The persistence of firms’ knowledge base: a quantile approach to Italian data

Abstract

The paper investigates the patterns of persistence of innovation and of the properties of firms’ knowledge base (KB) across a sample of Italian firms in the period 1998–2006. The analysis draws upon a theoretical representation of knowledge as a collective good, stemming from the recombination of knowledge bits that are fragmented and dispersed across economic agents. On this basis, we derived properties of the KB like the coherence, the cognitive distance and the variety, and investigated their patterns of persistence over time. The empirical analysis is implemented by exploring the autocorrelation structure of such properties within a quantile regression framework. The results suggest that the properties of knowledge are featured by somewhat peculiar patterns as compared with knowledge stock, and that such evidence is also heterogeneous across firms in different quantiles.

Keywords:

• persistence
• innovation
• knowledge coherence
• variety
• cognitive distance
• quantile regression
• autocorrelation

JEL Classification:

• L20
• L10
• 032

Acknowledgements

A preliminary version of this paper has been presented at theWorkshop “Un decennio perduto? Come innescare la competitività del sistema produttivo italiano”, held in Trento on the 6th and 7th December 2011, and at the SEIK seminar series organized by the BRICK at the Collegio Carlo Alberto on the 30th May 2012. We are particularly grateful to Antonio Accetturo, Cristiano Antonelli, Marco Guerzoni, Fabio Montobbio and Giuseppe Scellato for their insightful comments. The authors also acknowledge the financial support of the European Union D.G. Research with the Grant number 266959 to the research project ‘Policy Incentives for the Creation of Knowledge: Methods and Evidence’ (PICKME), within the context Cooperation Program/Theme 8/Socio-economic Sciences and Humanities (SSH) and of the Collegio Carlo Alberto with the framework of the research project IPER. We wish to thank Second Rolfo for making the access to the Bureau Van Dijk AIDA dataset at the CERIS-CNR possible.

Notes

1. The limits of patent statistics as indicators of technological activities are well known. The main drawbacks can be summarized in their sector-specificity, the existence of non-patentable innovations and the fact that they are not the only protecting tool. Moreover the propensity to patent tends to vary over time as a function of the cost of patenting, and it is more likely to feature large firms (Pavitt 1985; Griliches 1990). Nevertheless, previous studies highlighted the usefulness of patents as measures of production of new knowledge. Such studies show that patents represent very reliable proxies for knowledge and innovation, as compared with analyses drawing upon surveys directly investigating the dynamics of process and product innovation (Acs, Anselin, and Varga 2002). Besides the debate about patents as an output rather than an input of innovation activities, empirical analyses showed that patents and R&D are dominated by a contemporaneous relationship, providing further support to the use of patents as a good proxy of technological activities (Hall, Griliches, and Hausman 1986).

2. The total number of firms in Table 2 is slightly higher than that in Table 1 as the industrial classification field contains some missing values.

3. Different depreciation rates have been implemented, which provided basically similar results.

4. See Strumsky, Lobo, and van der Leeuw (2012) for a compressive discussion on the use of patents technological classes to study technological change.

5. It must be stressed that to compensate for intrinsic volatility of patenting behavior, each patent application is made in last five years.

6. It must be noted that by measuring the degree of technological differentiation, the calculation of information entropy is affected by the number of technological classes observed, but not necessarily by the number of technological classes in the classification itself. Indeed, the introduction of new technological classes that are not observed does not affect the calculations in that they would be events with zero probability. Entropy rises or falls according to the number of technological classes that are actually observed in the patent sample. It reaches the maximum if all events are equiprobable, i.e. if all technological classes show the same relative frequency. If probabilities are unevenly distributed, one can have very low values of information entropy even if a very large number of technologies are observed.

7. The function used to measure coherence is completely different from the one used to measure informational entropy. The fact that in both cases the co-occurrence of technological classes enters the calculations does not mean that both functions must lead to the same result. The informational entropy function measures the variety of the set, corresponding to the number of distinguishable entities it contains. The coherence function was introduced by Teece et al. (1994) to measure the coherence of a firm based on its products. Nesta and Saviotti (2005, 2006) have subsequently adapted it to measure the coherence of the knowledge base of a firm. The coherence function measures the extent to which the distinguishable entities in the set (in our case the types of knowledge corresponding to different technological classes) are used together irrespective of the number of entities contained in the set. The two functions are in principle independent since they use the same type of data to calculate different properties of the same system. The mathematical independence of the two functions does not imply that the evolution of the corresponding properties is independent. Thus, if new technological classes are introduced into the knowledge base of a sector (an increase in the number of distinguishable entities of the set) there is no reason to expect the capacity of firms to combine the new types of knowledge to be created instantly. We expect that as new types of knowledge are introduced into the knowledge base of a sector, the firms will slowly learn to combine them thus leading to a temporary fall in coherence.

8. To make it clear, informational entropy is a diversity measure which allows to account for variety, i.e. the number of categories into which system elements are apportioned, and balance, i.e. the distribution of system elements across categories (Stirling 2007). In this sense, entropy does not say anything about the relationships between technological classes, but provides a measure of the diversity of technological co-occurrences, suggesting whether in a sector most of the observed co-occurrences focus on a specific couple or on the contrary whether the observed co-occurrences relate to a large number couples. In this framework, related and unrelated varieties provide a measure of the extent to which observed variety applies to couples of technologies that belong to the same macro domain or to different macro-domains. One would expect established technologies to be characterized by relatively low variety of co-occurrences, insofar as the recombination focuses on relatively small numbers of technological classes that have proved to be particularly fertile. On a different ground, the coherence index is based on a normalized measure of how much each observed technology is complementary to all other technologies in the analyzed patents. In this sense, it cannot be understood as a measure of diversity. The relatedness index indeed provides a measure of the degree to which two technologies are actually jointly used as compared with the expected joint utilization. The index allows to establish a relationship of complementarity between the technologies in the analyzed patents. Based on the relatedness measure (tau), the coherence index provides an aggregate description of the degree to which the observed technologies in a given sector are complementary to one another.

9. Cognitive distance is the inverse of similarity or the equivalent of dissimilarity. The measure of similarity has been introduced by biologists and ecologists to measure the similarity of biological species and to understand to what extent they could contribute to biodiversity. The same measure has been applied by Jaffe (1986) to the similarity of technologies. It is not the only possible measure of similarity but it is the most frequently used one. The rational for its use starts from the assumption that when two technologies, i and j, can be combined with a third technology k, they are similar. We call this measure cognitive distance both because the two terms are used as synonyms in the biological literature and, even more so, because cognitive distance is a concept used by Nooteboom (2000) which has a number of very interesting implications for firm behavior and performance. In particular, the cognitive distance between different firms is expected to affect the probability that they form technological alliances. Intuitively, the need for a firm to learn a completely new technology (discontinuity) will lead to the incorporation into the firm's knowledge base of new patent classes, which would make the knowledge base recognizably different from what it was at previous times. The dissimilarity of the knowledge base can be expected to keep rising with respect to the pre-discontinuity knowledge base until the technology lifecycle has achieved maturity, at which stage the knowledge base of the firm will have stabilized, thus leading to a fall in cognitive distance.

10. For Engelsman and van Raan (1994), this approach produces meaningful results particularly at a ‘macro’ level, i.e. for mapping the entire domain of technology. An alternative approach to calculate technological proximity can be found in Sorenson and Singh (2007).

11. More on quantile regressions can be found in Koenker and Hallok (2001).