Measuring the impact of R&D on Productivity from a Econometric Time Series Perspective

Abstract

In this paper we argue that the standard sequential reduction approach to modelling dynamic relationships may be sub-optimal when long lag lengths are required and especially when the intermediate lags may be less important. A flexible model search approach is adopted using the insights of Bayesian Model probabilities, and new information criteria based on forecasting performance. This approach is facilitated by exploiting Genetic Algorithms. Using data on U.K. and U.S. agriculture the bivariate time series relationship between R&D expenditure and productivity is analysed. Long lags are found in the relationship between R&D expenditures and productivity in the U.K. and in the U.S. which remain undiscovered when using the orthodox approach. This finding is of particular importance in the debate on the optimal level of public R&D funding.