Natural language processing for evaluating initial public offerings and withdrawals: an analysis of the prospectus filings
Date
2023
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
In the United States, the Securities and Exchange Commission (SEC) requires firms issuing their Initial Public Offering (IPO) a.k.a. firms that are ‘going public’ to file a series of prospectus documents detailing the activities, financial position, and future goals of the firm. This dissertation utilizes word embeddings and topic analysis from the Natural Language Processing (NLP) literature to better extract, analyze, and explain information contained within the the initial prospectus, its subsequent amendments, the SEC’s response letter correspondence, and the final prospectus for its influence on the probability of withdrawal, the final offer price, and the first day return. ☐ Firstly, for the prospectus filings, a unique word embedding based metric for sentiment extraction performs significantly better at assigning separate significant sentiment factors than traditional word-count metrics, while still retaining interpretability and comparable performance as measured by adjusted R2 . Secondly, the word-embedding based topic analysis greatly aids withdrawal explanation, in the absence of industrial factors. Additionally, the current most widely published method, Latent Dirichlet Allocation (LDA), is unable to perform as well on withdrawal explanation. Thirdly, embedding based topic analysis of the SEC response letters is significantly stronger in explaining the final IPO price and first day return than previously published methods. Formally, this dissertation is a comprehensive review of existing works on IPOs and their withdrawals; no other known works including all of: withdrawal, sentiment analysis, topic analysis, every amendment filing, and the SEC response letters.
Description
Keywords
Initial public offering, Sentiment analysis, Topic analysis, Word embeddings, Natural Language Processing