Unveiling the entrepreneurial ecosystems black box: An interpretable machine learning approach

Abstract

Multidimensionality, interdependence, and non-linearity characteristics of complex systems make the measurement of Entrepreneurial Ecosystems a very challenging task. A known limitation of the Entrepreneurial Ecosystem frameworks is the difficulty of quantifying the impact of distinct elements. In this study, we propose a novel methodology based on recent advancements in Explainable Artificial Intelligence to fill this gap, analyzing the specific case of entrepreneurial talent transfer between regions. We find that Artificial Neural Networks, interpreted via Shapley values, achieve higher accuracy than traditional econometric techniques, accounting for the non-linearities of the data-generating process, and provide direct interpretation of each element. To validate and showcase our approach, we combine simulated results with the analysis of the case study of Lombardy, one of the most dynamic Italian Entrepreneurial Ecosystems. Our results offer an actionable framework for policymakers and practitioners, supporting the monitoring process of individual Entrepreneurial Ecosystem elements beyond composite indices.