Generative AI is now quite popular in quantitative finance for information extraction, signal discovery and time-series analysis. In this work, we detail our methodology of Estimation of Granular Geographical Risk Exposure of companies at country level given very sparse disclosure of revenue available from company financial statements. We use deep Bayesian networks in conjunction with large language models to blend the extracted features from company financial statement textual data with the disclosed higher-level exposures in a cross-sectional collaborative learning framework to impute geographical risk across a large set of global companies.
