Repository for Predicting Financial Shenanigans from Financial Statements
From the forthcoming preprint on ArXiv: [add link]
Every investor’s primary concern is (or should be) the prevention of the permanent loss of capital. There are many proprietary models available to investing professionals to identify companies that are engaging in financial shenanigans. In this project I develop a model that the average investor can use. With a step-wise AUC procedure in a Bayesian hieararchical modeling framework, I identified 3 Generally Accepted Accounting Principles items that predict false disclosures. These items include: basic shares outstanding from the income sheet, net purchases of PP&E on the cash flow sheet, and other equity from the balance sheet. The prediction model works over arbitrary time windows and had an accuracy of 0.9 when predicting fraudulent disclosure on a test data set. Interactions between these 3 variables revealed a high probability of fraudulent disclosure when all 3 are being manipulated. Investors may find this model useful when building a stock portfolio or to check on the status of their existing holdings. All code, data, and models are available on github for future researchers.