Publications et documents de travail
06/10/19
- documents de conférences
    David Bounie
Big Data, Research Perspectives in Digital Payments, and Concrete Applications
Publication at the congress "Big Data and Artificial Intelligence in Banking".
02/09/19
- documents de travail
    Sara Biancini and Marianne Verdier
Bank-Platform Competition in the Credit Market
The paper analyzes the equilibrium on the credit market when a bank and a platform compete to offer credit to borrowers. The platform does not manage deposit accounts, but acts as an intermediary between the borrower and the investor, offering a risky contract such that the investor is only reimbursed if the borrower is successful. We first characterize the optimal contracts proposed by the platform, depending on the two-sided structure of the market. Then, we study the impact of bank-platform competition on the average risk of bank loans and the relative level of interest rates. We derive the conditions on the lending and the deposit markets such that the bank accomodates platform entry.
23/08/19
    Olena Havrylchyk, Carlotta Mariotto, Talal-Ur- Rahim and Marianne Verdier
The Expansion of the Peer-to-Peer Lending
We use data from the two leading P2P lending platforms on the US consumer credit market, Prosper and Lending Club, to explore the main drivers of the expansion of consumer demand for P2P credit. We exploit the heterogeneity in local credit markets at the county level to test three hypotheses: 1) global financial crisis; 2) competition and barriers to entry; and 3) learning costs. Disentangling between these hypotheses is difficult because the financial crisis has triggered an increase in market concentration and the closure of bank branches. Our findings suggest that P2P lending platforms have partly substituted for banks in counties that were more affected by banks’ deleveraging in the wake of the financial crisis. High market concentration and high branch density appear to deter the entry and expansion of the P2P lending. Finally, we find a positive impact of variables that are correlated with lower learning costs, such as education, population density, high share of young population, as well as important spatial interactions.
11/03/19
- documents de travail
    Bertail Patrice, Bounie David, Clémençon Stephan and Waelbroeck Patrick
Algorithmes : biais, discrimination et équité
Les algorithmes s’immiscent de plus en plus dans notre quotidien à l’image des algorithmes d’aide à la décision (algorithme de recommandation ou de scoring), ou bien des algorithmes autonomes embarqués dans des machines intelligentes (véhicules autonomes). Déployés dans de nombreux secteurs et industries pour leur efficacité, leurs résultats sont de plus en plus discutés et contestés. En particulier, ils sont accusés d’être des boites noires et de conduire à des pratiques discriminatoires liées au genre ou à l’origine ethnique. L’objectif de cet article est de décrire les biais liés aux algorithmes et d’esquisser des pistes pour y remédier. Nous nous intéressons en particulier aux résultats des algorithmes en rapport avec des objectifs d’équité, et à leurs conséquences en termes de discrimination. Trois questions motivent cet article : Par quels mécanismes les biais des algorithmes peuvent-ils se produire ? Peut-on les éviter ? Et, enfin, peut-on les corriger ou bien les limiter ? Dans une première partie, nous décrivons comment fonctionne un algorithme d’apprentissage statistique. Dans une deuxième partie nous nous intéressons à l’origine de ces biais qui peuvent être de nature cognitive, statistique ou économique. Dans une troisième partie, nous présentons quelques approches statistiques ou algorithmiques prometteuses qui permettent de corriger les biais. Nous concluons l’article en discutant des principaux enjeux de société soulevés par les algorithmes d’apprentissage statistique tels que l’interprétabilité, l’explicabilité, la transparence, et la responsabilité.
07/01/19
- documents de travail
    Arash Aloosh and Jiasun Li
Direct Evidence of Bitcoin Wash Trading
Using data leaked by hackers from a major Bitcoin exchange, we find that more than 2% and up to 33% of all transactions are wash trades, a type of market manipulation in which traders clear their own limit orders to “cook” transaction records. Our finding provides direct evidence for the widely-suspected “fake volume” allegation against cryptocurrency exchanges, which are so far only backed by indirect inferences. While wash trades do not incur a significant positive impact on Bitcoin prices, they do increase transaction fees collected by the exchange. Wash trades also involve exchange insiders previously exposed in other price manipulations. The evidence is consistent with the hypothesis that the exchange commits wash trading itself – not to manipulate price but to inflate apparent trading volume so as to look more attractive to deceived customers and boost commission revenues. We further use our direct evidence to evaluate the indirect inference techniques proposed in the literature.
Afficher plus