Large Language Models for Unraveling the Causal Impact of Financing Constraints on Corporate Exports

This research utilizes the challenge of determining how the firm's financing constraints influence its export activity beyond the limitations of past studies by integrating both structured data and unstructured data. While Large Language Models (LLMs) demonstrate proficiency at producing an understanding of text, when applied to economics, there is room for hallucination due to, among other reasons, the likelihood of misinterpreting or misunderstanding the terms of the domain. We propose the application of the Prompt-Enhance LLM for Deep Economic Analysis (PE-LLM-DEA) whereby the PE-LLM-DEA is able to accommodate heterogeneous data by implementing a multi stage prompt engineering approach, namely, Constraint Identification, Export Outcome Identification, and Causal Reasoning. The model is follows the theory of economics producing convincing causal explanations. We also created a new Financing-Export Dataset (FED) that was used to examine three tasks: financing constraint classification, export scrap participation prediction and identification of causal mechanisms. In each task, PE-LLM-DEA achieves the state-of-the-art performance levels.