Patents and R&D: Is There A Lag?

This paper extends earlier work on the RID to patents relationship (Pakes-Griliches 1980, and Hausman, Hall, and Griliches,1984) to a larger but shorter panel of firms. The focus of the paper is on solving a number of econometric problems associated with the discreteness of the dependent variable and the shortness of the panel in the time dimension. We compare weighted nonlinear least squares as wellas Poisson-type models as solutions to the former problem. In attempting to estimate a lag structure on R&D in the absence of a sufficient history of the variable, we take two approaches: first, we use the conditional version of the negative binomial model, and second, we estimate the R&D variable itself as a low order stochastic process and use this information to control for unobserved R&D. R&D itself turns out to befairly well approximated by a random walk. Neither approach yields strong evidence of a long lag. The available sample, though numerically large, turns out not to be particularily informative on this question. It does reconfirm, however, a significant effect of R&D on patenting (with most of it occuring in the first year) and the presence of rather wide and semi-permanent differences among firms in their patenting policies.