Accepted and Published papers

  • Dufays, A., Houndetoungan, E. A., & Coën, A. (2022). Selective linear segmentation for detecting relevant parameter changes. Journal of Financial Econometrics, 20(4), 762-805.

    Paper WP

Working Papers

  • Count Data Models with Social Interactions under Rational Expectations, 2023 (Reject and Resubmit at the Journal of Econometrics)

    This paper proposes a peer effect model for counting variables using a game of incomplete information. I show that the game has a unique equilibrium under standard conditions. I also demonstrate that the identification argument in Bramoullé et al. (2009) extends to nonlinear models, particularly to the model of this paper. The model parameters are estimated using the Nested Partial Likelihood (NPL) approach, controlling for network endogeneity. I how that the linear-in-means/Tobit models with a counting outcome are particular cases of my model. However, by ignoring the counting nature of the outcome, these models can lead to inconsistent estimators. I use the model to evaluate peer effects on students' participation in extracurricular activities. I find that a one-unit increase in the expected number of activities in which a student's friends are enrolled yields an increase in the expected number of activities in which the student is enrolled by 0.08. This point estimate using the Tobit model is three times higher.

    Paper Online Supplement R Package
  • The Role of Child Gender in the Formation of Parents' Social Networks - with Asad Islam, Michael Vlassopoulos, and Yves Zenou

    Social networks play an important role in various aspects of life. While extensive research has explored factors like gender, race, and education in network formation, one dimension that has received less attention is the gender of one's child. Children tend to form friendships with same-gender peers, potentially leading their parents to interact based on their child's gender. Focusing on households with children aged 3-5, we leverage a rich dataset from rural Bangladesh to investigate the role of children's gender in parental network formation. We estimate an equilibrium model of network formation that considers a child's gender alongside other socioeconomic factors. Counterfactual analyses reveal that children's gender significantly shapes parents' network structure. Specifically, if all children share the same gender, households would have approximately 15% more links, with a stronger effect for families having girls. Importantly, the impact of children's gender on network structure is on par with or even surpasses that of factors such as income distribution, parental occupation, education, and age. These findings carry implications for debates surrounding coed versus single-sex schools, as well as policies that foster inter-gender social interactions among children.

  • Estimating Peer Effects using Partial Network Data, 2023 - with Vincent Boucher

    We study the estimation of peer effects through social networks when researchers do not observe the entire network structure. Special cases include sampled networks, censored networks, misclassified links, and aggregated relational data. We assume that researchers can obtain a consistent estimator of the distribution of the network. We show that this assumption is sufficient for estimating peer effects using a linear-in-means model. We provide an empirical application to the study of peer effects on students academic achievement using the widely used Add Health database and show that network data errors have a first-order downward bias on estimated peer effects.

    Paper Online Appendix R Package Vignette
  • Identifying Peer Effects on Student Academic Effort, 2023 - with Cristelle Kouame

    Peer influences on students' academic effort are often studied using GPA as a proxy variable for the effort when the latter is unobserved. We present an alternative method that does not require this approximation. Our identification strategy distinguishes unobserved shocks exerted on GPA without influencing the effort from unobserved students' preference shocks. We find that our estimate may be significantly different from the classical estimate (where the effort is approximated) if the network includes isolated students. Applying our approach to Add Health data shows that the peer effect estimate using a proxy for the effort is 1.6 times lower.

    Paper Online Appendix Codes (R & C++)
  • Healthcare Quality by Specialists under a Mixed Compensation System: an Empirical Analysis, 2023 - whith Damien Echevin and Bernard Fortin

    We analyze the effects of a mixed compensation (MC) scheme for specialists on the quality of their healthcare services. To do so, we exploit a major reform that was implemented in Quebec (Canada) in 1999. The government introduced a payment mechanism combining a fixed per diem with a reduced fee per clinical service. Using panel patient-doctor data covering the period 1996-2016 and including 320,441 patients, we estimate a multi-state multi-spell hazard model with correlated heterogeneity, analogous to a difference-in-differences approach. We compute three output-based quality indicators from our model. Our results suggest that the reform reduced the quality of MC specialist services as measured by the risk of re-hospitalization within 30 days after discharge and the risk of mortality within one year after discharge. These effects vary depending upon the specialty of the treating doctor.

  • Inference for Two-Stage Extremum Estimators - with Abdoul Haki Maoude (Draft available upon request)

    We present a simulation-based approach to approximate the asymptotic variance and asymptotic distribution function of two-stage estimators. We focus on extremum estimators in the second stage and consider a large class of estimators in the first stage. This class includes extremum estimators, high-dimensional estimators, and other types of estimators (e.g., Bayesian estimators). Importantly, the asymptotic distributions of both the first- and second-stage estimators may not be normal. Unlike resampling methods, our approach does not require multiple computations of the plug-in estimator. We show the effectiveness of our method using numerical simulations and an empirical application on network data.


Work In Progress

  • Quasi-Maximum Likelihood Estimator for Peer Effect Models with Partial Network Data

  • Physicians' Financial Incentives - whith Damien Echevin

  • Peer Effects in Active Labor Market Policies - with Jérémy Hervelin