LATEST TABLE MISPLACES
+6.6%
ONE YEAR
+21.2%
FOREIGN ECONOMIES MISPLACES
+25%
PAPER FINDS
+88%
Imagine you have a detailed recipe book for every dish a restaurant makes — not just ingredients, but exactly how much of each dish's output becomes an ingredient in every other dish. That recipe book is an input-output table: the wiring diagram of a national economy showing how steel feeds into cars, cars feed into logistics, logistics feeds into retail. Now imagine the restaurant changes its menu slowly over years — new suppliers, new techniques, shifting tastes. How long before last year's recipe book becomes dangerously wrong? And if you never had a recipe book for a particular restaurant, could you borrow one from a similar restaurant down the street? This paper answers both questions with surprising precision. Liu and Lin study the "kernel" of domestic intermediate transactions — the structural core of who-sells-to-whom that persists even after you adjust for changing industry sizes. Among 25 economies with annual national tables, projecting forward with the latest table misplaces 6.6% of this kernel after one year and 21.2% after ten. That's the decay rate of real economic structure. By contrast, borrowing the kernel from the five most structurally similar foreign economies misplaces 24–25% regardless of horizon. Your own stale table beats your neighbor's fresh one for roughly fifteen years before the two error curves cross. The fifteen-year crossing point is the headline finding, but the more unsettling discovery is what fills the gaps. Multi-country input-output databases (the OECD's ICIO tables are the main target here) need a national table for every country-year. Many country-years have no survey-based national table. The paper finds that 88% of country-years without a national table cluster near a single cross-country template — compared to only 3.6% of country-years that do have a real table. The databases are quietly papering over missing data with what amounts to a shared structural guess. That shared template isn't stable either. It covered 29 economies in 2019 and ballooned to 67 by 2022. The expansion likely reflects both genuine convergence in lighter-industry economies and the mechanical fact that more missing tables means more imputation from the same template. The paper doesn't disentangle these two forces, which is a gap worth noting — but the diagnostic itself is sharp. The method is refreshingly transparent. The authors decompose each input-output table into a "kernel" (the structural part invariant to rescaling) and industry totals (which change with GDP composition). They then measure misplacement — the share of kernel entries that land in the wrong bin when projected versus observed. This is a clean, interpretable metric that avoids the usual problem of aggregate statistics hiding structural errors. What makes this practically important is that input-output tables underpin a huge amount of policy analysis: carbon footprint accounting, trade-in-value-added calculations, supply chain resilience assessments, fiscal multiplier estimates. If the structural core feeding those models is a fifteen-year-old national survey — or worse, a cross-country template — the downstream analyses inherit that staleness silently. The paper doesn't model this propagation, but it lays the groundwork by quantifying the input error. The honest limitation is scope. Twenty-five economies with annual tables is a constrained sample, likely skewed toward richer countries with better statistical infrastructure. Whether the fifteen-year crossing holds for rapidly industrializing economies or commodity exporters is an open question. And the paper measures structural persistence, not economic consequence — we know the kernel drifts, but not yet how much that drift costs in misallocated policy.
The analogy
Think of it like Google Maps for a city. A fresh satellite scan gives you accurate roads, but the city changes — new construction, closed streets. After a year, the map is 93% right. After ten, it's 79% right. And if you never scanned that city at all and just pasted in a map from a similar-sized city nearby? That borrowed map is about 75% right on day one and never improves. Your own stale scan stays better than the borrowed one for about fifteen years.
City map
Satellite scan decays as roads change; a borrowed map from a similar city is always ~75% right
IO kernel
National table decays as production structure shifts; a cross-country template is always ~75% right
Honest assessment
This is a sharp diagnostic paper that pins concrete numbers on a problem the IO community has talked around for years: most missing country-years in major databases are template clones, and even real national tables have a roughly fifteen-year useful life. The paper doesn't fix the problem — it names it, measures it, and gives the field a reproducible metric. Most useful for researchers who consume OECD ICIO tables downstream and need to understand how much structural noise they're inheriting. The sample bias toward richer economies and the absence of alternative imputation comparisons are real limits.
What Is CLAIMS?
67
/ 100
BedrockNews applies a structured analytical framework called CLAIMS to academic papers and technical research releases.
How CLAIMS worksClaim
4/5
Ladder
3/5
Architecture
3/5
Integrity
3/5
Milestone
3/5
Successor
4/5
What is this paper actually committing to?
Genuine first quantification: how fast does a national IO table's structural core decay, and how much of the OECD database is template-imputed? Not a methods advance, but a sharp diagnostic nobody had pinned down with numbers.
The field fight
Are multi-country input-output databases capturing real national economic structure, or are they propagating imputed templates?
Side A
Imputation methods are good enough for most downstream analysis; the databases are reliable
Held by: OECD ICIO methodology, WIOD, Exiobase builders
Side B
Missing national tables inject systematic template bias that silently degrades trade, carbon, and policy analysis
Held by: IO table critics, this paper
Milestone tracker
Unlocks: Evidence-based refresh schedules for national statistical offices and defensible uncertainty bounds for trade-in-value-added and carbon accounting
Current
21.2%
kernel misplacement at 10yr
this paper
Next
<5%
downstream policy error from stale tables
propagation threshold
Unknown — requires separate study linking IO staleness to carbon/trade model error
Verified Facts (6)
The latest table misplaces 6.6 percent of the kernel one year ahead and 21.2 percent ten years ahead
The five most similar economies misplace 24.1 to 25.0 percent at every horizon
88 percent of country-years without a national table lie close to one cross-country template, against 3.6 percent of country-years with a table
The template spreads from 29 economies in 2019 to 67 in 2022
An own table's error meets its distance to that year's template near fifteen years
25 economies with a national table every year were studied
Practical Implications (4)
Researchers using OECD ICIO tables for trade-in-value-added or carbon footprint analysis should check whether their country-years have underlying national tables or are template-imputed — 88% of missing entries are clones
National statistical offices can use the fifteen-year crossing point to justify survey refresh cycles: a table older than 15 years is worse than a cross-country guess
Any supply chain resilience or fiscal multiplier analysis built on IO tables should include sensitivity analysis for kernel staleness, especially for tables more than 10 years old
The 29-to-67 template expansion (2019–2022) suggests recent ICIO vintages are less structurally diverse than they appear
More latest news
Read all