The gravity of knowledge

By Paul Walker 14/06/2013

Thanks to an email from Eric Crampton I have been alerted to a paper, with the title given above, by Wolfgang Keller and Stephen Ross Yeaple on the transfer of knowledge between countries. More precisely the transfer of knowledge between countries but within multinational companies. The paper appears in the American Economic Review, 2013, 103(4): 1414–1444, and the abstract reads,

We analyze the international operations of multinational firms to measure the spatial barriers to transferring knowledge. We model firms that can transfer bits of knowledge to their foreign affiliates in either embodied (traded intermediates) or disembodied form (direct communication). The model shows how knowledge transfer costs can be inferred from multinationals’ operations. We use firm-level data on the trade and sales of US multinationals to confirm the model’s predictions. Disembodied knowledge transfer costs not only make the standard multinational firm model consistent with the fact that affiliate sales fall in distance but quantitatively accounts for much of the gravity in multinational activity.

If I’m getting this right the paper assumes that knowledge can move over geographic space in one of two forms: embodied or disembodied. When knowledge moves in an embodied form the costs of moving it can be measured as the cost of goods trade. For disembodied knowledge it is harder to see movements of it and harder to calculate the costs of moving this type of knowledge. The paper claims to shed new light on this issue by casting the question in terms of the operations of multinational firms.

Multinationals have an incentive to endow offshore affiliates with their knowledge as efficiently as possible—after all, knowledge transfer costs raise overall costs and therefore reduce competitiveness. The paper model focuses on the difficulty of communicating knowledge from one person to another versus the costs of moving knowledge in goods. Knowledge can often not fully be codified, and communicating knowledge is prone to errors; just ask any teacher! The authors explain,

In the model, multinational firms produce final goods from individual intermediate inputs that vary in the extent that their production requires non-codified knowledge. Inputs highly dependent on noncodified knowledge are called knowledge intensive. Because not all knowledge can be codified, offshore production calls for communication between home country CEOs and affiliate managers. We assume that communication is more costly the more knowledge-intensive inputs are, but these costs are invariant to physical distance. Alternatively, the multinational can transfer knowledge by shipping ready-to-go inputs embodying the knowledge. This entails no communication costs since the input is produced near the expert at home, however shipping incurs trade costs that rise in geographic distance. The reason why multinational sales in knowledge-intensive industries suffer most strongly from gravity is that here disembodied knowledge transfer costs are highest, and to avoid them means embodied knowledge transfer whose costs rise in distance.

Two predictions follow from the model,

First, the knowledge intensity of production affects the level of affiliate sales around the world. The competitiveness of affiliates, measured in terms of their sales, falls as trade costs rise, and the effect of trade costs is strongest for knowledge-intensive goods, precisely because it is here that the scope for offshoring is most limited by costly disembodied knowledge transfer. Second, the knowledge intensity of production affects the composition of knowledge transfers that the multinational will employ. The affiliate’s cost share of imports gives the relative importance of embodied knowledge transfer. It falls more slowly with distance in knowledge-intensive industries than in less knowledge-intensive industries. As trade costs increase, multinational affiliates substitute away from importing inputs, but their ability to do so is constrained by how high disembodied knowledge transfer costs are. Therefore, trade costs have the weakest influence on affiliate imports in relatively knowledge-intensive industries.

The authors have a data set, from United States’ Bureau of Economic Analysis, involving information information on the sales and intermediate goods trade of individual multinationals. Testing the model the authors find

[ … ] strong support for both predictions using variation in multinational activity across industries and countries. Consistent with the model, there is evidence that both the level of the affiliate’s sales and its imports are affected by the ease to which knowledge can be transferred across space. A quantitative analysis based on these micro estimates shows that both market size and geography are central determinants of aggregate
foreign direct investment (FDI). This is in contrast to the benchmark model which largely ignores the geography dimension.

Moreover, the model predicts that as trade costs change relative to communication costs, the nature of trade in terms of its knowledge intensity changes systematically. Specifically, an increase in trade costs makes disembodied knowledge transfer more attractive so that the average knowledge intensity of inputs that continue to be traded increases. Despite the large body of work on the factor service content of trade, this is one of the few results on the knowledge content of trade of which we are aware. We find strong supportive evidence for this prediction from the international trade of US multinational firms.

One interesting implication of this work is to do with vertical integration. An observation in industrial organisation is that firms that are part of a domestic production chain do not transfer as many goods within the chain as theories of vertical integration would suggest they should. This outcome could be because knowledge inputs are the key inputs determining the firm’s organisational structure. Given this, it follows that the spatial organisation of a firm will depend on the spatial barriers to disembodied knowledge transfer. As such barriers fall the vertical links between firms will become increasingly invisible as there is less embodied knowledge transfer and more disembodied transfer.