An "incentive subsidy" policy for subsidizing private R & D is proposed that can be more efficient, from a social point of view, than subsidy policies in common use such as a "normal" subsidy policy (fixed amount granted at project start), and conditional loans (loan is repaid only if project is profitable).
The incentive subsidy compensates firms for any private loss and taxes away any gain in addition the firm receives a small fraction of the resulting invention' s social value. This mechanism comes close to being perfectly incentive compatible.
The firm chooses itself whether it wants to be covered under the incentive subsidy. Generally, the firm's choice coincides with three social aims: First, a project that the firm would conduct in any case should not be subsidized. Second, a project should not be subsidized if its social value is negative. Third, the subsidy should provide an incentive to maximize a project's social value.
Using a simulation over a range of hypothetical research projects it is shown that the efficiency of conditional loans and normal grants declines drastically as the government's information about project parameters becomes poorer, while the incentive subsidy performs consistently well.