Natural Language Processing Approach for Alternative Finance

This thesis explores the application of advanced Natural Language Processing (NLP) techniques to leverage soft information as a predictive tool for assessing firm performance, with a focus on early-stage companies where traditional financial data is often unavailable. In the context of alternative finance, especially equity crowdfunding, understanding key qualitative factors is essential for predicting outcomes. A review of NLP’s evolution highlightes the challenges of processing long, complex documents and summarizing the use of NLP in finance for deriving economic indicators. A unique dataset is then constructed, combining equity crowdfunding campaign metrics, firm-specific financial data, and detailed business plan evaluations, including both objective and subjective metrics from human evaluators. By employing a range of NLP models, we examine how different approaches perform in predicting campaign outcomes. Findings reveal that advanced contextual representations significantly improve predictive accuracy, underscoring the importance of soft information. Additionally, the thesis assesses the value of combining human insights with machine predictions, showing that while AI models excel in accuracy, human evaluators provide essential subjective insights. The fusion of these approaches enhances overall performance, highlighting the complementary nature of human intuition and computational analysis. This research contributes to the field of alternative finance by demonstrating the power of NLP in processing qualitative data to forecast firm outcomes.