Fig 1.
The network of countries, capabilities and products.
A visualization of the tripartite network between countries, capabilities and products. In the middle of the network are the countries with node sizes proportional to their diversity. They are linked to the capabilities they have (the capability nodes are with uniform size). For visualization purposes, we link each capability k to the products for which Bpk ≥ 0.3, see Methods Section for further details. The size of the product nodes is proportional to their ubiquity and they are colored according to the one digit SITC classification. Products for which there is no k such that Bpk ≥ 0.3 are isolated, and thus are not shown in the Fig.
Table 1.
Complete list of capabilities found by the S3R-IBP model in 2010 through the SITC classification.
Fig 2.
Correlations in the capability space.
Nodes correspond to inferred capabilities. The coloring is according to the meta-capability grouping (subsection “The Meta-Capability Space”). Edge width and intensity are proportional to the correlation strength. For better visibility, we only depict edges with correlation higher than 0.4.
Table 2.
Meta-features activity pattern.
Fig 3.
World heat map according to the presence of features associated with meta feature M-F1. Darker shade indicates presence of more capabilities that are associated with M-F1. Countries with the lightest blue shade only have M-F0 capabilities, whereas there is no data for the countries in white shade. The map was generated in the software R (Available at https://cran.r-project.org/) using the package “rworldmap” [49] and data from authors’ own calculations.
Fig 4.
Countries in the capability space.
The distance between two countries is an approximation of the Euclidean distance between their capability vectors estimated via Multidimensional Scaling. Node size is proportional to country’s diversity, whereas node color is region-based. (a) Calculated with data for 1995. (b) Same as (a) only for 2010.
Fig 5.
A zoom in on the positioning of Hong Kong, Ireland, Singapore and South Korea in the capability space in 2010. We also show the capabilities and link the countries to the capabilities they have. For simplicity, we exclude the capabilities that are absent in all four countries.
Fig 6.
Dynamics for selected countries.
Top row: number of active features per year for Chile (blue), Egypt (red) and Indonesia (green). Bottom row: activation of features for the same countries.
Fig 7.
Subset of the transition model.
Each node corresponds to the set of active features at this state; edges are weighted by the corresponding number of inferred transition in Z.
Table 3.
Monitoring products at risk.
Table 4.
Incorporating new products in the export portfolio.
Fig 8.
Global properties generated by the model.
a Adjacency matrix for the empirical country-product matrix. b Adjacency matrix for the inferred country-product matrix, c Adjacency matrix for the inferred country-capability matrix. d Adjacency matrix for the inferred capability-product matrix. d Comparison of the fitted diversity cumulative distribution between the baseline, S-IBP and S3R-IBP. and the empirical country-product networks 2010. f Same as e, only for ubiquity. a-d Countries are ordered according to their diversity dc, while products according to their ubiquity up. Darker shade indicates higher value.
Table 5.
Quantitative evaluation of accuracy and interpretability.
Fig 9.
Q-Q plots for the distribution inferred by the models.
A Diversity Q-Q plot for S3R-IBP. B Diversity Q-Q plot for the S-IBP. C Diversity Q-Q plot for the baseline model. D Ubiquity Q-Q plot for S3R-IBP. E Ubiquity Q-Q plot for S-IBP. F Ubiquity Q-Q plot for the baseline model.