Verónica C. Pérez

Economic uncertainty shapes investment, hiring, and asset prices, making its measurement from text a central task in economics and finance. While LLMs accurately measure uncertainty in financial texts such as earnings call transcripts, it remains unclear whether they develop coherent internal representations or merely pattern-match on surface lexical cues. In this study, we discover that LLMs linearly represent economic uncertainty with a single direction in the residual stream. Using activation patching on two synthetic datasets of contrastive earnings-call statements with varying linguistic styles, we localize this direction and find that models aggregate the uncertainty signal at the final token regardless of lexical patterns. The extracted direction perfectly separates held-out high- and low-uncertainty statements both within- and cross-dataset. Causal interventions show that adding or subtracting this direction monotonically flips uncertainty predictions, including cross-dataset transfer from templated to naturalistic text. Finally, in a downstream portfolio allocation task using real earnings-call excerpts, steering along the uncertainty direction shifts model investment toward safe assets, consistent with economic theory. Together, our results establish that LLMs encode economic uncertainty as a structured, causally active, and transferable representation, offering a foundation for interpretability-based auditing and control of LLMs deployed in financial analysis.

Since the 18th century, policymakers have debated the merits of industrial policy (IP). Yet, economists lack basic facts about its use due to measurement challenges. We propose a new approach to IP measurement based on information contained in policy text. We show how off-the-shelf supervised machine learning tools can be used to categorize industrial policies at scale. Using this approach, we validate longstanding concerns with earlier approaches to measurement which conflate IP with other types of policy. We apply our methodology to a global database of commercial policy descriptions, and provide a first look at IP use at the country, industry, and year levels (2010-2022). The new data on IP suggest that i) IP is on the rise; ii) modern IP tends to use subsidies and export promotion measures as opposed to tariffs; iii) rich countries heavily dominate IP use; iv) IP tends to target sectors with an established comparative advantage, particularly in high-income countries.