Dynamic count data models of technological innovation

Abstract

This paper examines the application of count data models to firm level panel data on technological innovations. The model we propose exhibits dynamic feedback and unobserved heterogeneity. We develop a fixed effects estimator that generalises the standard Poisson and negative binomial models allowing for dynamic feedback through both the firm's of knowledge and its product market power. By using the long pre-sample history of innovation information this entry stock estimator is shown to control for correlated fixed effects and is compared with an alternative nonlinear GMM estimator. We find evidence of history dependence in innovation activity although variables reflecting the company's economic environment are also found to play a major role. Technological innovation is an inherently dynamic and nonlinear process. Empirical models seeking to track its progress should share these features. Count data models, where the variable of interest is a non-negative integer, are commonly used to analyse innovation headcounts. However, they are not typically formulated to deal with the dynamic feedback that is suggested by theory. In this paper we examine alternative approaches to modelling innovation counts when there are important dynamics and unobservable heterogeneity. We model a count of the number of innovations commercialised by a firm in a year as a function of a firm's market power and its tangible and knowledge capital stock. There is clearly going to be some feedback mechanism between market power and innovation - a successful innovation is likely to lead to an increase in a firm's market share. In addition, any representative sample of companies is likely to display a wide range of innovative activity. The majority of companies make few innovations while a small group are involved in a high level of activity. This difference is unlikely to be solely attributable to observable differences across companies. Unobservable permanent heterogeneity is, therefore, an important feature of any empirical model of innovation activity.