I build forecasting models on real-time data: GDP nowcasting across major economies, recession probabilities, and institutional risk, all validated against only what was actually published at the time.
Experience applying economics, statistics, machine learning and AI to financial and economic data through predictive modeling, empirical research, and investment analysis. Interested in developing statistical models and analytical frameworks that improve forecasting, deepen quantitative insight, and support better investment and business decisions.
BA in Economics from the University of Rochester (2025). Senior thesis ranked in the top 3 of 75 papers in the department. Current work spans GDP nowcasting across 10 major economies, alternative-data equity signals, and multifactor risk model development.
An end-to-end system nowcasting current-quarter real GDP growth in 10 major economies from point-in-time macro data. Combines a mixed-frequency dynamic factor model with bridge regressions and machine learning in an inverse-variance ensemble. Validated out-of-sample and benchmarked against the Atlanta Fed's GDPNow; outputs directly informed global macro asset allocation decisions. Built as part of my research role at Versor Investments.
A vintage-honest forecasting pipeline built on ALFRED, the Fed's real-time data archive. Estimates US recession probability at nowcast, 3-, 6-, and 12-month horizons with a yield-curve probit, a mixed-frequency dynamic factor model, and a gradient-boosted classifier. Backtested over 27 years and 330 monthly forecasts using only data as published at the time.
View projectA survival-analysis pipeline predicting which US private nonprofit colleges will close 1 to 4 years ahead, built on a panel of ~1,200 institutions from 1998 to 2024 with every feature lagged to its real publication date. Beats the Department of Education's financial-responsibility score from public data alone (ROC 0.85 vs 0.74).
View projectTest-Optional Policies in College Applications: Applicant Behavior Before and After Widespread Adoption
This study investigates how test-optional admissions policies affect applicant behavior and the academic profiles of admitted students, leveraging the natural experiment created by the widespread adoption of these policies during the COVID-19 pandemic. I hypothesize that test-optional policies encourage lower-performing applicants to strategically withhold scores, leading to increased application volumes, higher 25th percentile test scores, and a compressed interquartile range of reported scores.
Using panel data from the Integrated Postsecondary Education Data System (IPEDS) covering 490 U.S. institutions from 2013 to 2023, I estimate difference-in-differences models that compare outcomes before and after policy changes across institutions that adopted test-optional policies at different times.
The results show that test-optional adoption during COVID led to a roughly 40 percentage point decline in score reporting, a nearly 40% rise in applications to highly selective institutions, a 50-point increase in 25th percentile test scores, and a 15-point narrowing of interquartile ranges, with effects that varied by institutional selectivity. In contrast, institutions that adopted test-optional policies before COVID experienced significantly smaller or statistically insignificant changes, suggesting that large-scale adoption, rather than isolated policy shifts, fundamentally reshaped admissions dynamics.