Publications

Uncovering Sparsity and Heterogeneity in Firm-Level Return Predictability Using Machine Learning
with Theodoros Evgeniou and Rodolfo Prieto   Article Internet Appendix SSRN Version
Journal of Financial and Quantitative Analysis, 2023, 58(8), 3384-3419
Abstract

We develop an approach that combines the estimation of monthly firm-level expected returns with an assignment of firms to (possibly) latent groups, both based on observable characteristics, using machine learning principles with linear models. The best-performing methods are flexible two-stage sparse models that capture group-membership predictive relationships. Portfolios formed to exploit such group-varying predictions based on a parsimonious set of characteristics deliver economically meaningful returns with low turnover. We propose statistical tests based on nonparametric bootstrapping for our results, and detail how different characteristics may matter for different groups of firms, making comparisons to the existing literature.

Working Papers

Soothing Investors: The Impact of Manager Communication on Mutual Fund Flows
Job Market Paper, solo-authored
Presented at NBER Behavioral Finance, FIRS PhD Sessions (awarded Travel Grant)
Abstract

I show that communication by fund managers to their investor clients fosters trust and encourages these investors to bear risk. Using an institutional setting that enables causal identification, I find that more detailed communication about risk encourages investors to increase their holdings in the market portfolio, driving flows into the stock market. I rule out learning about risk, returns or manager skill, and other potential explanations. Instead, my analysis shows this communication soothes investors' anxiety and alleviates their effective risk aversion, consistent with the money doctors framework of Gennaioli, Shleifer, and Vishny (2015).

 

Machine Traders, Human Behavior, and Model (Mis)Specification
solo-authored   SSRN Slides
Presented at SFS Cavalcade North America, Chicago Booth ML in Economics Summer Institute, Cambridge Conf. on Alternative Finance
Abstract

I examine how investors utilize data, exploiting a setting in which investors design machine-driven trading strategies under controlled yet realistic conditions. Investors disagree considerably in how they interpret identical information, leading to widely dispersed trading strategies and performance outcomes. Inexperienced investors underweight variables with predictive power for returns, and instead exhibit a bias towards variables with which they are more familiar. With experience, investors learn to overcome their bias, and benefit substantially from additional data availability. Investors' familiarity bias leads them to mis-specify their models of the world, and is encoded by the machine traders they design.

 

Mutual Fund Market Structure and Company Fee Competition: Theory and Evidence
with Richard Grice   SSRN Slides
Presented at FIRS, NSF Network Science & Economics Conference (poster)
Abstract

We investigate whether competition between the fund companies that offer mutual funds constrains individual fund fees. We document that a substantial fraction of individual fund fee variation is explained by company-wide components. Moreover, we show using SEC prospectus download data that company-level attributes influence investors' consideration of companies. We connect these facts with a model of fee competition between co-considered fund companies, characterising the competitive landscape and associated equilibrium fees. Calibrating the model, we derive a testable prediction for competitively constrained fees. The prediction successfully explains cross-sectional variation in company-level average fees, identifying the influence of company competition on fees.

 

Expectations and Attention to Experience
with Heiner Beckmeyer   SSRN Slides
Presented at CICF, Applied Young Economists Webinar
Abstract

We measure which past experiences determine investors' expectations about the market's future Sharpe Ratio. We first introduce a simple method to recover individuals' subjective Sharpe Ratios from a rich source of survey microdata. These subjective expectations are procyclical, extrapolative, cross-sectionally correlated with individual demographic characteristics, and well explained by a low-dimensional latent factor structure. We then use a customized machine learning estimation technology to estimate an economic model of experience effects that generalizes the lifetime weighting scheme of Malmendier and Nagel (2011). The model includes the influence of demographic characteristics in how past experiences determine individuals' future expectations, and succeeds in explaining a larger fraction of survey belief heterogeneity. We find that households' aggregate wealth share held in equities is strongly correlated with the share of investors who have experienced positive Sharpe Ratios (as measured by our model), thus confirming that experience effects drive investor flows. We also contribute a new set of facts on the role of demographic characteristics and the outsized influence of past recessions on how individuals learn from their experiences.

 

Trusting the Algorithm: A Decision Under Ambiguity
with Qiong Xia and Enrico Diecidue   SSRN
Presented at ESA, SPUDM, TIBER Symposium on Psychology & Economics
Abstract

We show that ambiguity attitudes influence decision-makers' (DMs') choices about whether to trust the forecasts of human and machine-learning (ML) financial analysts. DMs are similarly ambiguity-seeking and ambiguity-generated insensitive ("a-insensitive"; i.e., they insufficiently discriminate between changes in the likelihood of prediction accuracy) towards both analyst types. DMs hold more optimistic beliefs about the accuracy of ML analysts, which predicts higher trust in ML analysts over human analysts. However, DMs who are more a-insensitive are less likely to incorporate their beliefs into their choices. DMs' a-insensitivity increases with financial literacy, suggesting that financially literate DMs perceive greater ambiguity in prediction accuracy.

Other Research

Kernels for Time Series With Irregularly-Spaced Multivariate Observations
with Franz J. Király

Brief write-up of some machine learning methodology results from my UCL MSc dissertation.