Graph embeddings and firm diversification

Firms rarely diversify at random. Their next activities are shaped by capabilities, market proximity, institutional opportunities, and the history of what they already know how to do.

The new chapter on firm diversification uses public procurement data and graph embeddings to represent firms and activities in a way that machine learning models can use. Instead of describing a firm only through flat variables, graph embeddings place firms and activity sectors in a relational space where proximity can encode capability and opportunity.

This approach matters for economic complexity because it makes diversification more measurable and potentially more actionable. It can support better questions about which firms are likely to enter new markets, which activities are adjacent, and how public procurement data can reveal productive capabilities.