Firm level political risk measure and data

We describe in details the \citet{NBERw24029} firm-level political risk measure (PRiskit) and provide summary statistics for this measure as well as for the key independent variables in our study. As out tests move from the borrower-level to the lender-level, we organize our discussion of the sample selection procedure, the data sources, and the pertinent descriptive statistics accordingly. 

PRiskit

To arrive at a firm-specific time varying measure of political risk, \citet{NBERw24029} exploit the practice that publicly listed firms have quarterly earnings conference calls, in which financial analysts and other market participants discuss the current state-of-affairs with senior management. Applying a machine learning algorithm to the transcripts of these calls, the authors then determine how much of the conversation in the conference call centers on political topics. To determine what are political topics, the paper extracts all two-word combinations (“bigrams”) from training libraries that are indicative of discussion of political topics  and non-political topics ℕ. Specifically, they use an undergraduate text book on American Politics, supplemented with newspaper articles from the Domestic Politics sections of major US newspapers in addition to an undergraduate financial accounting textbook together with newspaper articles on corporate events. To construct the political risk measure, the number of occurrences of exclusively political bigrams in conjunction with a synonym for risk or uncertainty is counted and then divided by the total number of bigrams in the transcript (to adjust for the length of the transcript).