Cooperative Innovation: In Quest of Effective Partners. Evidence from Italian Firms

Abstract

In recent years, rapid technological change, shorter product life cycles and globalization have deeply transformed the current competitive environment. These changes are inducing firms to face stronger competitive pressures which push them to develop new products, improve production processes or implement new technologies. Thus, firms need to continually acquire new knowledge and innovate. At the same time, entrepreneurs have become aware that technological innovation is less and less dependent on an isolated effort of an individual firm. For small- and medium-sized enterprises (SMEs), R&D cooperation with sources of external knowledge is becoming increasingly essential for fostering innovation activities. Using firm-level data from the Community Innovation Survey for the years 2006–2008 (CIS 2008) and applying a Heckman probit model with sample selection, we analyze the determinants of cooperative innovation for the different types of partners (competitors, customers, suppliers, universities and government laboratories). Results show that internal and external R&D acquisitions, public financial support, as well as belonging to a scientific sector or to a business group are significant determinants of choice in collaborations, although with different magnitude across various types of collaborations.

Keywords:

• cooperation
• innovation
• R&D partnerships
• Community Innovation Survey (CIS)
• discrete choice
• sample selection bias

Notes

1. A basic difference in the ways innovative activities are structured and organized may be related to a fundamental distinction between Schumpeter Mark I and Schumpeter Mark II industries (see, e.g. Malerba & Orsenigo, 1996).

2. Innovative firms are specifically defined as those that carried out technological innovation during the chosen period.

3. This includes the following industries: manufacturers of coke and refined petroleum, chemical and pharmaceutical products, computer, electronic and optical products, machinery and equipment, motor vehicles, trailers and other transport equipment, telecommunications, computer programming, consultancy and related activities, information service activities, architectural and engineering activities, technical testing and analysis, scientific research and development. NACE classification codes are reported in Table 3.

4. As explained in the model section, these variables could also be relevant for innovation, but since they have censored observations they can only be used in the cooperative innovation equation.

5. In the section describing the model framework we explain that in the econometric Heckman model used for the analysis the selection equation should have at least one explanatory variable which doesn’t appear as regressor in the outcome equation. This guarantees the model identification.

6. A successful example is the ‘Regional Program for Industrial Research, Innovation and Technological Transfer’ launched in 2003 by the Emilia Romania region, which put into effect Regional Law no. 7/2002, art. 4 (see: Bollettino Ufficiale della Regione no. 64 of 14 May 2002 and Delibera della Giunta Regionale no. 2038 of 20 October 2003).

The program aims at sustaining firms’ industrial research and economic development in the region. Bronzini and Iachini (2012) prove the effectiveness of such a program.

7. If we denote with N the total number of firms in the sample, with M the number of the innovative firms and with N₁ the number of the firms that established a collaboration relationship, then the likelihood function can be written as follows.

8. This test, whose asymptotic distribution is a chi-squared variable with one degree of freedom, compares the likelihood of the full bivariate model with the sum of the likelihoods for the univariate probit models.

9. In this paper we apply these two methods because both of them have some drawbacks and there is not a dominant technique in the literature. On the one hand, the ML method, based on the strong assumption of normality in the error terms, may raise some problems of convergence. On the other hand, in the Heckman two-step method the heteroschedastic nature of the error term in the outcome equation (see the following note) should be duly taken into account when the standard errors of the estimated coefficients are computed.

10. As Heckman showed, sample selection (or incidental truncation) causes a specification error in the outcome equation which is given by the omission of a variable, namely the inverse Mills ratio. In fact it can be proved that.

where , called the inverse Mills ratio, is given by.

and is the normal density.

Accordingly, the cooperation equation reads as follows.

where ν_2i is a non systematic, heteroschedastic error term, that is.

and = [.