U.S. sanctions and firm offshore activities: A firm-country analysis using natural language processing

This study examines how U.S. multinational firms adjust their offshore activities following U.S. economic sanctions. Using more than five million firm-country-year observations, we link offshore disclosures from 10-K filings to country-level sanction data. We construct a sanction-risk measure using a supervised gradient boosting model that combines firm characteristics with FinBERT language embeddings extracted from risk disclosures. We find that sanctions are associated with reduced offshore engagement in sanctioned countries, with the magnitude varying across sanction types. These associations are not explained solely by disclosure behavior and remain consistent in event-window analyses around sanction onset. Sanction exposure is also associated with tighter financial constraints and lower profitability. Firms with higher ex ante sanction risk exhibit stronger adjustments in offshore activities, highlighting the relevance of AI-based textual analysis and machine learning for assessing geopolitical exposure. Finally, sanction exposure weakens the persistence of existing offshore networks, with effects varing by prior network size.