Abstract: It is of great significance for the formulation and implementation of subsidy policies to explore how the government can reasonably pick subsidy targets to achieve innovative growth of enterprises. Based on the large-scale innovation survey data set and considering endogeneity, this paper explores the mechanism of how government subsidies influence enterprise innovation from two aspects: “picking winners” and “building winners”. The empirical study finds that government subsidies have a “picking winner” strategy to select enterprises with high organizational learning ability. Empirical learning, cognitive learning and inter-organizational learning are the important selection criteria for government subsidies. The internal mechanism of “picking winners” is that enterprises with high organizational learning ability tend to adopt external technology acquisition mode to obtain stable innovation output in the short term, and it is easier to achieve the goal of government subsidies. On the basis of controlling selection bias, government subsidies have a significant “building winners” effect, which is the most, the less and the least pronouced respectively for universal subsidies, discretionary subsidies and selective subsidies. Further analysis by type of enterprises finds that empirical learning is not a selection criterion for the government to grant subsidies to state-owned enterprises, large enterprises and non-start-up enterprises, for which the “building winner” effect of government subsidies is more significant.

Key words: government R&D subsidy policy, picking winners, building winners, innovation performance