Research
Job Market Paper
This paper develops a control function estimator for a binary response model with a continuous endogenous regressor in the absence of exclusion restrictions. The proposed semiparametric approach utilizes heteroskedasticity to construct a control variable to address endogeneity. This paper establishes the large sample properties of the proposed Quasi Maximum Likelihood Estimator. In Monte-Carlo simulations, it performs well in finite samples. Using the Health and Retirement Study data, this paper employs a semiparametric model to estimate the heterogeneous effect of education on health without exclusion restrictions. Additionally, this paper estimates marginal effects based on the Average Structural Function.
Working papers
"A Score Test for a Constant Threshold Variance in Binary Response Models with and without Endogeneity" with Roger Klein
We propose a score test for heteroskedasticity in binary-response models with an endogenous continuous regressor. The test is constructed from quasi-maximum likelihood estimation of the binary model and ordinary least squares in the first stage. The resulting statistic is asymptotically standard normal under the null. Monte Carlo experiments indicate good size control and strong power in moderate samples. Since the procedure avoids parametric assumptions on the variance function, it provides a practical and robust diagnostic for heteroskedasticity in binary models with endogeneity.
This paper examines the dual role of financial literacy in mediating the effects of cognitive decline on fintech adoption and elder fraud among U.S. older adults. Drawing on a conceptual framework integrating bounded rationality and rational crime theory, we introduce the ``perception discount factor" (PDF) to model the convolution of literacy and cognitive decline. Using Health and Retirement Study data from 2002-2022, we first estimate the erosive impact of cognitive decline on financil literacy via a nonlinear mixed-effect panel regression, and then test $U$-shaped patterns via Heckman selection models to address biases toward high-functioning respondents. Empirical results confirm non-monotonic adoption for high-literacy elders, with rebounds exposing them to heightened fraud, while low-literacy individuals exhibit persistent disengagement. Cognitive impairment remains a key risk, but fraud targets educated, wealthy elders via overconfidence, challenging conventional vulnerability narratives. Protective factors like family guardianship and accurate risk perceptions mitigate victimization.
This paper investigates monopsony power in foreign-input markets—the ability of downstream firms to pay prices below the marginal revenue product of imported intermediates. We develop a structural estimator that incorporates factor-biased (nonHicks-neutral) technical change to recover firm-level markdowns. Using matched Chinese industrial firm and customs microdata from 2000–2007, we document economically large and persistent foreign-input markdowns. Over the same period, non-neutral technology expanded rapidly, reshaping effective input productivity. Ignoring this channel generates substantial bias in markdown estimates—overstating buyer power when technology is labor- or material-augmenting, and understating it when inputsaving forces dominate. Monte Carlo simulations confirm that the proposed estimator performs well in finite samples. Empirically, we find that the median firm-level markdown corresponds to a market power measure of about 2.81, while the average growth rate of factor-biased technology is approximately 1.40. These results underscore the importance of incorporating non-neutral technical change in measuring input-market monopsony and provide a tractable empirical framework for its identification.