Future Federalism

Federalism assigns tax bases, spending responsibilities, borrowing rights and bailout expectations across tiers of government. It is an unusually well-identified governance subject, because the interesting variation is written down: village council headships randomised by statute in India, a federal audit lottery in Brazil, oversubscribed school admissions in Venezuela, a staggered income-tax devolution in Spain, and equalisation formulas in three rich federations that publish their arithmetic to the euro.

The framing under test — that moving decisions closer to the people they affect produces better outcomes — comes back a tie, and the tie is the finding rather than a failure to reach one. Decentralisation improved allocation where money, mandate and local capacity arrived together; it degraded coordination and accountability where the institutions that discipline a local decision-maker were absent. Different outcomes, measured separately, both real. What survives is a change of question: not how much decentralisation, but what disciplines the decentralised actor.

Established State the tie first, because every claim below sits on one side of it. On allocation: Bolivia's post-1994 municipal spending moved towards measured local need; India's randomised council reservations changed what got built; Venezuelan school autonomy raised the poorest students' scores by about a quarter of a standard deviation. On coordination and accountability: immunisation coverage fell, infant mortality rose where national parties were weak, and pandemic response was slower in more self-ruling systems across 93 countries. Both sets are real and they are about different things.

Established The most informative result in the field is that party discipline sets the sign of decentralisation's effect, not merely its size. Enikolopov and Zhuravskaya, on up to 95 countries over 1975–2000, interact fiscal decentralisation with the age of national parties, a proxy for how institutionalised the party system is. In developing countries a 10% increase in revenue decentralisation raises infant mortality by 0.6 percentage points where party age is half a standard deviation below the mean; the effect at the mean is about +0.2 points; and half a standard deviation above it reverses, lowering infant mortality by 0.2 points. Ten additional years of party age at mean decentralisation cut infant mortality by 0.2 points on its own. The same policy is harmful, neutral or beneficial according to one institutional variable that has nothing to do with proximity.

Established Immunisation behaves differently in the same paper. There the same 10% increase reduced coverage by 11 percentage points at low party age and by 5 points at high — both negative, so strong parties halve the damage rather than reversing it. One design, two outcomes, two structures: sign reversal on mortality, damage limitation on coverage. A summary giving one shape for both is describing a paper that was not written.

Established The same variable reappears twenty years later on a different sample and outcome. Lago-Peñas, Martinez-Vazquez and Sacchi regress cumulative COVID deaths per million on institutions across up to 113 countries: the federalism dummy is positive and marginally significant, the party-nationalisation measure negative and significant in every specification, democracy shows no robust difference, and the standard regional-authority index is significant only in the earliest period and called “not robust” by the authors — decentralisation researchers reporting that federations did worse. Established Tselios prices the delay directly: across 93 countries, each point of self-rule adds 0.8 to 1.3 days between first confirmed case and first containment measure. That scale runs 0 to 27.970, so the full range implies roughly 22 to 36 days — three to five weeks. Frontier Both are observational; whatever makes a country decentralised may also make it slow. Their value is agreeing from different measurements.

Established The other half of the tie is just as well measured. Bolivia's 1994 Popular Participation Law doubled municipal funding to 20% of national tax revenue, switched allocation to strict per-capita and created 198 new municipalities. The three largest cities' share of devolved funds fell from 86% to 27%. Central investment had been 73% transport, hydrocarbons and energy; municipal investment was 79% education, urban development and water and sanitation, and after the reform spending on those became positively correlated with illiteracy, non-connection and malnutrition rates. Frontier The author advocates decentralisation; the reallocation is budget data, the needs-correlation is the interpretive step. Established The cleanest causal result here comes from changing who decides: India reserves every third Gram Panchayat on a serially ordered list for a woman pradhan, and across 161 councils and 322 villages reserved councils delivered 9.09 more drinking-water facilities built or repaired (SE 4.02) and a 0.18 (SE 0.06) higher probability of roads in good condition.

Established School-level autonomy is where devolution is most clearly progressive, and it is the opposite of the fiscal case. Allcott and Ortega compared Venezuela's decentralised Fe y Alegría network with public schools on the national entrance exam — 2,237 against 46,460 students, propensity-score matched on an oversubscribed-admissions design. Average effects are modest, 0.09 SD verbal and 0.08 SD maths; the heterogeneity is not, at 0.242 and 0.249 SD for the lowest social classes, roughly three to four times the effect for advantaged students, at comparable per-pupil spending. Established Argentina ran the other experiment: transferring national secondary schools to provinces produced the verbatim conclusion that bringing decisions closer to the people “may help the good get better” while “the already disadvantaged may not receive these benefits”. Fiscal decentralisation to a weak province and managerial decentralisation to a school principal are opposite-signed interventions that the cross-country literature codes identically.

Established Local accountability is not produced by locality. It is produced by a local information market. Brazil's federal audit lottery covered 669 municipalities between July 2003 and June 2005, auditing about R$5.5m per municipality per year; at least 73% had a reported corruption incident, averaging 1.74 violations. Releasing an audit before an election cut a mayor's re-election probability by roughly 7 points per violation at two and 14 at three — but only where the information could travel. With a local radio station, three violations cost 29 percentage points and a clean audit gained the incumbent 17; without local radio the penalty collapsed to 3.7 points.

Established Does the flypaper effect survive identification? In the one large programme with a credible instrument, no. Theory says an unconditional grant should raise subnational spending by the income effect, on the order of 5–10%; the literature has reported estimates near 100%. Knight instruments federal highway aid with the political power of a state's congressional delegation across 47 states, fiscal 1983–1997, 705 observations. The fixed-effects grant coefficient is 0.0327 (SE 0.0559), indistinguishable from zero; instrumenting moves it by more than an order of magnitude and makes it significant, at magnitudes of about 0.88 in 2SLS (SE 0.4199) and 1.12 in LIML (SE 0.4915) against private-income coefficients of 0.0094 and 0.0100. In words: grants displace state highway spending close to dollar for dollar, with little effect on combined spending. Established The sentence everyone quotes — that after correcting for endogeneity “constituent private income and grants have similar effects on public spending” — belongs to the working paper, Federal Reserve FEDS 2000-49, not to the journal article.

Frontier The field's own 2024 survey of intergovernmental grants declines to give a headline effect size at all, offering only a comparison income effect of 15–20% and stressing that magnitude turns on conditionality, political institutions, fiscal illusion and agency slack. Its authors have spent careers arguing decentralisation works, which makes the refusal informative. There is no verifiable meta-analytic flypaper magnitude in the public literature.

Frontier Does decentralisation raise growth? Nobody has identified it. Two decades of panels regress growth on the subnational spending share; the field's own meta-analysis, in Economic Inquiry in 2016, closes its abstract on open questions rather than a signed effect. Speculative The effect-size distribution and publication-bias tests could not be obtained from any fetchable version, so even the direction is treated here as unestablished.

Frontier Does devolution buy off separatism? The quantitative work says “through a channel running the wrong way”. Brancati, on 30 democracies over 1985–2000, finds two channels of opposite sign: decentralisation directly reduces ethnic conflict and secessionism by widening participation, and indirectly increases both by encouraging regional parties that reinforce regional identity and mobilise for conflict. The net effect turns on whether the settlement strengthens region-specific parties — which is what asymmetric federalism is designed to do.

Established Three areas are recorded as empty, because a searched-for and unfound literature is a finding. Metropolitan and city-region governance: prominent in every “future of federalism” discussion, and this research pass located zero credible causal estimates. Transnational fiscal capacity, the EU's Recovery and Resilience Facility read as a proto-federal instrument: zero. Both are design proposals with advocacy behind them and no evaluation literature. Speculative And most specifically: no quantitative estimate of a devolution settlement's effect on separatist vote share was found for Scotland, Catalonia, Belgium or Quebec — not a weak one, none at all. The most politically consequential claim made for federalism is the one with no measurement behind it.

Established The standard measure of decentralisation does not measure decentralisation. “Subnational share of expenditure” does not separate own-source tax revenue from earmarked central transfers, and the panels keep transfers in the numerator because removing them is not comparable across countries. A country whose local governments spend 40% of the budget while raising 5% of it is coded as highly decentralised. Malaysia is constitutionally federal and scores 0.19. Established The same coding cannot separate fiscal from managerial autonomy — the Argentina-and-Venezuela contrast — so a coefficient on decentralisation averages over two interventions that pull opposite ways.

Established The causal arrow in the popular story points the wrong way. Garrett and Rodden find market integration associated with fiscal centralisation: a ten-percentage-point rise in trade as a share of GDP is associated with a 2% fall in the subnational share of public spending.

Frontier The instruments are weakest where the results are most quoted. Fisman and Gatti's much-cited finding — subnational expenditure share negatively associated with survey-measured corruption, −0.42 (t = −2.97) across 55 countries, about 0.3 standard deviations of improvement per standard deviation of decentralisation, robust across indices at −0.22 to −0.46 — is a cross-section instrumented with legal origin, which correlates with essentially every institutional driver of corruption. Established Their most interesting result is the one nobody quotes: the federal-constitution dummy goes insignificant once fiscal decentralisation is controlled for.

Established The next bottleneck is physical rather than statistical: the tax base moves. Spain's 2010 reform let autonomous communities keep half the personal income tax base and set their own top brackets; by 2014 regional top rates differed by 4.5 percentage points, Madrid the low outlier. Agrawal and Foremny estimate a stock elasticity of top taxpayers with respect to the net-of-tax rate of about 0.85 and, conditional on moving, a 1% rise in a region's relative net-of-tax rate raising the probability of moving there by 1.7 points. Madrid's 0.38-point cut: mechanical loss €50m, migration-induced gain €9m, taxable-income response €4m — a net loss of about 0.42% of personal income tax revenue, migration recouping 18%. Mobility is real and nowhere near self-financing. Frontier And central transfers are empirically a brake on it: instrumenting US pandemic aid with small-state overrepresentation in Congress, an extra $1,000 per capita made a state 13 percentage points less likely to cut its top corporate rate. That is an argument for equalisation with nothing to do with equity.

Established Nothing on this map produces a result this brief waits on. Federalism waits on two political facts, recorded as typed constraints below: an equalisation formula donor jurisdictions keep consenting to, and subnational debt rules that still bind when tested. Germany's marginal clawback rates and Australia's legislated GST floor are both formulas visibly losing consent; Spain's liquidity facility is the live test of the second. Established What it waits on from research is narrow: a measure separating own-source revenue from earmarked transfers, a measure separating fiscal from managerial autonomy, and a party-system variable better than party age — currently the most explanatory variable in the field, on a proxy nobody designed for the job.

Speculative Three areas are empty rather than thin. Metropolitan and city-region governance: no credible causal estimate located. Transnational fiscal capacity, including the EU facility usually called proto-federal: no credible causal estimate located. The effect of a devolution settlement on separatist vote share: no quantitative estimate located at all. Each is a live policy conversation running ahead of any evaluation literature.

Established The cheapest high-value experiment is to pre-register the evaluation of an equalisation reform before the formula changes. The dates are announced years ahead and the outcome data exist: Canada's programme renews 31 March 2029, Australia's Productivity Commission reports finally on 31 December 2026, Germany reformed in 2020 with parameters published in advance. Three staggered natural experiments in rich federations with audited fiscal data, and no announced design attached to any of them.

Established Second: replicate the audit lottery outside Brazil. It is the strongest accountability design in the field, it is cheap, and its interesting result is an interaction — 29 percentage points with a local radio station against 3.7 without. Whether the mechanism is radio, local media generally, or something correlated with both is unresolved and directly testable. Frontier Third, and related: randomise managerial autonomy separately from fiscal transfer. The Argentina and Venezuela results differ in sign and the cross-country literature cannot tell them apart; varying hiring, firing and budget authority at school or clinic level while holding the funding tier fixed would settle whether “closer to the people” is about money or about management.

Frontier Fourth: build a synthetic control for separatist vote share around a devolution settlement. Scotland in 1999, Catalonia's 2006 Statute and its 2010 partial annulment, and Belgium's successive state reforms all have hard dates and long electoral series; the most conspicuous hole in the subject is not a hard study to run. Speculative One credible causal estimate in metropolitan governance would likewise be new evidence in a field that has none. Frontier And China's debt swap runs whether anyone designs for it or not: whether converting off-budget borrowing into cheaper on-budget debt changes provincial behaviour or merely re-prices it will show in provincial accounts within a few years, and the pre-period is already recorded.

Established Equalisation formulas are not neutral machines; they are political settlements, and the buying-off is the largest measurable fact about them. Canada's is a pure federal expenditure — Ottawa pays, no province contributes — measuring fiscal capacity across five revenue bases at national-average rates against a ten-province standard, at about $25.3bn in 2024–25, some 4.7% of federal spending, with statutory renewal due 31 March 2029. Established Two features mean it no longer does what its name says: resource revenue enters at 50% or 0%, whichever pays the recipient more — an artefact of a 2006–07 collision between an expert panel and a pre-election promise — and since 2009 the total envelope grows with a three-year moving average of nominal GDP rather than with measured disparity, so it keeps growing when disparities shrink and is capped when they widen. It is a GDP-indexed grant wearing a formula's clothes.

Established Germany's system works as compression, and the compression is large. The 2020 reform abolished the horizontal Länderfinanzausgleich and folded equalisation into the VAT distribution, raising the weight on municipal tax capacity from 64% to 75%. Ministry figures for 2020: €14.8bn moved horizontally through VAT surcharges and deductions and €6.6bn in general federal supplementary grants, plus €1.2bn in municipal-tax-capacity grants, €642m for political administration costs and €268m for eastern structural unemployment; Bavaria's deduction alone was €7.8bn, and North Rhine-Westphalia flipped from net recipient to net contributor. A spread of roughly 50 percentage points in relative fiscal capacity — Bavaria 121.5%, Bremen 71.8% — is compressed to about 10 points, and to under 2 among the weak Länder.

Established That compression has a price, and it is the most precisely quantified perverse incentive in this brief. Scherf computes the marginal burden on a Land raising additional own revenue: up to 92.6% at Land level for fiscally weak Länder receiving the general supplementary grant; about 117.1% for the combined Land-plus-municipal household; about 136% at Land level for recipients of the municipal-tax-capacity grant; and a theoretical maximum of 121.6% on business tax for the combined household. Rates above 100% mean a Land that succeeds economically ends up with less money. Scherf's name for it is Armutsfalle — a poverty trap for governments.

Established Australia bought stability instead, and published the invoice. On the 2025–26 determination — dated deliberately, because the Commission's 2026–27 update would have landed around March 2026 and is not reflected here — the GST pool is about $95.15bn and relativities run from Western Australia at 0.75000 to the Northern Territory at 5.15112. WA's 0.75 is not an assessment; it is a legislated floor, paid for in cash through “no worse off” payments of $4.938bn in 2025–26 plus a $1.03bn pool top-up. Frontier WA's own submission concedes the scale of what the floor replaced: an assessed share of about 10% of its population share in 2023–24, and, before the reform, “almost 90% of Western Australia's iron ore, lithium and nickel royalties” redistributed to other states.

Established The Productivity Commission's 2026 interim assessment of those arrangements is blunt. Cumulative cost to 2024–25 of nearly $23bn — more than four times the original projection — running at about $6.4bn a year and potentially $12bn a year if iron ore prices rise, because the guarantee is indexed to the very commodity cycle it was meant to insure against. The reforms “have largely not achieved their goals”, produced “two sets of rules”, and should be unwound; final report due 31 December 2026. Established Read the three together: Australia bought consent for about $23bn, Germany bought it with marginal rates above 100%, and Canada bought it by indexing the envelope to GDP and letting recipients choose the resource treatment that pays them more. Compression works; incentive-compatible compression has not been achieved anywhere.

Established On the borrowing side the bailout is the norm and the refusal is the exception — and the exception is 180 years old. US state debt outstanding in 1841 was $198m. Eight states and the Territory of Florida defaulted between 1841 and 1843 and five repudiated all or part of their debts. Congress debated federal assumption and refused, on grounds of unequal treatment, moral hazard and burden on federal finances, and eleven states then adopted new constitutions between 1842 and 1852 carrying procedural debt restrictions — typically a tax raised alongside a bond issue and approved by voters. Frontier States that imposed those limits subsequently traded at higher average bond prices than those that did not: the market priced the constraint. Established Brazil then ran the textbook version: Law 9,496 of 1997 refinanced state and municipal debt at 30 years, 6% real, service capped at 13% of revenue with the excess capitalised, collateralised on participation-fund transfers; subnational revenues rose 2.5% of GDP between 1998 and 2002; and the Fiscal Responsibility Law of 4 May 2000 banned future federal refinancing. Frontier Whether that ban held is a separate question, and states have been back at the federal window since.

Frontier Spain is the live test, and it has to be stated precisely because the usual summary overstates it. Royal Decree-Law 21/2012 of 13 July 2012 created the Fondo de Liquidez Autonómico, a central credit line administered through the state credit institute so that autonomous communities “do not have to finance their debt in the markets” — structured as loans, not transfers. On 2 September 2025 the Council of Ministers approved an anteproyecto de ley under which the State would assume €83.252bn of autonomous-community debt: capped at 50% of each region's 2023 debt, a minimum 19% forgiveness for every region, 75% distributed by adjusted population, €80.31bn of identified “crisis over-indebtedness” from comparing debt growth in 2009–13 with 2019–23, and a claimed €6.7bn of freed interest. Established That is a draft bill. It requires a further approval and passage through the Cortes, and enactment could not be verified. The correction cuts against the argument and is carried anyway: the tidy line that a subnational soft budget constraint is only ever a matter of time is weakened, because the constraint has not yet legally softened. Established here are the intention and the arithmetic, not the outturn.

Established China is running the largest subnational soft-budget experiment in the world, on a single asset class. On-budget local government debt reached 38% of GDP as of October 2025, up more than 15 percentage points of GDP since 2020; off-budget financing-vehicle debt was estimated at 46% of GDP in 2023; total direct government debt was 61% of GDP in 2024, local government 58% of it. Beijing launched an RMB 10 trillion debt swap in November 2024. Land lease revenue — the fiscal basis of the whole arrangement — fell 7.4% year-on-year over the first ten months of 2025, with accumulation most rapid in poorer northern and western provinces. Frontier This is revenue autonomy delegated to one asset class, and the asset class is contracting.

Within this map: Future Democracies, which owns the electoral machinery party discipline runs through; Distributed Governance, the same question asked of non-territorial structures; Institutional Design, where the incentive-compatibility problem measured here as clawback rates is stated generally; Future Public Administration, whose evaluation-is-discretionary finding explains why equalisation reforms are never evaluated; Global Cooperation Models, federalism's problem one tier up and with no fiscal capacity at all; and Sustainable Megacities, which inherits the metropolitan-governance vacuum recorded here.

Outside it: public finance and the theory of fiscal federalism; the political economy of party systems, which supplies the operative variable; development economics, which supplies the randomisations; and sovereign-debt law, which supplies the bailout record.

Established This subject is unusually well documented because the money leaves a trail that has to be published — equalisation arithmetic to the euro, municipal audit results, per-jurisdiction payment tables. That is why this brief can state rates rather than anecdotes, and why the gap is conspicuous: the governments publishing the inputs almost never publish an evaluation of the change. Established Several of the best sources are interested parties and it is load-bearing to say which way: Germany's finance ministry reports on its own reform, Western Australia's treasury is the sole beneficiary of the floor it defends, and the Bolivian evidence comes from an author who advocates decentralisation. Established Two findings run against their authors' research programme — the COVID panel showing federations did worse, and the grants survey declining to give the effect size its own field is famous for — and both are weighted more heavily for it.

Established The constraint that actually binds is consent, and it is measurable in the formulas themselves. A donor facing a marginal clawback above 100% is being asked to consent to a rule that punishes it for succeeding; a floor legislated to stop one state's relativity falling below 0.75 is a formula that has already lost the argument once. Frontier No system has found a compression rate that is simultaneously equitable, incentive-compatible and politically durable. Frontier The second constraint is a debt rule that binds under stress: one credible refusal in 1842 followed by constitutional limits the market priced, and since then a bailout, a ban on future bailouts, and a bill that has not yet passed. Both are recorded as typed requirements below — neither is a research result, and both are things a legislature could choose to supply.

Established The distributional finding is the one most often left out of the case for devolution. Argentina's school transfer produced the conclusion that bringing decisions closer to the people “may help the good get better” while “the already disadvantaged may not receive these benefits”: decentralisation can be regressive across regions even where it raises the average, and the average is what gets reported. Established The countervailing case is about capacity rather than tier — where autonomy reached the level that actually manages the service, the poorest gained most, 0.242 and 0.249 SD against 0.09 and 0.08 on average.

Established Corruption does not evaporate under decentralisation; it restructures. Indonesia's 2001 “big bang” is the largest single decentralisation event ever run. Coding 190 newspaper-reported corruption cases (1,960 actors, 28,725 dyads) across 2001–04 and 2005–13: cases involving third-party intermediaries rose from 46% to 69%, cases resting on kinship and friendship ties doubled from 15% to 30%, and detected cases rose from 448 to 629 a year between 2010 and 2014. Corruption became less hierarchical and more networked — what one would predict if devolution multiplies the veto points that can be bought. Frontier Detected cases are not incidence; a decentralisation that also improved detection would look similar in this data.

Established And equalisation raises a consent question its designers rarely state plainly. A resource-rich jurisdiction can watch most of a royalty stream redistributed; a fiscally weak Land can face a marginal rate above 100% on its own success. Both are formulas working exactly as designed, and both generate the political energy that eventually breaks them. The ethical argument for equalisation is strong; the ethical argument for these parameters is a different argument, usually made as though it were the same one.

Established The terminal position is an explicit tie, and it is worth stating twice because almost nobody states it once. Decentralisation improved allocation — Bolivia's post-reform spending shifted to local education, water and sanitation and tracked measured need; India's randomised reservations changed the goods delivered; Venezuelan school autonomy raised the poorest students' scores by about a quarter of a standard deviation. Decentralisation degraded coordination and accountability — immunisation coverage fell in both institutional environments, infant mortality rose where national parties were weak, pandemic response slowed by three to five weeks across the range of self-rule in 93 countries, and corruption networks densified after Indonesia's devolution. Different outcomes, measured separately, both real. Picking a side requires ignoring half the measurements.

Established What follows is a change of question, not a compromise between the answers. “Closer to the people” is a claim about information. It says nothing about whether the local decision-maker internalises the externalities their decision imposes on everyone else, and nothing about whether anyone can punish them for not doing so. Where a national party controls careers, or a local radio station exists to broadcast an audit, the information advantage is realised. Where neither exists, proximity is just distance from oversight. Established That is why the operative variable is the discipline rather than the tier: the same factor sets the sign on infant mortality in a 95-country panel covering 1975–2000 and is the only significant institutional coefficient in a 113-country COVID panel from 2022 — two samples, two decades apart, two outcomes, one variable, which is as close to replication as this field gets.

Frontier The long-run risk is not that decentralisation fails but that its fiscal settlements expire. Every equalisation formula in the record is bought: with money in Australia, with above-100% marginal rates in Germany, with GDP indexation and a recipient-favourable resource rule in Canada. Each purchase has a term. Canada's renews in 2029, Australia's is under review with a final report at the end of 2026, and Germany's clawback arithmetic is unchanged since 2020. A federation whose formula loses consent does not become unitary; it becomes a federation arguing about the formula, which is a different and worse operating state.

Speculative And the largest open question is whether party systems can be built deliberately. The evidence says nationally integrated parties are what makes decentralisation safe. Nothing in it says how a country acquires them, whether electoral rules can engineer them, or whether the causation runs the other way from state capacity to party institutionalisation. A design recommendation that reduces to “have better parties” is not yet an actionable one, and this brief does not pretend otherwise.

These horizons track statutory renewal dates, debt maturities and electoral cycles rather than technology:

• 10 yr: Established Three hard dates dominate: Australia's Productivity Commission reports finally on 31 December 2026, Canada's equalisation programme renews on 31 March 2029, and Spain's €83.252bn assumption either passes the Cortes or does not. Frontier Expect Germany's clawback arithmetic to survive unchanged, because the Länder who would have to consent to fixing it are the ones a fix would cost. Frontier China's RMB 10 trillion swap becomes measurable in provincial accounts inside this window, and land revenue falling 7.4% year-on-year is the variable to watch rather than the swap headline.
• 25 yr: Speculative The plausible split is that identification improves where randomisation is already possible — service delivery, audits, school and clinic management — and stays absent for questions needing country-level variation, because 95-country panels cannot be fixed by adding countries. Speculative A transnational fiscal capacity in Europe either becomes permanent, in which case there is finally something to evaluate, or lapses and the literature stays empty. Handwave Forecasting which is asserting a political outcome, not extrapolating a measurement.
• 50 yr: Speculative If party nationalisation really is the operative variable, the long-run distribution of decentralisation outcomes tracks the long-run distribution of party systems — and no federation has shown it can choose that. Speculative The metropolitan tier is the likeliest site of genuinely new federal structure, because that is where population is and where the evidence base is empty. Handwave Both are extrapolations from a single institutional correlate.
• 100 / 250+ yr: Handwave Beyond useful forecasting. The only data point at that horizon is the United States refusing to assume state debts in 1842 and the constitutional debt limits that followed, still binding today. Handwave One episode, 180 years old, in one federation, is a story rather than a base rate.

• Depends on Nothing on this map. Federalism waits on no result another brief produces: the evidence base is already better identified than most of this category, and the binding constraints are political settlements rather than discoveries. No typed depends-on edge is claimed.
• Requires (not on this map) An equalisation formula donor jurisdictions keep consenting to. Both large formulas in the record are losing that consent: Germany's marginal clawback on additional own revenue runs to 92.6% at Land level for fiscally weak Länder receiving the general supplementary grant, about 117.1% for the combined Land-plus-municipal household and about 136% at Land level for municipal-tax-capacity grant recipients — above 100% means succeeding economically leaves a Land with less money — while Australia legislated a floor of 0.75 under one state's relativity and has paid nearly $23bn in “no worse off” compensation to 2024–25, more than four times the original projection, with its own Productivity Commission finding the reforms have largely not achieved their goals and should be unwound. And subnational debt rules that still bind when tested: the record holds exactly one credible refusal — the United States Congress declining to assume state debts in 1842, after which eleven states wrote constitutional debt limits and the market rewarded them with higher bond prices — against Brazil's 1997 refinancing followed by a statutory ban on future refinancing that states have tested since, and Spain's liquidity facility, written as loans in 2012, with a draft bill approved on 2 September 2025 that would convert €83.252bn of them into an assumption by the State. That bill has not passed the Cortes, so the constraint has not yet legally softened — the one piece of evidence running the other way. Neither requirement is a research result; both are things a legislature could choose to supply and has so far chosen not to.
• Enables In principle, any devolved delivery this map assumes — metropolitan infrastructure, regional energy and transport programmes, subnational adaptation — rests on a fiscal settlement that holds. No typed enabling edge is claimed, and the reason is itself a finding: the metropolitan and city-region governance literature searched for this brief contains no credible causal estimate, so the enabling relationship has never been measured.
• Adjacent Public finance and the theory of fiscal federalism; the political economy of party systems, which supplies the variable that sets the sign; development economics, which supplies the randomisations; sovereign-debt law, which supplies the bailout record; and within this map Future Democracies, Distributed Governance and Institutional Design.

Frontier “The flypaper effect means grants raise spending by roughly 40 cents on the dollar.” No such number is established. It is a robust anomaly, not a measured parameter: theory predicts an income effect of 5–10%, the literature has reported estimates near 100%, the field's own 2024 survey declines to give a headline magnitude, and the best-identified US study finds that instrumenting the grant makes the anomaly disappear. Established Do not quote a flypaper coefficient; no verifiable meta-analytic magnitude exists in the public literature.

Frontier “Decentralisation reduces corruption.” The canonical result — −0.42 across 55 countries — is a cross-section instrumented with legal origin, which correlates with practically every institutional determinant of corruption. Established The same paper's finding that the federal-constitution dummy goes insignificant once fiscal decentralisation enters is a warning about what the decentralisation variable is absorbing, not a result about constitutions. Established And the best structural evidence, from Indonesia, shows networks restructuring rather than shrinking. Decentralisation redistributes corruption; it has not been shown to reduce it.

Frontier “The literature shows decentralisation raises growth.” It does not. The field's own meta-analysis closes its abstract on open questions rather than a signed effect, and the underlying panels inherit a measure that cannot separate own-source revenue from earmarked transfers — a country spending 40% locally while raising 5% is coded highly decentralised, and constitutionally federal Malaysia scores 0.19. Speculative Anyone citing “the literature” for a growth dividend is citing work whose own meta-analysts declined to draw a conclusion from it.

Established “Globalisation is devolving power downward.” The fiscal data show the opposite sign: a ten-percentage-point rise in trade as a share of GDP is associated with a 2% fall in the subnational share of public spending. Established And “formal federalism is what matters” fails on the same evidence — the federal-constitution dummy is insignificant once fiscal decentralisation is controlled for. Only where the money actually moves does anything get measured.

Frontier “Devolution defuses separatism.” The quantitative decomposition gives two channels of opposite sign, with the indirect one — devolution builds regional parties, regional parties reinforce regional identity — pushing the other way. Speculative And the honest statement is stronger than the qualification: this pass found no quantitative estimate of a devolution settlement's effect on separatist vote share in Scotland, Catalonia, Belgium or Quebec.

Frontier “Subnational fiscal rules never survive contact with a crisis.” Mostly true, and this brief resists rounding it up. Brazil's 2000 law explicitly banned future federal refinancing and states have been back at the window; Spain wrote a liquidity facility as loans in 2012 and its government approved a draft bill in September 2025 that would assume €83.252bn of them. Established But that is an anteproyecto de ley, not an outturn — approved by the Council of Ministers, still needing a further approval and passage through the Cortes, with enactment unverified. The strongest modern evidence for “a soft budget constraint is only a matter of time” is currently an intention rather than a completed conversion, and saying so weakens the claim. The one credible refusal in the record remains 1842.

Established Three figures in circulation are deliberately not printed here, and naming them stops their absence reading as an oversight. The amount Australia's GST relativities are said to redistribute away from equal-per-capita shares could not be verified against the determination itself. A cumulative disbursement total for Spain's liquidity facility circulates widely and could not be verified against any primary Spanish government source. And a state commission's estimate that a transfer-duty rate rise returns a large share of the extra revenue through higher grants comes from a single interested party and could not be corroborated. Speculative Modelled long-run population effects of alternative grant arrangements, which circulate in the same debate, are computable general-equilibrium outputs and are not measurements of anything.

Established And the framing itself: “closer to the people” is an information claim doing duty as an accountability claim. Proximity gives a local decision-maker better information about local preferences. It does not give anyone the ability to punish that decision-maker, and it does not make them bear the cost their decision imposes elsewhere. The measured evidence says the punishing and the cost-bearing are supplied by nationally integrated parties and by a local media market — and where those are absent, the information advantage is not realised at all.