Supply Chain Disruptions: Evidence from the Great East Japan Earthquake

GAICIL Insights

A factory flooded in one prefecture, and an assembly line stalled hundreds of miles away.¹ Supply Chain Disruptions: Evidence from the Great East Japan Earthquake turns that kind of story into one of the clearest measurements ever made of how shocks travel through an economy. Exploiting the exogenous and regional nature of the 2011 earthquake, Vasco M. Carvalho, Makoto Nirei, Yukiko U. Saito and Alireza Tahbaz-Salehi quantify the role of input-output linkages as a mechanism for propagating and amplifying shocks.

The findings are precise. The disruption propagated upstream and downstream along supply chains, hitting the direct and indirect suppliers and customers of disaster-stricken firms. Firms far from the affected region felt the blow through their trading partners. A general equilibrium model of production networks then estimates the full macroeconomic impact including these propagation effects: a 0.47 percentage point decline in Japan's real GDP growth in the year after the disaster.

The value of that number is everything it contains. An estimate counting only directly affected firms would miss the losses that traveled through supply chains. The network approach captures them. As the conclusion notes, a growing literature argues that input linkages, precisely because of their vital role in production, can serve as a channel for propagating and amplifying shocks. Using data on supply chain links among Japanese firms, this paper supplies the large-scale evidence that the channel is real and quantitatively significant.

Now turn the lens to artificial intelligence, whose own supply chain is long and narrow: semiconductors, cloud computing, foundation models, the applications built on top. A disruption at a key node, from a natural disaster, a cyberattack or a policy shock, would ripple upstream and downstream exactly as this paper documents. Mapping the network of AI production is a prerequisite for knowing how fragile it is. Few governments have that map today.

The same wires carry good news. Linkages that transmit negative shocks transmit positive ones. A productivity gain at a firm that supplies many others flows downstream to customers and their customers in turn. AI deployed at central nodes of production networks may therefore have amplified effects on aggregate growth. The network view implies that where adoption happens matters as much as how much of it happens. An earthquake and an innovation obey the same physics.

The methodology carries its own lesson. Firm-level supply chain data made it possible to trace the shock. Such data are rare and valuable for policy. Machine learning can help build and update supply chain maps from transaction records, invoices and administrative sources, giving governments and firms a clearer view of interdependencies. Better maps support faster responses, from targeted help to rapid identification of alternative suppliers.

Care is needed with the policy conclusions. Supply chain vulnerability is often used to justify reshoring and protectionism. The paper's evidence concerns propagation through existing linkages and does not by itself show that shorter or domestic chains would be more resilient. Specialized links exist because they raise productivity. Policies that keep the gains from specialization while improving the capacity to switch suppliers after shocks, through information, flexible contracting and open markets, address vulnerability without sacrificing efficiency. Resilience can be built from substitutability as well as from proximity.

The estimates also line up with official planning. The upper range is consistent with an earlier government assessment of the maximum economic loss from a major earthquake, which included damage from production stoppage, transportation disruption and propagation outside the disaster area. That consistency builds confidence in the network approach as a planning tool.

The broader message is that an economy cannot be understood one firm at a time. Shocks, good and bad, travel through relationships. As AI becomes woven into those relationships, its effects will travel too. Policymakers fluent in production networks will be better placed both to protect the economy from disruption and to amplify the gains from innovation across every link of the chain. The question to ask of any AI strategy: which nodes, if they fail or flourish, move everything else?

Introduction

Exploiting the exogenous and regional nature of the Great East Japan Earthquake of 2011, this article provides a quantification of the role of input-output linkages as a mechanism for the propagation and amplification of shocks. We document that the disruption caused by the disaster propagated upstream and downstream along supply chains, affecting the direct and indirect suppliers and customers of disaster-stricken firms. Using a general equilibrium model of production networks, we then obtain an estimate for the overall macroeconomic impact of the disaster by taking these propagation effects into account. We find that the earthquake and its aftermaths resulted in a 0.47 percentage point decline in Japan’s real GDP growth in the year following the disaster.

Takeaway

At the backbone of any modern economy is an entangled web of specialized production units, ensuring the flow of goods and services from suppliers to other producers and final consumers. Precisely due to their vital role in the production process, a growing literature has argued that these input linkages can serve as a channel for the propagation and amplification of shocks throughout the economy.¹

Using a general equilibrium model of production networks, we then obtain an estimate for the overall macroeconomic impact of the disaster by taking these propagation effects into account. At the micro level, our findings suggest that, when faced with a supply-chain disruption, individual firms are unable to find suitable alternatives in order to completely insulate themselves from the shock (at least in the short run). This is consistent with an emerging literature (e.g., Bernard, Moxnes, and Saito (2019b), Eaton, Kortum, and Kramarz (2018), and Barrot and Sauvagnat (2016)) that emphasizes the role of search frictions and relation-specific investments along supply chains.² However, it should be noted that our results are indicative of relatively small firm-level effects on average—corresponding to roughly a 2 to 3 percentage point decline in firm-level annual sales growth in our baseline specification—specially when compared to average firm-level sales growth volatility (which, as per Davis, Haltiwanger, Jarmin, and Miranda (2007), is arguably one order of magnitude larger). At the macro level, our results point to the structure of production network linkages as a key driver of aggregate fluctuations. In particular, they indicate that, even if average firm-level effects are not necessarily large, the potential propagation of shocks over the economy’s production network can propagate to a significant fraction of firms, thus resulting in movements in the aggregates.

We view our paper as a step towards a systematic empirical investigation of the role of inputoutput linkages as a mechanism for propagation and amplification of shocks. Several important issues, however, remain open to future research. First, while binary information on the presence of firm-level linkages enabled us to obtain estimates for the extent to which shocks propagate, using more detailed information on the value of firm-to-firm transactions would pave the way for a more complete analysis of the macroeconomic implications of production networks. Administrative firmlevel data from countries with value-added tax (such as the dataset constructed by Dhyne et al. (2015) and Magerman et al. (2016) for Belgium) would serve as the ideal dataset for such a study.³^,⁴ Second, even though the input-output network in our model is assumed to be exogenous, in reality, firms decide on the set of suppliers and customers that they transact with. It is reasonable to expect that the extent to which firms can form new linkages has first-order implications for the nature and intensity of shock propagation (in particular, in horizons longer than what we considered in the paper). While recent work, such as Oberfield (2018) and Acemoglu and Azar (2020), have focused on how firm-level decisions shape productivity and the organization of production, developing a comprehensive framework for endogenous formation of networks would be crucial for the theoretical and empirical investigation of the role of input-output linkages in the propagation and amplification of shocks.⁵^,⁶

Last but not least, while our theoretical framework treats all supplier-customer relationships as homogeneous—in the sense that any specific supplier is as easy to substitute away from as any other—real world supply chains are likely to exhibit important heterogeneity across inputoutput linkages. Arguably, firm-to-firm linkages are better approximated by the coexistence of easily-substitutable suppliers alongside suppliers producing customized inputs that are not easily substitutable. This, in turn, suggests the existence of key bottleneck firms with no short-run, satisfactory alternatives, whose outright exit may result in the destruction of relation-specific surplus between firms. On the theoretical side, developing a framework to capture the implications of movements in the extensive margin for the relationship-specific surplus that is at the heart of supply chains may require a departure from the competitive analysis that is prevalent in most of the literature. On the empirical side, documenting the relationship between patterns of firm failures and the ensuing rewiring process of the economy’s production network would be a natural next step in understanding the role of firm-level production networks in propagating shocks.

Article

Notes

Vasco M. Carvalho et al., Supply Chain Disruptions: Evidence from the Great East Japan Earthquake, 136 Q.J. Econ. 1255 (2021), https://doi.org/10.1093/qje/qjaa044. Takeaway source: Conclusions, reproduced from a free author or repository copy. ↩
Barrot, Jean-Noël and Julien Sauvagnat (2016), “Input specificity and the propagation of idiosyncratic shocks in production networks.” Quarterly Journal of Economics, 131(3), 1543–1592.. ↩
Dhyne, Emmanuel, Glenn Magerman, and Stela Rubı́nová (2015), “The Belgian production network 2002–2012.” National Bank of Belgium Working Paper 288. di Giovanni, Julian, Andrei A. Levchenko, and Isabelle Méjean (2014), “Firms, destinations, and aggregate fluctuations.” Econometrica, 82(4), 1303–1340.. ↩
Magerman, Glenn, Karolien De Bruyne, Emmanuel Dhyne, and Jan Van Hove (2016), “Heterogeneous firms and the micro origins of aggregate fluctuations.” National Bank of Belgium Working Paper 312.. ↩
Oberfield, Ezra (2018), “A theory of input–output architecture.” Econometrica, 86(2), 559–589.. ↩
Acemoglu, Daron and Pablo D. Azar (2020), “Endogenous production networks.” Econometrica, 88(1), 33–82.. ↩

Cite as: Vasco M. Carvalho et al., Supply Chain Disruptions: Evidence from the Great East Japan Earthquake, 136 Q.J. Econ. 1255 (2021), https://doi.org/10.1093/qje/qjaa044

Header image: Fiber Optic cable inside a box at McKinney, TX.jpg by Ranch9613, CC0, via Wikimedia Commons.