Ex Ante Predictability of Rapid Growth: A Design Science Approach

Abstract

We examine how machine learning (ML) predictions of high-growth enterprises (HGEs) help a budget-constrained venture capitalist source investments for a fixed size portfolio. Applying a design science approach, we predict HGEs 3 years ahead and focus on decision (not statistical) errors, using an accuracy measure relevant to the decision-making context. We find that when the ML procedure adheres to the budget constraint and maximizes the accuracy measure, nearly 40% of the HGE predictions are correct. Moreover, ML performs particularly well where it matters in practice—in the upper tail of the distribution of the predicted HGE probabilities. JEL Classification: C53, D22, L25

Scaling‐up: Building, Leading and Sustaining Rapid Growth Over Time

“…Nevertheless, recent advances in machine learning (ML) methods have proven to be effective in predicting phenomena that so far have been rather challenging (Mullainathan and Spiess, 2017). While recent studies have explored the predictability of HGEs using machine learning (ML) methods (Coad and Srhoj, 2020; Hyytinen et al, in press; Kaiser and Kuhn, 2020; Weinblat, 2018), their findings have been rather mixed because these studies relied on rather generic financial and non‐financial predictors such as firm size, age, assets and cash flow. To advance the literature on scaling and HGFs and to enable more accurate predictions of HGFs in the future, we suggest moving beyond generic characteristics and including indicators about capabilities and behaviours.…”