With Veronika Rockova, PhD, Professor of Econometrics and Statistics
Booth School of Business
University of Chicago
Generative AI provides a rich, albeit imperfect, source of probabilistic information that may be leveraged to enhance statistical inference. In econometrics, for example, AI has been used to design in silico experiments, form expectations from historical news, predict stock price reactions, or elicit investment preferences. This talk develops a unified approach to integrating AI-generated information into both likelihood-based and likelihood-free Bayesian analysis. We consider nonparametric formulations in which a Dirichlet process prior is placed on the data-generating distribution, with AI outputs informing the baseline measure and augmenting the observed sample via imputed pseudo-data. We further extend this framework to models defined by moment conditions, where inference proceeds by projecting posterior draws of the empirical distribution onto the constrained model using exponentially tilted empirical likelihood. The resulting procedure is computationally tractable trough parallelizable projection steps and admits rigorous asymptotic guarantees, including Bernstein–von Mises results under regimes where prior influence either vanishes or persists asymptotically. Empirical studies illustrate that incorporating AI-generated auxiliary information may substantially improve predictive performance and uncertainty quantification, particularly in settings with limited observed data.
The Department of Epidemiology & Biostatistics welcomes all participants to our events. If you need a reasonable accommodation to participate in this event because of a disability, please contact Liz Buggs ([email protected]).