patient experimentation, with the expectation of finding winning combinations of technologies, as where the technological relatedness lies may be unclear a priori
2
Granstrand et al., 1997.
3. The data
In order to analyse corporate diversification and to investigate the coherence issue through the multi-dimensional type of relatedness suggested above, we
considered both product and technological diversification straategies of the world’s largest firms. Several studies have extensively shown that large multi-divisional
firms are the largest single source of the new technological knowledge. Indeed, they perform most of the RD activities, employ most of the qualified research scientists
and engineers, perform and publish most of the corporate basic research, and maintain the closest links with academic research Pavitt, 1999. They also con-
tribute to the development of knowledge and products for their suppliers of production equipment, components and software Rosenberg, 1963; Patel and
Pavitt, 1994.
Corporate diversification has been examined using data on world’s largest firms sales and patenting activities in the United States for the period 1977 – 1995.
Specifically, we considered 248 US, European and Japanese worlds’ largest indus- trial companies as listed in Dunning and Pearce 1985 with a few additions of
companies apparently missed from the Fortune listings. The sample chosen is stratified in order to include firms from the whole sectoral and geographical
spectrum. The consolidated firms are allocated to their primary industry of output according to the product distribution of their sales, so that firms have been divided
into 18 industrial groups see Appendix A. In order to analyse business diversifica- tion, product distribution of sales has been allocated over 42 sectors 26 manufac-
turing and 16 services sectors, see Appendix B. Such data have been collected at Politecnico di Milano.
Technological diversification of firms relates to their patenting activity in the USA. The corporate patenting was divided into 56 technological fields see Ap-
pendix C, derived from the US patent class system. Patent statistics present a potentially very rich source of empirical evidence on
questions related to technology. The advantages and disadvantages of using US patents as an indicator of technological activity are well known and quite widely
discussed in the literature e.g. Schmookler, 1950, 1966; Basberg 1983, 1987; Pavitt, 1985, 1988; Jaffe, 1986; Archibugi, 1992. Though recognising some potential
limitations of the US patenting measure, mainly related to the fact that patents measure codified knowledge whereas a high proportion of firm-specific competence
2
For instance, firms from certain chemical sectors sometimes move into areas of competence associated with the manufacture of specialist chemical equipment a mechanical technology, to a greater
extent than into a more closely related science-based technologies within the chemical field Cantwell and Fai, 1999b.
is tacit or non-codified knowledge, studies like Patel and Pavitt 1997 have shown that other measures that embody tacit knowledge such as RD expenditure, or
judgement of technological peers give results that are very similar to those obtained using US patenting.
Sectoral and geographical characteristics of firms constituting our sample are reported in Table 1. The comparison with the relevant characteristics of the
Fortune 500 firms also reported in Table 1 allows to assess the balance of the sample.
4. The three dimensions of relatedness
