The Network Origins of Aggregate Fluctuations

GAICIL Insights

For decades macroeconomists waved away sectoral shocks with a statistical shrug.¹ With hundreds of sectors, idiosyncratic shocks should average out and leave aggregate output untouched. The Network Origins of Aggregate Fluctuations, by Daron Acemoglu, Vasco M. Carvalho, Asuman Ozdaglar and Alireza Tahbaz-Salehi, shows that the shrug fails in the presence of intersectoral input-output linkages. Microeconomic shocks, propagated through those linkages, can indeed drive aggregate fluctuations.

The central result is a characterization. As the economy becomes more disaggregated, the rate at which aggregate volatility decays is set by the structure of the network of input-output linkages. What matters is each sector's importance as a supplier to its immediate customers and as an indirect supplier to chains of downstream sectors. These higher-order interconnections capture cascades in which a productivity shock to one sector propagates to its customers and onward through the economy.

The key condition is asymmetry. Sizable aggregate fluctuations arise from micro shocks only when sectors play significantly asymmetric roles as direct or indirect suppliers. High variability in sector degrees, and high-degree sectors sharing common suppliers, slow the decay of aggregate volatility. The sparseness of the input-output matrix, by contrast, is unrelated to the nature of aggregate fluctuations. The analysis, the authors write, provides a fairly complete answer to the debate between Dupor and Horvath: Dupor's critique applies to economies with balanced structures, and in general the sectoral structure of the economy may have a defining impact on aggregate fluctuations.

Now the AI angle. Artificial intelligence is turning into the textbook asymmetric supplier. AI services, and the semiconductors and computing infrastructure behind them, feed a rapidly growing number of downstream sectors. In network terms, AI-producing sectors are acquiring high degree. The paper's framework says their shocks will not wash out. Productivity gains in AI will cascade through the economy, and so will disruptions. The diversification shrug does not apply to a sector everyone buys from.

That symmetry between upside and downside is where policy should start. The upside is large: a sector supplying many others transmits its improvements widely, lifting output in industries far from its own. This is the network mechanism by which a general-purpose technology becomes aggregate growth. The downside demands resilience planning: a technological, geopolitical or regulatory disruption at a central supplier propagates the same way.

Regulatory shocks deserve a special warning label. A rule that impairs a central supplier can produce economy-wide effects that a sector-by-sector impact assessment would never catch. A compliance cost imposed on an AI hub is, through the network, a cost imposed on everyone downstream. Impact assessments for AI rules should be network assessments.

The prescription favors competition and redundancy at central nodes. Where a few sectors or firms occupy central supplier positions, exposure to their shocks rises. Encouraging entry, interoperability and multiple sources of supply in AI inputs diversifies the network and lowers aggregate risk, while intensifying the competition that drives innovation. Here stability and dynamism point in the same direction.

The authors call for more systematic quantitative work and for extensions to dynamic multisector economies, and later research has answered, making the framework a foundation for measuring how shocks spread. For AI, the natural next step is to map how AI inputs enter the input-output structure and to track their centrality as adoption grows. Centrality could become a headline statistic, reported alongside output.

The paper is also a model of how theory settles arguments. By identifying the exact network property that decides whether shocks average out, it moves the debate from intuition to measurement. Policymakers can check input-output data and see whether the economy is growing more or less asymmetric.

The AI era is likely to make production networks more asymmetric, at least for a while, as a new general-purpose input spreads. This paper explains why that matters for both the business cycle and growth. The architecture of supply relationships shapes the destiny of the aggregate economy, and keeping central nodes competitive and diversified protects stability while preserving the gains from innovation.