Research

Working Papers

Learning and Forgetting in the Knowledge Economy: Implications for Organizational Design
       This paper explores the individual-level dynamics of learning by doing, leveraging a uniquely detailed dataset from a freight forwarding firm. In contrast to existing literature, which primarily examines learning and productivity at the firm level, we use high-frequency data on employee interactions with an internal webapp to estimate learning trajectories, information depreciation, and the effects of collaboration. Our findings show that for specific roles, new employees can reach productivity levels comparable to experienced workers within weeks, and that information depreciation is substantial, with only 10-30% of information retained after one month. These results challenge the prevailing view that reducing employee attrition is essential for maintaining productivity, suggesting that in certain settings, the costs of retraining new employees may be lower than previously assumed. This study contributes to the literature on productivity and human capital by offering new insights into the micro-level mechanisms of learning in modern, knowledge-intensive industries, with implications for managerial policy and future research on incumbency advantages.

In Progress

Adaptive Estimation for Nonparametric Treatments in Double Machine Learning (Joint with Manu Navjeevan)
       We extend the Double Debiased Machine Learning (DML) framework by proposing a method for estimating marginal treatment effects without requiring a pre-specified functional form for the treatment-to-outcome relationship. Drawing on techniques from sieve estimation and generalized cross-validation, the approach adaptively tunes the functional mapping from treatments to outcomes while maintaining valid inference. The method allows for the flexible modeling of treatment effects over the entire support of treatment levels, addressing challenges posed by complex and high-dimensional confounders. We establish the asymptotic properties of the estimator, demonstrate its consistency, and apply the framework to a benchmarked example. By leveraging automatic parameter tuning over the sieve space and adapting post-model selection inference techniques, this work contributes to the literature on causal inference and policy evaluation, providing a robust tool for estimating treatment effects in high dimensional settings with continuous and complex treatments.

Solutions for Congestion in Matching Markets (Joint with Andrew Tai)
       We develop a novel matching mechanism for reducing the congestion in the Gale Shapley Deferred Acceptance matching algorithm. Simulations demonstrate the effectiveness of the modified mechanism, reducing the congestion problem by orders of magnitude. We are establishing a theoretic framework to prove these results without relying on simulations.

Learning to Bid in Dynamic Auctions with Asymmetric Information
       This study examines dynamic bidding strategies in auctions with both common and private value components, where bidders accumulate information through experience, leading to asymmetric beliefs about the common value. Using a dynamic extension of the mineral rights model, bidders adjust their strategies over time as they learn and reduce exposure to the winner’s curse. The model incorporates learning-by-bidding, where the precision of private information improves via repeated auction participation. Nonparametric identification of the asymmetric conditionally independent private information (ACIPI) model is established in both static and dynamic settings, addressing challenges such as asymmetric experience levels and non-monotonic bidding strategies. I hope to apply the framework to first-price sealed-bid real estate auctions, where first-time homebuyers often compete against more experienced bidders.