Machine Learning and Survival Prediction Models for Decision Support in the Private Equity Market

The object of this thesis is the development of predictive algorithms for the support of investment decisions in Private Equity. In particular, the developed methodologies attempt to estimate the probability associated to a future state of a company, distinguishing among four possibilities: being acquired, go bankrupt, stay private, go public. Further, we model the evolution in time of the probability for a company to undertake an Initial Public Offering (IPO), which is one of the major events in the Private Equity market. The first objective has been realized by means of Machine Learning techniques, using Random Forest and Neural Networks models, in both the Multilayer Perceptron and Long Short-Term Memory configurations. The second objective was implemented by exploiting Survival Models, which are commonly used in the bio-medical environment, in particular Kaplan-Meier estimate, Cox models and the Accelerating Failure Times (AFT) model. Extensive numerical tests have been performed with R and Python, on the basis of an historical dataset available from Thomson Reuters.