We study government optimal subsidy policies for research programs in the face of servere infor- mation asymmetry-when firms have private information about the likelihood of project viability but the government cannot form a unique prior belief about this likelihood. The paper makes two con- tributions. First, we show that the way in which R&D is subsidized matters. Under both monopoly R&D (i.e., a single firm conducts R&D in isolation) and R&D competition, different types of subsi- dies (e.g., earmarked, unrestricted subsidies, and pure matching subsidies) have significantly different effects on firms' R&D investment incentives. Second, we show that a simple subsidy scheme works even when the government is unable to form a unique prior belief about the firm's private information on project viability. If the shadow cost of public funds is zero, under monopoly R&D, there exists a pure matching subsidy that induces the firm to follow the first-best R&D policy irrespective of its prior beliefs about the viability of the project, meaning it is a (belief-free) ex post equilibrium policy; under R&D competition, the first-best outcome can also be achieved through a simple combination of a matching subsidy and an unrestricted subsidy. If the shadow cost of public funds is positive, an ex post equilibrium in general does not exist either under monopoly or competition. We then consider two alternative policy decision criteria that are appropriate for belief-free games: rationalizability and max-min criteria. We argue that the max-min criteria is preferable in our context, and by way of doing so establish that the set of max-min subsidy policies under either monopoly or competitive R&D consists entirely of simple pure matching subsidies. We further establish that allowing firms to form an R&D consortium reduces the matching rate for the highest max-min subsidy, suggesting that cooperative R&D has the potential to economize on the shadow costs of public funding of subsidies. Keywords: Research and development, subsidies, dyanmic stochastic game, aymmetric information, belief-free game social cost of public funds, research consortia.
JEL classification: 038, D60, D82, H2, C73.