The Anatomy of Machine Learning-Based Portfolio Performance (with Philippe Goulet Coulombe, Erik Christian Montes Schütte, and Sander Schwenk-Nebbe)

Abstract: Asset return predictability is routinely assessed by economic value: based on a set of predictors, out-of-sample return forecasts are generated—increasingly via “black box” machine learning models—which serve as inputs for portfolio construction, and performance metrics are computed over an evaluation period. We develop a methodology based on Shapley values—the Shapley-based portfolio performance contribution (SPPC)—to directly estimate the contributions of individual or groups of predictors to a performance metric. We illustrate the SPPC in an empirical application measuring the economic value of cross-sectional stock return predictability using a large number of firm characteristics and machine learning.

[Presented at the 2024 CEMFI Workshop on Big Data in Asset Management | Presented at the 2024 European Economic Association and Econometric Society European Meeting (EEA-ESEM) | Presented at the 2024 International Symposium on Forecasting | Presented at the 6th Future of Financial Information Conference | Presented in the Applied Machine Learning, Economics, and Data Science (AMLEDS) Webinar Series | Subject of Machine Learning & Quant Finance blog post]

Sparse Macro-Finance Factors (with Guofu Zhou)

Abstract: We estimate sparse principal components from a large set of macro-finance variables. Each component is a sparse linear combination of the underlying variables, enhancing economic interpretability and yielding sharper signals for asset pricing. Innovations to the components constitute a set of sparse macro-finance factors. Robust tests show that sparse factors tied to housing, yields, and credit spreads earn significant risk premia. Among the top 20 factor models formed from leading characteristic-based factors and mimicking portfolios for the housing, yield, and credit spread factors, the latter three are always selected, highlighting the importance of sparse macro-finance factors for capturing systematic risks.

[Internet Appendix (in progress) | Sparse Macro-Finance Factors website | Best Paper Prize, INQUIRE UK & Europe Spring 2019 Residential Joint Conference]

The Anatomy of Out-of-Sample Forecasting Accuracy: A Shapley-Based Approach (with Daniel Borup, Philippe Goulet Coulombe, Erik Christian Montes Schütte, and Sander Schwenk-Nebbe)

Abstract: We introduce the performance-based Shapley value (PBSV) to measure the contributions made by each of the individual predictors in fitted time-series forecasting models to the out-of-sample loss. The PBSVs for the individual predictors sum to the out-of-sample loss, so our new metric produces an exact decomposition of out-of-sample performance. In essence, the PBSV anatomizes out-of-sample forecasting accuracy, thereby providing valuable information for interpreting fitted time-series forecasting models. The PBSV is model agnostic—so it can be applied to any fitted prediction model, including “black box” models in machine learning—and it can be used for any loss function. We also develop the TS-Shapley-VI, a version of the conventional Shapley value that gauges the importance of predictors for explaining the in-sample predictions in the entire sequence of fitted prediction models that generates the time series of out-of-sample forecasts. We then propose the model accordance score to compare predictor ranks based on the TS-Shapley-VI and PBSV, thereby linking predictors’ in-sample importance to their contributions to out-of-sample forecasting accuracy. We illustrate our new metrics in an application forecasting US inflation using a large number of predictors and a variety of machine-learning models.

[Online Appendix | Python package anatomy | Presented at the 12th European Central Bank Conference on Forecasting Techniques | Earlier version: Federal Reserve Bank of Atlanta Working Paper 2022-16]

Going Supranational: Anomaly-Market Links and New Dimensions of Market Efficiency (with Xi Dong, Yan Li, Yanran Li, and Guofu Zhou)

Abstract: We connect cross-sectional anomalies to time-series market return predictability using data from 44 non-US countries. While a large set of representative anomaly returns show limited predictive power for market returns at the country level, they exhibit strong predictive ability when aggregated to the supranational level. We develop an international analytical framework to explain this difference: cross-sectional mispricing corrections in one country can propagate into market-wide corrections in another, enhancing supranational predictability precisely when mispricing is more country-specific than global. We further decompose anomaly-market links into three analytically grounded market (in)efficiency measures of broad relevance: systematic mispricing, overpricing dominance, and price randomness. Supported by data, they govern the strength and nature of anomaly-market links across global markets.

[Presented at the 2025 Midwest Finance Association Meeting]

Economic Fundamentals and Short-Run Exchange Rate Prediction: A Machine-Learning Perspective (with Ilias Filippou, Mark P. Taylor, and Guofu Zhou)

Abstract: This paper establishes the out-of-sample predictability of monthly exchange rates based on economic fundamentals using country characteristics, global variables, and their interactions. Previous work does not find consistent evidence of short-horizon predictability, likely due to using a small set of fundamentals and inadequately capturing time variation and nonlinearities in predictive relations. By employing a large set of economic fundamentals and global variables in conjunction with machine-learning techniques, we are able to consistently and significantly outperform the stringent no-change benchmark forecast. We find stronger predictability during periods of crisis and recession. The exchange rate forecasts are also economically valuable, as they generate sizable utility gains for an investor in the context of foreign currency portfolios. To enhance our understanding of the economic drivers of exchange rate predictability, we identify the most relevant predictors for forecasting exchange rates in the fitted machine-learning models.

[Internet Appendix | Presented at the 2021 Vienna Symposium on Foreign Exchange Markets]

Cryptocurrency Return Predictability: A Machine-Learning Analysis (with Ilias Filippou and Christoffer Thimsen; older version—currently being revised)

Abstract: We investigate the out-of-sample predictability of daily cryptocurrency returns using modern machine-learning methods. We consider a large number of cryptocurrencies (41) and a rich set of predictors relating to network value and activity, momentum, technical signals, and online activity. We find that return predictability is an important feature of the cryptocurrency market: machine-learning methods significantly improve the statistical accuracy of cryptocurrency return forecasts and provide substantial economic value to investors. Predictors relating to momentum, size, and value stand out as important determinants of future cryptocurrency returns. Nonlinearities also play a significant role in improving cryptocurrency return predictability.

[Online Appendix | Presented at the 2024 Econometric Society Interdisciplinary Frontiers (ESIF) Economics and AI+ML Meeting (slides) | Presented at the 2023 Northern Finance Association Annual Conference | Presented at the INQUIRE UK 2023 Autumn Residential Seminar]

WORK IN PROGRESS

Improved Hedge Fund Return Prediction: Dealing with Missing Data via Deep Learning (with Ilias Filippou, Ioannis Psaradellis, and Lazaros Zagrafopoulos)

Accepted for presentation at the 2025 Alpine Finance Summit

Is the Phillips Curve Alive (in Any Form)? (with David Chester and Tao Zha)

Forecasting Inflation: The Role of Financial Market Information (with Nikolay Gospodinov and Bin Wei)

OTHER ITEMS

Forecasting Asset Returns: The State of the Art

Invited Lectures at CEMA/CUFE on July 12–13, 2016

[Slides for Lecture 1 | Slides for Lecture 2]

Forecasting Asset Returns in Realistic Environments

Invited Presentation at the CFA Montreal Asset Management Forum on October 8, 2015 (other presenters: Andrew Ang, Mark Carhart, Craig Bodenstab, Philip Tetlock)

[Slides | CFA Research Foundation Brief: Portfolio Structuring and the Value of Forecasting]