This paper develops a joint estimator for a shared dense component and row-sparse task heterogeneity. It establishes near-minimax rates under collective identification, even when some source tasks are not identified on their own. Sparse target correction yields confidence sets and regret guarantees for sequential decisions.
This paper studies semiparametric inference in short panels when estimated unit-level coordinates are reused within clusters. It identifies derivative–influence and curvature biases of order T−1 that remain after first-order orthogonality. A no-splitting correction supports valid √NT inference with growing sieves.
This project develops a smoothed panel quantile framework with heterogeneous slopes and multi-way fixed effects. It derives analytical corrections for incidental-parameter bias in static and dynamic models.
This project asks how biased group beliefs shape reputational incentives and when strategic responses can sustain self-confirming performance gaps.