Relatedness in the Era of Machine Learning

TL;DR

This paper introduces a new method called Continuous Projection Space (CPS) for measuring relatedness between countries' industries, significantly improving prediction accuracy over traditional co-location approaches by leveraging multi-product correlations and network embedding techniques.

Contribution

The paper proposes CPS, a novel network embedding method for better relatedness estimation, and demonstrates its superior performance over existing co-location-based methods in predicting industry relatedness.

Findings

CPS outperforms co-location approaches in relatedness prediction.

Multi-product correlations improve forecast quality.

Traditional co-location methods perform worse than trivial auto-correlation strategies.

Abstract

Relatedness is a quantification of how much two human activities are similar in terms of the inputs and contexts needed for their development. Under the idea that it is easier to move between related activities than towards unrelated ones, empirical approaches to quantify relatedness are currently used as predictive tools to inform policies and development strategies in governments, international organizations, and firms. Here we focus on countries' industries and we show that the standard, widespread approach of estimating Relatedness through the co-location of activities (e.g. Product Space) generates a measure of relatedness that performs worse than trivial auto-correlation prediction strategies. We argue that this is a consequence of the poor signal-to-noise ratio present in international trade data. In this paper we show two main findings. First, we find that a shift from two-products correlations (network-density based) to many-products correlations (decision trees) can dramatically improve the quality of forecasts with a corresponding reduction of the risk of wrong policy choices. Then, we propose a new methodology to empirically estimate Relatedness that we call Continuous Projection Space (CPS). CPS, which can be seen as a general network embedding technique, vastly outperforms all the co-location, network-based approaches, while retaining a similar interpretability in terms of pairwise distances.