Working Papers
Presentations: Econometric Society NASM (Jun 2026) · CES China (Jul 2026) · Singapore Economic Review Conference (Jul 2026) · Midwest Econometrics Group (Oct 2026) · Canadian Econometrics Study Group (poster, Oct 2026) · Econometric Society Australasia Meeting (Nov 2026) · ANZAE AE2 Conference (Dec 2026)
Conventional heteroskedasticity- and cluster-robust standard errors plug OLS residuals in for the unobserved errors and correct for its own leverage. Each residual, however, also carries the errors of other observations, and this projection spillover biases the classical corrections. For inference on $a'\beta$, we propose a level factor, computed from the design alone, that measures the size of this bias, which need not vanish as the sample grows. We estimate the variance of $a'\hat\beta$ directly rather than correct the residuals: since both the variance and the expected residual products are linear in the error covariance, estimation is a linear inverse problem, solved by a matrix-free algorithm. The resulting Riesz variance estimator is exactly unbiased whenever the design identifies the variance, and is consistent and yields an asymptotically normal $t$-statistic with many controls and growing clusters. It exists where block-inversion corrections (CR2, CR3, and leave-cluster-out) do not. A general theorem extends it to any estimator linear in the errors and any dependency graph, including multi-way clustering and network, spatial, and banded serial dependence. A degrees-of-freedom correction and restricted covariance classes handle designs with few clusters. For weighted and generalized least squares, the within-cluster products of the weighted residuals and the full set of products no longer give the same estimator, and a change of coordinates restores the OLS benchmark. In simulations the Riesz estimator attains nominal coverage where classical estimators do not. Two applications, one with many fixed effects and one with household weights, show it can change published conclusions.
Regression residuals are projections of the errors, not observations of them. This paper characterizes what residuals identify about the error covariance. Residualization acts on the covariance through a known linear map, and under an unrestricted covariance model the directions it destroys are exactly those generated by the column space of the regressors. A linear covariance functional is identified if and only if it does not load on these lost directions, and restricting the covariance model, for instance to diagonal or clustered form, shrinks them. The central result separates \emph{projection loss} from \emph{moment-selection loss}: projecting the residual covariance onto the maintained covariance structure loses no identifying information, so cross-cluster, long-lag, and long-distance residual products add no identifying directions. They can still improve conditioning, since restricting to the aligned moments squares the condition number. This equivalence relies on the orthogonal-projection geometry of OLS, and generally fails for non-orthogonal transformation. Quadratic statistics in the outcome that are free of the regression coefficients additionally identify the covariance between the fitted and residual components, but not the covariance within the span of the regressors. Because the OLS sandwich variance depends only on that block, it is not identified by any such statistic under an unrestricted covariance model; robust variance estimation draws its identification from maintained restrictions on the covariance structure.
Can Latent Error Clusters be Recovered from Regression Residuals?
(with Maoran Xu, draft coming soon)
Cluster-robust inference presumes the partition of units into dependent clusters is known; we study when it can be recovered from regression residuals. Because the partition is discrete and the projection is known, recovery is an inverse problem on cluster memberships. Our criterion uses residuals alone, so coefficients may drift arbitrarily, and its convex relaxation certifies global optimality when tight. We prove exact recovery for a fixed cross-section as the number of periods grows, under conditions on the design.
Work in Progress
Exact Robust Variance Estimation for Anderson–Rubin Test
(with Keli Xu)
Labor Mobility Networks and Latent Firm Clusters
(with Isaac Hacamo)
Is the ESG Premium a Large-Firm Premium?