Complexity is a powerful multi-disciplinary idea that combines insights from both the natural and social sciences, especially economics and economic geography, to study the dynamics of complex systems of heterogeneous agents, the multiple interactions between them, and the aggregate behaviours that emerge from those interactions. Arguably, the most prominent empirical approach to complexity is that of the Atlas of Economic Complexity, developed by Ricardo Hausmann and Cesar Hidalgo (H&H) at Harvard’s Growth Lab. The Atlas itself is a remarkable – and strikingly beautiful – online resource, which uses network theory methods to provide a snapshot of a country’s productive structure, as well as a measure of the complexity and diversity of its production and those of individual products. This ‘Product Space’ approach is rooted in the idea that ‘countries become what they produce’. It is a view of economic development as the accumulation of productive capabilities of increasing sophistication. As countries develop, they produce more and more products and those products attain higher and higher levels of complexity, embodying more and more productive knowledge. This in turn provides more capabilities to produce yet more products and so creates a virtuous circle. Countries in which more detailed policy analysis has been carried out within the last two years include Sri Lanka, Uganda, Rwanda, Panama, Algeria, Mexico and Peru.
This paper contextualises H&H’s work within the recent resurgence of interest, in both academic and policy circles, in industrial policy. Comparing and contrasting the Product Space approach with other contemporary approaches to industrial policy, from authors including Justin Lin, Dani Rodrik, Joseph Stiglitz, Ha-Joon Chang and Mushtaq Khan, this paper sets out the strengths and weaknesses of conceptualising industrial development in terms of increasing economic complexity and diversity. In the technical section, the paper critiques the mathematical methods by which complexity is defined and rewrites the concept in terms of Markov chains on weighted graphs. This alternative formulation permits the application of Markov chain concepts, such as convergence to the stationary state, similarity and spectral clustering, in order to reinterpret complexity and other spectral data. These techniques are then related to economic aspects of complexity, such as production, technological change and capabilities. It is argued that the changing nature of production poses particular challenges to the ‘Product Space’ approach as well as other modes of industrial policy, and on this basis an ‘industrial ecosystems’ approach is outlined.
Written with Ha-Joon Chang (University of Cambridge) and Antonio Andreoni (SOAS)