The struggle for existence in the world market ecosystem

Abstract

The global trade system can be viewed as a dynamic ecosystem in which exporters struggle for resources: the markets in which they export. We can think that the aim of an exporter is to gain the entirety of a market share (say, car imports from the United States). This is similar to the objective of an organism in its attempt to monopolize a given subset of resources in an ecosystem. In this paper, we adopt a multilayer network approach to describe this struggle. We use longitudinal, multiplex data on trade relations, spanning several decades. We connect two countries with a directed link if the source country's appearance in a market correlates with the target country's disappearing, where a market is defined as a country-product combination in a given decade. Each market is a layer in the network. We show that, by analyzing the countries' network roles in each layer, we are able to classify them as out-competing, transitioning or displaced. This classification is a meaningful one: when testing the future export patterns of these countries, we show that out-competing countries have distinctly stronger growth rates than the other two classes.

Fig 1. An example of a displacement relation.

From left to right we observe a pattern in the yearly car import data from the US. In the first year, only Italy is present. In the second year, Japan appears in the market. In the third year, Italy disappears from the market. This pattern from the trade data is represented in the competition network as a directed edge from the displacer (Japan) to the displaced (Italy). The edge is labeled with its layer: the car market in the United States.

Fig 2. Timeline of Japan’s and Italy’s exports in the US car market in the 60s.

Fig 3. Out- and in-degree distributions.

(Left) Countries are sorted and ranked in the x-axis according to their out-degree, i.e. the number of times they out-compete another country—reported in the y-axis. (Middle) The same plot, replacing the out-degree measure with the in-degree one, i.e. the number of times the country was displaced from a niche. (Right) The relationship between out-degree (x-axis) and in-degree (y-axis). Each observation is a country.

Fig 4

(Left) All possible triangles in a directed graph. (Right) Frequency of different types of directed triangles in the multilayer network.

Fig 5. Distribution of number of displacements per one digit SITC product.

Fig 6. The energy landscape of the parameter space (projected over the omitted dimension).

From left to right: δ-κ (λ omitted); δ-λ (κ omitted); κ-λ (δ omitted).

Fig 7. The average role-feature scores per cluster in our example.

O = Out-competing, D = Displaced, T = Transitioning.

Fig 8. The distribution of growth rates for countries classified in the different clusters.

Box plots report the 10th, 25th, 50th, 75th, and 90th percentile. Outliers are reported with circles. From left to right: 1981-90 growth in SITC product 8 (Miscellaneous manufacturing); 1991-2000 growth in SITC product 1 (Beverages and tobacco); and 2011-13 growth in SITC product 3 (Mineral fuels, lubricants and related materials).

Fig 9. The aggregate coefficient values across all product across all decades (y axis) for increasing decade lag (x axis).

A decade lag equaling one means that we predict the decade after the data used to calculate the cluster (i.e. the main result of the paper). A decade lag equaling two means we predict two decades after the cluster data: if we had cluster data from 1960 we predict 1980 growth; if we had cluster data from 1980, we predict 2000 growth.

Fig 10. Robustness tests for k_max (left) and α (right).

For different choices of these parameters, we report the effect in the R² of the prediction. We focus on product 1 in the 1960-1970 decade.

Fig 11. The relationship between the R² export growth prediction using the role clusters (x-axis) and using the logarithm outdegree/indegree ratio (y-axis).

Each observation is a decade-product combination: the x-axis values are the R² values reported in Table 1, excluding product 3 and the 2000-2010 decade. The black line is the identity line: observations below the line are the ones for which the role clusters performed better than the log degree ratio.