Export Trade Clusters

This post, as with the prior ones on trade clusters, aims to help visualize patterns of trade in the OECD from 50 years of partner trade statistics. The data is rich, meaning we should be able to develop rich intuition by exploring it visually.

These slides follow the method laid out in Jong-Eun Lee, “Two Maps for the World’s Trade Integration,” Applied Economics Letters, 11:4 (2004). All computations were performed in R.

Visualizing Dyadic Trade Flows

I gave a talk at the Fletcher School today on my work on dyadic trade flows (slides).

In a nutshell, the talk argues that cartograms and dendrograms can give students and practitioners a better understanding of the patterns of trade among partner contries, both for teaching and for research. We have thousands of observations of dyadic relationships in panel datasets. Most often these datasets are presented as aggregates: total annual world trade, top exporters in world trade, top exporters, top exporters in an industry sector, top exporters to a political union (such as the EU), top exporters within a geographic area, etc. What these statistics ignore is the information in the dyadic trade flows: who trades with whom?

What I offer is a way to crunch down the total number of country dyads into manageable graphics that can appear on a single slide. We can look directly at the dyadic patterns of trade using hierarchic clustering (dendrograms). We can compare partner trade flows across countries and time periods using cartograms. The techniques are not new; what is new is the presentation of rich international trade datasets in relatively complete format that can be digested by inspection, rather than with complex and poorly understood statistical techniques. Complete annual sets of cartograms and dendrograms give scholars the power to explore the distribution of dyadic trade and discover hypotheses that are worth testing more carefully, either with quantitative or qualitative methods.

One of the reasons trade courses have focused so much on models, theorems, and policy of international trade is that it is hard to describe trade patterns in any meaningful and comparable terms. My slides suggest how to do exactly that: present changes to global trade patterns in a succinct, visual format that enables rich comparisons across time and space.

Unilateral trade clusters using raw import flows

This set of dendrograms, again, is based on raw partner import flows from OECD. The dendrograms show complete linkages (all countries in a cluster exceed the threshold value for mutual trade flows), but the dyad is measured by the greater of the two trade flows.

This gallery shows an annual series of dendrograms using that dataset back to 1993.

Bilateral import clusters using raw trade flows

As promised, here are a new round of dendrograms using OECD trade data as a reciprocal distance measure among countries reporting. In trade, relationships matter, and these dendrograms show which relationships matter the most. Clusters are drawn by complete linkages, using the lesser of the two pairwise trade flows (greater notional distance).

The important thing about these dendrograms, relative to the ones posted the last few days, is that they take the raw trade flows themselves–not normalized for population, or total imports, or GDP–as the unit of analysis. This is actually a much more useful picture of trade than the normalized flows, because is suggests which relationships ought to draw the most water in trade politics.

The cluster algorithm isn’t a perfect way to capture the data; a few outliers can skew the presentation of the data somewhat. But it is the only good way I have seen to present cross sections of country-dyad data at a glance. It’s a very useful tool for presentation of descriptive statistics on international trade.

Unilateral import clusters in international trade

As with yesterday, these graphics depict complete clusterings in international trade, treating the partner country’s share of total home country imports as a raw distance measure. The greater the share, the closer the two countries are. For visual clarity, I have used logarithmic scales; so the scale at left doesn’t have any concrete meaning.

The clusters in this dendrogram indicate complete linkages, meaning that all of the country dyads in each cluster share a unilateral import concentration greater than the threshold value for the cluster. At 100% concentration, no country has a partner providing 100% of imports; so all the countries are separate at the bottom of the scale. At 0% concentration, countries all have at least some trade with one another; so one giant supercluster exists at the top of the scale.

Bilateral import clusters in international trade: insights

Major insights from the dendrograms in the previous post:

1. Among OECD countries, New Zealand and Australia cleave closest to the Pacific alliance: NAFTA, Japan and Korea.
2. Scandinavian countries have undergone a durable shift in trade patterns, away from France and the Iberian peninsula, and toward the British Isles.
3. Greece, Italy, and Turkey constitute a durable trade cluster over many years. In the early 1990s this group traded heavily with the UK and Ireland; now more so with central Europe and Switzerland.
4. Dutch trade patterns oscillate more than other countries; suggesting durable, close ties to some member(s) of both the Scandinavian and Mediterranean clusters.

Bilateral import clusters in international trade

One big problem in the visualization of bilateral statistics is that maps don’t work very well.

Dendrograms give a very attractive and intuitive solution to the problem by dispensing with geographic information altogether. The dendrogram shows how clusters of countries form along a continuum of closeness. For international trade, I’ve presented below a cluster analysis of the OECD by share of total imports sourced from the partner country. Here, the linkages are bilateral, meaning that if countries A and B source 10% and 20% of their total imports from one another (respectively), the A-B partnership would join the same cluster at a threshold of 10% of imports.

At the top of the screen (0% of imports), all countries belong to a single supercluster. At the bottom of the screen (100%), no country sources 100% of its imports from any trade partner. Moving up the screen from the bottom, countries join the same cluster when the lesser of their mutual import shares falls above the threshold value. Technical notes below.

Tomorrow, I will post dendrograms for unilateral linkages (essentially the same as above, but using the greater of the two numbers). The following day, I will post dendrograms using the raw import flows, rather than the share of total imports. Following that, expect images depicting GDP weighted import flows. Next week, I will post the same series of images for imports.