Machine Learning Goes Global - Cross-sectional Return Predictability in International Stock Markets

Sep 17, 2026

Uploaded by Asher

Machine learning is used to evaluate return predictability across 46 international stock markets by transforming 148 firm characteristics into inputs for a diverse model set. The study finds predictive signals are mainly captured by widely used factor types such as momentum, reversal, value, and size. While individual models deliver sizable economic gains, combining models is especially effective. Performance is robust overall but varies with firm size, recent-information availability, and internationally across listed-firm counts and idiosyncratic risk that constrains arbitrage.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages57

Reading Time2 h 24 min

AccessFree to download

Abstract

Introduction

Research backdrop and motivation

Data and methodology overview

Machine Learning Goes Global: Cross-Sectional Return Predictability in International Stock Markets* Nusret Cakici †, Christian Fieberg ‡, Daniel Metko §, Adam Zaremba 􀟜⁑ Abstract We examine return predictability with machine learning in 46 stock markets around the world. We calculate 148 ﬁrm characteristics and use them to feed a repertoire of diﬀerent models. The algorithms extract predictability mainly from simple yet popular factor types—such as momentum, reversal, value, and size. All individual models generate substantial economic gains; however, combining them proves particularly eﬀective. Despite the overall robustness, the machine learning performance depends heavily on ﬁrm size and availability of recent information. Furthermore, it varies internationally along two critical dimensions: the number of listed ﬁrms in the market and the average idiosyncratic risk limiting arbitrage. Keywords: machine learning, return predictability, international stock markets, the cross-section of stock returns, forecast combination, asset pricing, ﬁrm size JEL Codes: C52, G10, G12, G15 This version: 30 June 2023 * We thank Vitor Azevedo, Turan Bali, Hendrik Bessembinder, David Blitz, Guillaume Coqueret, Guanhao (Gavin) Feng, Matthias Hanauer, Fabian Hollstein, Tobias Hoogteijling, Theis Ingerslev Jensen, Tobias Kalsbach, Herald Lohre, Alexandre Rubesam, Laurens Swinkels, Andea Tamoni, Pim van Vliet, and Guofu Zhou for helpful comments and suggestions, as well as seminar participants at Technical University of Munich, Robeco, Ono Academic College, American University of Sharjah, and American University in Dubai. This research was funded in whole or in part by National Science Center of Poland [Grants no. 2019/33/B/HS4/01021, 2022/45/B/HS4/00451] . For the purpose of Open Access, the author has applied a CC-BY public copyright licence to any Author Accepted Manuscript (AAM) version arising from this submission ⁑ Corresponding author. † Nusret Cakici, Gabelli School of Business, Fordham University, 45 Columbus Avenue, Room 510, New York, NY 10023, USA, tel. +1-212-636-6776, email: [cakici@fordham.edu](cakici@fordham.edu). ‡ Christian Fieberg, 1) City University of Applied Sciences, Bremen, Germany, [christian.fieberg@hs-bremen.de](christian.fieberg@hs-bremen.de) ; 2) University of Luxembourg, Luxembourg, Luxembourg, [christian.fieberg@ext.uni.lu](christian.fieberg@ext.uni.lu) ; 3) Concordia University, Montreal, Canada, [christian.fieberg@concordia.ca](christian.fieberg@concordia.ca). § Daniel Metko, University of Bremen, Bremen, Germany, [metko@uni-bremen.de.](metko@uni-bremen.de.) . 􀟜 Adam Zaremba, 1) Montpellier Business School, 2300, avenue des Moulins, 34185 Montpellier cedex 4, France, email: [a.zaremba@montpellier-bs.com](a.zaremba@montpellier-bs.com) ; 2) Department of Investment and Capital Markets, Institute of Finance, Poznan University of Economics and Business, al. Niepodległości 10, 61-875 Poznań, Poland, [adam.zaremba@ue.poznan.pl](adam.zaremba@ue.poznan.pl). 1. Introduction A recent merger between machine learning and asset pricing research has raised hopes for a solution to the factor zoo problem found within stock markets . So far, the evidence has been promising. Studies from the U.S. and China have demonstrated that machine learning models can eﬀectively explain and predict the cross-section of stock returns.1 Nevertheless, the evidence from individual markets may not necessarily hold elsewhere.2 Return predictability depends on the region or country characteristics that aﬀect input data and model performance.3 Furthermore, the complexity of machine learning techniques—as well as the diversity of their design—augments the risk of data dredging. To alleviate these concerns, extensive out-of-sample evidence across many countries and models is necessary. Against this backdrop, this study combines machine learning with asset pricing research to comprehensively examine cross-sectional return pre

FAQs

What is the main research focus of the document?

It examines how machine learning can predict cross-sectional stock returns across 46 international markets and identifies drivers of performance differences across countries.

How are return predictability models constructed and evaluated?

The study computes 148 firm characteristics from three decades of CRSP and Compustat data, then feeds them into a set of 11 representative algorithms, including forecast combination, and assesses predictive ability and economic gains.

Which factor types and conditions most influence predictive performance?

Predictability is extracted mainly from popular factor types such as momentum, reversal, value, and size. Machine-learning performance also depends heavily on firm size and the availability of recent information, and it varies internationally with market breadth and idiosyncratic risk limiting arbitrage.

Sep 17, 2026

Uploaded by Asher