Regional entrepreneurship and innovation in Chile: a knowledge matching approach

Michelacci (2003) imposes this restriction which assures the existence of equilibrium research efforts. Intuitively, without some “saturation” of the knowledge stock, research effort has constant or increasing returns to scale in the probability of matching with an entrepreneur, and so such efforts would tend to increase. Eventually, the gains from additional effort would have to decrease as the probability of a match approaches one.

Isla de Pascua (Easter Island), Juan Fernández and Antartica.

Data are available at http://​ion.​inapi.​cl:​8080/​Patente/​ConsultaAvanzada​Patentes.​aspx.

The database included invention patents (67.3 % of all applications), utility models (14.6 %), industrial designs (17.8 %) and industrial drawings (0.1 %).

An additional complaint with respect to using patent counts is that they are noisy even as measures of intermediate outputs, greatly differing in their relevance and quality. This problem has stimulated the use of quality-weighting schemes to account for such heterogeneity, making use of data on patent citations (e.g., Aghion et al. 2005). We do not have enough information to establish the quality of each innovation, but we argue that patent applications published by the INAPI entail a non-trivial innovative effort. All applications in the INAPI dataset passed a first technical assessment conducted by INAPI’s experts and thus qualified for the external examination process. This first filter ensures that the applications meet minimum standards of: (1) novelty, (2) industrial applicability and iii) inventive level. Applications include at least a descriptive report (summary and a review of the state of the art), a description and justification of the innovation to be protected and the technical sketches. The detailed content in the application and the costs involved in its preparation (technical advisory, preparation, publication) ensure that the applicant and the government body share what Griliches (1990, p. 1669) succinctly refers to as a “non-negligible expectation to its ultimate utility and marketability.”

As pointed by one of the reviewers, many theoretical models of endogenous growth equate new firms to innovations themselves, so this variable could well be in the left-hand side of our estimation equation. Given the low-levels of R&D investment and the poor performance of the Chilean economy with respect to the production of knowledge and innovation (OECD 2012), it is more realistic to assume that only a small share of new firms in Chile are actually created to implement some new innovation. Therefore, the number of new firms would be a closer indicator of regional entrepreneurship (as usually has been accepted in the regional science literature) rather than of regional innovation levels.

http://​www.​sii.​cl/​estadisticas/​inicio_​actividades.​htm (version with date of extraction: 13/06/2011).

We thank an anonymous referee for suggesting the CASEN data source and the interpolation approach used to build the research effort variable.

http://​www.​sinim.​gov.​cl/​.

http://​www.​conicyt.​cl/​fondecyt/​sobre-fondecyt/​.

The database included information of the following specific instruments of the program: Regular Contest, International Cooperation, Research Initiation and Postdoctoral Studies.

http://​www.​conicyt.​cl/​fondef/​sobre-fondef/​que-es-fondef/​.

1 U.F. is around $US 45.

http://​www.​sii.​cl/​estadisticas/​empresas.​htm (version with date of extraction: 17/08/2011).

We thank an anonymous referee for suggesting this variable.

http://​www.​sii.​cl/​estadisticas/​inicio_​actividades.​htm (version with date of extraction: 25/05/2011).

http://​www.​sii.​cl/​estadisticas/​empresas.​htm (version with date of extraction: 17/08/2011).

During the exploratory analysis, we fitted other count data models, such as pooled and panel Poisson and negative binomial. Also, pooled zero-inflated Poisson (ZIP) and negative binomial (ZINB) models that are frequently used with patent data (Agrawal and Cockburn 2003; Andersson et al. 2009). Using these approaches, parameter estimates were consistent with expectations for specifications that included variables available for the period 2002–2008 (1–4), but were very unstable across the various model specifications and estimation methods when the other variables were included. Other estimations included “true” fixed-effects ZINB models (Allison and Waterman 2002), for which convergence of the log-likelihood function was not achieved.

As suggested by one of the reviewers, we also estimated the models at a different spatial scale. We worked at the level of labor market areas built by Berdegué et al. (2011) following the methodology by Killian and Tolbert (1993). Labor market areas (or functional regions) reflect areas of high intraregional economic and social interaction (Karlsson and Olson 2006), usually delimited by workers’ commuting flows. In Chile, they encompass 3.3 comunas in average. Results (available upon request) for specifications 1–4 with 430 observations indicate that the results for the KM and KPF1 variables remain qualitatively unchanged. The autoregressive parameter, however, becomes negative in some of the specifications. In addition, results for specifications 5–10, based on less than 250 observations, yielded unstable and mostly counterintuitive coefficients. All these results can be attributed to small-sample biases of the WW estimator, reported by Blundell et al. (2002).

Since \(\beta^{s} = \frac{{{\text{d}}\ln (m)}}{{{\text{d}}\ln (s)}} = 1 - \frac{{{\text{d}}\ln (p)}}{{{\text{d}}\ln (k/s)}}\).

In unreported results, we estimated models (3) to (10) using a spatial weight matrix with a bandwidth of 300 km. Available upon request.