Link to paper

The full paper is available here.

You can also find the paper on PapersWithCode here.

Abstract

• Problem of inferring technological structure of production processes unsolved
• Empirical literature focuses on outputs instead of transformative processes
• Method developed to quantify technological sophistication
• Method measures degree of synergistic interaction between inputs
• Synergistic technologies more sophisticated
• Synergy scores predict export-based measures of economic complexity

Paper Content

Introduction

• Transformative process of turning inputs into outputs is present in various systems
• Different disciplines analyze it through distinct frameworks and tools
• Technological sophistication is a building block in the study of economic complexity
• Quantifying degree and structure of technological sophistication is important to understand production systems
• Developed a method to quantify technological sophistication of production processes
• Method empirically infers nature of input-input and input-output relationships
• Method facilitates quantification of technological sophistication and estimation of interaction networks
• Method is of general purpose and can be adapted to other contexts

Knowledge gap

• Technological change can be described as a recombinant process
• Evidence supports this process
• Technological sophistication is related to this process
• Empirical evidence is domain-specific
• Generating reliable estimates requires big data
• Production models assume input-input and input-output interactions
• Limitations of this approach have been discussed
• Supply-chain surveys have been conducted to determine dependence on inputs
• Systems engineering has developed the concept of Design Structure Matrices
• Data-driven approach needed to infer interaction structures
• Economic complexity literature has emerged
• Economic Fitness Index and Economic Complexity Index are gold standard
• These frameworks provide proxy for technological sophistication
• Production processes remain obscure
• Quantification of technological sophistication is a problem for multiple disciplines
• Need for generalizable quantitative methods to address production process

Framework and data

• Estimate mutual information between pairs of inputs in a given industry
• Decompose contribution to output into different modes of information sharing
• Focus on synergistic information, i.e. information that can only be obtained from interaction between inputs
• Measure degree of synergy between inputs and estimate a synergy score
• Higher synergy score means more novel information during production process
• Synergy score quantifies technological sophistication

The synergy score

• Data Y captures changes in output of industry
• Data X1 and X2 capture changes in inputs 1 and 2
• Total mutual information measures amount of uncertainty in Y reduced by knowing X1 and X2
• Partial information decomposition (PID) decomposes mutual information into synergistic type and other types
• Synergistic information can only be obtained from interaction between X1 and X2
• Synergy score is used to compute all unique pairs of inputs in data

Data

• Dataset used to capture technological sophistication is Eora26, which contains input-output tables for 26 industries across 148 countries from 1995-2020
• Data is transformed into log-fluctuations (growth rates)
• Method tested using OECD IO tables, covering 66 countries across 45 unique industries
• Synergy scores tested to see if they contribute to prediction of economic complexity indices
• Economic complexity indices constructed using independent dataset on export flows

Descriptive features

• Synergy networks capture the average technological structure of a group of countries
• Synergy networks show that the same industry can have different interaction structures in different countries
• Quantifying and understanding the internal structure of production processes is important for measuring technological sophistication
• Diversity of synergies is important for measuring technological sophistication
• Highly synergistic interactions are less abundant than low-synergy ones

Results

Predicting economic complexity

• Investigated relationship between synergy scores and complexity indices
• Found clear positive correlation between synergy score and three indices
• Synergistic interactions may predict proxies of economic sophistication
• Method focuses on input-input and input-output interactions
• Results show synergy score contributes to prediction of EFI
• Coefficient of log synergy remains positive and statistically significant
• Results are robust when using GENEPY index and ECI as dependent variable

Synergistic interaction networks

• Construct a synergistic interaction network for an industry using synergy scores of every pair of inputs
• Each node in the network represents an industry providing an input
• Edges are weighted according to pairwise synergy score
• 26 nodes in each network
• Extract backbone of each network using disparity filter
• Estimate measures of centrality, clustering, and mixing probabilities using binary version of backbone networks

Discussion and conclusions

• Quantifying technological sophistication in production processes is a difficult problem
• This paper introduces a method to measure input-input and input-output interactions
• Estimate amount of synergistic mutual information between inputs of production process
• Construct synergistic interaction networks to reveal features of industrial sophistication
• Results provide a major step towards understanding microfoundations of economic complexity

Methods

• Partial information decomposition requires a large number of observations per variable
• Data-clustering procedure based on technological similarity of countries in a given industry
• Workflow consists of two major steps: clustering and estimation
• Clustering involves grouping countries with similar production patterns in a target industry
• Estimation uses PID to estimate synergy scores for each pair of inputs in the target industry

Synergy score

• Information theory has been used to quantify interdependencies between variables in complex systems
• Shannon entropy measures the variability of a system in a given state-space
• Mutual information measures the dependence between two variables
• Synergy score quantifies the interdependencies between an industry’s inputs
• Shannon entropy is highest when the variable explores all possible states with equal probability
• Mutual information is 0 if variables are independent
• Mutual information between two Gaussian random variables can be written as an equation
• Synergy score decomposes mutual information into Synergistic, Redundant and Unique information
• Synergy score is a distance measure and can be encoded in bits
• Bias correction procedure is used to remove numerical artefacts

Data and prepossessing

• Input-output tables are obtained from the Eora26 dataset, which is the largest in terms of country coverage.
• Eora26 is used in the study of global value chains and material footprint.
• International trade is measured using the BACI dataset.
• Economic complexity index is calculated using the BACI dataset.
• Three auxiliary development indicators are used to improve clustering procedure.
• Input and output log-fluctuations are collated across years and countries for data-augmentation purposes.

Clustering similar technologies

• Clustered countries with similar technologies to increase data size
• Used marginal product to measure technological similarity
• Log-transformed marginal product to account for output differences
• Used k-means-constrained clustering algorithm to define four country clusters