“Privatization and Productivity in China.” the RAND Journal of Economics, 52:4 (Winter 2021), 884–916, with Yuyu Chen, Mitsuru Igami, and Mo Xiao. (WP version)


We study how changes in ownership affect the productivity of firms. Privatization of state-owned enterprises (SOEs) was a major economic reform during China's rapid growth, but its true impact remains controversial. Although private firms seem more productive than SOEs, the government selectively privatized (or liquidated) non-performing SOEs, which complicates the measurement of productivity. We address this selection problem by incorporating endogenous ownership change into a nonparametric estimation method and exploiting a lag structure in data. Results suggest privatization conferred both short-run and long-run productivity gains. The private-SOE productivity gap is larger among older firms and in less economically liberal regions.

Working Papers


In regression discontinuity designs, manipulation threatens identification. A known channel of harmful manipulations is precise control over the observed assignment, but this channel is only an example. This study uncovers the only other channel: sample selection by deciding manipulation precisely based on the given assignment status. For example, in the assignment design of a qualification exam, self-selection by allowing test retakes precisely based on failing the exam is a precise decision. This precise decision harms identification without precisely controlling the final assignment. For instance, retaking the test never ensures passage, but it distorts the qualification assignment because some students that failed then pass. However, students that have already passed, never fail. This novel channel redefines the justification for identification. Furthermore, under a new auxiliary condition, McCrary (2008)'s test is able to confirm identification and the existing worst-case bounds are nested within our new bounds. In a replication study, another sample selection by analysts appears critical in the robustness of their original conclusion.

"Joint diagnostic test of regression discontinuity designs: multiple testing problem"

Joint work with Koki Fusejima and Takuya Ishihara. Current version, 2022, May 10.


Diagnostic tests with many covariates have been the norm to validate regression discontinuity designs (RDD). The test procedure with many covariates lacks validity due to the multiple testing problem. Testable restrictions are designed to verify a single identification restriction, therefore, a single joint null hypothesis should be tested. In a meta-analysis of the top five economics publications, the joint null hypothesis of local randomization is massively over-rejected. This study provides joint testing procedures based on the newly shown joint asymptotic normality of RDD estimators. Simulation evidence demonstrates their favorable performance over the Bonferroni correction for dimensions fewer than 10 covariates. However, neither Bonferroni correction nor our procedure guarantees its size control with a larger number of covariates.

"Win/Loss Data and Consumer Switching Costs: Measuring Diversion Ratios and the Impact of Mergers"

Joint work with Jeff Qiu and Gloria Sheu. Current version, 2021, November 6. Revision requested at the Journal of Industrial Economics.


The diversion ratio is a key input to indicators of merger harm like up- ward pricing pressure. Measuring the diversion ratio, however, is challenging in the presence of consumer switching costs. We propose an identification strategy for diversion that relies on win/loss data from the two merging firms, a type of data that antitrust authorities can frequently obtain. First, we show that win/loss data from the merging firms and market shares for all firms in two periods are sufficient to identify the diversion ratios between the merging partners. Second, we show that win/loss data from the merging firms are sufficient for partial identification, and we construct a lower bound that provides a good approximation to the diversion ratio when switching costs are high. We demonstrate the performance of our method with numerical simulations and with an application to the Anthem/Cigna merger.

“Noncompliance in randomized control trials without exclusion restrictions.”

Current version, 2021, June 2

(Formerly circulated as "Identification and Inference of Post-Treatment Subgroup Effects.")


This study proposes a method to identify treatment effects without exclusion restrictions in randomized experiments with noncompliance. Exploiting a baseline survey commonly available in randomized experiments, I decompose the intention-to-treat effects conditional on the endogenous treatment status. I then identify these parameters to understand the effects of the assignment and treatment. The key assumption is that a baseline variable maintains rank orders similar to the control outcome. I also reveal that the change-in-changes strategy may work without repeated outcomes. Finally, I propose a new estimator that flexibly incorporates covariates and demonstrate its properties using two experimental studies.


We propose an estimation procedure for discrete choice models of differentiated products with possibly high-dimensional product attributes. In our model, high-dimensional attributes can be determinants of both mean and variance of the indirect utility of a product. The key restriction in our model is that the high-dimensional attributes affect the variance of indirect utilities only through finitely many indices. In a framework of the random-coefficients logit model, we show a bound on the error rate of a l1-regularized minimum distance estimator and prove the asymptotic linearity of the de-biased estimator.

Works in Progress

"Complementarity in Couples’ Retirement : The Effect of Mandatory Retirement Age Extension"(with Mika Akesaka)

"Targeted Meta-Analysis" (with Toru Kitagawa)