Decomposing technological change at the twilight of the twentieth century: evidence and lessons from the world's largest innovating firms.

The following analysis is based on data extracted from the SPRU database using US Patent Office (USPTO) information. Our database reports accumulated patent counts for 463 of the world's largest manufacturing companies across 14 in dustries and 34 technology classes for the years 1981-85, 1985-90, and 1991-96. Patents were assigned to the primary class for which they were granted by the examiners. These were then allocated to one of 34 broader patent classes of the SPRU database (see Appendix 1).

Some patent classes were a simple process of aggregation i.e. where a USPTO class clearly fell into a single class within the broader SPRU classification scheme e.g. patents registered under US Patent Class 435 Chemistry: Molecular Biology and Microbiology were allocated to the SPRU category 7 Drugs and Bioengineering in their entirety. However, occasionally patents within a USPTO class were split and allocated to two or more broader SPRU categories e.g. some of the patents registered under US Patent Class 424 Drug, Bio-Affecting and Body Treating Compositions were allocated to SPRU class 3 Agricultural Chemicals, and others were assigned to class 7 Drugs and Bioengineering where appropriate.

The construction of the industry data involved a tremendous effort of consolidation of 4500 subsidiaries and divisions of firms: different assignee names, kept or bought by the 463 up to 1992, were identified using Whom Owns Whom of 1992 as a basis for allocation to their parent companies. The parent companies have then been allocated to one of SPRU's 14 industrial classes (see Appendix 2) according to their primary production output.

Although patents have become a hugely popular innovation indicator their proper use remains a non-trivial matter. The problems of this indicator are significant, but will not be discussed here. The concerned reader is directed to the methodologically oriented literature, now quite mature and extensive (Pavitt, 1985; Narin and Olivastro, 1988; Griliches, 1990; Smith 2005).

Following the multi-technology literature we know that large industrial firms are technologically active (i.e. they claim patentable knowledge at the frontier of given knowledge fields) even outside their core domains traditionally linked to the generation of their industrial output. Table 1 illustrates the correspondence between industrial sector and core and non-core technological fields within the SPRU patent database. The final column of the Table 1 shows the proportion of patents registered by firms in each industry in technological fields outside of those identified as core to each industry. For instance, whilst ICTs related technologies are core technical fields for the Computer and Electrical/Electronics industries, 25.1% and 39.3% of patents are taken out in other technologies such as drugs and bioengineering by firms in these two sectors respectively. Thus, we have a way to measure the extent and dynamics of the technological diversification behaviour.

Figure 1 presents the same data with an inter-temporal perspective. The Computer and Electrical/ Electronics sectors, appear to be registering fewer patents in technologies outside of their core fields, or equivalently, are focusing more on their core technological competencies, over time. In contrast, Photography and Photocopy demonstrate a sharp increase in the patents granted outside its core technical fields. We interpret this as a transition towards a richer ensemble of technological activities and hence a broadening of the knowledge base of this industry. Other industries displaying similar tendencies include: Motor Vehicles and Parts, Machinery, Pharmaceuticals, and so a more limited extent, Food, Drink and Tobacco.

Despite cross-industry variability in the level and rhythms of technological diversification, almost half of the patents generated by our population are generated outside each industry's core domain of technological expertise (Mendonca, 2003).

4. METHOD

The patent data is analysed using structural decomposition analysis (SDA). SDA is derived from constant market share analysis as used in empirical studies of trade (Tyszynski, 1951; Fagerberg and Sollie, 1987; Laursen 1999). Tyszynski (1951) looked at change in the export performance of a nation in terms of its market shares at the end of the period compared to that at the start. He broke this down into two elements. He calculated what the nation's market share of exports would have been at the end of the period if the nation's initial shares across the basket of commodities did not change over time (i.e. using Laspeyres weighted indices). The difference between the initial share and this hypothetical end share is the structural effect because it reflected the changes in a nation's share of trade that was attributable to structural changes in its trading environment. The residual, or remaining difference between the hypothetical end share and the actual end share (i.e. Paasche weighted index) he put down to a competitiveness effect because it reflected changes in a nation's share of trade that was attributable the nation's changing competitive strength.

[FIGURE 1 OMITTED]

Fagerberg and Sollie (1987) strengthen Tyszynski's basic analysis and demonstrate that by using initial weight (Laspeyres) indices throughout their methodology, the residual effect which Tyszynski attributed entirely to the competitiveness effect can actually be broken up in to two separate effects: the reported competitiveness effect, but also a commodity adaptation effect. In other words, this third effect allows for the possibility that a nation's export performance might improve overtime because it can alter the composition of its 'basket' of export commodities so as to adapt with any changes in the broader composition of commodities in world export markets. Laursen (1999) borrows this methodology and applies it not only to export markets but also sectors of technological opportunity. This paper in turn borrows from Laursen's (1999) application of the methodology to technological opportunity but brings it down to an industry-level analysis. The logic of the methodology is given below:

i = a technological field (1 ... 34)

j = an industry (1 ... 14)

t-1, t = subscripts for initial year and final year of the period under consideration

Let

M = industry j's share of all patents

a = industry j's share of all patents in technology i

b = technology i's share of all patents

M can be written as the inner product of the vector a and vector b: M = ab or,

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

The change in industry j's share of patents in an industry over time is:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Where the third term in the final line indicates the degree to which an industry has succeeded in adapting its own technological profile to the changes in the broader technological environment in which it operates. It is the technology adaptation effect.

Fagerberg and Sollie caution that a zero technology adaptation effect does not indicate that no adaptation occurred, but that the rate of the industry's adaptation is exactly the same as the rate at which the broader environment's technological profile is changing. Thus a positive adaptation effect suggests the industry is adapting well relative to the pace of change in the environment and a negative adaptation effect suggest it is not adapting well. However, following Laursen (1999), the reason for a positive value of the adaptation effect has two bases: the industry appears to be adapting well because it is entering areas of growing technological opportunity, or because it is leaving areas of stagnating opportunity. Laursen therefore breaks up the third term--technology adaptation effect, into two parts: the technological growth adaptation effect which is positive if the industry moves into technological areas providing more opportunities for growth and the technological stagnation adaptation effect which is positive if the industry moves out of areas of declining opportunities. Thus following on from above, the full equation for the structural decomposition model is given by:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Thus, the change in a firm's patent share consists of four elements:

(TS) the technology share effect isolates the extent to which an industry has gained or lost shares of total patents through its endogenous patent growth into new areas, assuming a fixed technological structure at the broader technological environment level across the period.

(ST) the structural technology effect isolates the extent to which an industry has gained or lost shares of total patents because the technological structure of the broader environment has shifted to more closely or less closely resemble the balance of the industry's own technological composition as it was at the start of the period.

(GA) the technology Growth Adaptation measures the extent to which an industry has gained shares of total patents through the movement into the 'right' or more influential technological fields (positive sign), or equivalently, out of the 'right' technological fields (represented with a negative sign).

(SA) the technology Stagnation Adaptation effect isolates the extent to which an industry has benefited from moving out of the 'wrong' or stagnating technological fields (positive sign), or equivalently, into stagnating technological fields (negative sign).