Innovation Propensity and Unobserved Firm-Heterogeneity: Estimating a Varying-Coefficients Binary Choice Model Using the Generalized Maximum Entropy Method


Marc Tiri, Limburg University Center, Diepenbeek

Even though a wide variety of theories on firm behavior and performance explicitly recognize the existence of considerable inter-firm heterogeneity, many innovation studies and econometric specifications developed herein, do not account for firm-specific behavior. In particular, when working with a single cross-section of firms, parametric specifications are used that generate coefficients that are identical to all agents. The innovation propensity of firms is generally modeled through a binary logit or probit specification and estimated by ML. This renders the analysis particularly sensitive to a variety of model inadequacies, specification errors and omitted-variable bias. Furthermore, to the extent that heterogeneity is present, neglecting it may lead to badly biased estimators and wrong inferences.
In this essay we demonstrate the valueo of information-theoretic models to control for unobserved heterogeneity. We analyze empirically the innovation propensity of a cross-section of firms from the CIS-3. Specifically, we develop a varying-coefficients binary choice Generalized Maximum Entropy (GME-VC) specification, which allows for coefficients to vary randomly across firms in order to capture inter-firm differences in the innovation propensity.

Date: 22 March-00 0000


UNU-MERIT