Bottom line:
Understand measurement before studying production networks, as it shapes what is observed.
Differences across datasets often reflect reporting rules, not real structure.
🎯 Benchmark for:
• Working with firm-level networks
• Calibrating macro models
• Creating synthetic data
There are several studies on specific countries – but what do we (really) know about nation-wide production networks?
What have we learned from these studies?
Why are there differences or similarities between them?
Could major differences only be data artefacts? – As it turns out, yes!
Some robust features we find:
Distributions of the num of customers & suppliers: heavy-tailed, but very different exponents.
Largest firms have a huge num of customers, but far fewer suppliers.
Excess clustering & reciprocity: firms form dense clusters in which they trade with each other.
....
We systematically study the effect of the reporting threshold and show that
👉 Reporting thresholds systematically distort network statistics, and we can often identify the direction of the bias.
❗️Calibrating economic models using incomplete datasets can lead to targeting biased moments.
By contrast, the tail exponent of the distribution of transaction values is not affected by the reporting threshold.
Distribution of firm-to-firm transaction values spans more than 10 orders of magnitude, even in mid-sized countries.
Sales, expenses, & the num of customers & suppliers are all positively correlated. We report precise and non-obvious elasticities.
Many large firms actually have very few partners.
For example, the average number of trading partners can vary by an order of magnitude across datasets – largely driven by reporting thresholds rather than real economic differences.
This is definitely not the time to delay or dial back climate policy. The urgency has never been greater.
We are approaching dangerous tipping points, almost 200 scientists warn. 🌊
news.exeter.ac.uk/faculty-of-e...
Andrea Bacilieri
Andrea Bacilieri
Many studies on national supply chains: It’s time to take stock!
1st to use rare admin data to establish a benchmark 🧵
🔗Paper tinyurl.com/3wf77bkw
🔥Code tinyurl.com/v9c63cub
@francoislafond.bsky.social A Borsos M Hoefer @pastudillo.bsky.social @inet-complexity.bsky.social @inetoxford.bsky.social
