UC Santa Cruz

The objective of this research is to design, develop and implement an intelligent decision support system (IDSS) for making rational private equity investment decisions. (Private equity investments are capital investments in enterprises that are not traded on public equity markets; they include Equity Buy-Out, Venture Capital, and the new Equity Crowd Funding (ECF) asset classes). The design and development of the IDSS requires the integration of investment science (valuation theory, portfolio theory, efficient market theory) and information science & technology (statistical learning, software engineering).

The IDSS architecture is built around the two primary tasks in the due diligence process for evaluating private equity investments, namely, evaluating whether a prospective investment can generate returns commensurate with its risk, and then verifying that the enterprise has the appropriate attributes or success factors to achieve that rate of return. These two tasks are implemented as stages in the IDSS: the “Evaluate” stage (Stage 1), and the “Verify” stage (Stage 2). The motivating application for this IDSS is the emerging ECF asset class, which is applied in this research to demonstrate the financial engineering and statistical learning principles and processes developed.

A fundamental investment science result of this research is the derivation of the Capital Market Line (CML) for the private equity market, based on modern portfolio theory, efficient market theory, and historical financial performance metrics. The CML enables quantification of the target Internal Rate of Return (IRR) for individual assets in the market as a function of risk. Results show that the target IRR for ECF investments places significant constraints on the operating requirements of companies seeking ECF financing.

Achieving the target IRR requirement identified in Stage 1 is a necessary but not sufficient condition for making a positive investment decision. In the Verify stage statistical learning methods are applied to verify whether a prospective investment has the necessary attributes in terms of team competencies, product value proposition, and market dynamics to achieve the target IRR. This dissertation describes the processes developed and results achieved in this research for attribute selection and extraction, model selection, and model validation, which together form the core elements of Stage 2. The results of this part of the research demonstrate that the k-nearest-neighbors (kNN) algorithm yields the lowest loss function value of the set of 8 supervised, unsupervised and meta classifiers tested.