A Simulation-Based Approach to R&D Measurement Error and Spillover Effect

Measurement of R&D activities is fuzzy because of different opinions of recorders as to what counts as R&D. This can lead to downward bias of the estimates due to the classical errors-in-variables (CEV) problem. To solve this problem, one can compare the panel data estimates with cross-sectional estimates and remove the bias. However, panel data are not usually available at the firm level. Moreover, there can be substantial externalities of innovation subsidies due to the spillover effect of R&D across firms. This is ignored in the estimated marginal effects based on firm-level data. To deal with both issues, we propose a novel approach based on simulations of the data with different measurement errors. We find that the firm-level marginal effect of innovation subsidies on productivity is similar to the industry-level spillover effect