Research
I work on panel data econometrics with my supervisors, Eric Renault and Kenichi Nagasawa.
Granular Instrumental Variables under Weakness
Granular Instrumental Variables can be weak in the absence of dominant units (lack of granularity). I study and characterize instrument strength when granularity is suspect. To do this, I model unit sizes $\mathcal{s}_i$ having a power-law tail, $\Pr(\mathcal{s}_i>s)=c s^{-\mu}$, so that the tail index $\mu$ governs the prevalence of dominant units, and thus the strength of the GIV. Three regimes emerge. (i) When $\mu\in(0,1)$, tails are very thick and the instrument is strong: GIV is consistent and asymptotically normal at the standard $\sqrt{T}$ rate. (ii) When $\mu\in(1,2)$, identification is near-weak: validity requires $N/T\to 0$, and asymptotic normality obtains at the slower rate $\sqrt{T}/N^{\,1-1/\mu}$. (iii) When $\mu>2$, tails are too thin to sustain instrument strength and the GIV estimator is inconsistent.I propose a weakness-robust inference using an Anderson-Rubin statistic in this regime.
Granular Instrumental Variables (with Eric Renault and Kenichi Nagasawa)
The methodology of Granular Instrumental Variables (GIV) pioneered by Gabaix and Koijen (2022) offers an exciting way to estimate structural parameters on endogenous variables. Our paper adds on to this growing literature with two main contributions regarding asymptotic inference. First, in the context of a known matrix of factor loadings, we set the general asymptotic theory of efficient GIV, by resorting to the theory of optimal instruments associated to conditional moment restrictions. Second, in the case of unknown factor loadings, we stress that in contrast with efficient estimation of standard conditional moment restrictions, a first step estimator of optimal instruments has an impact on the asymptotic distribution of estimators of structural parameters.
Interactive Fixed Effects with Heterogenous Parameters (with Kenichi Nagasawa)
Bai(2009)'s popular Interactive Fixed Effects estimator fails when the parameter of interest varies with individual units. Our research explores the consistent estimation of these heterogenous parameters.