Nearly 30 years ago, Paul Romer published a paper exploring the economic value of knowledge. In that paper, he argued that, unlike the classical factors of production (capital and labor), knowledge was a “non-rival good.” This meant that it could be shared infinitely, and thus, it was the only thing that could grow in per-capita terms.
Romer’s work was recently recognized with the Nobel Prize, even though it was just the beginning of a longer story. Knowledge could be infinitely shared, but did that mean it could go everywhere? Soon after Romer’s seminal paper, Adam Jaffe, Manuel Trajtenberg and Rebecca Henderson published a paper on the geographic diffusion of knowledge. Using a statistical technique called matching, they identified a “twin” for each patent (that is, a patent filed at the same time and making similar technological claims).
Then, they compared the citations received by each patent and its twin. Compared to their twins, patents received almost four more citations from other patents originating in the same city than those originating elsewhere. Romer was right in that knowledge could be infinitely shared, but also, knowledge had difficulties travelling far.
What made knowledge sticky? Following the steps of Romer and Jaffe, scholars mapped the co-authorship networks of inventors. This showed that it was an inventor’s professional network, not other aspects of geography (such as the institutional environment or shared culture) that explained the limited diffusion of knowledge. Despite patent fillings and publications, an inventors’ knowledge of their field was limited by the horizon of their own collaboration network. In a couple of decades, we had understood why knowledge was at the center of economic value, but also, why it was a honey that everyone wanted but few had.
When Romer published his seminal paper on economic growth I was only 10. Sixteen years later, I was doing my PhD at the University of Notre Dame. Unlike Romer, I had troves of data. I had data on mobile phone calls tracing social networks and human mobility. I had data on international trade, summarizing countries’ patterns of production with exquisite detail. This last data set was the flour we needed to create empirical measures of knowledge, extending Romer’s ideas to the world of big data.
The first measure of knowledge we published is now known as a measure of relatedness. It measures the knowledge an economy has regarding a specific activity. Here an activity is a broad concept. It could be an industry (shirt manufacturing), a product (a shirt), a technology (weaving machinery) or even an area of research (non-woven textiles). Relatedness measures the “potential” of an economy to develop an activity that is not yet present in it. Relatedness honors an important property of knowledge, the fact that it is not easily transferrable among activities. Being an expert at music doesn’t make you good at sports. Similarly, an economy that is good at exporting electronics may be inexperienced at mining.
Measuring relatedness is quite simple. First, you need to build a network connecting similar products. In our case, we connected products that tended to be exported together: shirts and blouses, apples and pears, buses and cars. Then, you focus on a product and use this network to count the fraction of “sister products” already exported by each country. If that fraction is large, you predict the country is more likely to start exporting that product. And that’s exactly what the data show. Economies are more likely to enter an activity when they are in presence of related activities. This is true for countries and products, regions and industries, cities and patents and even universities and research areas. This principle of relatedness, is as robust as an economic principle gets.
A few years later we published a second metric measuring the total knowledge in a country, region, or city. This measure focused on the intensity of knowledge—the fact that knowledge cannot be simply added, since it has overlaps and comes in discrete chunks. The basic idea was that the knowledge of a place was expressed in the activities present in it, and the knowledge of an activity was expressed in the places where that activity was present. This allowed us to define knowledge in a completely circular manner using either recursions or a mathematical technique related to principal component analysis. The good news was that this made no assumptions about which places or activities were most knowledge intense. We called this metric the Economic Complexity Index.
But did economic complexity vindicate Romer’s vision? The answer was a resounding yes. Countries that were more knowledge-intense were richer and less unequal, and when they had an excess of knowledge per unit of GDP per capita, they grew faster. The magic metric predicted the rise of East Asia, the crisis of Greece and the stagnation of Latin America. Yet, these findings still told us little about how knowledge got into new places. This is where research is going next.
When I present my work to scholars, entrepreneurs, ministers and public servants, they usually ask me the same question: “What’s the list?!” What they mean is that they want to know the list of the activities that are most related to their location—a list of where to focus their industrial development efforts. But I never liked “the list.” So recently, with Aamena Alshamsi and Flávio Pinheiro, we wrote a paper in which we explored millions of lists, instead of focusing on just one. The math showed that following a list in decreasing order of relatedness was actually suboptimal.
This was because the list contained products that were highly related, but were also dead-ends (that is, products that were not connected to other products). Dead-ends can rank high on “the list,” but sometimes, it is better to focus on products that are harder to develop but that open new paths. Moreover, the math showed that there was a narrow window of opportunity when it was optimal for countries to deviate from the most related activities. Being too ambitious too early led to failed development projects. Being too conservative during the optimal window wasted an opportunity.
But could we ever accelerate the flow of knowledge? Better data and methods are allowing us to put the flow of knowledge under the microscope. We can observe how knowledge moves as workers switch jobs or become unemployed. We can see how changes in communication and transportation technologies affect knowledge diffusion: from the introduction of the printing press in early modern Europe, to the speeding up of trains in China. We can study the role of migration on knowledge flows. We can even use patents to explore the relatedness and complexity of innovative activities.
What will the study of knowledge bring us next? Will we get to a point at which we will measure Gross Domestic Knowledge as accurately as we measure Gross Domestic Product? Will we learn how to engineer knowledge diffusion? Will knowledge continue to concentrate in cities? Or will it finally break the shackles of society and spread to every corner of the world? The only thing we know for sure is that the study of knowledge is an exciting journey. The lowest hanging fruit may have already been picked, but the tree is still filled with fruits and flavors. Let’s climb it and explore.