[Paper Review] Evaluating the principle of relatedness: Estimation, drivers and implications for policy

Yang Li, Frank Neffke|arXiv (Cornell University)|May 5, 2022

Regional Economics and Spatial Analysis4 citations

TL;DR

This paper evaluates the principle of relatedness in urban economic development by testing tens of thousands of model specifications to identify robust measures of inter-industry relatedness using co-occurrence data from US cities. It finds that relatedness based on industry density (Marshallian externalities) outperforms variety-based measures (Jacobs-type), and while relatedness predicts stagnation better than growth, it is most valuable for diagnosing developmental bottlenecks rather than identifying growth opportunities.

ABSTRACT

A growing body of research documents that the size and growth of an industry in a place depends on how much related activity is found there. This fact is commonly referred to as the "principle of relatedness". However, there is no consensus on why we observe the principle of relatedness, how best to determine which industries are related or how this empirical regularity can help inform local industrial policy. We perform a structured search over tens of thousands of specifications to identify robust -- in terms of out-of-sample predictions -- ways to determine how well industries fit the local economies of US cities. To do so, we use data that allow us to derive relatedness from observing which industries co-occur in the portfolios of establishments, firms, cities and countries. Different portfolios yield different relatedness matrices, each of which help predict the size and growth of local industries. However, our specification search not only identifies ways to improve the performance of such predictions, but also reveals new facts about the principle of relatedness and important trade-offs between predictive performance and interpretability of relatedness patterns. We use these insights to deepen our theoretical understanding of what underlies path-dependent development in cities and expand existing policy frameworks that rely on inter-industry relatedness analysis.

Motivation & Objective

• To resolve methodological ambiguity in measuring inter-industry relatedness for urban economic development research.
• To identify robust, out-of-sample predictive specifications of the principle of relatedness across diverse data sources and model configurations.
• To assess whether relatedness reflects shared capabilities or other locational factors, especially in non-tradable or resource-based sectors.
• To reframe the principle of relatedness as a diagnostic tool for identifying developmental bottlenecks rather than a growth predictor.
• To improve policy frameworks by integrating uncertainty and interpretability trade-offs into relatedness-based industrial policy recommendations.

Proposed method

• Conducts a structured, large-scale specification search across tens of thousands of model configurations to estimate relatedness effects on industry size and growth in US cities.
• Uses co-occurrence data from establishments, firms, cities, and countries to derive multiple relatedness matrices based on portfolio similarity.
• Employs out-of-sample prediction performance as the primary criterion to evaluate model robustness and select optimal relatedness measures.
• Applies OLS regression models to assess the predictive power of different relatedness specifications on industry growth and presence.
• Residualizes industry and city size data to isolate the effect of relatedness from size effects, focusing on relative performance.
• Compares relatedness measures derived from different portfolio types (e.g., city-level, country-level) to assess their predictive and interpretative value.

Experimental results

Research questions

• RQ1 Which specifications of inter-industry relatedness yield the most robust out-of-sample predictions for industry growth and presence in US cities?
• RQ2 What is the relative contribution of industry density (Marshallian externalities) versus related variety (Jacobs-type externalities) to the predictive power of the principle of relatedness?
• RQ3 Why does the principle of relatedness hold in sectors where capabilities are not a primary driver of location decisions, such as public services or resource-based industries?
• RQ4 To what extent does relatedness predict actual growth, or is it better at identifying industries unlikely to grow?
• RQ5 How can the principle of relatedness be reoriented from a growth-prediction tool to a diagnostic tool for identifying developmental bottlenecks in local economies?

Key findings

• Relatedness measures based on the mass of related industries (i.e., industry density) significantly outperform those based on the variety of related industries in predicting industry size and growth.
• The principle of relatedness performs better at identifying industries that are least likely to grow than those most likely to grow, suggesting it is more effective as a diagnostic tool than a predictive one.
• The principle of relatedness persists even in sectors where capabilities are not a primary locational determinant—such as public services, resource-based industries, and nontraded services—challenging the dominant capability-based explanation.
• The leading explanation for relatedness—related diversification minimizing new capability acquisition—cannot fully account for the pattern in non-capability-driven sectors, implying alternative mechanisms such as shared infrastructure or institutional factors may be at play.
• The study reveals a key trade-off between predictive performance and interpretability: highly predictive relatedness measures are often less interpretable, suggesting caution in policy use.
• The principle of relatedness is most useful not for identifying growth opportunities, but for detecting anomalies in a city’s industrial portfolio—potential developmental bottlenecks due to missing capabilities or structural constraints.