Published works
Estimating Nonlinear Network Data Models with Fixed Effects [Link] (2025, Journal of Econometrics)
This paper considers estimation of a directed network model in which outcomes are driven by dyad-specific variables (such as measures of homophily) as well as unobserved agent-specific parameters that capture degree heterogeneity. I develop a jackknife bias correction to deal with the incidental parameters problem that arises from fixed effect estimation of the model. In contrast to previous proposals, the jackknife approach is easily adaptable to different models and allows for non-binary outcome variables. Additionally, since the jackknife estimates all parameters in the model, including fixed effects, it allows researchers to construct estimates of average effects and counterfactual outcomes. I also show how the jackknife can be used to bias-correct fixed effect averages over functions that depend on multiple nodes, e.g. triads or tetrads in the network. As an example, I implement specification tests for dependence across dyads, such as reciprocity or transitivity. Finally, I demonstrate the usefulness of the estimator in an application to a gravity model for import/export relationships across countries.
Example code: basic jackknife code
Efficient Bias Correction for Cross-section and Panel Data, with Jinyong Hahn, Guido Kuersteiner and Whitney Newey (2024, Quantitative Economics) [link]
Bias correction can often improve the finite sample performance of estimators. We show that the choice of bias correction method has no effect on the higher-order variance of semiparametrically efficient parametric estimators, so long as the estimate of the bias is asymptotically linear. It is also shown that bootstrap, jackknife, and analytical bias estimates are asymptotically linear for estimators with higher-order expansions of a standard form. In particular, we find that for a variety of estimators the straightforward bootstrap bias correction gives the same higher-order variance as more complicated analytical or jackknife bias corrections. In contrast, bias corrections that do not estimate the bias at the parametric rate, such as the split-sample jackknife, result in larger higher-order variances in the i.i.d. setting we focus on. For both a cross-sectional MLE and a panel model with individual fixed effects, we show that the split-sample jackknife has a higher-order variance term that is twice as large as that of the “leave-one-out” jackknife.
Working papers
Measuring Misallocation with Experiments, with Jeremy Majerovitz [link]
Misallocation of inputs across firms has been proposed as a reason for low levels of development in some countries. However, existing work has largely relied on strong assumptions about production functions in order to estimate the cost of misallocation. We show that, for arbitrary production functions, the cost of misallocation can be expressed as a function of the variance of marginal products. Using an RCT that gave grants to microenterprises, we estimate heterogeneous returns to capital by baseline characteristics, and provide a lower bound on the total variance of returns to capital. This lower bound is a nonlinear function of the parameters from a linear IV model, and we show that standard methods (e.g. the delta method or projection) fail in this setting. We provide novel econometric tools that provide uniformly valid confidence intervals for nonlinear functions of parameters. We find evidence for sizable losses from misallocation of inputs across the firms we study, although the magnitude depends critically on which inputs we allow to be reallocated. We estimate that optimally reallocating capital would increase output by 22%, while optimally reallocating all inputs would increase output by 301%.
Synthetic Controls with Many Time-varying Covariates
This paper presents a new method for incorporating many time-varying covariates into the synthetic controls method. As introduced in Abadie and Gardeazabal (2003) and Abadie, Diamond, and Hainmueller (2010), the standard synthetic control method uses a small number of predictors that are not time varying (or are constructed to be such by averaging over time). We propose a model in which the untreated potential outcomes and covariates have a shared factor structure. Unit-specific loadings on these factors then form suitable variables that can be used in the construction of synthetic control weights. We estimate these factor loadings from the product of the untreated potential outcomes matrix Y (N units by T0 periods) and the covariate tensor X (N units, K covariates, T0 periods). We provide a rate for the estimation of the factor loadings from this tensor, and show how they can be used in a variety of synthetic controls estimators.
The Role of Collateral in Borrowing, with Nicholas Garvin and Jose-Luis Peydro [RBA Discussion Paper]
Work in Progress
Estimating Linear IV Models with Many Endogenous Regressors [link]