Jérémy Fouliard

GitHub
Jérémy Fouliard

Welcome! I am a PhD candidate in Economics at the Paris School of Economics, advised by Pierre-Cyrille Hautcoeur and Eric Monnet, and a senior French civil servant, member of the State corps of Ingénieurs de la statistique, de l'économie et de la donnée.

My research is in macroeconomic and financial forecasting and in economic measurement. Both ask what the data available today can reliably tell us about an economy that we observe imperfectly and with delay. I currently work as a research methodologist in the National Accounts Department of the French National Institute of Statistics and Economic Studies (INSEE), on the measurement of public sector output, profit shifting by multinational firms and crypto-assets. Before my PhD, I worked at London Business School on Hélène Rey's ERC project and at the Bank for International Settlements.

Job Market Paper

  • Slow to Catch Up: Return Predictability on the Paris Bourse, 1845–2025
    Abstract

    I study the real-time predictability of stock returns over 180 years with a new security-level dataset for the Paris Bourse (DFIH, 1845–1976, matched with EUROFIDAI, 1977–2025). Forecasts combine models as they become available through Bernstein Online Aggregation, a rule with a bounded regret that assumes nothing about the return-generating process. The market index is barely predictable: macro-financial predictors explain 0.5% of its half-monthly variance, mostly in crises, and the larger predictability of lagged index returns is a stale-price artefact. Individual stocks are more predictable, by about 0.6%, mostly because they react to market-wide news with a delay that is strongest in small stocks, not explained by stale prices, and unchanged over 155 years. Momentum earns 7% a year net of costs in almost every decade since 1900, while the historical volatility and beta premia vanish after 1977. Markets became better at pricing risk, not faster at absorbing information.

Working Papers

  • Answering the Queen: Machine Learning and Financial Crises — with Michael Howell (CrossBorder Capital), Hélène Rey (BIS) and Vania Stavrakeva (LBS)
    New version, 2026
    Earlier versions: NBER WP 28302 · CEPR · BIS
    Abstract

    Early-warning systems for financial crises face a problem that the literature leaves aside: whether a quarter belonged to a pre-crisis period is known only years later, when it is too late to act. We address this delayed feedback by combining online learning with sequential change-point detection. We build one predictive model per channel of vulnerability (credit, real estate, external imbalances and the global financial cycle), treat each model as a sensor, and shift weight in real time towards the channel whose signal is changing. The procedure carries a formal performance guarantee that requires no assumption on the data-generating process. Applied to France, Germany, Sweden and the United Kingdom, with every parameter fixed on pre-2001 data, it raises an alarm at least nine quarters before the 2007–2008 crises in all four countries, with an average out-of-sample AUROC of 0.90. The channel that carries the alarm differs across countries, which is why an aggregation based on detection rather than on past performance is needed.

  • Volume Output Measurement of Public Services: The Case of Education — with M. Le Floch (INSEE) and P. Mendras (INSEE)
    Revise and resubmit, Review of Income and Wealth
    Coverage: JMS · UNECE
    Abstract

    Measuring the volume of public services output is a fundamental challenge for national accounts, due to the absence of market prices. In France, the volume of non-market education is measured using an output approach based on student enrollment. As this indicator mainly reflects demographic trends, the recent decline in the number of students mechanically reduces the estimated volume of education, even if quality teaching could have improved, for example through smaller class sizes. This study compares alternative volume measurements, and examines different quality adjustments to capture other features of the service provided. It introduces a theoretical framework that incorporates quality adjustments into output volume measures based on the contribution of quality to human capital, drawing on estimates consistent with the existing literature. Between 2013 and 2022, the adjustment based on PISA scores lowers annual growth rates (−0.056 percentage points on average) while the class-size adjustment slightly raises annual growth (+0.015 percentage points on average). Overall, quality adjustments have a modest effect on the estimated volume of education output, but moderate effects could accumulate over time and result in visible change in trend.

  • Quality-Adjusted Output and Productivity of the Non-Market Health Sector in France
    Presented at the OECD Working Party on National Accounts, 2026
    Abstract

    Despite an extensive literature on health care productivity, measuring quality-adjusted output in non-market settings remains an open challenge. This paper derives a welfare-based, quality-adjusted output index from a social planner's cost-minimisation problem, and identifies the distinct sources of productivity growth. Quality adjustment rests on a disease-based output specification in which each treatment contributes to aggregate health through two channels: a survival component, measured by severity-adjusted mortality rates, and a health-related quality-of-life component, measured by QALY gains per episode. Rather than importing QALY values from external tariffs, we estimate them directly from individual-level administrative data from the French Système National des Données de Santé (SNDS), which covers virtually the entire French population over 2012–2022. This exhaustive coverage enables a counterfactual analysis that isolates hospital-produced health gains from secular trends in population health. The framework is directly portable to any health system with diagnosis-related group tariffs and linked administrative records.

  • Profit Shifting and the Measurement of Value Added: Evidence from a Distribution of French National Accounts by Type of Firm — with U. Di Nallo (INSEE)
    Presented at the INSEE National Accounts Seminar and the OECD EG-MG Conference, 2026
    Abstract

    National accounts record value added in the economy where the producing unit is resident, but multinational groups can separate the location of their profits from the location of their production. We develop a method that distributes the production account across any partition of firms, following each legal unit from its tax return to published value added, and apply it to the breakdown recommended by the 2025 SNA: foreign-controlled corporations, domestic multinational groups and other domestic corporations. Combining country-by-country reports with tax returns for 2019–2023, we estimate profit shifting at the group level and pass this estimate through the compilation of French national accounts, to assess whether a correction for profit shifting can be produced within official statistics.

  • Is This Time Different? Financial Follies across Centuries — with Hélène Rey (BIS) and Vania Stavrakeva (LBS)
    Abstract

    Carmen Reinhart and Ken Rogoff write that "no matter how different the latest financial frenzy or crisis always appears, there are usually remarkable similarities with past experience from other countries and from history". Can we really use the same models to predict the 20th century Great Depression and the 21st century Great Recession? We find that we can predict out-of-sample the 1929 Great Depression and all the 20th and 21st century systemic financial crises in a panel of countries. We also generally do not over-predict crises during the Bretton Woods period.

Work in Progress

  • AI and the Transatlantic Productivity Gap: Who Captures the Gains from US Innovation?
    Abstract

    Frontier AI is produced mainly in the United States and China, while Europe, which produces little of it, will be a large user. Focusing on the transatlantic link, I study what this asymmetry implies for the productivity gap between Europe and the United States. I build a two-country model in the spirit of Baqaee and Farhi, in which production is a network of an AI sector, AI-using traded sectors and a sheltered sector, calibrated on international input-output tables. Innovation is concentrated in the United States and financed by the rents of AI producers, European adoption is endogenous and disciplined by occupational exposure to AI, compute is an accumulated capital stock, and the exchange rate is determined in equilibrium. The model separates the productivity gains that Europe obtains from AI from the value added it captures, the difference being the rent paid to US producers and the change in the terms of trade. It also shows how hedonic deflation and the capitalisation of AI spending shape measured European productivity, and when the constant-price and current-PPP readings of the transatlantic gap can each be right.

Teaching

  • Programming for Economists — Paris School of Economics, graduate course (Python, Git workflows), 2026
  • Economic Measurement — Sciences Po, graduate course, co-taught with E. Richard (DREES), 2026
  • Teaching Assistant, Microeconomics and Statistics — Sciences Po, second-year undergraduate students, 2022–2024

References

  • Hélène Rey — Economic Adviser and Head of the Monetary and Economic Department, Bank for International Settlements
  • Eric Monnet — Professor of Economics, Paris School of Economics
  • Pierre-Cyrille Hautcoeur — Professor of Economics, Paris School of Economics
  • Sébastien Roux — Programme Director, Augmented National Accounts, INSEE