Expropriation risks and innovation: Estimating the relationship between corruption and innovation inputs and outputs using machine learning

We develop a growth-theoretic framework to analyze how expropriation risk, exemplified by corruption, affects innovation incentives at the country level. Our model, building on Romer (1990), predicts that higher expropriation risk reduces R&D expenditure, diminishes the share of human capital engaged in R&D, lowers patenting and scientific publication rates, and slows technical progress and economic growth. We test these predictions using a novel dataset spanning nearly two decades and apply a machine learning-based instrumental variable approach – IV-LASSO – to address endogeneity in corruption. Our empirical results provide robust evidence that greater corruption (i.e., higher expropriation risk) significantly hampers innovation inputs (lower R&D spending and research personnel) and innovation outputs (fewer patent applications, scientific publications, and a lower Economic Complexity Index). These findings underscore the detrimental impact of corruption on innovation and highlight the importance of strong institutions and anti-corruption policies to foster innovation-led economic development. JEL Classification: O30, O43