Detection of stock price manipulation using quadratic discriminant analysis

In this paper, we use a discriminant analysis (DA) based model to identify stocks that are potentially manipulated. Earlier researchers have used Linear Discriminant Function (LDF), a type of DA, without validating the assumption governing the model. We tested the assumptions on data from the Indian capital market and found that the assumptions do not hold good. We identified the Quadratic Discriminant Function (QDF) as the appropriate DA based classification technique for instances where the data does not meet the stated assumptions of LDF. We developed the LDF Classifier equation using key market data variables that capture the characteristics of the stock. In a market like India, where there are about 5000-plus listed securities, it becomes extremely difficult to monitor all for potential market abuse. The proposed model helps investigators to arrive at a shortlist of potentially manipulated securities which could then be subject to further detailed investigation, if required.