In our paper, "AI-Generated Production Networks" (2024), joint with with @trfetzer.com Bennet Feld @prashantgarg.bsky.social , we used AI to build a granular input/output network across 5,000+ traded goods. This public dataset can be used to study the knock-on impacts of trade shocks.

Ostensibly the goal of this move is to support domestic solar production, a highly complex production process which requires key components targeted in this action. Our input/output network can be used to identify all the production linkages for this, and 5,000+ other products:

🚨 BREAKING: Biden to announce major tariff hikes on critical production components (e.g. tungsten, polysilicon, solar wafers) - @financialtimes.com

Our research provides a lens to study the ripple effects of this US-China trade escalation. A brief 🧵 (1/6)

Pooling over all products using product-dyads in a DiD setting, there is clear evidence of "onshoring" i.e. increased import demand for upstream/critical products linked to shock-exposed final goods.

Finally, by identifying all the input/output relationships we can empirically test for vertical substition patterns i.e. "onshoring". We take a large trade shock to a small open economy and follow demand for both the focal product and it's basket of production inputs.

Another neat feature of our granular production network is the ability to identify emerging critical inputs, such as Chips, Lasers, Lithium - v. helpful to inform industrial strategy!

Across broad BEC product groups, "Capital Goods" and "Industrial Supplies" have become ever more critical parts of the production network.

We also leverage the AIPNET network structure to measure shifts in import composition towards more critical/central/upstream inputs. We see lots of divergence e.g. 🇨🇳China: Building domestic capabilities 🇺🇸US: Reducing upstream input dependence.

Some products are incredibly complex, like "Wind powered generators" which require huge amounts of complex parts and materials:

The AI-built Production Network (AIPNET) maps 5,000+ products, connecting them based on production processes. Each node is a product, connections show input/output links. E.g. "Full-fat milk".

🧵 New paper uses AI to map global production networks & study recent shifts in global trade: "AI-Generated Production Networks" by @trfetzer.com @econopete.bsky.social @prashantgarg.bsky.social and Bennet Feld.