I am a Ph.D. student in the Joint Program in Financial Economics at the University of Chicago, jointly based at the Booth School of Business and the Kenneth C. Griffin Department of Economics. My research interests are in applied AI methods, asset pricing, and behavioral finance, with a focus on how agents process information, allocate attention, and make decisions.
Before Chicago, I earned a B.Sc. and M.Sc. in Economics from the University of Mannheim and interned at the Deutsche Bundesbank and the European Parliament.
My research has been supported by the Alberto Giovannini UniCredit Scholarship, the Liew Fama Miller Fellowship, and the Yiran Fan Memorial Fellowship.
Research
-
This paper develops a theory-grounded AI model that mimics analysts' cognitive process of selecting, interpreting, and forecasting from corporate disclosures. The model uses analyst reports to identify economically relevant information in earnings calls and transforms the resulting attention-weighted call representations into horizon-specific cognitive embeddings. These embeddings are behaviorally disciplined cognitive-state proxies: model-implied representations of call content constrained by the analyst's own writing and validated against realized outcomes. On analyst-disjoint test sets, the model explains the term structure of EPS forecasts, from short- to medium- and long-term horizons, as well as long-term growth forecasts. The recovered attention measures also explain forecast errors, linking differences in analysts' written interpretations of common disclosures to differences in subsequent forecast performance. Analysts make different forecasts, and systematic forecast errors, because they attend to different parts of the same public disclosure. More broadly, the framework provides a general empirical approach for recovering model-implied information selection from behavioral text. While the paper focuses on analysts and earnings calls, the same architecture can be applied to other expectation-formation environments, including macroeconomic forecasting, retail trading, managerial communication, and investor-relations settings.
-
Transformer Learning Associative Memory
I build a novel dataset linking 290,000 equity analyst reports to their I/B/E/S forecasts and measure analysts' realized recall from historical references in the text. I show that transformer models are equivalent to standard psychological models of contextual retrieval and leverage this equivalence to model recall in the field. The resulting model explains up to 43% of recall behavior, compared to 8% for traditional decay-based models. A 10 p.p. increase in the constructed contextual recall probability implies a 9.3 p.p. increase in the actual recall probability extracted from reports, indicating an approximately one-to-one mapping from simulated recall probabilities to realized recall in the data.
-
Shock-Restricted Markov Switching Structural VARs
The relationship between uncertainty and economic activity remains unresolved in both theory and empirical research: does uncertainty trigger economic downturns, or do economic downturns fuel uncertainty? This paper introduces a novel identification strategy that integrates the flexibility of a Markov-switching VAR model with external shock restrictions to examine the transmission mechanism of uncertainty shocks. These shocks are both statistically and economically significant, exhibiting long-term negative effects on the U.S. economy. Moreover, macroeconomic and financial uncertainty are closely linked to economic activity. Supporting the growth options theory, macroeconomic uncertainty shocks are shown to result in an immediate increase in industrial production.
Teaching
Chicago Booth
- TA for Quantitative Portfolio Management, EMBA/MBA, Ralph Koijen Summer 2024-2026, Winter 2025-2026
- TA for Asset Pricing III, PhD, Stefan Nagel Spring 2025
- TA for Machine Learning in Finance, MBA/PhD, Leland Bybee Spring 2025