Working Papers

Abstract: Social interaction with coworkers is common in the workplace. This paper explores how coworkers affect inequality through labor market sorting and on-the-job learning. Using matched employer-employee data from Italy, I first document two sets of empirical evidence by estimating an econometric model that incorporates coworkers in a wage regression with a novel estimation method. I find two main mechanisms through which coworkers affect wages: production complementarity and learning from coworkers. I also show that coworkers explain a substantial fraction of wage inequality, similar to that firm heterogeneity explains. To account for wage dynamics induced by these two channels and the subsequent impact on lifetime income inequality, I incorporate coworkers into a labor search model with worker and firm heterogeneity. I find that half of the lifetime income variation is explained by workers' initial ability. Firm heterogeneity explains around 15 percent of the remaining unexplained part, while coworker production complementarity and learning contribute to another 15 percent and 30 percent, respectively. 

"Time Aggregation in Health Insurance Deductibles" (with Corina Mommaerts)

Conditionally Accepted, American Economic Journal: Economic Policy

First Draft: January 2021; NBER Working Paper

Abstract: Health insurance plans increasingly pay for expenses only beyond a large annual deductible. This paper explores the implications of deductibles that reset over shorter timespans. We develop a model of insurance demand between two actuarially equivalent deductible policies, in which one deductible is larger and resets annually, and the other deductible is smaller and resets biannually. Our model incorporates borrowing constraints, moral hazard, mid-year contract switching, and delayable care. Calibrations using claims data show that the liquidity benefits of resetting deductibles can generate welfare gains of 3-10% of premium costs, particularly for individuals with borrowing constraints.

Abstract: This paper proposes a framework for estimation and inference in a panel data model of peer and spillover effects.  We consider a linear-in-means model that may include social influences through three channels: outcomes, observed characteristics, and an unobserved individual-specific characteristic. To learn about the magnitude of effects in models that include some but not all of these social influences, the existing econometrics literature has adopted estimators based on least squares, maximum likelihood, and two-step instrumental variables ideas. Neither of these approaches will, in general, yield consistent estimators in a panel data context. Instead, we propose estimation and inference based on a novel objective function. We illustrate the ideas using the universal transcript data from the University of Wisconsin-Madison and explore the classroom peer effects during the semester when the university switched its learning mode to online. We show that the existing method estimates a positive and significant peer effect, while our method finds it to be close to zero and statistically insignificant.

Abstract: We study a critical driver of future wages: peers. Using linked employer-employee data for Italy, we explore peer effects in two directions. First, accounting for the endogenous sorting of workers into peer groups, we estimate that a 10 percent rise in peer quality increases one’s wage in the next year by 1.8 percent. The effect decreases gradually over time and becomes about 0.7 percent after five years. Second, using an event study specification around mobility episodes, we show that hiring high-quality workers, separating from low-quality workers, and moving into high-quality peer groups, are significant drivers of wage growth.

Abstract:  This paper empirically shows that the imbalance between an ethnic group’s political and military power is crucial to understand the likelihood that a group engages in a conflict. We develop a novel measure of a group’s military power by combining machine learning techniques with rich data on ethnic group characteristics and outcomes of civil conflicts in Africa and the Middle East. We couple this measure with available indicators of ethnic groups’ political power as well as with a novel proxy based on information about the ethnicity of cabinet members. We find that groups characterized by a higher mismatch between military and political power are approximately 30% more likely to engage in a conflict against their government. We also find that the effects of power mismatch are nonlinear, which is in agreement with the predictions of a simple model that accounts for the cost of conflict. Moreover, our results suggest that high-mismatched groups are typically involved in larger, longer, and centrist conflicts.

Abstract: This paper examines the effect of the New Rural Pension Scheme (NRPS) on labor supply among the aged population in rural China. Using a difference-in-difference specification, I find the introduction of NRPS has increased the (intensive) labor supply for both pensioners and contributors by more than 10 percent. Heterogeneity analyses suggest that the potential mechanisms are different for the pensioner and the contributor. For pensioners, the program has elevated effective labor productivity through health improvement and credit constraint alleviation, which leads them to work more. On the other hand, pension contributors, especially those who are hand-to-mouth, increase labor supply because the annual contribution is an additional financial burden to them.

Work in Progress

“Dynamic models with many regressors” (with Mikkel Sølvsten


"GiniInc: A Stata Package for Measuring Inequality from Incomplete Income and Survival Data" (with Guido Alfani, Chiara Gigliarano, and Marco Bonetti). The Stata Journal, 2018.

The package is available to download here or by simply typing - ssc install giniinc - in Stata.

Abstract: Often, observed income and survival data are incomplete because of left- or right-censoring or left- or right-truncation. Measuring inequality (for instance, by the Gini index of concentration) from incomplete data like these will produce biased results. We describe the package giniinc, which contains three independent commands to estimate the Gini concentration index under different conditions. First, survgini computes a test statistic for comparing two (survival) distributions based on the nonparametric estimation of the restricted Gini index for right-censored data, using both asymptotic and permutation inference. Second, survbound computes nonparametric bounds for the unrestricted Gini index from censored data. Finally, survlsl implements maximum likelihood estimation for three commonly used parametric models to estimate the unrestricted Gini index, both from censored and truncated data. We briefly discuss the methods, describe the package, and illustrate its use through simulated data and examples from an oncology and a historical income study.