DeePre: Delisting Predictor
- 0 Collaborators
Market downturns are an unavoidable reality. There are substantial commonalities in the trends of companies that face being delisted from the stock market that show up as far as one year before the actual removal of the stock. We use DL/NLP on financial statements to perform risk assessment. ...learn more
Project status: Under Development
Intel Technologies
Intel Python
Overview / Usage
[The final version of this project will be submitted for academic journal or conference review]
Financial statements have been ubiquitously used by corporations throughout the world. Governments and investors inspect these statements for a broad range of purposes. However, the numeric data in these statements provided by these corporations can be easily manipulated to mislead consumers or even the companies themselves. Current systems for inspecting the legitimacy of financial statements are based on restrictive methods that use numeric analysis and simple rule-based AI. Countermeasures for falsified financial statements have major problems; malicious actors may easily bypass the inspectors' countermeasures by creating seemingly or borderline legitimate data that may be overlooked. Furthermore, the malicious actors may have broad experience in generating seemingly legitimate data, causing for further confusion.
Methodology / Approach
Many companies that have been delisted from the stock market have experienced the same patterns in their financial statements; it is overwhelmingly the case that these patterns are predictable even so far as one quarter or even one year before the company is delisted. Therefore, while most legitimacy inspection tools are targeted at numbers, we are focused on terminology used in the statements.
Cosine Similarity
Technologies Used
Google's Bidirectional Encoder Representations from Transformers (BERT) can be used in finance for finding important words in sentiment analysis.
Run on Intel Python for accelerated training