Economic Resilience

Economic resilience is the capacity of an economy to absorb a shock without a durable loss of output, employment or supply: trade linkages, inventories, strategic reserves, bank balance sheets and the machinery that propagates a local disruption into an aggregate price. The framing under test is that resilience can be designed into an economy.

The epistemic problem comes first, because it governs everything else. Resilience is a property observable only in a shock. Between shocks, a resilient economy and a merely lucky one are observationally identical. That is not a caveat; it is the structural reason this subject has an unusually poor evidence base relative to its policy prominence. Almost every resilience intervention in force today has been tested against zero, one or at most two events, and the events were not drawn from the distribution the intervention was designed against. Bank capital rules calibrated to a credit shock met an interest-rate-and-deposit-speed shock; gas storage mandates designed against a pipeline cut-off met a mild winter; petroleum reserves designed against an embargo were used against a price spike.

The finding that most cuts against the framing is that the single most widely adopted resilience policy of the last decade is, on the best available evidence, resilience-reducing. A review of the global supply-chain literature finds that chains attenuated past shocks rather than amplifying them, that diversification of foreign suppliers lowers volatility, and that reshoring raises aggregate volatility by reducing source diversification. An OECD general-equilibrium model finds localisation cuts global real GDP by more than 5% and produces larger price swings under shock. And the measured outcome of the actual policy is a documented import-price increase with indirect exposure intensifying.

This brief owns the failure behaviour of economic systems — trade linkages, inventories, reserves, bank balance sheets and price propagation. The failure behaviour of built systems belongs to Infrastructure Resilience; catastrophes severe enough that no functioning outside remains belong to Civilization Resilience Planning; and supply-chain automation as a technology belongs to Autonomous Supply Chains, where this brief owns supply-chain structure as an exposure. The 2023 bank failures appear here as a resilience test rather than as a capital-markets question, which belongs to Future Capital Markets.

Frontier The headline finding of the empirical supply-chain literature runs against the policy response it produced, and the review says so in its opening framing. Baldwin and Freeman note that pandemic disruptions created the impression that shortages would have been less severe if global supply chains were shorter and more domestic, or more diversified — and then ask whether that is right. Their answer, from the recovery record of past shocks: global supply chains have helped attenuate shocks rather than amplify them; diversification of foreign suppliers lowers volatility; and reshoring increases aggregate volatility by reducing source diversification. They also establish that direct exposure measures badly understate dependence: US auto-sector reliance on Chinese inputs rises from roughly 5% to nearly 20% once indirect supply-chain dependencies are counted. China's manufacturing share tripled over 2004–2018 to over 28%, and every major economy increased its reliance on Chinese inputs between 2005 and 2015. Full decoupling would reduce national GDPs by 3–70% depending on the country. This is a well-supported review of a heterogeneous literature rather than a single identified estimate, and it is flagged as such.

Frontier The modelling result points the same way from a different method. Using the OECD's METRO model, Kowalski, van Tongeren and Arriola find that a shift to localised production would decrease global real GDP by more than 5% relative to a post-pandemic baseline — and, the important part, that under supply shocks the localised regime produces larger price swings and greater variability of incomes, because domestic markets must absorb more of the adjustment. Their conclusion is that localisation delivers neither greater efficiency nor greater security of supply. A computable general-equilibrium result is only as strong as its elasticities, and that is the honest caveat. Frontier But two independent lines — a literature review and a structural model — converge on the same counterintuitive claim, and the convergence is what raises the weight.

Established What the policy actually achieved is reallocation, not decoupling, and it has been measured cleanly. Alfaro and Chor find China's share of US imports fell from a peak of 21.6% in 2017 to 16.5% in 2022 — about five percentage points. Vietnam gained roughly 2pp, Mexico rose from about 13% to 14%, and Taiwan, Korea and India gained modestly. Established But the gaining countries are the ones most tied to China. Vietnam's imports from China rose from 9% in 1994 to about 40% by 2022, with China gaining 5.5pp during 2017–2022 alone. Mexico's imports from China rose from 1% in 1994 to 20% by 2022, and China gained the most import share in Mexico over that same window. Chinese direct investment in Mexican manufacturing grew roughly fivefold, from $31.6 million in 2017 to $151.5 million in 2022, concentrated in computer equipment and motor vehicle parts. Meanwhile China's import share rose 2.7pp in Europe over 2017–2022 while the US share rose only 0.4pp.

Frontier And it was not free. Products where China's import share fell show unit-value increases of 9.8% for Vietnam imports per 5 percentage points of China share lost, 3.2% for Mexico and 2.3% for high-wage Asian suppliers. The authors' own conclusion is that US indirect supply-chain links to China remain intact and along some dimensions have been intensifying. The measured result of the largest supply-chain resilience policy of the decade is a documented price increase and an undocumented change in actual exposure. The trade shares are established; the price attribution is a model-assisted estimate.

Frontier The behavioural response was real, small, and possibly temporary — and the one firm-level measurement declines to call it structural. Doan and Zhang, on Japanese firm-level data, find that importers significantly and persistently increased inventories relative to domestic-only purchasers from Q2 2020 through Q2 2021, across materials, work-in-process and finished goods, with the largest response among firms with high ex-ante import reliance. Their own caveat is the finding: it is premature to conclude that manufacturing firms have completely shifted from just-in-time to just-in-case production, because the data run only to Q2 2021. Established This matters because “just-in-time is dead” became a stylised fact in 2021–22 on the basis of surveys of intentions. The one firm-level measurement available found a real but bounded effect and refused the structural claim.

Established The semiconductor shortage is the canonical fragility case, and what it actually measured is a propagation result rather than a shortage. Cleveland Fed commentary records transportation-equipment manufacturing capacity utilisation at 68.5% in May 2021, 6.7 percentage points below its 2019 average; the PCE price index for new motor vehicles rising from −0.5% in April 2020 to +3.6% in May 2021, the highest since the mid-1990s; and the CPI for used cars and trucks up almost 30% over the twelve months to May 2021, with used cars accounting for about one-third of the overall monthly CPI increase in May 2021. Established Their forecast that the effect on new car prices would subside within six to nine months was wrong, and it is cited here as a documented miss rather than as an estimate. Frontier The useful lesson is the propagation: a single input constraint in one sector produced roughly a third of a month's headline consumer inflation in the largest economy in the world, from a component that is a trivial share of vehicle cost. That micro-to-macro mechanism is explicitly named by the field's own review article as one of its principal unknowns.

Frontier Strategic reserves are the one intervention with a real test, and the estimate of its effect diverges by a factor of three along the line of the estimator's interest. The 2022 release was the largest strategic reserve action on record: 180 million barrels announced by the United States in March 2022 plus 60 million from IEA partners, with the reserve falling by more than 20% between November 2021 and July 2022. A Treasury analysis, as reported by a non-partisan data organisation, puts the effect at 13 to 31 cents per gallon over March–July 2022, against an actual gasoline price rise of about 68 cents in the same window. Frontier The releasing administration's own fact sheet claims the release reduced gasoline prices by up to about 40 cents per gallon, citing 240 million barrels of combined drawdown. That is the maximally interested party, and “up to about” is doing a great deal of work. Frontier The honest statement: a 240-million-barrel coordinated release produced an estimated effect of 13–40 cents per gallon in a market where prices rose 68 cents anyway. The reserve moved the price in the intended direction; it did not prevent the shock. And because there is no counterfactual, the range is an artefact of model choice, with the wider figure coming from the party that authorised the release.

Established The 2023 US bank failures are the only recent live test of post-2008 resilience rules, and the supervisor's own review is unusually direct. Its four takeaways: management failed to manage risk; supervisors did not fully appreciate the extent of the vulnerabilities; when they did identify them they did not take sufficient steps; and the tailoring approach adopted in response to statutory change, together with a shift in the stance of supervisory policy, impeded effective supervision. The facts: the firm grew 271% from end-2018 to end-2021 against 29% industry growth, reached about $212 billion in assets, carried about $15 billion of unrealised held-to-maturity losses by December 2022, saw over $40 billion of deposits withdrawn on 9 March 2023 and management expecting a further $100 billion the following day — about 85% of the deposit base. The review's own assessment is that higher requirements might not have prevented failure but would likely have bolstered resilience. Note the standing: a regulator publishing a finding of its own supervisory failure is interest running strongly against the finding, which raises its weight considerably.

Established And the vulnerability did not go away. The Office of Financial Research reports aggregate unrealised securities losses at US banks of $481 billion as of 31 December 2024, equal to 19.9% of aggregate equity at banking subsidiaries and about 8.6% of the fair value of aggregate securities holdings, having peaked in Q3 2022. Its assessment is a continuing vulnerability that could amplify stress and increase the chance of lack-of-confidence runs at some banks. Frontier What 2023 actually tested was not capital adequacy. It was deposit-run speed. A framework calibrated on an experience of asset-quality losses met a shock consisting of duration risk plus a run that moved 85% of a deposit base in two days. That is the same structure established in Infrastructure Resilience on redundancy standards and dynamic failure: a system can satisfy its resilience standard and remain exposed to a class of failure the standard does not test. The shared finding is stated once here and attributed there.

Established Energy security has the clearest documented case in this brief of a resilience mandate imposing a measurable cost and then being loosened because of it. The EU's gas storage regulation required 90% capacity by 1 November for 2023–25 with binding intermediate targets, across 18 member states and about 105 billion cubic metres of capacity. The original regulation contributed to increases in gas prices, particularly sharp in the third quarter of 2022, as member states bought simultaneously; a consultancy study cited in the same paper found the filling obligation was the key reason summer–winter spreads were negative in Q1 2025, reaching about −€8 per megawatt-hour in late January 2025 — which destroys the commercial rationale for refilling storage. Some governments bought gas on behalf of market participants. The 2025 amendment made the 90% target achievable at any time between 1 October and 1 December and downgraded intermediate targets to indicative, with permitted deviations of 10% plus 5%, and a further 5% at the Commission's discretion. The publishing institute is substantially industry-supported and its funding is not disclosed in the paper, so treat interpretive framing as potentially interested. Frontier The cost here was not an externality; it was the mandate's own mechanism. A legally required simultaneous buyer inverts the seasonal spread that makes storage economic.

Frontier Food security is where resilience is most clearly non-composable, and the price effects have been estimated. World Bank analysis counts 34 countries imposing restrictive export measures on food and fertilizers as of 2 June 2022, within 135 announced or implemented measures of which 74 were export-restrictive, and estimates the price contribution of export restrictions at rice +12.3%, wheat +9%, citrus +8.9%, soya bean oil +14% and maize +6.1%. For comparison, 36 countries imposed export restrictions in 2008–2012, associated with staple price surges well over 30%. The attribution to restrictions specifically is model-based. Established The structural finding is that national food resilience policy is a negative-sum aggregate: each restricting country improves its own supply and worsens everyone's price, including — through import prices — its own.

Frontier And the largest national food-resilience programme has a price tag and an uncomfortable trend. A think tank with a stated strategic interest in Chinese behaviour reports China holding about half of global grain reserves as of 2022 and nearly five times US corn ending stocks in 2024; a 95% grain self-sufficiency target set in 1996 and reaffirmed in 2019, against overall food self-sufficiency that fell from 94% in 2000 to 66% in 2020 and is projected toward 59% by 2030; over 80% soybean import dependence; and $10 billion spent on public stockpiling in 2023, roughly twenty times what all OECD countries spent collectively, with $215 billion of food imported in 2023 as the world's largest food importer since 2021. Read as a resilience case study: the world's most expensive national food-security programme has coincided with self-sufficiency falling by 28 percentage points in two decades.

Frontier Finally, the aggregate cost of the fragmentation that resilience policy produces, with a spread that is itself the finding. The IMF's range for geoeconomic fragmentation runs from as low as 0.2% of world GDP in a mild scenario to as high as 7% in an extreme one, with around 2% from foreign-direct-investment fragmentation alone in a two-bloc model, and low-income countries facing up to four times the global output loss. The measured evidence since 2022: quarterly trade growth between US- and China-leaning blocs almost 5 percentage points below the 2017–2022 rate; trade between blocs declining roughly 12% and 20% more than flows within blocs, against intra-bloc trade down only about 2pp; and new trade restrictions more than tripled since 2019. Established A 35-fold spread between the mild and extreme scenarios means the model is not constraining, and the distributional point is the one usually dropped: resilience purchased by advanced economies through fragmentation is paid for disproportionately by low-income countries that had no role in the policy.

Frontier “Reshoring increases resilience.” The dominant policy hypothesis of the last eight years across the US, EU, India and Japan — and the weight of available evidence is against it. For: intuitive appeal, and the observation that shortages arrived through international links. Against: a review finding reshoring increases aggregate volatility by reducing source diversification; a general-equilibrium model finding localisation produces larger price swings and greater income variability; and a measurement finding the achieved reallocation raised import prices 2.3–9.8% while leaving indirect China exposure intact or growing. What would settle it: a shock arriving after a completed reshoring, with a matched comparison.

Frontier “Diversification increases resilience; concentration is the risk, not internationalisation.” The main academic alternative, and on the evidence assembled here the best-supported hypothesis in the brief — the same two sources read positively. Against it: diversification is costly and nobody has priced it. What would settle it: firm-level data on supplier counts against disruption outcomes in a common shock, which is feasible with customs microdata and has not been done at scale.

Frontier “Just-in-time was the culprit and just-in-case is the fix.” Widely asserted in 2021–22. For: a real, persistent inventory increase among importers. Against: the authors' own refusal to call the shift complete on data ending Q2 2021, and the structural point that inventory is a buffer against duration, not against a permanent loss of a source. What would settle it: inventory-to-sales measurement through a subsequent disruption.

Frontier “Strategic reserves work.” Held by the IEA framework and every member state. For: the 2022 release moved prices in the intended direction by an estimated 13–40 cents per gallon. Against: prices rose 68 cents anyway, the estimate range is threefold, the wider figure comes from the releasing administration, and refill carries fiscal cost. What would settle it: a pre-registered evaluation methodology agreed before the next release — which is cheap, uncontroversial in principle, and has never been done.

Frontier “Post-2008 bank capital rules made the system resilient” against “the 2023 failures were idiosyncratic and prove nothing about the framework” — and the honest move is to declare the tie. The first is disconfirmed in part by the supervisor itself: tailoring impeded effective supervision, supervisors did not fully appreciate the vulnerabilities, and $481bn of unrealised losses persisted to end-2024 at 19.9% of subsidiary equity. The second has real support too: a single-sector depositor concentration, 271% growth, and specific management failures the review documents first. Both hypotheses carry evidence, the aggregate figure is not idiosyncratic, and what would settle either is another run.

Speculative “Resilience mandates impose costs that eventually repeal them.” One clean documented case: a storage obligation that raised prices in Q3 2022, inverted the summer–winter spread by Q1 2025, and was loosened in 2025. The mechanism is coherent and the base rate does not exist — nobody has computed the rate at which resilience mandates are loosened over a decade. Frontier “Resilience is non-composable: national resilience policies aggregate to systemic fragility.” Strong for the food case — 34 countries restricting exports in 2022 with estimated staple price effects of 6–14%, each acting rationally and all worse off, with a 36-country precedent in 2008–2012. Speculative Weak as a general claim, because not all resilience policies are rival: a bank capital buffer depletes nobody else's. What would settle it is a formal decomposition of resilience policies into rival and non-rival classes, which nobody has published.

Frontier “Fragmentation is itself the resilience risk.” The IMF's position, and the direction is well supported: inter-bloc trade growth 5pp below trend, inter-bloc flows down 12–20% more than intra-bloc, restrictions more than tripled since 2019, low-income countries bearing up to four times the loss. The magnitude is not supported: a 0.2%-to-7% range is a 35-fold spread and does not constrain a decision. Speculative “Resilience should be built into market structure rather than into stockpiles.” A minority position in supply-chain research proposing mandatory dual sourcing and disclosure of tier-2 and tier-3 dependencies rather than inventory requirements. Its appeal is that the measured failure — direct exposure understating true exposure roughly fourfold — is an information failure, and information remedies are cheap. Evidence: essentially none. No jurisdiction mandates deep-tier disclosure.

Speculative “Resilience is unmeasurable in principle, and every claim to have designed it is unfalsifiable until it fails.” The strongest sceptical position and one this brief takes seriously. Its policy implication is uncomfortable: resilience spending cannot be appraised by cost-benefit analysis and should be treated as an insurance premium, priced by willingness to pay rather than by expected value. What would settle it: nothing, which is the point. Handwave And the fringe worth covering: “autarky is achievable and the costs are overstated.” It is not empty — the review's own decoupling range runs from 3% to 70% of GDP depending on the country, and the 3% end is a real number for a large, resource-rich continental economy. Frontier What is unsupported is the extension to a resilience benefit: both the model and the review find localisation raises volatility. The claim persists because its cost premise is defensible for the largest economies and its benefit premise is not defensible for anyone.

Established The identification problem is not incidental to this field; it is the field. A resilience intervention can only be evaluated against a shock, shocks are rare, and the shock that arrives is never the one the intervention was designed against. That means the effective sample size for most resilience policy is one, and the one observation is off-distribution. No amount of methodological care fixes a sample of one, which is why so much of this subject is argued from mechanism rather than from measurement.

Established The second bottleneck is that nobody has priced the diversification-versus-efficiency trade-off, and the absence is the most important finding in this brief. A general-equilibrium model gives a global GDP cost of localisation above 5% and a trade study gives an import price effect of 2.3–9.8%, but no jurisdiction has published an ex-ante price for the resilience it claims to be buying, in the form “we accept X% higher input costs in exchange for Y reduction in expected disruption days”. Handwave Without that statement, resilience policy has no cost-benefit test at all, and any claimed price is asserted rather than computed.

Frontier The third is that exposure is measured at the wrong tier. Direct exposure understates true dependence roughly fourfold in the best-documented case — auto-sector reliance on Chinese inputs rising from about 5% to nearly 20% once indirect links are counted. A policy targeted on direct import shares can therefore reduce the measured exposure while leaving the actual exposure intact, which is precisely the pattern the post-2018 reallocation produced through Vietnam and Mexico.

Frontier The fourth is that the micro-to-macro propagation mechanism is explicitly unknown. A single input constraint produced about a third of a month's headline consumer inflation in the largest economy in the world, and the field's own review names the relationship between micro shock and macro outcome as one of its principal open questions. Until that is characterised, nobody can say which concentrated inputs matter and which do not, so stockpiling and dual-sourcing policy is targeted by salience rather than by exposure.

Established And the fifth is that resilience mandates interact with the markets they operate in, sometimes destroying the mechanism they rely on. A legally required simultaneous buyer of gas inverted the summer–winter spread to about −€8 per megawatt-hour, which is the price signal that makes commercial storage viable in the first place. Frontier That is a general design hazard rather than a European accident: any mandate that compels many parties to transact on the same date converts a price signal into a cost, and no jurisdiction appears to have tested its mandates against that failure mode before enacting them.

Established Nothing on this map produces a result this brief waits on. What it waits on is measurement that nobody is funding and disclosure that nobody is requiring: an ex-ante published price for resilience, deep-tier dependency data, and a pre-registered evaluation design for the next reserve release. All three are choices rather than discoveries, and they are recorded as typed requirements below.

Established One shared finding runs to an existing brief and is attributed once rather than restated. Infrastructure Resilience established that a system can satisfy its resilience standard and remain exposed to a class of failure the standard does not test, on redundancy compliance and dynamic failure in built networks. This brief observes the identical structure in bank capital rules meeting a deposit-run-speed shock, and the convergence of the two is stronger evidence than either alone.

Frontier What this brief waits on from research is specific and mostly feasible. Firm-level supplier counts matched to disruption outcomes in a common shock, which customs microdata would support; inventory-to-sales series carried through a subsequent disruption rather than stopping in 2021; and a decomposition of resilience policies into rival and non-rival classes, which would tell a policymaker in advance whether a given measure aggregates to fragility. Speculative None of those is technically hard. Each requires somebody to decide that resilience policy should be evaluated at all.

Established The cheapest high-value experiment in this brief costs nothing and has never been run: pre-register the evaluation of the next strategic reserve release before it happens. Two published estimates of the 2022 release differ by a factor of three and the wider one comes from the party that authorised it. A methodology agreed in advance — comparator markets, price windows, model specification — converts the next release from an advocacy exercise into a measurement. The reserve exists, the release will happen, and the design cost is a memorandum.

Frontier Second: mandate deep-tier dependency disclosure in one sector and measure what it changes. The measured failure is informational — direct exposure understating true exposure roughly fourfold — and no jurisdiction requires tier-2 or tier-3 disclosure. Speculative A single-sector pilot with mandatory reporting of second- and third-tier suppliers would produce, for the first time, a dataset in which exposure can be computed rather than inferred, and would test whether firms respond to knowing their own dependencies. The hypothesis that information is the cheap remedy is coherent, matched to the measured failure mode, and entirely untested.

Frontier Third: match supplier counts to disruption outcomes across a common shock. Customs microdata identify how many sources a firm had before an event; production and price data identify what happened to it afterwards. This is the direct test of the diversification hypothesis, the best-supported claim in the brief, and it has not been done at scale. The obstacle is data access rather than method.

Established Fourth: extend the inventory measurement past 2021 and through the next disruption. The one firm-level study found a real and persistent increase but stopped at Q2 2021 and declined the structural claim. Frontier Re-running the same design on the same panel through a later event settles whether just-in-case was a policy shift or a shock response, and it is a replication rather than a new study.

Speculative Fifth: run an ex-post evaluation of a food-security stockpile against its own stated objective. The largest programme in the world has a cost of about $10 billion in 2023 and a self-sufficiency trend from 94% to 66% in two decades, and no jurisdiction has published a counterfactual for a stockpile. Handwave The reason is not methodological difficulty; it is that stockpile programmes are justified on security grounds, and security justifications are the class of claim that evaluation is least often applied to.

Established The gas storage case is the clearest piece of mechanism design in this brief and it is worth stating mechanically. Commercial storage is financed by the summer–winter spread: buy cheap in summer, sell dear in winter, and the spread pays for the capacity. A regulation requiring 18 member states to reach 90% fill by a fixed date across about 105 billion cubic metres creates a legally compelled simultaneous buyer, which raises the summer price and compresses the spread — to about −€8 per megawatt-hour by January 2025, at which point refilling is loss-making and someone has to be compelled or compensated to do it. Frontier The 2025 amendment’s fix is a widened window — any time between 1 October and 1 December — plus indicative rather than binding intermediate targets and deviations of 10% plus 5%, with a further 5% at discretion. That is a design change from a date to a range, and it is the correct one: the harm came from simultaneity, not from the level.

Established The banking case has an equally specific mechanism, and it is an accounting one. Securities held to maturity are not marked to market on the balance sheet, so a duration loss accumulates invisibly until a sale forces recognition. About $15 billion of unrealised held-to-maturity losses sat against a $212 billion balance sheet, and the run that forced the question moved over $40 billion in a day with a further $100 billion — about 85% of deposits — expected the next. Established The aggregate version of the same mechanic is still in place: $481 billion of unrealised securities losses at end-2024, 19.9% of aggregate banking-subsidiary equity and 8.6% of securities fair value. Frontier A capital framework can be satisfied throughout, because the loss is not in the capital calculation until it is realised, and the run speed determines whether it ever is.

Frontier The trade case has a measurable engineering signature too: substitution shows up in unit values. Where China's share fell five percentage points, unit values rose 9.8% for Vietnamese sources, 3.2% for Mexican and 2.3% for high-wage Asian suppliers — the price of the second-best supplier, which is what a diversification requirement buys. Established And the exposure did not move as much as the shares did, because the new suppliers themselves deepened their Chinese input dependence over the same window, Vietnam to about 40% and Mexico to 20%. The engineering conclusion is that supplier substitution at tier 1 is not exposure reduction unless tier 2 is measured, and tier 2 is not measured anywhere.

Frontier Reserve mechanics are the last piece, and their limitation is arithmetic. A 240-million-barrel coordinated drawdown against a global market moved price by an estimated 13–40 cents a gallon while price rose 68 cents. Speculative A reserve is a rate instrument, not a level instrument: it changes the speed at which a market clears rather than where it clears, which is why it can be simultaneously effective on its own terms and unable to prevent the shock. Designing to the second expectation guarantees disappointment.

Within this map: Infrastructure Resilience, which owns built systems and supplies the shared standard-versus-failure-class finding; Civilization Resilience Planning, which owns catastrophes severe enough that no functioning outside remains, where this brief owns shocks the economy survives; Autonomous Supply Chains, which owns supply-chain automation as a technology where this brief owns structure as an exposure; Future Capital Markets, which owns intermediation; Civilization Scale Investment, which owns the capital formation any resilience programme would need; and Future Ports and Shipping, where the physical chokepoints sit.

Outside it: international trade economics and the global-value-chain literature; macroprudential regulation and bank supervision; energy market design, which supplies the storage mechanism; agricultural economics, which supplies the export-restriction record; and the general methodology of evaluation under rare events, which is the discipline this subject most needs and least uses.

Established The most valuable source in this brief is a regulator publishing a finding of its own supervisory failure, and the direction of that interest is why it is weighted heavily. Its verbatim takeaways — that supervisors did not fully appreciate the vulnerabilities, did not take sufficient steps when they did, and that a tailoring approach impeded effective supervision — are the kind of statement institutions do not make about themselves without cause. Established The opposite standing applies to the reserve estimate at the top of the published range, which comes from the administration that authorised the release, and to a storage analysis from an institute that is substantially industry-supported and does not disclose its funding in the paper.

Established The institutional requirement that would change this subject most is a published price. Nobody has stated, in advance and in public, what a jurisdiction is willing to pay in higher input costs for a given reduction in expected disruption. Until a government writes that sentence, every resilience programme is unfalsifiable by construction, and the debate is between people asserting that a policy is worth it and people asserting that it is not.

Frontier The second is a disclosure requirement that reaches past tier 1. The measured failure is that direct exposure understates true exposure roughly fourfold, and no jurisdiction mandates second- or third-tier dependency reporting. This is the cheapest available remedy matched to the best-documented failure, and its complete absence from the policy record is striking given how much has been spent on the expensive remedies.

Frontier The third is coordination, because the food case shows the aggregate is negative-sum. Thirty-four countries restricting exports in 2022 each improved their own position and worsened the common price, with a 36-country precedent a decade earlier. Speculative A standing prohibition on export restrictions with a compensation mechanism is the obvious design and nothing resembling it exists, which is itself evidence about how binding the coordination problem is.

Frontier The distributional fact is that resilience purchased by rich economies is paid for by poor ones. The IMF's fragmentation scenarios put low-income countries at up to four times the global output loss, from policy decisions taken in jurisdictions where they have no vote. Established The food case makes the same point at higher resolution: export restrictions by 34 countries raised staple prices by an estimated 6–14%, and the incidence of a staple price rise falls hardest on the households spending the largest share of income on food.

Established The second ethical fact is that the costs of resilience policy are paid diffusely and its benefits are claimed specifically. An import price increase of 2.3–9.8% is borne by every purchaser and appears in no programme evaluation; the reduction in China's US import share appears in every speech. Frontier That asymmetry is not corruption; it is a measurement asymmetry — the share is countable and the price effect requires a study — but it produces the same result as a bias, and it is why the published ex-ante price matters more here than any individual policy.

Speculative Third: if the sceptical position is right, resilience spending is insurance, and insurance has an ethics of its own. Treating a programme as an insurance premium priced by willingness to pay rather than by expected value is defensible — that is how households buy insurance — but it removes the discipline that cost-benefit analysis provides, and it makes the size of the premium a matter of political preference rather than of evidence. Saying so plainly is more honest than the current practice of citing a benefit estimate that no jurisdiction has produced.

Frontier And there is a domestic distributional point that the trade literature usually leaves implicit. The gains from a reallocation of import sourcing accrue to specific producers and regions; the costs appear as a higher price level for everyone. An economy can therefore run a resilience policy that is popular, measurably costly in aggregate, and never evaluated, which is a reasonable description of the last eight years.

Frontier The terminal position is that the framing survives only in a weakened form. Resilience can be purchased, sometimes, at prices nobody has published — and the leading design in current use has an evidence base pointing the wrong way. Two independent lines find reshoring resilience-reducing; the achieved reallocation raised import prices while indirect exposure intensified. The gap between the confidence of resilience policy and the evidence for resilience policy is the widest of any subject in this category.

Frontier The deeper structural finding is non-composability, and it generalises beyond food. Where a resilience measure is rival — a claim on a globally traded supply — every country taking it worsens the aggregate, and the equilibrium is worse than the starting point for everyone including the actors. Where it is non-rival — a capital buffer, a redundancy standard, better information — the aggregate improves. Nobody has published that decomposition, and it is the single most useful thing this field could produce, because it would tell a policymaker in advance which class of measure they are choosing.

Established And a system can satisfy its resilience standard and remain exposed to a class of failure the standard does not test. That was established in built infrastructure and it recurred exactly in banking, where a framework calibrated on asset-quality losses met duration risk plus a two-day run. Two independent domains producing the same structure is the closest thing this subject has to a law, and its practical content is that a resilience standard should be read as a list of the failures somebody has already thought of.

Speculative The long-run risk is that the policy converges on the measurable rather than on the binding. Import shares are countable, so import shares get targeted; deep-tier dependencies are not, so they do not. An economy that optimises the observable exposure while the unobservable one grows is not becoming more resilient; it is becoming more confident. Handwave That is a mechanism claim rather than a measurement, and stating it as a forecast would be assertion — but it is the pattern the Vietnam and Mexico numbers describe.

These horizons track shocks, statutory reviews and data availability rather than technology:

• 10 yr: Frontier The next reserve release and the next significant bank run both arrive inside this window on any reasonable base rate, and whether either is evaluated against a pre-registered design is a decision available now. Established The loosened European storage regime faces its first genuinely cold winter, which is the test the original mandate was written for and the amended one has not had. Frontier Expect the inventory question to be settled by the next disruption rather than by a study, and expect deep-tier disclosure to remain unmandated absent a shock that runs through tier 3 visibly.
• 25 yr: Speculative If fragmentation continues at the measured rate — inter-bloc trade growth about 5 percentage points below trend and restrictions more than tripled since 2019 — the two-bloc counterfactual stops being a scenario and becomes the baseline, at which point the IMF's 0.2%-to-7% range resolves by observation rather than by model. Speculative The diversification hypothesis is either tested against a real shock with matched data or it is not, and there is no third path. Frontier The one thing that would visibly change the field is a published ex-ante resilience price by a major jurisdiction; nothing in the record suggests it is coming.
• 50 yr: Speculative On the sceptical hypothesis, this subject remains a set of insurance purchases with no expected-value discipline, and the literature grows without converging. Speculative On the optimistic one, a decomposition of rival from non-rival resilience measures gives policy a rule, and the negative-sum class is handled by treaty in the way export restrictions are not now. Handwave Which arrives is a political question and any confident answer is assertion.
• 100 / 250+ yr: Handwave Beyond useful forecasting. The only base rates at this horizon are that trade openness has oscillated over centuries rather than trended, and that the periods of retreat have been associated with conflict rather than with policy design. Handwave Two very coarse regularities are a story, not a forecast, and neither has anything to say about whether a specific mandate works.

• Depends on Nothing on this map. This brief waits on no result another brief produces: what it lacks is a published price, a disclosure requirement and an evaluation design, none of which is a discovery. No typed depends-on edge is claimed. One shared finding runs to Infrastructure Resilience — that a system can satisfy its resilience standard and remain exposed to a failure class the standard does not test — and it is attributed there rather than claimed as an edge.
• Requires (not on this map) Three constraints, none of them a research result. First, a published ex-ante price for the resilience a jurisdiction claims to buy, in the form “we accept X% higher input costs in exchange for Y reduction in expected disruption days”. No government has written that sentence. The costs are measurable and measured — localisation cuts global real GDP by more than 5% in the OECD model, and the achieved US reallocation raised unit values by 9.8% for Vietnamese sources, 3.2% for Mexican and 2.3% for high-wage Asian suppliers — while the benefit has never been stated as a quantity, which leaves resilience policy with no cost-benefit test at all. Second, mandatory deep-tier dependency disclosure: US auto-sector reliance on Chinese inputs rises from roughly 5% to nearly 20% once indirect links are counted, so a policy targeted on tier-1 import shares can cut the measured exposure while the real one is untouched — which is what happened, with Vietnam's imports from China rising to about 40% and Mexico's to 20% over the same period the US share fell 5 percentage points. No jurisdiction requires tier-2 or tier-3 reporting, and this is the cheapest remedy matched to the best-documented failure. Third, and market-shaped rather than institutional: a storage buyer of last resort that does not invert the seasonal spread. The EU obligation to reach 90% fill by a fixed date across 18 member states and about 105 billion cubic metres created a legally compelled simultaneous buyer, contributed to sharp price rises in the third quarter of 2022, and drove summer–winter spreads to about minus €8 per megawatt-hour by January 2025 — destroying the commercial rationale for the very refilling the mandate required, and forcing a 2025 amendment that widened the date to a range. Any mandate compelling many parties to transact on one date converts a price signal into a cost, and nobody appears to test for it before enacting.
• Enables Every programme on this map that assumes a functioning supply of inputs, capital or energy through a disruption inherits this brief's uncertainty, and the honest statement is that the inheritance is unpriced. No typed enabling edge is claimed, because what would enable a downstream claim is a cost-benefit test that no jurisdiction has published rather than a result this brief could supply.
• Adjacent International trade economics and the global-value-chain literature; macroprudential regulation and bank supervision; energy market design; agricultural economics; the methodology of evaluation under rare events; and within this map Infrastructure Resilience, Autonomous Supply Chains and Civilization Resilience Planning.

Frontier “Reshoring increases resilience.” This is the dominant resilience policy of the decade and the available evidence points the other way. A supply-chain review finds reshoring increases aggregate volatility by reducing source diversification; a general-equilibrium model finds localisation delivers neither greater efficiency nor greater security of supply and produces larger price swings under shock; and the measured reallocation raised import prices by 2.3–9.8%. Established Two independent lines and one measurement, all pointing the same way, against a policy adopted across four major economies. That is the central finding of this brief and it should be stated flatly.

Established “The post-2018 trade policy achieved decoupling.” It achieved reallocation. China's share of US imports fell from 21.6% to 16.5% while Vietnam's imports from China rose to about 40% and Mexico's to 20%, Chinese direct investment in Mexican manufacturing grew roughly fivefold, and China's import share in Europe rose 2.7pp. Frontier The authors' own conclusion is that indirect links remain intact and along some dimensions are intensifying — so the headline share fell, the exposure did not, and nobody has measured by how much because tier-2 data does not exist.

Frontier “Just-in-time is dead; just-in-case replaced it.” The claim became a stylised fact in 2021–22 on surveys of intentions. The one firm-level measurement available found a real and persistent inventory increase among importers and explicitly declined to call the shift complete, on data ending Q2 2021. Established And inventory is a buffer against duration, not against a permanent loss of a source — which means it is the wrong instrument for the failure mode that reshoring policy is nominally addressing.

Frontier “The 2022 reserve release prevented the price shock.” It is estimated to have reduced prices by 13–40 cents a gallon while prices rose about 68 cents, with the top of that range published by the administration that authorised the release and the bottom by a Treasury analysis reported second-hand. Handwave “Up to about 40 cents” is a phrase carrying a great deal of weight, and there is no counterfactual behind either figure — the threefold spread is an artefact of model choice, not a measurement of uncertainty.

Frontier “Post-2008 bank capital rules were shown adequate in 2023.” The supervisor's own review concludes that tailoring impeded effective supervision and that supervisors did not fully appreciate the vulnerabilities, and $481 billion of unrealised securities losses persisted to end-2024 at 19.9% of aggregate banking-subsidiary equity. Frontier The counter-claim — that the failures were idiosyncratic — also has real support, and the tie should be declared rather than resolved: single-sector depositor concentration and 271% growth are genuinely unusual, and an aggregate $481bn figure is genuinely not.

Established “Gas storage mandates are costless resilience.” The obligation contributed to a price spike in Q3 2022, inverted the seasonal spread that makes commercial storage viable to about −€8 per megawatt-hour by January 2025, and was loosened in 2025 for that reason. Speculative Whether this generalises to a rule — that resilience mandates impose costs which eventually repeal them — is one clean case with a coherent mechanism and no base rate, and nobody has computed the rate at which such mandates are loosened.

Frontier “National food stockpiling secures food supply.” The largest programme in the world spends about $10 billion a year, roughly twenty times all OECD countries combined, holds about half of global grain reserves, and has coincided with overall food self-sufficiency falling from 94% in 2000 to 66% in 2020 with a projection toward 59%. Established That is a cost and a trend, not an evaluation — no jurisdiction has published a counterfactual for a stockpile — and the figures come from a think tank with a stated strategic interest in the subject.

Handwave “Autarky is achievable and the costs are overstated.” This persists across the political spectrum and it is not empty: the decoupling cost range runs from 3% to 70% of GDP depending on the country, and 3% is a real number for a large, resource-rich, continental economy. Frontier What is unsupported is the resilience benefit, which is where the argument does its work by assertion: both the model and the review find localisation raises volatility. The claim survives because its cost premise is defensible for a handful of economies and its benefit premise is defensible for none.

Established And the largest misconception in the subject is that anybody has priced it. No jurisdiction has published an ex-ante price for the resilience it claims to be buying. The cost side is measured — above 5% of global real GDP for localisation, 2.3–9.8% on import prices for the achieved reallocation — and the benefit side has never been stated as a quantity by anyone. Speculative That is not a gap in the literature; it is the reason the literature cannot settle anything, and it is the finding a reader should leave with.