Estimating Return on Initial Public Offering Using Mixtures of Regressions

Abstract

Financial advisors working in a stock exchange market are often faced with a situation to convince a client of merits of investing in a company that just entered the market. To predict company's return based on its revenue, a simple linear regression may be used. This thesis finds that a model based on a mixture regressions is superior over a simple linear regression. The error term in each regression component is assumed to follow standard Gaussian distribution. The data is tested on 116 companies that entered the market as Initial Public Offering (IPO). A 2-component mixture regressions is found to provide the best fit for the data. A simulation study is conducted to verify the performance of this model. Optimum number of components is found using Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) as well as the parametric bootstrapping of the likelihood ratio test statistics.