Background

The integration of national economies into global trade has been one of the most important developments of the past century[1]. At the same time, trade disruptions due to wars[2], geopolitical frictions[3], climate change[4], cyber-attacks[5], strikes[6] and pandemics[7] have highlighted their broad and unexpected repercussions. Within the EU, the departure of Britain from the Union, Covid-19 and the war in Ukraine had far-reaching implications affecting many aspects of our lives. Measuring the regional consequences of trade-shocks and predicting future disruptions requires granular data that closely monitor trade-flows and uncover the interdependencies across countries. Until now, the EU lacks a detailed database that would help in this direction and to fill this data gap, we propose a framework that allows us to construct Multi-Regional Input-Output tables (MRIO) based on local statistics and transport flows between NUTS-2 areas.

Research progress 

The main objective of estimating trade flows is to determine the likelihood of trade between regions and sectors. In this study, our central idea is to infer the joint distribution between regions while constraining the edge distribution. Our aim is to optimize this process and resemble as much as possible the regional traffic freight patterns, which we use as a proxy for regional trade flows.

To achieve this, we start by collecting inter-country input-output (ICIO) tables and regional information, such as gross value added, gross fixed capital formation, and households' income. We then disaggregate the national marginal variables to the regional level to ensure that our aggregated multi-regional input-output tables (MRIOs) are consistent with the ICIOs. The next step is to estimate the input-output coefficients matrix based on a region’s industrial specialization. In this setting when one region employs more people in a sector than the national average, we assume that the region can either fully supply itself and/or export to other countries or regions; alternatively if its Location Quotient (LQ) is smaller than 1, the same region will need to import goods or services for this sector from other places. Next, we use regional freight transport data as a proxy for regional trade flows and apply the cross-entropy method to minimize the difference between our estimates and the proxy. The cross-entropy method is essentially a trial-and-error approach that involves generating random solutions, evaluating their performance against the proxy, and using this information to update the set of solutions we consider in the next round.

To validate the accuracy of our MRIO, we compare it with formerly estimated MRIOs for the same regions (through the EU REGIO database) as well as “ground truth” data originating from local surveys in Austria, Finland, and Scotland. Using a range of performance indicators[8], we find that our MRIO does well, outperforming the previous MRIO used by the EU almost a decade ago (EU REGIO).  To further test our method, we use our MRIO predictions to infer regional carbon emissions for EU regions through Monte Carlo simulations and compare our outcomes with the actual emissions observed over time. Our MRIO scores quite well in this test[9] and for the majority of regions and industries its accuracy is higher than the common RSEs of the MRIO literature. Across all these validation exercises it also becomes clear that additional micro survey data including freight and commodity information along with flight data would likely improve our results.

All in all, our framework provides a tool to capture the interaction between regions and sectors in European regions and can be used more broadly to address the gap in official statistics.

Fig.1 : European multi-regional input-output table 

Future directions

Given the multitude of trade disruptions our dataset can be expanded to cover more regions beyond the EU provided the necessary data are available. The rather obvious next step would be to focus on specific commodities or services that are deemed essential for the EU and beyond. With this approach and a granular regional setup we can derive more specific outcomes for commodities of interest.

This dataset can also be used as a starting point for a range of scenarios of regional or national supply shocks. Using the MRIO as a basis, researchers could simulate how each disruption would propagate through the network.  This research can reveal critical regions and sectors in the EU that are vulnerable to targeted or broad disruptions.

All these research directions combined (additional regions for the MRIO, specific commodity data and shock propagation) can help policymakers prepare for a range of scenarios and potentially limit their repercussions.

[1] Ortiz-Ospina, Esteban, Diana Beltekian, and Max Roser. "Trade and globalization." Our World in Data (2018).

[2] https://www.iisd.org/articles/policy-analysis/russia-ukraine-trade-implications 

[3] https://www.ecb.europa.eu/pub/pdf/scpwps/ecb.wp2503~ca71d98a53.en.pdf 

[4] https://www.wto.org/english/res_e/booksp_e/wtr22_e/wtr22_ch2_e.pdf 

[5] https://icg.citi.com/icghome/what-we-think/citigps/insights/the-cyber-problem 

[6] https://www.ft.com/content/b434a2ad-fe74-46bd-84e0-5067d3a2a6ec 

[7] https://www.oecd.org/coronavirus/policy-responses/international-trade-during-the-covid-19-pandemic-big-shifts-and-uncertainty-d1131663/

[8] Including mean absolute deviation (MAD), Isard-Romanoff similarity index and Pearson’s correlation

[9] With Relative Standard Errors (RSE) less than 10% for all regions and simulations. In most cases RSE is below 5%.