Research
Work in Progress
A Pairwise Differencing Distribution Regression Approach for Network Models. 2025. • Download paper
Abstract
This paper proposes a novel estimation method for distribution regressions in a network setting, considering the effects of covariates on the entire outcome distribution rather than just on the mean. I adopt a semiparametric approach, taking into account two-way unit-specific effects that are treated as fixed parameters to be estimated. Thus, I extend the standard distribution regression approach to a network setting by estimating multiple binary choice models with twoway fixed effects for different thresholds of the distribution. Instead of using bias-correction methods to address the incidental parameter problem, as previously proposed in the literature, I propose to employ a conditional maximum-likelihood approach (Charbonneau (2017), Jochmans (2018)) that differentiates out the unit-specific effects. This method yields consistent point estimates that converge at a parametric rate and remain asymptotically unbiased in the tails of the outcome distribution, where the underlying network can be seen as sparse. Monte Carlo simulations validate these findings for both single cut-offs and the overall outcome distribution. The empirical application focuses on gravity equations for bilateral trade, demonstrating the effectiveness of the proposed approach in cases where the outcome variable is bounded below at zero.Bounds on Average Effects in Network Formation Models (with Cavit Pakel).
Working papers
On the Use of the Synthetic Difference-in-Differences Approach with(-out) Covariates: The Case Study of the Brexit Referendum (with Artūras Juodis and Esther de Brabander). 2023. R&R Econometric Reviews • Download paper • Download supplementary appendix • Video of online seminar
Abstract
The Synthetic Control (SC) method has been a popular and dominant method to evaluate treatment and intervention effects in the last two decades. The method is powerful yet very intuitive to use both for empirical researchers and policy experts, but is not without shortcomings. As a response to this, the new Demeaned SC (DSC) and Synthetic Differencein-differences (SDID) approaches were introduced in the literature. In this paper, we evaluate the relative benefits of using DSC and SDID using in-sample placebo analysis on the real data on the Brexit referendum, as well as an extensive Monte Carlo study. Overall, using the SDID methodology, we find that the estimated effect of the Brexit referendum on UK GDP at the end of 2018 and 2019 is higher than previously documented in the literature.On the Use of U-statistics for linear dyadic interaction models. 2023. • arXiv