Working Papers & Research Experience
My research uses Bayesian time-series and structural methods to study credit markets, macro uncertainty, and monetary policy transmission.
U.S. hours worked has a pronounced spectral peak at business-cycle frequencies. I show that the shock explaining most business-cycle variation in hours reproduces this peak only when macro uncertainty and credit spreads are included in the information set. The shock combines two forces with distinct propagation: higher macro uncertainty causes a broad contraction followed by gradual recovery, while a credit-related shock generates an expansion followed by a reversal and accounts for the spectral peak in hours. I develop and estimate a model in which firms borrow against collateral to finance their wage bill. Uncertainty about enforcement tightens financing conditions immediately, whereas a favorable but noisy signal about future productivity initially relaxes them. When the signal is revealed to be noise, inherited debt and lower collateral values turn the expansion into a contraction. The model is solved globally with a Deep Equilibrium Network, allowing the borrowing constraint to switch between binding and slack regimes.
Shows that roughly 10% of high-frequency policy shocks can be predicted from forward information already held by the Fed and forecasters, challenging the assumption that these shocks are fully unexpected. Constructs a new shock series orthogonal to forward information and finds materially larger, more statistically significant real effects than earlier high-frequency studies report.
U.S. hours worked, unemployment, the investment-output ratio, and credit spreads exhibit a recurring cycle of about nine years. The model of Beaudry, Galizia, and Portier (2020) generates this cycle because lenders price one-period household credit using current employment, creating a feedback between employment, borrowing costs, and spending. We change only how credit is priced: lenders price default risk over the life of the loan using expected future employment. This change causes the endogenous cycle to disappear once loan duration extends slightly beyond one quarter. Re-estimation preserves the cycle only for loans lasting about 1-1.5 quarters; with longer loans, the best-fitting model relies on persistent external shocks.
Extends a two-sided platform model to allow agents to experience congestion from members of their own group. Congestion raises platform profits when tolerance is low under single-homing, while the condition reverses under multi-homing.
Studies how a monopoly platform chooses prices and first-party content when sellers compete on the platform. Seller competition can lead the platform to restrict seller participation and can reverse standard predictions about first-party-content investment under unfavorable expectations.
Compares pure intermediation with a platform that also competes in sellers’ product market. Entry lowers sellers’ membership fees and can benefit buyers, sellers, and the platform under monopoly; with competing platforms, buyer gains depend on product differentiation and perceived quality.
- Built an automated workflow to extract, translate, and structure data from more than 1,000 scanned Hindi PDFs
- Developed a machine-learning-assisted pipeline that converted more than 3,000 heterogeneous credit card statements into structured, privacy-safe transaction data
- Supported more than 10 academic projects with data acquisition, cleaning, statistical analysis, and modeling
- Developed and optimized Python algorithms for a high-dimensional fixed-point problem in a macro-labor equilibrium model, improving convergence and computational efficiency
- Supported the assessment of a $150 million external commercial borrowing for compliance with India’s foreign-exchange regulations