Governments deploy research and development (R&D) incentives to promote economic growth and directly enhance innovation. The presence of R&D spillovers and adverse selection in the financing market for R&D justify these policies. At the federal level, the U.S. has direct subsidies for R&D, allows deductions for R&D expenditures, and gives a R&D tax credit. States also have their own incentives which frequently include R&D tax credits. These state R&D tax credits are often a large source of outlays. For example, in 2005 California invested over $900 million in R&D tax credits (Franchise Tax Board, 2008).

This study estimates two-way fixed effects models to identify the price elasticity of R&D and has two important differences from the extant literature. With regards to measurement, I gather new data from state session laws to model the representative firm's marginal cost of R&D as how the marginal dollar of R&D affects the firm's total tax burden (i.e., the proportion of R&D subsidized by the government at the margin). This marginal cost takes into account federal tax law, state tax law, and the interaction effects between state and federal tax law.

Additionally, to mitigate concerns over state level policy endogeneity, I take advantage of plausibly exogenous sources of variation in the state level tax burden of R&D: changes in U.S. federal tax laws. In two-way fixed effects models, time dummies absorb the variation that affects all units in each time period equally. However, certain changes in federal tax law have heterogeneous impacts across states as a function of preexisting state tax laws, which allows for identification of the policy variable when including time dummies in the regressions. In the context of R&D, these features of state tax laws include how states treat the deductibility of R&D expenditures, the basis for state taxable income, and state R&D credit piggybacking.

This paper's identification strategy of exploiting variation in the marginal cost of R&D driven by federal changes assumes the degree of treatment from federal laws is uncorrelated with state level conditions that also affect R&D. Under this identification assumption, estimates of the policy variable will be unbiased. With both state and federal variation, OLS applied to two-way fixed effects models gives inelastic estimates of the price elasticity (standard error) of around -0.55 (0.40). These estimates suggest if governments were to increase R&D subsidies by 1%, then R&D would increase by 0.55%. However, models exploiting the variation across states from only federal laws, which should be exogenous to state level conditions that would otherwise affect state level policies, indicate a more elastic price response with a representative estimate (standard error) of -3.62 (1.21). The results suggest serious bias towards finding R&D tax incentives are ineffective when relying on variation in laws implemented at a level concurrent with the unit of observation.