L’exclusion sociale. Reconstruire les communs, Collection Essais, Bréal, March 2022 (expected)
L’exclusion sociale, Thèmes et Débats, Bréal, 2017 [link]
“Answering the Queen : online machine learning and inancial crises”, 2020, with M. Howell (CrossBorder Capital) and H.Rey (London Business School)), NBER Working Paper [Link]
Abstract : Financial crises cause economic, social and political havoc. We use the general framework of sequential predictions also called online machine learning to forecast crises out-of-sample. Our methodology is based on model averaging and is “meta-statistic” since we can incorporate any predictive model of crises in our set of experts and test its ability to add information. We are able to predict systemic financial crises twelve quarters ahead out-of-sample with high signal-to-noise ratio in most cases. We analyse which experts provide the most information for our predictions at each point in time and for each country, allowing us to gain some insights into economic mechanisms underlying the building of risk in economies.
“Confidence interval in online predictions, 2021, with P. Alquier (Riken Center for Artificial Intelligence).
Abstract : Online prediction with expert advice has become a popular approach in machine learning. It also received a beautiful theoretical treatment, which shows that it is possible to predict as well as the best convex combination of expert without any stochastic assumption on the data. However, it does not include confidence intervals, and indeed, without stochastic assumptions, this notion does not even make sense. In this paper, we propose to study empirical and theoretical implications of aggregating a number of finite models coming without confidence intervals as well as with confidence intervals in the framework of conformal predictions.
“Is this time different? Crises across centuries”, 2021, with H.Rey (London Business School)
Abstract : Carmen Reinhart and Ken Rogoff have written 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 21rst century Great Recession? We find we can predict out-of-sample the 1929 Great Depression, the other 20th century systemic crises as well as the 21rst century ones for a panel of countries by aggregating models optimally using an online learning approach. Models are selected based exclusively on information of the 19th and beginning of the 20th centuries (up to 1910) and then optimally weighted as information unfolds year by year.