The PPP That Never Ends

Why the asset does not come back, why the second contract costs three times the first, and what a law of mandatory re-tendering does to fiscal accounting

Author: David Duarte Arancibia, Austral Fiscal & Infrastructure Intelligence Date: September 2026 Audience: ministries of finance and budget offices; PPP units and infrastructure ministries; fiscal councils and supreme audit institutions; the IMF and multilateral development banks; academics in public economics and infrastructure regulation Series: paper 22 of the Austral series on the fiscal risks of public-private partnerships. It occupies the slot planned as The Last Five Years and closes the lifecycle arc the series opened with The Recognition Rule; the title changed when the evidence did (§2). Disclosure: the author was head of the public-private partnership unit at Chile's Ministry of Finance and later head of the Public Finance Division at DIPRES, Chile's budget office; he is a co-developer of PFRAM, the IMF/World Bank PPP Fiscal Risk Assessment Model, and was the World Bank's lead on PPP fiscal-risk assessment. This paper argues that a model he helped build embeds an assumption that Chilean law contradicts. He is also the founder of Austral, which sells fiscal-risk tools; readers should weigh that interest. Every empirical claim below rests on public documents — supreme decrees, tender specifications, published price indices — identified so they can be checked.

Abstract

This paper set out to measure a maintenance cliff: the well-known prediction that a concessionaire stops maintaining an asset once the remaining contract term is shorter than the life of the intervention. That measurement could not be made, and the reason is reported here as a finding rather than buried — in Chile, the country with the longest continuous concession programme in the developing world, no per-contract conservation-spending series exists. The word "conservation" appears in 52 of 60 audited annual reports of concession companies, but as the boilerplate of the contract's legal object, not as an expenditure line.

What the record does show is more consequential. Under Article 25 of Chile's concessions law, when a concession expires the works "shall be granted in concession again by the Ministry of Public Works for their conservation, repair, expansion or exploitation," and the new tender must be timed so that "no gap exists between the two concessions." Reversion to the State is the exception, available only where the works have fallen into disuse or re-tendering would harm the State, and it requires a reasoned decree of the President. The asset does not come back. As a matter of law, it is not meant to.

Across 18 Chilean assets that have been concessioned more than once — 41 contract generations, 3,687 documents — the pattern is consistent. Of 17 generation-to-generation transitions with an observable term, fourteen lengthen and the median multiplies by 1.72; but that figure is reported here with its own correction attached, because the instrument changed between generations. First rounds were auctioned on offered tariff with a fixed term; 13 of 17 second rounds are least-present-value-of-revenue auctions with a variable term, so for most pairs the comparison sets a realised fixed term against a cap that may not bind. The direction of the bias is known and the paper does not claim 1.72 as realised duration. What survives precisely is sharper: maximum contractual exposure lengthens, two second rounds now sit exactly at the statutory ceiling of fifty years, and the end date of a variable-term concession is not knowable in advance at all — which matters more for fiscal accounting than any multiple. The two transitions that shorten are not counterexamples: they are 48-month bridge contracts over "existing installations," which exist precisely because Article 25 forbids a gap. On the money — a comparison unaffected by term design — five corridors have a comparable official budget in both rounds, and all five rise, by a median of 3.25 times. Normalising by kilometre barely moves the result (median 3.07× against corridor length growing 1.06×), and on the cleanest case — 144.158 km in the first concession and 144.158 km in the second — the multiple is 3.72.

None of this measures deterioration, and the paper is emphatic on the point. The increase decomposes into four legitimate drivers, three of which Article 25 itself contemplates: capacity expansion for demand the first concession helped create; safety standards that did not exist at first tender (the tender specifications require pre-existing works "that do not meet 100 km/h" to be modified, while new works are designed for 120 km/h); real construction-cost inflation above the CPI-indexed unit in which budgets are written, measured here at 8.2% over 9.25 years using the Ministry's own contract-adjustment index — a small share; and environmental and social requirements that did not exist. On the last, the dates settle it: the five Ruta 5 sections now being re-tendered were awarded between June 1995 and May 1996, eleven to twenty-two months before environmental impact assessment became mandatory in Chile and thirteen years before indigenous consultation. The first round did not compete under the rules the second must satisfy.

Two consequences follow. First, fiscal accounting is wrong as a matter of law, not of practice: PFRAM, IPSAS 32 and ESA 2010 all treat the concession period as the life of the arrangement, whereas Chilean statute makes renewal the default and reversion the exception. The obligation does not extinguish at term; it renews, longer and larger. Second, the headroom that an affordability envelope shows opening as contracts expire is legally pre-committed to be re-tendered on the same asset — it is not discretionary space.

Spain is the counterfactual, and it is not a happy one. Having no equivalent obligation, Spain let the AP-1, AP-4, AP-7 and AP-2 revert between 2018 and 2021 with tolls falling to zero; from January 2027 the State absorbs 993 km more at €80 million a year charged directly to the budget. But conservation there now competes annually against new construction and loses — a structural outcome, not an administrative failure, because a politician is not remembered for a well-maintained road. And the €2,504 million of network reset that concessionaires financed, equal to 2.3 times the ministry's road-creation programme, lost its funding source on reversion.

The policy conclusion is uncomfortable for the common political claim on both sides of the Atlantic. The idea that once a concession is paid off the road becomes free is an illusion. At expiry the asset is not paid off: it is old, under-dimensioned for the demand it helped create, built to a superseded safety standard, and unbuilt to environmental and social requirements that now exist. Bringing it current costs between 1.8 and 3.9 times the original budget per kilometre. The real question is not tolls versus free access. It is who finances the next investment cycle, and whether that decision is visible when it is made.

Keywords: public-private partnerships · concessions · handback · re-tendering · asset reversion · fiscal accounting · IPSAS 32 · ESA 2010 · PFRAM · least-present-value-of-revenue auctions · road maintenance · political economy of public investment · Chile · Spain · United Kingdom.

Feedback & discussion. Published openly for challenge. Comment at https://austral-intelligence.com/research; email [email protected].

Contents

1. The end of the contract, which the literature skips
2. A null result, and the object it revealed
3. What the statute actually says
4. The record — term, budget, and kilometre
5. Why the second round costs more, and why none of it is deterioration
6. Why a contract maintains better than a budget
7. Spain — the counterfactual where the asset does come back
8. What this does to fiscal accounting
9. The ceiling that is running out
10. The illusion to abandon
11. Limitations
12. The policy lead     Appendix A — data, sources and reproducibility · Appendix B — glossary

1. The end of the contract, which the literature skips

A concession contract is designed, tendered, financed, built and supervised with enormous care for twenty-five years. The literature follows it closely for most of that time. There is a large body of work on selection and value for money, on risk allocation, on financing structures, on renegotiation — this series has contributed to several of those. There is almost nothing on what happens when the contract ends.

The gap is not obviously a gap. Termination looks administrative: the term runs out, the asset returns to the State, the contingent liabilities extinguish, the file closes. Every fiscal-accounting framework in use encodes exactly that picture. The PPP Fiscal Risk Assessment Model projects the arrangement over the concession period and stops. IPSAS 32 recognises a service concession asset and a corresponding liability that unwinds across the term of the arrangement. ESA 2010 and the Eurostat guidance built on it allocate the asset to one balance sheet or the other for the duration of the contract, after which the question is moot.

The blind spot is not for want of a literature on PPPs. Engel, Fischer and Galetovic's The Economics of Public-Private Partnerships (Cambridge, 2014) is the standard economic treatment; Guasch's work on renegotiation defined a field; there is a substantial empirical literature on selection, on financing, and on the fiscal treatment of these contracts, to which this series has contributed. What that literature covers thinly is the terminal phase — and the exception proves how recent the interest is. The UK's National Audit Office reported on it only in 2020, and the Infrastructure and Projects Authority issued handback guidance later still, both prompted by the fact that more than 700 Private Finance Initiative contracts begin expiring from 2025.

This paper began from a narrower suspicion inside that picture: the maintenance cliff. The prediction is standard and intuitive. A concessionaire's incentive to maintain an asset collapses once the remaining term is shorter than the useful life of the intervention. A resurfacing lasts ten years; with four years left on the contract, it does not get done. If that is right, the State receives an asset in worse condition than the contract nominally requires, the performance bond is smaller than the cost of restoring it, and the gap is a liability that materialises precisely when nobody is looking at it.

That is a good hypothesis. It is also, as far as this paper could establish, unmeasurable with existing data — and the attempt to measure it is what redirected the work.

2. A null result, and the object it revealed

Chile is the natural test bed. Its concession programme began with the Túnel El Melón, awarded in April 1993, and has run continuously for more than three decades. Contracts, tender specifications, award decrees and modifications are published. Several assets have completed a full cycle. If a conservation-spending profile by residual contract year can be built anywhere, it can be built there.

It cannot. The attempt is documented here because a null result about data availability is itself a finding about fiscal transparency.

The word conservación appears in 52 of the 60 audited annual reports of Chilean concession companies filed with the financial regulator. In substantially all of them it appears as the boilerplate of the contract's legal object — the standard formula "Ejecución, Conservación y Explotación de la obra pública fiscal" — and not as an expenditure line that can be read off a period. The regulator's filings consolidate operating costs at a level that does not separate routine maintenance from periodic maintenance from major rehabilitation. A dataset of accrued flows exists for a handful of urban concessions but measures a different object. Of 146 supplementary agreements between the State and concessionaires, five mention maintenance at all.

So the series does not exist, and it cannot be reconstructed from what is published. This matters beyond the present paper. The single variable that would let a government verify whether its concessionaires are maintaining assets as the contract requires is not collected in a form anyone outside the supervising agency can use — in the country with the longest and best-documented programme in the developing world. Any government that intends to manage handback needs to start collecting it, and §12 returns to this.

The hypothesis is not idle — where it has been audited, it bites

The Chilean data cannot test the maintenance cliff. British data have, and the answer is not comfortable.

The National Audit Office's Managing PFI assets and services as contracts end (5 June 2020) examined the United Kingdom's preparation for the expiry of more than 700 PFI contracts, the bulk of which begin expiring from 2025. Of the nine PFI projects that had by then expired and transferred back to public bodies, four public authorities were dissatisfied with the condition of the assets they received. The NAO warned that infrastructure could return to the public sector in an "unsatisfactory condition" and that services could be disrupted absent a strategic approach, and found handback provisions themselves to be a source of the problem — vague, gap-ridden and open to interpretation, and therefore disputable. Public bodies, it concluded, were underestimating the time, resources and complexity of managing the end of these contracts.

Four in nine is not a rate to generalise from, and the NAO does not present it as one. But it establishes that the terminal-condition problem is real where it has been looked for, and it makes the Chilean data gap harder to excuse: the country that audited the question found a problem in nearly half the cases it could observe; the country that cannot audit it has the longest programme in the region.

There is also a mechanism-level reason to expect the two jurisdictions to differ, and §6 develops it. UK PFI is overwhelmingly availability-based social infrastructure — schools and hospitals, where the operator is paid for making the asset available regardless of how many people use it. Chilean road concessions are predominantly user-financed, where revenue depends on traffic and traffic depends on the state of the road. The economic literature predicts stronger maintenance incentives in the second case. That is a prediction this paper can motivate but not test, and it is offered as such.

What the Chilean record does contain, in abundance, is the transition itself: the documents of assets that have been concessioned more than once. That turns out to be a better object than the one the paper set out to measure, because it answers a question the maintenance cliff only approaches obliquely. Not how much was left undone — but what the State actually does when the contract ends.

The answer, in Chile, is that it does what the law tells it to do.

3. What the statute actually says

Article 25 of Decree with Force of Law No. 900 of 1996, the consolidated text of Chile's Concessions Law, is short, and it is decisive.

Its first paragraph caps the term: concessions "shall have the duration determined by the award decree, which in no case may exceed fifty years."

Its third paragraph governs the end:

"Once the term of the concessions has concluded, the works shall be granted in concession again by the Ministry of Public Works for their conservation, repair, expansion or exploitation, whether isolated, divided, or integrated together with other works. The corresponding tender shall be carried out with the anticipation necessary so that no gap exists between the two concessions."

Its fourth paragraph provides the only exit: where the works have fallen into disuse, or where re-concessioning would be "inadmissible, inconvenient or harmful to the State of Chile," the President of the Republic may say so by reasoned decree and release the Ministry from the obligation.

Three things follow, and each changes how the end of a concession should be read.

Re-tendering is mandatory, not discretionary. The verb is deberán — shall. The Ministry does not choose whether to put the asset back on the market; it chooses the scope and the terms. Reversion to direct State operation is the exception, and it is an exception that must be justified at the highest level of government.

The new concession is expressly for expansion, not only for upkeep. The statute lists "conservation, repair, expansion or exploitation" as the objects of the second grant. Chilean law therefore anticipates that a re-tendered asset will carry new investment. This matters enormously for §5: an analyst who finds that second rounds cost more than first rounds has not found a scandal, they have found the statute working.

And continuity is a legal requirement. The obligation to tender early enough that no gap opens explains a feature of the data that would otherwise look like discretion, and §4 returns to it.

Is the exception ever used? This is the strongest empirical objection to reading Article 25 as binding: a duty with a wide escape valve is not a duty. The corpus was searched for it. The disuse-or-harm clause appears five times, and in every instance it is the statute being quoted — including in a concessionaire's own 2020 legal filing — not an applied case. No presidential decree releasing the Ministry from the obligation was found. That is supporting evidence, not proof: a decree declaring an asset not to be re-tendered would have no reason to sit in a concession dossier, which is what this corpus is built from. The honest statement is that the exception is invoked in argument and was not observed in practice here, and that a definitive answer would require searching the Official Gazette rather than the contract files.

The provision is live, not historical. Recent supreme decrees quote it verbatim, and so does the Comptroller General. In Official Letter E557569 of 25 October 2024, the Comptroller General's office addressed a twelve-month extension of an airport concession on the ground that the tender, evaluation and award of the successor contract "require a period longer than the term of the current concession," and that it was in the public interest "to ensure the continuity of both concessions" — citing Article 25 and quoting its third paragraph. The statute is doing operational work in 2024 and 2026.

4. The record — term, budget, and kilometre

Chile has 132 concession contract dossiers in the corpus assembled for this paper, covering 109 distinct assets. Eighteen of those assets have been concessioned more than once, across 41 contract generations documented by 3,687 files. Two airports — Diego Aracena in Iquique and El Tepual in Puerto Montt — are on their fourth concession. El Loa in Calama and La Florida in La Serena are on their third.

4.1 The term lengthens — but the instrument also changed, and that matters more

Seventeen generation-to-generation transitions have a contractual term observable on both sides, taken from the award decree or the tender specifications with the citation preserved. On the face of it, fourteen lengthen, one is unchanged, and two shorten; the median ratio is 1.72.

That 1.72 must be read with a large caveat, and the caveat is the more interesting finding. The two sides of the comparison are not the same kind of number. In the source data, the first-round term is recorded as a contract term in 13 of the 17 transitions; the second-round term is recorded as a maximum term in 15 of 17. And the bid variable explains why: where it is recorded, the first rounds were auctioned on offered tariff — a fixed-term, fixed-price concession — while 13 of the 17 second rounds were auctioned on "VPI / ITC (variable term)".

That is a least-present-value-of-revenue auction: bidders compete on the present value of revenue they require, and the concession ends when that value is collected, not on a calendar date. The mechanism is Engel, Fischer and Galetovic's, proposed in the Journal of Political Economy in 2001 and adopted in Chile — the country whose programme they were writing about.¹

So for most pairs the table compares a realised fixed term against a cap that may not bind. Under LPVR a concession with strong traffic ends well before its maximum. The direction of the bias is known: the 1.72 overstates the true lengthening, and this paper does not claim it as an estimate of realised duration. What can be said precisely is narrower and still substantial:

• The maximum contractual exposure lengthens. For a State projecting liabilities, the cap is the relevant number for a worst case, and it has risen.
• Two second rounds sit at the statutory ceiling of fifty years, which under a variable-term contract is the longest the State can be held, not the expected duration.
• And the instrument itself changed between generations. Chile did not merely re-tender these corridors; it re-tendered them under a different auction design, moving demand risk from the concessionaire back toward a variable term.

This last point does more work in §8 than the multiple does. Under a variable-term concession the end date is not knowable in advance at all — it is a function of realised traffic. Any fiscal framework that projects "over the concession period" is projecting over a random variable and treating it as a date.

The two transitions that shorten confirm the legal mechanism rather than contradicting it. Both are 48-month contracts, and the tender specifications say what they are for: the concession "shall be extinguished at the end of forty-eight (48) months counted from the Provisional Commissioning of the Existing Installations or Phase 1." These are bridge contracts — short grants to operate an airport as it stands while the expansion is tendered. They exist because Article 25 forbids a gap between concessions. Iquique's sequence is 15 years, then 4, then 25. Not even in the interval does the asset return to State operation.

4.2 The budget multiplies

Five corridors have an official estimated budget observable in both generations, drawn from the award publication in the Official Gazette for the first round and from the technical datasheet or award decree for the second. All five rise. The median multiple is 3.25; in aggregate, 36.6 million UF becomes 115.2 million.

A methodological note that matters for anyone reproducing this. Chilean tender documents report two magnitudes that are easy to conflate and must not be: the official estimated budget, which is capital expenditure, and the ITC (Ingresos Totales de la Concesión), the present value of total concession revenue that serves as the bid variable in variable-term auctions. Mixing them inflates ratios by an order of magnitude. Only the former is used above.

4.3 And it is not because more road is being concessioned

The obvious objection is scope: perhaps the second contract simply covers more kilometres. It does not. Normalising by corridor length barely moves the result.

The median per-kilometre multiple is 3.07 against a gross multiple of 3.25, with corridor length growing by a median of just 1.06×. Collipulli–Temuco is the clean case: 144.158 km in the first concession and 144.158 km in the second — the same corridor, kilometre for kilometre, at 3.72 times the official budget.

So the money is not buying more road. It is buying the same road, delivered with more capacity, to a different standard, at different prices. Those are three distinct claims, and §5 separates them.

5. Why the second round costs more, and why none of it is deterioration

It would be easy, and wrong, to read a 3.25× multiple as the price of deferred maintenance — as evidence that the first concessionaire ran the asset down and the State is now paying to fix it. The maintenance cliff of §2 is a real hypothesis, and nothing here refutes it. But it is not what these numbers measure, and four identifiable drivers account for the increase. Three are named in Article 25 itself.

5.1 Capacity

Demand grew for twenty-five years, in substantial part because the first concession built the road. A corridor tendered in 1996 for the traffic of 1996 needs third lanes, grade-separated junctions, bypasses and interchanges for the traffic of 2026. The statute contemplates this directly: the second grant is for "conservation, repair, expansion or exploitation."

Talca–Chillán shows the scale. Its second concession covers 193.31 km of existing trunk road plus a new 55.8 km Talca bypass — a fresh dual-carriageway alignment that did not exist before, with eight bridges including a new 1,300-metre crossing of the Maule river, thirteen interchanges, six grade-separated crossings and a footbridge. None of that is restoration.

5.2 A safety standard that did not exist

This is visible in the tender documents rather than inferred. The specifications for Talca–Chillán require that pre-existing works "that do not meet the speed of 100 km/h shall be modified in such a way that they comply with the required speed," while new works — the Talca bypass — are designed "for a design speed of 120 km/h."

The second concession therefore carries two upgrades at once: bringing the old alignment up to a standard it never met, and building the new one to today's. A design-speed change is not a signpost. It propagates into horizontal and vertical geometry, sight distances, lateral clearance, barrier systems and carriageway separation — the expensive parts of a road.

5.3 Real construction-cost inflation

Chilean budgets are written in Unidades de Fomento, a unit indexed to consumer prices. If road construction costs rise faster than the CPI, an identical work costs more UF than it did, without a single specification changing. That excess must be netted out before attributing anything to capacity or standard.

The Ministry of Public Works publishes the index it uses to adjust its own road contracts — the reajuste polinómico series maintained by its Planning Directorate. The published file carries two sheets, both on a December 2020 = 100 base, which chain without rebasing: March 2016 to December 2020 (87.17 → 100.00) and December 2020 onward (100.00 → 143.53 in June 2025). That gives 112 monthly observations over 9.25 years.

Two things stand out. The real excess is post-pandemic: over the first 4.75 years it was 1.9%; almost all of it accrues after 2020. And even so, over 9.25 years the accumulated real excess is 8.2% — 0.86% a year.

Netting that out:

Price explains a minor share. What remains — between 1.8 and 3.9 times — is capacity, standard, and the fourth driver.

5.4 Environmental and social requirements that did not exist

What passed without comment in 1996 — an alignment splitting a town, a wetland crossed, a community not consulted — now has a procedure, a timetable and a cost.

Classifying tender documents by the award year of the contract (not by folder, which mixes eras within a single dossier) gives a clean before-and-after across road concessions:

(n = 47 / 103 / 202 tender documents from road contracts.)

The social dimension is what is genuinely new; the environmental dimension was already installed by the 2000s. Environmental assessment rises between the 1990s and the 2000s and then plateaus, while citizen participation goes from absent to nearly a fifth of documents, and indigenous consultation and climate simply did not exist as categories.

The dates settle it more sharply than the percentages:

The five Ruta 5 sections now in their second concession were awarded between eleven and twenty-two months before mandatory environmental assessment existed in Chile, and thirteen years before indigenous consultation did. Their second concessions are tendered with both fully in force.

That is not a nuance. It is a change of regime. The first round did not pay a cost the second must pay, and the budget differential cannot be read as though both had competed under the same rules.

Method note. The table above measures the presence of a requirement in a document, not its depth or its cost. It is a floor. Counting mentions does not establish how much each requirement weighs in a budget, and this paper does not claim otherwise; §11 returns to the point.

6. Why a contract maintains better than a budget

Before turning to the counterfactual, one mechanism needs stating, because it explains what follows and it is political-economic rather than technical.

A road whose maintenance competes for an annual appropriation is systematically more exposed to under-provision than one whose maintenance is a contractual obligation.

The reason is not that private operators are better managers. It is that the two claims on money have different standing. A concessionaire's conservation standard is an enforceable obligation with a counterparty, a supervising agency and a dispute mechanism. A ministry's conservation line is a discretionary appropriation that must be re-won every year against every other claim on the same budget — and above all against new construction.

That contest has a predictable direction. A politician is rarely remembered for having maintained a road well; they are remembered for the new one. The incentive at every budget cycle is therefore to build without maintaining. A cut to conservation is invisible in the year it is made: the road looks the same in December as it did in January. The deterioration arrives distributed across a decade, by which time it is someone else's administration and someone else's problem.

This is not a novel claim, and it should not be presented as one. Glaeser and Poterba, surveying the economics of infrastructure investment, put the underlying asymmetry plainly: project-by-project cost-benefit analysis "typically yields only modest returns for most new large-scale infrastructure projects. Returns for maintenance of existing infrastructure are typically much higher," and for the United States "the highest social returns come from maintaining existing infrastructure rather than from new projects" — a conclusion they note "has been a mantra for microeconomic transportation economists ever since" Gramlich.² Higher-return activity that is nonetheless under-provided is the signature of an incentive problem, not an information problem. And there is direct evidence that maintenance responds to electoral incentives: a 2023 study in Public Choice using mayoral election timing finds that local election cycles shift the timing and location of road maintenance projects.³ That study establishes political responsiveness rather than a build-versus-maintain bias as such, and is cited here for the former.

The economics literature also identifies the second mechanism, and it is not the one this paper is arguing. Engel, Fischer and Galetovic's point, as Glaeser and Poterba summarise it, is that in a PPP "charging users for infrastructure access creates incentives for better maintenance since the private provider doesn't get paid unless the roads are used, and the roads do not get used if they are in bad shape."⁴ Singh (2018) finds private road providers in India deliver smoother roads, with the difference attributed to the same operator bearing responsibility for construction and later maintenance.

Those are two distinct mechanisms — a revenue incentive and a commitment device — and they should not be conflated, because they have different failure modes. The revenue incentive works only where quality affects revenue. Glaeser and Poterba state the limit directly: private managers "have stronger incentives to cut costs, which can both reduce waste and reduce quality, especially when quality reductions do not lead to losses in revenues." That condition fails in two places that matter here. It fails in availability-based contracts, where the operator is paid regardless of usage — which is most of UK PFI, and where §2's handback problem was found. And it weakens as the term runs out, when revenue foregone from a deteriorating road falls below the cost of preventing it — which is precisely the maintenance cliff this paper could not measure.

So the two mechanisms are complements with a gap between them, and the gap is the terminal phase. The commitment device survives to the last day of the contract because the obligation is enforceable to the last day. The revenue incentive does not.

What would reverse this claim. It predicts that where maintenance sits on an annual appropriation, spending should fall below the level the asset requires, and should be sensitive to the budget cycle in a way that contractual maintenance is not. A jurisdiction that reverted assets to direct public operation and then sustained conservation at or above the contractual standard through a full political cycle would falsify it. Spain (§7) is a live test rather than a settled case: the reversion is recent and the 2027 wave has not happened yet. A reader who expects that test to come out the other way should say so now and check in 2032.

With that qualification, the contractual obligation is doing real work. It is not merely a legal formality that happens to attach to private operation; it is a commitment device that removes maintenance from annual discretion. This series has elsewhere criticised the rigidity of PPP commitments as a loss of fiscal flexibility — The Useful Silence argues that governments choose these contracts partly because their obligations escape measurement, and The Pipeline Under the Ceiling shows what happens when committed payments crowd a budget. Both criticisms stand. This is the other side of the same property: the rigidity that makes a commitment fiscally awkward is the same rigidity that protects the asset from the budget cycle.

An honest assessment has to hold both. And Spain provides the test.

7. Spain — the counterfactual where the asset does come back

Spain is the precise counterfactual to Chile because it has no Article 25. When a Spanish motorway concession expires, there is no statutory obligation to grant it again, and the asset genuinely reverts.

After those releases, 1,435 km of directly tolled motorway remained in the State network. A second wave is arriving: ten concession contracts awarded in 2007 over first-generation dual carriageways expire in December 2026, so that from 1 January 2027 the State assumes direct conservation of 993 km — 9.5% of its toll-free network — financed entirely from the general budget, at an estimated €80 million a year.

Three findings come out of the contrast, and the third is the one that matters.

First, without a duty to re-tender, the problem Article 25 prevents actually occurs. The expiry of the AP-1 on 30 November 2018 left the administration without a conservation contractor at the moment of transfer. The lesson was absorbed explicitly: for the 2026 wave, the stated objective is to run the tenders before the concessions end so as not to repeat it. That is, almost word for word, what Chilean law has required since 1996 — that the tender be held "with the anticipation necessary so that no gap exists between the two concessions." Chile legislated in 1996 what Spain learned the hard way in 2018.

Second, reversion does not remove the cost; it moves it. The toll disappears for the user. The conservation does not. It becomes a permanent budget line competing with everything else in the ministry — and it enters a budget already short, with the annual gap for adequate conservation of the State network estimated in the order of €540 million.

Third, and decisively, the concession model was financing a network reset that the budget does not replace. On the concessionaires' own figures, those corridors mobilised €2,504 million in "reset and network replacement" investment — equivalent to 2.3 times the Ministry of Transport's entire road infrastructure creation programme.

That is precisely the mechanism observed in Chile as a 2-to-4× second round. Re-tendering is the vehicle through which an ageing asset is financed up to current standard. Spain had that vehicle for as long as the concessions ran, and on reversion it lost it — at the same moment it moved maintenance into the one place where, per §6, it structurally loses.

None of which makes Spain's choice wrong. Tolls are regressive in incidence and politically costly; abolishing them on a paid-off road is a defensible decision that a democratic government is entitled to take. The point is narrower and harder: the decision was presented as the end of a cost, and it was a transfer of a cost — from users to taxpayers, and from an enforceable obligation to an annual appropriation.

8. What this does to fiscal accounting

Return to the frameworks. PFRAM projects the fiscal effects of a PPP over the concession period. IPSAS 32 recognises a service concession asset and unwinds the corresponding liability across the term of the arrangement. ESA 2010 and the Eurostat guidance assign the asset for the contract's duration. All three treat the end of the term as the end of the arrangement.

In Chile that is incorrect as a matter of law. Article 25 makes renewal the default and reversion the exception — an exception requiring a reasoned presidential decree on grounds of disuse or harm to the State. The economic substance is not a contract that terminates but a rolling arrangement: the term of the contract is not the term of the commitment.

Two consequences are worth separating.

There is a second, independent problem that §4.1 uncovered. Where a concession is awarded by least-present-value-of-revenue auction — 13 of the 17 second rounds here — the end date is endogenous to realised traffic. The contract ends when the bid revenue is collected, which may be years before the stated maximum. A framework that projects "over the concession period" is therefore not merely assuming a termination that will not happen; for a large and growing share of Chilean contracts it is treating a random variable as a date, and using its worst-case realisation as the point estimate. That is a modelling defect independent of Article 25, and it applies wherever LPVR-type auctions are used.

For liability recognition, the profile is wrong in shape, not merely in level. A framework that shows an obligation decaying to zero at year 30 is describing a legal event that, on the evidence of 41 contract generations, does not occur — and when the successor contract arrives it is longer (median 1.72×) and larger (median 3.25×). The State's exposure to the corridor does not end; it is re-struck on new terms.

For affordability, the consequence is sharper still, and it connects to the companion paper on the affordability envelope. An envelope built from signed contracts shows commitments running off as those contracts expire, and therefore shows headroom opening. Under Article 25, that headroom is legally pre-committed to be re-tendered on the same asset. It is not discretionary space for new projects. A ministry reading the declining path as room to commit elsewhere is reading a number that the statute has already spent.

This is not an argument that the frameworks are badly designed. It is an argument that they encode a jurisdiction-specific legal assumption as if it were a universal accounting fact, and that the assumption should be tested rather than inherited. The test is simple and every PPP unit can run it: does the governing statute make re-tendering mandatory at expiry? In Chile it does. In Spain it does not. The correct fiscal treatment differs accordingly, and no current framework asks the question.

Peru is a third case worth noting, because it has moved in Chile's direction on term. Law 32441, in force since December 2025, allows concession terms of up to 60 years — ten more than Chile's ceiling — while capping the accumulated stock of firm and quantifiable contingent commitments, net of revenue, at 12% of GDP. A 60-year term is more than the working life of most of the officials who will administer it.

9. The ceiling that is running out

Chile's fifty-year cap has an implication that the record is now beginning to show.

If the second concession is the mechanism through which an ageing asset is financed to current standard, then the term is a lever: a longer concession spreads a larger investment over more years of toll revenue, which is how a 3× budget increase is made bankable without a proportionate increase in tariffs or state payments.

That lever has a statutory limit, and on the busiest corridors it is nearly exhausted. Two second-round concessions are already at exactly fifty years — Ruta 5 Collipulli–Temuco and Ruta 5 Río Bueno–Puerto Montt, both at 600 months, the maximum Article 25 permits. Five more second rounds sit between forty and fifty years.

Under a variable-term auction the fifty years is a ceiling on duration rather than a schedule, and a contract that meets its revenue target early will end sooner. That does not soften the constraint; it sharpens it. A concession written to the statutory maximum is one whose sponsors judged that even fifty years of revenue might be needed to finance the works — which is a statement about how large the investment has become relative to what the corridor can earn. When the cap binds at the design stage, the term has stopped being available as a lever before the contract is even signed.

When the term can no longer absorb additional investment, only two levers remain: the tariff paid by users, or a direct payment from the State. Both are visible and both are politically expensive, which is precisely why the term lever was used first. The design question arrives sooner than it appears, and it arrives on the corridors that carry the most traffic — which are also the corridors where a tariff increase is most contested.

10. The illusion to abandon

There is a claim, made across the political spectrum in Chile and made explicitly in Spain, that once a concession has been paid for the road should become free. It is intuitive: the private party has recovered its investment and its return, the asset is built, the debt is repaid, and continuing to charge looks like rent extraction.

The evidence in this paper says the premise is wrong. At expiry the asset is not paid off in any sense that matters. It is:

• old — twenty-five to thirty years of traffic on a pavement and structures designed for a fraction of it;
• under-dimensioned — for demand that the concession itself helped generate;
• built to a superseded safety standard — designed when 100 km/h was the requirement, in a network now specified at 120;
• and unbuilt to environmental and social requirements that did not exist — awarded, in the Ruta 5 cases, before mandatory environmental assessment and thirteen years before indigenous consultation.

Bringing it current costs, on the Chilean record, between 1.8 and 3.9 times the original budget per kilometre, net of measured real construction-cost inflation. That is a measurement, not a position.

So the real question at the end of a concession is not tolls versus free access. It is who finances the next investment cycle, and by what route:

• Chile charges the upgrade to the users of the successor concession. Where the corridor can carry it on tolls, it does not appear as budget expenditure at all — nor, as §8 showed, as occupied space in the affordability envelope. Where it cannot, it returns as a subsidy commitment, which does appear.
• Spain transferred it to taxpayers and put it into the budget, where it is visible and where it competes — and, per §6, where it loses.

Neither route is free. The difference is who pays and whether it can be seen when the decision is made. A government that abolishes a toll while telling voters the cost has ended has not eliminated an obligation; it has moved one, from a contract that could be enforced to an appropriation that must be re-won every year against a new bridge someone would rather cut a ribbon on.

11. Limitations

The maintenance cliff remains unmeasured in Chile. This paper reports that the data to test it do not exist in usable form (§2); it does not claim the hypothesis is false — and the UK's National Audit Office found a condition problem in four of the nine PFI projects it could observe. If a supervising agency holds per-contract conservation spending by year — and Chile's does, internally — the test is a week's work and should be run.

The term comparison is not like-for-like, and the paper says so where it makes it. In 13 of 17 transitions the first-round figure is a fixed contract term and the second-round figure a maximum under a variable-term auction (§4.1). The 1.72 median therefore overstates realised lengthening by an unknown amount, and is not offered as an estimate of duration. Closing this would require the actual termination dates of concessions that have already ended under LPVR, which are knowable and should be collected. The claims that do not depend on it — the budget multiples, the statutory ceiling, and the whole of §3 and §8 — are unaffected.

The UK comparison is suggestive, not matched. UK PFI is largely availability-based social infrastructure and Chilean concessions are largely user-financed roads. §6 argues that this difference should matter for maintenance incentives, and that argument is drawn from the literature rather than tested here. Reading the NAO's four-of-nine as a prediction for Chilean roads would be exactly the error the section warns against.

The matched samples are small. Seventeen transitions have an observable term on both sides; five have a comparable official budget; five support a per-kilometre comparison. The corpus contains 41 generations and 3,687 documents, so the constraint is extraction effort, not availability, and the sample should be extended before these multiples are treated as parameters rather than as a documented pattern.

The cost index does not span the gap. The published road-cost series begins in March 2016 and cannot cover 1996–2016, where most of the price effect between the two rounds must lie. The 8.2% real excess measured here is a floor. Reversing the conclusion of §5.3 would require the real excess of 1996–2016 to have been of an entirely different order from that of the last decade, which is possible but would need to be shown.

Environmental and social requirements are measured by presence, not weight. The classification in §5.4 counts whether a requirement appears in a tender document. It does not measure how much it costs. The direction is unambiguous and the regime dates are decisive; the magnitude is not established.

One corridor is excluded on data quality. Santiago–San Antonio (R-78) has a first-round length recorded as 5.0 km in the source dataset, which is impossible for a corridor of roughly 130 km. Rather than normalise by a figure known to be wrong, it is excluded from the per-kilometre table and reported only at gross.

And a trap worth flagging for anyone using this corpus. Chilean executive reports describe the current concession, not the historical one. The report filed under the folder for the first concession of Ruta 5 Talca–Chillán states the budget and length of the second. First-round figures here come only from Official Gazette publications of the period.

The escape valve was not exhaustively searched. §3 reports no applied instance of Article 25's disuse-or-harm exception in the corpus, but the corpus is built from concession dossiers and would not necessarily hold a decree declining to re-tender. A negative finding there is weaker than the positive findings elsewhere in this paper, and it is the single claim most worth an independent check.

Finally, Chile-dependence. The core finding is a reading of one country's statute and one country's record, with Spain as counterfactual and Peru and Colombia as reference points. That is a proof of existence — that a legal duty to re-tender exists, is operative, and produces a measurable pattern — not a cross-country regularity. Whether other concession regimes carry equivalent provisions is an open and answerable question, and §12 makes it a recommendation.

12. The policy lead

For ministries of finance and budget offices.

1. Ask whether your law makes re-tendering mandatory. It is a one-paragraph question with large consequences for how PPP liabilities should be projected, and no current fiscal-risk framework asks it. If the answer is yes, the concession term is not the life of the obligation and your models are shaped wrongly at the end.
2. Stop treating expiry as freed fiscal space. Where a duty to re-tender exists, the headroom that opens as contracts run off is already spoken for. An affordability envelope that shows it as discretionary is overstating room to commit.
3. Budget the re-tender as an investment decision, not an administrative one. On the Chilean record it costs 1.8 to 3.9 times the original per kilometre. That is a capital programme arriving on a predictable schedule, and it should be planned as one.

For PPP units and infrastructure ministries.

1. Collect conservation spending per contract per year, and publish it. This is the missing variable of §2. Without it no government can verify whether its concessionaires are maintaining assets as required, and no handback can be priced.
2. Measure the asset before the term ends, not after. The condition survey that determines what the successor contract must fix should exist while there is still a counterparty with an obligation and a bond.
3. Decompose the successor budget explicitly — capacity, standard, real cost, environmental and social requirement — and publish the decomposition. It is the only way to answer the accusation that re-tendering is a subsidy to incumbents, and on this evidence the answer is usually favourable to the State.

For standard-setters — IPSAS, Eurostat, and the PFRAM maintainers.

1. Handle variable-term concessions honestly. Where the term is set by a revenue target rather than a date, the concession period is a random variable. Projecting over its maximum and reporting the result as a point estimate overstates duration and understates uncertainty. The fix is to model the distribution, or at minimum to state which convention was used.
2. Make the reversion assumption explicit and testable. All three frameworks currently assume the arrangement ends with the term. Where statute says otherwise, the treatment should follow the statute. At minimum the assumption should be a documented input rather than an unstated default — a change that costs nothing and makes a real difference wherever a Chilean-style provision exists.

And for the public debate.

1. Retire the claim that a paid-off concession should be free. It is the most common thing said about concessions at the end of their term, in Chile and in Spain, and it is not true. The asset needs a new investment cycle. The legitimate argument is about who funds it and how visibly — and that argument is worth having on its own terms, rather than under the cover of a promise that the cost has ended.

Appendix A — data, sources and reproducibility

Corpus. 132 concession dossiers covering 109 assets, assembled from the Ministry of Public Works concessions portal and from award decrees retrieved from LeyChile, the official legal database. 18 assets have more than one generation; those 18 span 41 generations and 3,687 documents.

Term data. plazos_bases_ds.csv — contractual term in months for 130 contracts, each with the source (award decree or tender specifications) and the quoted passage preserved. Generation pairing and the transition table are produced by generaciones.py, output data/generaciones_transiciones.csv.

Budget data. presupuesto_oficial.csv (official estimated budget from Official Gazette award publications) and presupuesto_fichas.csv (from Ministry technical datasheets). The ITC field in plazos_bases_ds.csv is Ingresos Totales de la Concesión — concession revenue, not capital expenditure — and is reported separately and never mixed with budget figures.

Length data. longitudes_corredores.csv, curated from the kilometre-post definitions in tender specifications, with second-round lengths taken from current executive reports and technical datasheets. Per-kilometre analysis: por_kilometro.py, output data/por_kilometro.csv.

Construction cost. Índices de reajuste polinómico MOP, Planning Directorate, "Infraestructura vial y portuaria / General", published at mop.gob.cl; two sheets on a December 2020 = 100 base chained without rebasing, 112 monthly observations, March 2016 to June 2025. UF values from the Chilean statistical series for the same dates. Analysis: costos_construccion.py, output data/costos_construccion.csv.

Legal sources. Decree with Force of Law No. 900 of 1996 (consolidated text of the Concessions Law), Article 25. Law 19.300 of 9 March 1994. Supreme Decree 30 of the General Secretariat of the Presidency, 27 March 1997, published 3 April 1997 (regulation of the environmental assessment system). ILO Convention 169, in force in Chile from September 2009. Comptroller General Official Letter E557569 of 25 October 2024. For Peru, Law 32441, published 16 September 2025, in force from 25 December 2025 with Supreme Decree 316-2025-EF. For Colombia, Law 1508 of 2012, Article 26.

Spanish figures are drawn from the Ministry of Transport and Sustainable Mobility's published information on State-dependent tolls and from Spanish economic press reporting on the 2018–2021 reversions and the expiry of the ten first-generation concessions in December 2026. The €2,504 million network-reset figure is as declared by the concessionaires and should be cited as such, not as an audited or official number.

Literature and external evidence. Glaeser, Edward L. and James M. Poterba, "Economic Analysis and Infrastructure Investment," NBER Working Paper 28215, December 2020 (quoted passages verified against the working paper text). Engel, Eduardo, Ronald Fischer and Alexander Galetovic, "Least-Present-Value-of-Revenue Auctions and Highway Franchising," Journal of Political Economy 109(5), 2001; and The Economics of Public-Private Partnerships: A Basic Guide, Cambridge University Press, 2014 — the latter's maintenance-incentive argument is quoted here as characterised by Glaeser and Poterba (2020), not from the book directly, and is flagged as such. "Road maintenance over the local election cycle," Public Choice 198(1), 2024, doi:10.1007/s11127-023-01115-3. National Audit Office, Managing PFI assets and services as contracts end, HC 369, 5 June 2020.

Null result. The absence of a conservation-spending series was established across 60 audited annual reports of concession companies filed with Chile's financial regulator, a flow-accounts dataset covering five urban concessions, and 146 supplementary agreements between the State and concessionaires. The negative finding is reported in §2 rather than omitted.

Appendix B — glossary

Article 25 — the provision of Chile's Concessions Law capping concession terms at fifty years and requiring that works be granted in concession again on expiry, without a gap.

Bridge concession — a short contract (48 months in the cases here) covering existing installations while a larger successor concession is tendered, used to satisfy the no-gap requirement.

ESA 2010 — the European System of Accounts, which governs whether a PPP asset sits on the government balance sheet in EU statistics.

Handback — the return of a concession asset to the grantor at the end of the contract, together with the condition standards that govern it.

IPSAS 32 — the international public sector accounting standard on service concession arrangements.

ITC (Ingresos Totales de la Concesión) — present value of total concession revenue; the bid variable in Chilean variable-term auctions. Not capital expenditure.

LPVR (least present value of revenue) — an auction design in which bidders compete on the present value of revenue they require and the concession ends when that value is collected, producing a variable term. Recorded in Chilean tender data as "VPI / ITC (plazo variable)".

PFI (Private Finance Initiative) — the United Kingdom's PPP programme, largely availability-based social infrastructure.

PFRAM — the PPP Fiscal Risk Assessment Model developed by the IMF and the World Bank.

Re-tendering / re-concessioning (relicitación) — granting an expired concession again to a new or the same operator, under a new contract.

SEIA — Chile's environmental impact assessment system, mandatory from April 1997.

UF (Unidad de Fomento) — Chilean unit of account indexed to consumer prices, in which concession budgets and payments are denominated.

Vigencias futuras — Colombian authorisations to commit future budget years, the instrument through which APP commitments are approved.