A data-driven machine learning framework for forecasting institutional dynamics and structural breaks: Evidence from ASEAN economies

Abstract Institutional conditions are central to investment risk and development outcomes in emerging economies, yet they are typically assessed through retrospective indicators that offer limited insight into how institutional environments evolve or when they become unstable. This study examines whether institutional dynamics display systematic patterns of short-horizon predictability, and where the boundaries of such predictability lie. Drawing on a panel dataset covering ASEAN economies from 2009 to 2023, we construct a composite Business Environment Index integrating governance quality, regulatory conditions, and policy uncertainty. Using a strict rolling out-of-sample design, we evaluate the one-year-ahead predictability of institutional conditions with several machine-learning models and benchmark econometric specifications. Three core findings emerge. First, under relatively stable conditions, institutional dynamics exhibit pronounced path dependence, allowing historical information to generate meaningful short-term forecasts with low absolute errors. Second, forecast performance deteriorates sharply during periods of major disruption, including political crises and the COVID-19 pandemic, indicating clear limits to data-driven predictability when institutional trajectories are subject to structural breaks. Third, investment-related variables, particularly foreign direct investment flows, contain forward-looking information that precedes changes in conventional governance indicators, suggesting that market behavior embeds early signals of institutional stress. Together, these results demonstrate that institutional environments are neither fully predictable nor entirely opaque. Instead, they operate across distinct regimes in which predictability is conditional on the stability of underlying political and economic structures. By clarifying both the scope and the limits of institutional forecasting, this study contributes to emerging work on institutional risk monitoring and offers a data-driven perspective on how structural breaks in governance environments may be anticipated. JEL Classification : C53; C45; F21; O17; P48