A Bayesian approach to ranking private companies based on predictive indicators

approach to ranking private companies based on predictive indicators. AI Communications. Private equity investors seek to rank potential investment opportunities in growth stage private companies within an industry sector. The sparsity of historical investment transaction data for many growth stage private companies may present a major obstacle to using statistical methods to discern industry specific features associated with successful and failed companies. This paper describes a Bayesian ranking approach based on (i) extracting and selecting features; (ii) training support vector machine classifiers from feature pairs of labeled companies in an industry; (iii) non-parametric estimation of posterior probabilities of success and failure; and (iv) ranking unlabeled companies within a cohort based on scores derived from posterior probability estimates. We anticipate that this approach will not only be of interest to statisticians and machine learning specialists with an interest in venture capital and private equity but extend to a broader readership whose interests lie in classification methods where missing data is the primary obstacle. Silicon Valley is currently host to company growth rates and exit valuations of unprecedented levels. Take, for example, the recent purchase of Instagram by Facebook for $1 Billion, representing a 20x return on total investment in only two years. In the absence of any empirically substantiated general formulaic approach to pick the best companies to invest in, investors base investment decisions on a wide set of considerations influenced, in part, by their prior experiences and oftentimes fundamental analysis of companies within an industry sector. Growth stage private companies, however, often have investment transaction histories from which characteristics associated with successful and failed companies (labeled companies) may be discerned using statistical methods. One of the primary challenges in pursuing this approach is the sparsity of the historical data. This papers describes a four step approach based on (i) feature extraction and selection; (ii) classification; (iii) derivation of a score using a novel Bayesian approach to estimate posterior probabilities of success and failure; and (iv) ranking companies within a cohort based on their scores. Focusing on agricultural companies in the cleantech sector, we demonstrate how this approach can be used to rank companies based on a set of features. Statistical techniques for learning predictive indicators and patterns have been used extensively in the public capital markets. One area of significant traction is in high frequency trading, which uses statistical techniques to generate an estimated $8 billion a year in trading …