Predicting Private Company Exits Using Qualitative Data

Private companies backed by venture capitalists or private equity funds receive their funding in a series of rounds. Information about when each round occurred and which investors participated in each round has been compiled into different databases. Here we mine one such database to model how the private company will exit the VC/PE space. More specifically, we apply a random forest algorithm to each of nine sectors of private companies. Resampling is used to correct imbalanced class distributions. Our results show that a late-stage investor may be able to leverage purely qualitative knowledge of a company's first three rounds of funding to assess the probability that (1) the company will not go bankrupt and (2) the company will eventually make an exit of some kind (and no longer remain private). For both of these two-class classification problems, our models' out-of-sample success rate is 75% and the area under the ROC curve is 0.83, averaged across all sectors. Finally, we use the random forest classifier to rank the covariates based on how predictive they are. The results indicate that the models could provide both predictive and explanatory power for business decisions.