom… The mental models in this book describe the investment world as full of uncertainty, wild randomness, unpredictability, and pitfalls. There’s no easy path. But mental models that embrace reality—that take the world as it is, not how we think it is or want it to be—will make you a better investor and a better consumer of investment advice.” Based on his many years of wealth management experience, especially during the 2007-2008 Financial Crisis, he concludes that:

Keep Reading

For the last half-century, financial researchers have fallen short of scientific rigor, focusing on associations not supported by theory and not amenable to falsification. Is there is hope for finance to become a scientific field? In his April 2023 paper entitled “The Hierarchy of Empirical Evidence in Finance”, Marcos Lopez de Prado proposes a hierarchy of empirical evidence that gives greatest scientific weight to methods allowing falsification of causal claims. He addresses:

• Why associations alone do not constitute scientific knowledge.
• The importance of causality and how statistical methods enable the falsification of causal claims.
• How to extract causal effects from observational studies in fields like finance with inherent barriers to controlled experimentation.
• An example of examining causality in finance.

He ultimately translates the distinction between associational and causal claims into a hierarchy of empirical evidence in finance. Based on the philosophy of science and the constraints of studying complex financial systems, he concludes that: Keep Reading

Which industries are most exposed to disruption by artificial intelligence (AI) language models such as ChatGPT? In the April 2023 version of their paper entitled “How will Language Modelers like ChatGPT Affect Occupations and Industries?”, Edward Felten, Manav Raj and Robert Seamans focus the previously developed AI Occupational Exposure (AIOE) measure on models such as by ChatGPT, relating model capabilities to 52 human abilities and thereby to human occupations and industries. Applying this adaption, they find that:

Keep Reading

How can researchers assess the true value and robustness of new stock return anomalies (predictors) in consideration for addition to the factor zoo? In their January 2023 paper entitled “Assaying Anomalies”, Robert Novy-Marx and Mihail Velikov present a protocol/tool set for dissecting and understanding newly proposed cross-sectional stock return predictors. The tools address the most important issues involved in testing asset pricing strategies, including some machine learning techniques. They pay particular attention to implementation costs that prevent exploitation of predictors with good gross returns (as with high turnover and/or overweighting small stocks). The tool set, including automated report generator, is available as a free web application and a public github repository. Key aspects of reports generated by this tool set are:

Keep Reading

Should investors presume that, in the absence of falsifiable theories, the body of factor investing research is largely spurious? In the January 2023 version of his paper entitled “Causal Factor Investing: Can Factor Investing Become Scientific?”, Marcos Lopez de Prado reviews the current state of confusion about causality in factor investing research and discusses ways to resolve that confusion. Specifically, he addresses:

• Differences between association and causation.
• Why the study of association alone does not create scientific knowledge.
• How observational studies, natural experiments and simulated interventions support investigation of causality.
• The current state of causal confusion in econometrics and factor investing studies.
• How to transform factor investing into a truly scientific discipline.

Based on many references and the logic of the scientific method, he concludes that:

Keep Reading

Given that finance is ultimately tied to human emotions, does the body of research support belief that abilities of machine learning to handle large amounts of data, non-linearities and variable interactions will revolutionize investing? In their January 2023 paper entitled “How Can Machine Learning Advance Quantitative Asset Management?”, David Blitz, Tobias Hoogteijling, Harald Lohre and Philip Messow study application of machine learning to investing from the perspective of a prudent practitioner. They define machine learning models as initially guided software programs that subsequently learn by themselves and then make predictions about unseen (out-of-sample) data. They describe benefits and pitfalls of machine learning versus classical (linear regression) econometrics. They discuss critical design choices for applying machine learning models to asset management. They focus on the ability of machine learning to beat the stock market, but also discuss asset risk forecasting, optimal portfolio construction and trading optimization. Based on the body of research and their collective experience in asset management, they conclude that: Keep Reading

If a published theory is correct, its empirical results should hold for years after original test samples end. Are peer-reviewed, theory-supported (risk-based) academic studies of stock return predictors thereby superior to other streams of predictor research, and thereby especially useful to investors? In their December 2022 paper entitled “Peer-Reviewed Theory Does Not Help Predict the Cross-section of Stock Returns”, Andrew Chen, Alejandro Lopez-Lira and Tom Zimmermann employ out-of-sample tests to address these questions in two ways:

1. Read the arguments used in papers from finance, accounting and economics journals reporting discoveries of 202 firm-level stock return predictors. Label each predictor as risk-based (resting on theory), mispricing-based (behavioral) or agnostic. Validate categories via software that counts ratios of risk-related words to mispricing-related words. Compare out-of-sample returns of the three categories, with expectations that: (a) risk-based returns will persist; and, (b) mispricing-based returns will decay as investors learn about them and alter their behaviors.
2. Compare out-of-sample returns of the 202 predictors from peer-reviewed studies to predictors matched on in-sample return statistics from a set of 18,240 trading strategies generated by brute-force sorting of firms on simple combinations of 240 accounting variables.

They calculate predictor returns monthly from periodically reformed equal-weighted or value-weighted portfolios that are long (short) the tenth, or decile, of stocks with the highest (lowest) expected returns. Using data from 45 years of cross-sectional asset pricing research for the 202 predictors from formal studies and returns for the 18,240 mined trading strategies during both in-sample and fixed out-of-sample intervals, they find that: Keep Reading

How should investors think about, and perhaps exploit, asset return volatility? In his December 2022 paper entitled “A Stylized History of Volatility”, Emanuel Derman reviews how generations of financial modelers have quantified volatility and ultimately created tradable volatility-based assets. He also discusses some general modeling considerations. Based on the body of research and his experience, he concludes that: Keep Reading