Machine Learning to Assess Relatedness: The Advantage of Using Firm-Level Data

Figure 2

Comparison using different metrics between random forest built on firms and built on countries when we predict future exports of both firms (a) and countries (b). The orange bars refer to a random forest that has been trained on firm-level data, the blue bar refers to a random forest trained on country-level data, and the green bar refers to a product space built on firms (a) or country (b) data. The best assessment of relatedness at the country and at firm level is given by the relative random forest. However, the relatedness computed at the firm level performs better than the product space also at the country level.

(a)

(b)