Research
You can find my C.V. here !
Published Paper
(with Professors Robert Clark, Decio Coviello, and Art Shneyerov)
We study the impact of an investigation into collusion and corruption to learn about the organization of cartels in public procurement auctions. Our focus is on Montreal’s asphalt industry, where a police investigation was launched in 2009 to examine the allegations of bid rigging, market segmentation, and bribes to bureaucrats. We collect procurement data and use a difference-in-difference approach to compare outcomes before and after the investigation in Montreal and in Quebec City, where there had been no allegations or evidence of collusion or corruption. We find that, following the investigation, entry and participation in auctions increased, and that the price of procurement decreased. We then decompose the price decrease to quantify the importance of two aspects of cartel organization, coordination and entry deterrence, for collusive pricing. We find that the latter explains only a small part of the decrease.
Work Under Review
(with Professors Achyuta Adhvaryu, Anant Nyshadham, and Jorge Tamayo)
We study some aspects of the day-to-day functioning of India’s largest garment producer. We focus on the relational contracts between managers of production lines. We show that managers form informal partnerships with other managers, independently from upper management, in order to smooth the effect of worker absenteeism shocks to their production lines. They do so by borrowing (lending) workers when their absenteeism shock is relatively larger (lower) than that of their partners. Managers cultivate relationships with only a few other managers, leaving many potentially beneficial trades unrealized. Moreover, they tend to form trade partnerships with managers that are similar to them in terms of demographics and in terms of productivity level. The results are consistent with our relational contracting model. While the behavior of managers ultimately helps the firm limit the impact of worker absenteeism, we show by way of simulations that the firm can better leverage these informal relationships to improve profit and productivity by incentivizing partnerships between high and low productivity managers.
(with Professors Teresa Molina, and Anant Nyshadham)
We study the role of labor misallocation (i.e., suboptimal sorting of households across sectors) in explaining low productivity in developing countries. We estimate a generalized earnings equation with dynamic correlated random coefficients, allowing households to learn about their relative productivity across sectors. Estimates show that households select into non-farm enterprise on the basis of comparative advantage, but learn and converge slowly over time, with many households spending substantial amounts of time in a suboptimal sector. Roughly 35% of households are misallocated to start, earning nearly 50% less on average than they could have if they were properly sorted across sectors.
Working Paper
I study the effect of India's local minimum wages on the production structure of firms in the formal economy. I compile data on the country's numerous minimum wages which vary at the state, year, and industry level, and show that changes to these wages have important effects on firm-level capital investment and employment of different types of employees. The effects depend on the firms' ability to automate and offshore certain tasks. Using a difference-in-difference approach, I show that firms in the average industry, that is, firms in industries neither intensive in routine nor offshorable tasks, continue to invest in machinery and computers at a rate of 8% per year following a minimum wage hike. However, they substitute payroll workers with managers and contract workers less likely to be bound by the minimum wage. Firms in industries intensive in routine tasks that are easier to automate invest 6.1% more in machinery and 4% more in computers, at the expense of payroll workers. Firms in industries intensive in tasks easier to do remotely continue to invest in machinery and computers, but the rate of investment in computers falls by 6.2% following a minimum wage hike, and payroll worker employment falls as well. This suggests that some tasks that combine workers and computers, like data analysis, may be offshored. These results support the predictions of a task-based production model, and indicate that minimum wages have a strong effect on the structure of production at the firm level, leading some towards increased rates of automation and offshoring.
Work in Progress
Great Expectations: Responses to Current and Future Transfers for Low-Income Individuals (draft coming soon!)
(with Professors Achyuta Adhvaryu, Pamela Jakiela, and Dean Karlan)
Cash transfers constitute a primary tool for income redistribution in most countries around the world. Accordingly, the impact of such transfers has received much academic as well as policy attention. Yet despite their importance, we still know little about the effects of the prospect of future transfers on life cycle outcomes. Do future beneficiaries adjust their current behaviors related to consumption, saving, labor supply, health, and the like in response to the promise of future transfers? Or are constraints on borrowing, saving, time, etc. too strong for future beneficiaries to adjust their behavior optimally? We conduct a randomized controlled trial in Uganda to study these questions. Our trial design features two treatment groups who received transfers immediately (one after a light-touch mental planning program), a novel group who was promised (and then indeed received) a transfer one year in the future, and a pure control group who did not receive any transfers. We find that both contemporaneous transfers and the promise of future transfers increased labor supply, consumption, entrepreneurship, and health in the short and medium run. Our findings are rationalized by a forward-looking life cycle model of consumption with productive health stock. Rich outcome data coupled with the trial's novel design allows us to reject many other standard versions of the life cycle model. Our results also raise important questions regarding the validity of interpreting delayed receipt of treatment as a control group in RCTs.
Using AI to Expand the Job Search of Displaced Workers in the Aftermath of the Covid-19 Crisis-Supported by funding from the Russell Sage Foundation
(with Professors Achyuta Adhvaryu, Anant Nyshadham, and Jorge Tamayo)
To date, more than 38 million workers have filed for unemployment during the COVID-19 crisis. Early evidence suggests that the worst economic impacts will be concentrated among entry-level service workers, who are disproportionately young, female, and from marginalized groups. These populations are often slow to be reintegrated into the economy after economic contractions due at least in part to discrimination and imperfect information, with full recovery from the last decade’s financial crisis, for example, taking more than 7 years. This inequality in employment outcomes is caused in part by job search frictions that disproportionately affect the ability of low-income minority workers to find good matches. In the proposed study, we design a labor market experiment to test how machine learning predictions of performance can help job seekers in their search. To do so, we used machine learning models to predict performance from psychometric profiles on a wide range of entry-level occupations from previous work. Then, we recruit over 2000 U.S. participants seeking jobs in entry-level positions to take part in a Randomized Controlled Trial (RCT) where all participants are given access to an online job search platform that we designed. Participants in treatment groups receive additional information about their predicted performance in different entry-level occupations, see a reduction in their search cost, or both. We measure the treatment effects on the participants’ search outcomes including unemployment spell, accepted wage, satisfaction with the accepted offer, and the number of occupations they search from survey and usage data from the platform.
Optimal Contracting with Psychometrics
(with Professors Achyuta Adhvaryu, Anant Nyshadham, and Jorge Tamayo)
The skills, personality traits, and the cost of effort of workers partly explain variation in responses to pay incentives. However, the workers' “type” and realized effort are hard to observe and, therefore, hard to contract on for employers. As the demand for routine tasks decreases, personality traits or “soft skills” are increasingly valuable on the labor market. Sophisticated firms have internalized this and increasingly use psychometric tools to screen workers for hiring and promotion decisions. However, screening is costly and imperfect. We study the potential gains from psychometric screening to enable personalized performance pay contract in the context of a digital labor market experiment where such contracts are the norm. To do so, we measured a broad array of personality traits and cognitive skills of 350 workers on Amazon's Mechanical Turk digital labor market. Workers were then hired to perform three different customer service tasks varying in difficulty and skill requirements. We experimentally vary performance pay within worker and within task type which allow us to nonparametrically recover the mapping between realized effort and worker type. We simulate gains for a sophisticated firm capable of screening workers and offer personalized contracts in competition with unsophisticated firms who do not screen and offer a common contract. We use machine learning methods to predict the workers' type from the psychometric data which allows us to characterize which type of workers provides the greatest opportunity for gains in different tasks and to study how noise from the psychometric data erodes gains.
(with Yuqing Gu, and Professors Achyuta Adhvaryu, Anant Nyshadham, and Jorge Tamayo)
We introduce a cost-sensitive machine learning framework that can minimize hiring costs under different labor market conditions for firms using machine learning to screen and escalate applicants to interviews. We show that the optimal algorithm depends on the abundance of quality applicants in the market, the interview cost, and the value to the firm of filling vacancies. We provide ways to train the machine learning models in order to provide a fair and nondiscriminatory treatment of all applicants regardless of their identity.
Using Screening Technologies to Target Trainings and Improve Managerial Productivity: Evidence from a Large-Scale Experiment in Ready-Made Garment Factories in India
(with Professors Achyuta Adhvaryu, Anant Nyshadham, and Jorge Tamayo)
Good managers are extremely important for both worker satisfaction and well-being (and consequent workplace outcomes like attendance and retention) as well as for enabling workers to achieve higher productivity (and as a result to earn higher wages). However, identifying good managers, both among candidates at the time of hiring as well as among incumbents for the purposes of rewarding best practices and targeting training in deficient skills, is costly and difficult, particularly for low-margin labor-intensive manufacturing firms in developing countries. Over the last 3 years, in a large garment manufacturing firm in India, we have developed and validated an instrument to comprehensively measure managerial skills, traits, and practices; estimated the contributions of each dimension of managerial quality to productivity; simulated gains from enhanced screening and training policies; and designed and tested training to remediate managerial deficiencies. We translated the aforementioned measurement and training tools into a tablet app-based platform that allows for self-administration and efficient scaling. In a large-scale experiment, we randomize the access to our skill measurement and training tool across incumbent managers and new hires. We, then, test the benefit of using the tool on productivity and the well-being of workers.