[Next] [Previous] [Top] [Contents]

Modeling R&D Strategy as a Network Search Problem

1 The setup and its rationale


It is customary in formal models of technological change to treat R&D as a process of directed search. In some analyses, of which the work by Nelson and Winter [3] is seminal, there is an exogenous development path described by a logistic function reflecting a view that the early stages of a line of technological evolution is difficult but becomes easier as the technology becomes better understood and then more difficult again as the benefits of the core technology are exhausted. Rosenberg [4, 5] and Dosi equation [1] argue for a connectedness in the development of a technology based in Rosenberg's view on the experience of the innovators and in Dosi's view on innovators' experience and some coherence of the" technological paradigm" on which the stream of innovations are based.

A further set of issues turns on such technological complementarities as in airframe and aeroengine design. Nelson and Winter equation [3] have argued that improvements in (say) airframe design to allow airliners to fly higher and faster yield potential benefits that can only be realised once improvements in aeroengine design and materials are available actually to drive the airliner to the higher altitudes and at faster speeds.

One way of thinking about this connectedness within a technological paradigm or core technology is in terms of the diffuseness of the entry points to the technology and the range of developments which follow from each state of technological understanding as well as the complementary technological developments essential to realise a new innovation or its benefits. There is no obvious reason to believe that technological development paths are unique so that any representation of technological possibilities should allow us to express the diversity of development paths leading to a given technology or set of similar technological developments.

Additionally, some developers might give greater emphasis to in-house while others are willing to buy in new developments either by licence or embodied in new plant and equipment. Other characteristics of R&D strategies are important. We know from our work with major UK-based companies that R&D project teams scan a range of possibilities and try to keep more than one alive as they pursue a main line of development in case that line turns out to be a technological cul de sac. Moreover, some companies are likely to consider a wider range of possible developments than are others.

Based on these, and other, observations we would expect a model of technological change to have the following characteristics.

  1. New technological developments are dependent on the existing technologies in a very specific manner. The knowledge about the general level of technological sophistication is not enough, you also need to know the structure and relatedness of those technologies, i.e. the feasibility of technical innovation is context- and path-dependant.

  2. The future uses of technologies to develop other new technologies is largely unpredictable. Firms need to explore the possibilities in some way. For example, they can attempt to develop a large technological base in-house or opportunisticaly look for technologies that can be quickly developed based on existing technologies in the market.

  3. There is a difference between technologies that are developed in-house and those that are bought in or licensed, especially when there is a significant gap between the in-house capability and the externally acquired technology.

  4. There are at least two type of knowledge that are important to technological innovation: knowledge of the relationship between technologies and the know-how necessary to implement a technology. The point of pilot-studies is primarily to gain knowledge of the former sort.

In order to capture these issues, we will represent the underlying core technology or technological paradigm as a network over which R&D teams search. The network is acyclic on the grounds that experienced R&D teams will not typically discover exactly the same technological advance more than once. Each network node is given a value which is the realization of a random number in the unit interval and represents the productivity increase achieved when an agent acquires that node. Similarly, each link between nodes is given a random value in the [0,10) interval indicating the cost of traversing from a node to the parent node.

We set as parameters the number of nodes in the network, the number of leaf nodes (i.e. the number of nodes with links to, but none from, them) and the maximum span of (i.e. the maximum number of links from) each node which is not a leaf node. In the models reported here, the number of nodes was 1000 and the maximum span was 5. The number of leaf nodes in each model was different.

The networks are created by the following algorithm:

  1. Generate an ordered list of nodes N = [n1, n2,...,nn].

  2. Let NP = [n1, n2,...,np], where p=n-c and c is the number of leaf nodes

  3. Starting with n1, for each ni xce NP in turn

This algorithm gives us a network such as in Figure 1. The leaf nodes of the network, n16,...,n20 are the possible points of entry into the technology network for R&D teams of the enterprises. All leaf nodes are labelled with 0s to indicate that they are free to acquire but, once acquired, confer no productivity improvements or production cost reductions. they are simply entry points to the network for individual R&D teams. Nodes with children are acquired, and therefore confer the labelled value of productivity improvements, once all of their children have been acquired. So, for example, starting from node n16 in Figure 1, node n6 is then acquired automatically and node n1 can be visited from node n6. However, to acquire node n1 in the sense of gaining its implied productivity improvements would require the prior acquisition of node n3 which could not itself be acquired before the acquisition of nodes n7 and n8 (since n6 must already have been acquired). But, to acquire node n7, for example, requires the prior acquisition nodes of n12 and n13 and their children.

In this paper we distinguish between visiting and acquiring a node. When a node is visited, the R&D team discovers the value of the node and the links to and from that node. A node is acquired when it has been visited and all of its children have been acquired. We assume that R&D teams can visit any node which has a link from any node they have already visited. But in any time period these visits can be made only by traversing one link from a previously visited node. In the example given above, there is an easy path from the leaf node n16 to node n1 but to acquire n1 turns out to be difficult, expensive and time-consuming.


Modeling R&D Strategy as a Network Search Problem - 12 APR 96
[Next] [Previous] [Top] [Contents]

Generated with CERN WebMaker