Artificial Intelligence and Managerial Language: Evidence on Firm Outcomes

This dissertation studies how managerial language reveals latent firm states that matter for corporate behavior and valuation. Across three essays, I use large language models to turn unstructured corporate disclosures into forward-looking measures of (i) financial constraint pressure, (ii) executive confidence bias, and (iii) the credibility of long-horizon climate targets. The first essay constructs a Financial Covenant Pressure Index (FCPI) from covenant-related disclosures and shows that higher covenant pressure increases the likelihood of earnings misstatements before any realized covenant breach. The second essay develops a prompt-based framework that measures executive confidence bias in earnings calls along three psychological dimensions and finds that more confident managerial language predicts lower valuation multiples, higher risk, and moderately more aggressive corporate policies, especially where managerial discretion is higher. The third essay studies whether markets value net-zero targets and, crucially, whether markets price delivery after the target is announced. Using net-zero pledges extracted from sustainability reports, a firm-specific benchmark path for counterfactual emissions intensity, and an instrumental-variables strategy that addresses endogenous timing, the results show that announcing is associated with a modest valuation premium, while delivery is economically large: firms that fail to deliver face sizeable valuation penalties and firms that make sustained progress are rewarded, with stronger effects in hard-to-abate industries and in later years.