The Silent FX Liability — Multi-Currency PPPs as Unhedged Ministry-of-Finance Exposure

A planner's framework for pricing and placing the currency mismatch in PPP and IPP contracts — retain, hedge, guarantee, or push back — with a reference implementation on the Austral platform and post-mortems of six devaluation cascades

Author: David Duarte Arancibia, Austral Intelligence Date: June 2026 Audience: Ministries of Finance and debt-management offices in frontier and emerging markets; reserve managers; sovereign rating analysts; MIGA (the Multilateral Investment Guarantee Agency), multilateral development bank (MDB) and development finance institution (DFI) guarantee desks; supreme audit institutions Companion software: Austral platform — https://platform.austral-intelligence.com Series position: Paper 6 of the Austral series. It extends the orchestrator partition of Paper 3 (From Gatekeeper to Orchestrator) and the planner's-perspective decision logic of the Third-Party Guarantee paper to the one exposure those papers left implicit: the currency mismatch a multi-currency PPP carries onto the public balance sheet.

Abstract

A public-private partnership (PPP) in a frontier or emerging market is usually financed in hard currency — the lenders want US dollars or euros — while its revenue is local: a tariff, a farebox, an availability payment set in the domestic unit. The gap between the two is an open foreign-exchange position, and someone holds it. This paper is about who, and at what price.

The position is silent. It does not appear on the debt-stock report, it does not trip a fiscal-rule threshold, and — under the cash-basis and risk-classification regimes in actual frontier use, and only partly under full-accrual IPSAS (International Public Sector Accounting Standards) — the fiscal-commitments machinery flags an explicit FX guarantee as a contingent liability but books an FX-indexed availability payment, which carries the same currency exposure, as ordinary spending. The exposure surfaces only when the currency falls — and then it is large, sudden, and already on the sovereign's account.

We frame the Ministry of Finance's choice at the gate as a single decision: how to price the mismatch ex ante and how to place it. There are four placements — retain the open position; hedge it with a cross-currency swap; write a government FX guarantee; or push it back to the sponsor through tariff indexation — and this paper's central finding is that they are not what they look like. We built a reference implementation, now shipped in the Austral platform, that prices all four side by side from a Monte-Carlo distribution of the FX loss, and it makes the point in the numbers: retain and a government FX guarantee are the same FX exposure — identical expected payout and identical tail, holding the contracted obligation fixed — and differ chiefly in whether the liability is booked. A guarantee does not transfer currency risk off the public balance sheet; it relocates it from invisible to visible. Only a hedge or a genuine push-back moves the risk off the sovereign at all, and each costs money: the swap, where a market exists, or the sponsor's risk premium, where it does not. On a US$1.2 billion dollar-indexed obligation with no swap market, for instance, the instrument prices an expected FX loss of US$433 million — 37% of the open position — and a 95th-percentile loss of US$661 million (56%), and, with no hedge to be had, recommends booking the exposure as an explicit guarantee rather than holding it silently. The same exposure has a bankability face: because the debt is serviced in hard currency from local revenue, a devaluation erodes debt-service coverage directly — a project designed at a 1.40× DSCR breaches a 1.20× lock-up covenant on a ~17% devaluation, and only near-full tariff FX-indexation or a hedge keeps it bankable (§3.4).

We demonstrate the instrument on three sectors — a power independent power producer (IPP), an availability-payment hospital, and a toll road — and anchor it in post-mortems of six real devaluation cascades (Argentina 2018, Ghana 2022, Sri Lanka 2022, Egypt 2016/2022, Pakistan 2022, Turkey 2021–22), every figure carried with its verification status. The cases sort into two archetypes that the framework predicts: where the contract placed the FX risk on the state by design (Turkey's euro correction factor, Pakistan's dollar-indexed capacity payments, Egypt's indexed tariff, Argentina's World Bank–backstopped dollar PPAs), a devaluation lands on the public budget rather than on the project; where the mismatch was simply unhedged (Ghana, Sri Lanka), it crystallised as arrears and utility losses — and, where the sovereign was already fragile, helped tip the path to default. And the asymmetry now runs backwards: because capable ministries already price and provision their explicit guarantees, the fiscal machinery's incentive rewards the unflagged indexed payment — the very exposure least safe to leave silent. The policy lead is the same in both: make the silent position visible, price it, and book it — because the alternative is not avoidance, it is surprise.

Keywords: foreign-exchange risk · multi-currency PPP · independent power producer · availability payment · cross-currency swap · government guarantee · tariff indexation · fiscal commitments and contingent liabilities · Ministry of Finance · sovereign balance sheet.

Feedback & discussion. This paper is published openly so that Ministries of Finance, debt-management offices, rating analysts, MDB/DFI guarantee desks, and the academic community can challenge and use it on real programmes. We especially welcome additional devaluation case studies and corrections to any figure carried here as single-source.

• Comment in-page at https://austral-intelligence.com/research
• Email the author: [email protected]
• Live demo or walkthrough on request

Table of contents

1. The decision problem — the silent gap
2. Why the mismatch is silent — and why that is the problem
3. The instrument — pricing the open position
4. The four placements — and where each one actually lands the risk
5. The decision rule
6. Wiring it into the fiscal envelope
7. Architecture and reproduction in ten minutes
8. Three worked examples
9. The evidence — six devaluation cascades
10. Limitations
11. The policy lead — make the silent liability visible

Appendices A, B, C, D follow the page break.

1. The decision problem — the silent gap

Why this matters: the moment a Ministry of Finance lets a multi-currency PPP is the moment it acquires a currency position it did not choose and rarely prices. Unlike a procurement red flag, the mismatch is not a question of integrity; it is a question of arithmetic. The hard-currency obligations are fixed; the local-currency funding is fixed at a spot rate that will move. The gate's job is to see the position, price it, and decide where to put it — before a devaluation decides for it.

1.1 The position

Consider the financial anatomy of a typical frontier-market PPP. The sponsor raises debt from international lenders, who require repayment in US dollars or euros — the currencies in which they fund and are regulated. Debt service, and often a capacity payment or an availability payment, is therefore denominated in hard currency. The project's revenue, by contrast, is local: a toll, a tariff, a farebox, a budget line, all set and collected in the domestic unit. At financial close the two are reconciled at the prevailing spot rate, and the contract looks balanced.

It is not balanced. It is matched at a point and open over time. The hard-currency obligation runs for twenty-five years; the exchange rate that funds it from local revenue will not stay at its closing value. The difference between the hard-currency obligation and the local-currency funding, as the rate moves, is an open FX position — a long-dated, one-sided exposure to the domestic currency falling. Someone in the structure holds that position: the lender, the sponsor, or — most often, as §9 shows — the state.

1.2 Who holds it

The position can sit in three places, and which one depends entirely on the contract:

• Directly on the Ministry of Finance, when the MoF is the offtaker and the obligation it owes is hard-currency-indexed — the clearest case being an availability-payment PPP whose payments carry an explicit FX or euro correction factor, funded from a local-currency budget.
• Contingently on the Ministry of Finance, when a private offtaker bears the obligation but the MoF has written a guarantee — a minimum-revenue guarantee, an FX guarantee, a take-or-pay backstop — that pays out when the currency falls.
• On the sponsor, when the contract genuinely leaves the FX risk with the private party, who funds the hard-currency obligation from a tariff that does not fully index to the exchange rate.

The first two are the same economically — both leave the loss with the public — and only the third moves it. That equivalence, developed in §4, is the analytical core of this paper.

1.3 The gate decision

At the fiscal-approval gate, the Ministry of Finance must decide how to handle this position before it underwrites the deal. We frame the decision as a choice among four placements:

• Retain — hold the open position and fund the shortfall if and when the currency falls.
• Hedge — buy a cross-currency swap and convert the floating FX cost into a certain one.
• Guarantee — let a private party be the nominal payer but stand behind the FX gap with an explicit guarantee.
• Push back — require the contract to leave FX risk with the sponsor, who prices it into the tariff.

The right choice depends on the size of the exposure, the cost and availability of a hedge, the price the sponsor would charge to bear the risk, and the state's own capacity to absorb a tail. The rest of this paper builds the instrument that prices these four side by side (§3–§4), states the decision rule (§5), wires the chosen placement into the fiscal envelope (§6), and tests the whole thing against three sectors (§8) and six real crises (§9).

2. Why the mismatch is silent — and why that is the problem

Why this matters: a risk that is priced and visible can be managed; a risk that is neither is simply waiting. The defining feature of the FX mismatch is not that it is large — guarantees and sovereign debt are large too — but that the standard fiscal machinery looks straight past it. Closing that blind spot is most of the contribution.

2.1 It is off every report the gate reads

The mismatch is invisible to the instruments a Ministry of Finance actually consults. It is not a debt instrument, so it does not appear on the public debt-stock report. It is not a realised cash outflow at close, so it does not register against a deficit or a fiscal-rule ceiling. And it is embedded inside a contract — an availability-payment schedule, a capacity-charge formula, a tariff-indexation clause — rather than stated as a number. The exposure exists from the day of signature, but nothing on the gate's desk says so.

2.2 The recording asymmetry — and how far it goes

Two contracts that place the same one-sided currency exposure on the public balance sheet are recorded differently. A written FX or government guarantee is a recognised contingent liability — reported as a memorandum item in the government-finance statistics and disclosed under IPSAS 19⁴. An FX-indexed availability or capacity payment is a firm liability — under PFRAM an on-balance-sheet PPP becomes "a firm liability for government… classified as loans in public sector debt statistics"³ — and its embedded currency exposure is surfaced nowhere by default. That the two carry the same exposure is not our claim alone: the practitioner literature already observes that "if a project's revenues are indexed to the exchange rate, a currency swap is effectively built in to the contract… the currency risk is transferred to the buyer, often a state utility or government entity."⁵ The equivalence is older still in statistical doctrine — the IMF Balance of Payments Committee holds that a foreign-currency-indexed instrument should be "treated as though [it is] denominated in that foreign currency," precisely to "treat identically instruments that have economically equivalent characteristics"⁶ — though there it is drawn for external-debt classification, not the contingent-liability register, and with no recording-asymmetry implication. We take this indexation equivalence as an established premise; the increment here is to pin it to a government FX guarantee and tie it to the recording asymmetry that follows (the G20 infrastructure work likewise now names FX-indexed payments as a source of "potentially substantial fiscal and foreign exchange liabilities,"⁷ without pricing the placement or drawing the inversion of §2.4).

How completely the indexation leg escapes depends on the reporting regime, and the honest statement is regime-scoped:

• Cash-basis reporting — the frontier and emerging-market majority — books the FX-indexed payment as annual cash expense; its embedded FX is surfaced nowhere. The asymmetry holds fully.
• ESA 2010 / Eurostat classifies a PPP on- or off-balance-sheet on risk allocation alone — the test turns on construction plus availability or demand risk, leaving exchange-rate and indexation effects outside the classification¹⁹ — so the embedded FX is invisible to the headline deficit and debt. (ESA binds EU members, not the emerging-market sovereigns this paper addresses; we cite it as the reference framework.)
• Full-accrual IPSAS captures it only partially. IPSAS 4 retranslates the recognised concession liability at each period's closing rate — but that is a backward-looking spot revaluation through the surplus/deficit, not a forward present value of the expected loss; and the future availability-payment service leg is not a recognised liability under IPSAS 32, so its multi-decade currency exposure rides off-balance-sheet²⁰. (FX-indexation is not a separable embedded derivative to be fair-valued: IPSAS 41 treats an FX-denominated stream inside a host debt instrument as closely related and never bifurcates it²⁰.) Accrual narrows the asymmetry; it does not close it.
• Even PFRAM 2.0, the leading fiscal-risk tool, auto-calculates an FX-indexed tariff's adjustment only once the analyst sets the per-tariff exchange-rate-indexed flag; and in its macro exchange-rate shock, "this is not automatically modeled… the user will be able to manually link various cost and revenue components to the exchange rate," the manual states³. The firm-liability classification carries no FX flag by default, so the devaluation exposure is dropped from the sensitivity analysis unless someone remembers to set the toggle or wire the shock by hand.

The doctrine prices the guarantee leg and, across the regimes in actual use, lets the indexation leg slip — and §9 shows that the indexation leg is where some of the largest losses have landed.

2.3 The cost of silence is surprise

The practical consequence is that the mismatch is discovered, not managed. It does not enter the gate's verdict, the medium-term fiscal framework, or the contingent-liability provision, because no instrument forced it to. It then surfaces, all at once, in a devaluation — when the local-currency cost of the hard-currency obligation jumps and the funding set at the old rate no longer covers it. The six cascades of §9 are, in this reading, not six accidents but six instances of the same omission: a position that was always there, never booked, suddenly due. The instrument this paper ships exists to convert that surprise into a line item at the gate.

2.4 The asymmetry now points the wrong way

The recording asymmetry between recognised firm obligations and disclosed contingent ones is not itself new — it is the foundation of fiscal-risk doctrine⁴. What is new is its direction. The canonical warning runs the opposite way: guarantees are dangerous precisely because they look free. A guarantee "remains something of a free good" until it is called⁴; it is "more attractive… because of [its] hidden nature," and the fiscal-risk matrix that sorts liabilities into recognised and contingent cells exists to drag that hidden exposure into view⁸. It is the kind of "stealth support" that bypasses the scrutiny built into the budget process⁹; and the modern transparency literature presses Ministries of Finance to recognise, value, and provision it¹⁰. The capable ministries have done exactly that — Chile prices its exchange-rate guarantees with an option model and carries them in a published contingent-liability register; Colombia funds them through a dedicated contingent-liabilities account⁴ — and the discipline for doing so is now codified¹¹. The effect is that the explicit FX guarantee, in a serious Ministry of Finance, is now the scrutinised instrument.

The FX-indexed firm payment is not. It carries the same currency exposure, but nothing forces it onto the contingent-liability register, no option model values it, no fund provisions it — and, as §2.2 showed, most reporting regimes in use do not surface it at all. So the incentive runs backwards: the structure a Ministry of Finance is rewarded for choosing — the one that draws the least fiscal-accounting attention — is the silent indexed payment, which is the exposure it can least afford to leave unmanaged. We have not found this inversion stated anywhere in the fiscal-risk literature — which runs the classic direction above and treats the indexation channel only in passing if at all: that the maturation of explicit-guarantee discipline is exactly what now rewards the unflagged indexed payment. Naming that inversion, and pricing the embedded leg on the same footing as the guarantee, is the contribution of this paper. The instrument of §3–§5 is what makes that leg visible; this is why it matters.

2.5 Prior work, and what is new here

Three of the ingredients here are established, and the paper cites them as such. That an FX-indexed obligation carries the same currency exposure as an FX-denominated one is settled — in balance-of-payments statistical doctrine⁶, in project-finance practice⁵, and in the G20's infrastructure work, which now names the channel a potential fiscal-and-FX liability⁷. That recognised firm obligations and disclosed contingent ones are recorded asymmetrically is the foundation of fiscal-risk doctrine — the fiscal-risk matrix⁸, the contingent-liability literature⁹, the recognition-versus-disclosure analysis¹⁰ — and the accounting standards confirm it (GFSM, IPSAS 19/32, ESA/Eurostat¹⁹). And valuing a guarantee as an option, capping it, and minimising a tail measure are standard tools¹³¹⁴¹⁵.

What is new is the join. No prior work we have found (a) proves that silent retention and a fee-free, uncapped FX guarantee are the identical loss random variable, with the fee/cap dominance corollary (§4.2); (b) applies the recording asymmetry to the embedded FX leg of a firm indexed payment — surfaced nowhere by default, while the equivalent guarantee is captured (§2.2); or (c) draws the inversion, that the very maturation of explicit-guarantee discipline¹¹ now rewards the unflagged indexed payment (§2.4). Engel–Fischer–Galetovic supply the one explicit normative answer for who should bear currency risk — the concessionaire, by default¹² — which this paper prices as one of four placements rather than assuming. The rest of the paper makes the silent leg visible and puts a number on it.

3. The instrument — pricing the open position

Why this matters: a decision framework is only as good as the number it puts in front of the decision-maker. The instrument's first job is to take a contract's hard-currency obligation stream and a view on the currency, and return the one figure the gate never sees: the present value of the FX loss the public is carrying, with its tail.

3.1 What it takes and what it returns

The placement-comparison layer — shipped as a module in the Austral platform's financial-model package — takes a normalized FX-exposure record: the hard-currency obligation stream year by year (the debt service, capacity payment, or availability payment whose value is at FX risk), the hard and local currency codes, the closing spot rate (informational — the loss is driven by the depreciation path, not the spot level), a discount rate, and a view on the currency expressed as a mean and volatility of annual depreciation. It returns, for each of the four placements, an expected cost and a tail cost in hard-currency-equivalent present value, a flag for which party holds the risk, a feasibility flag, and a recommended placement. The same object that produced the worked examples in §8 is the one a Ministry of Finance would call at its gate.

3.2 The FX-loss distribution

The engine's core is the distribution of the FX loss on the open position. We model the exchange rate as a path of annual depreciations drawn from the gate's view (a mean trend and a volatility around it), and we measure, for each simulated path, what the currency move costs the public.

The measurement is the one subtlety worth stating precisely. At close, the local funding is fixed to cover the obligation at the spot rate. If the currency depreciates so that the cumulative depreciation to year t is dₜ (the rate has risen by a factor of 1 + dₜ), the local funding now buys only 1 / (1 + dₜ) of the hard-currency obligation due that year. The shortfall the public must find, valued in hard currency at the prevailing rate, is therefore the obligation times dₜ / (1 + dₜ) — a quantity bounded by the obligation itself: you can lose at most the whole thing if the currency collapses to nothing, and you gain if it appreciates. Discounting that shortfall stream and averaging across simulated paths gives the expected FX loss; the chosen percentile gives the Value-at-Risk — the loss in a bad-but-not-worst devaluation. These two numbers — a mean and a tail — are what the four placements are then priced against.

3.3 Why a distribution and not a point

A single devaluation scenario would understate the decision. The reason the FX mismatch is dangerous is precisely its tail: the expected loss may be tolerable while the 95th-percentile loss is not, and a Ministry of Finance with thin reserves cares about the second number more than the first. The instrument therefore reports both, and the decision rule of §5 turns on the gap between them. (The engine is self-contained and seed-reproducible; it composes with the platform's Monte-Carlo guarantees module, which prices a defined FX guarantee against a fuller macro path, but does not require it.)

3.4 The bankability break-point — the mismatch as a coverage problem

The open position is not only a silent liability; it is a bankability problem, and the two are one exposure read two ways. The SPV services its debt in hard currency from local revenue, so a devaluation inflates the local cost of debt service while the tariff — unless it is FX-indexed — does not follow. Debt-service coverage therefore erodes with the un-passed-through share of the move: with a tariff that passes through a fraction φ of each depreciation, DSCRₜ = (design DSCR) × ∏ₛ (1 + φ·depₛ)/(1 + depₛ), which at φ = 0 is simply the design ratio divided by the cumulative depreciation¹⁶. The break-point is brutally low. A project sized to a 1.40× DSCR against a 1.20× lock-up covenant breaches on a 16.7% cumulative devaluation (design ÷ covenant − 1, path-independent), and the cascades of §9 ran 50–100%+ — so coverage does not merely breach, it collapses (a 50% move takes 1.40× to 0.93×; a doubling, to 0.70×). The cure runs on the same axis as the placements: only near-full FX-indexation of the tariff (φ → 1, the push-back of §4.4) or a hedge holds the covenant, and the open share 1 − φ is exactly the silent position this paper prices. So bankability bounds the menu from below — a project funded with un-indexed local revenue against hard-currency debt is barely bankable on retention once the currency view is realistic (§8.6).

4. The four placements — and where each one actually lands the risk

Why this matters: this is the analytical heart of the paper, and the place where the instrument earns its keep. The four placements look like four different things a Ministry of Finance can do with the FX risk. Priced properly, two of them are the same thing wearing different clothes, and the framework's job is to stop a gate from mistaking a relabelling for a transfer.

4.1 Retain

Retaining the position is the default that happens when nobody decides. The Ministry of Finance — as offtaker, or as the ultimate funder behind a public offtaker — holds the open position and pays the FX shortfall as it arises. The cost is the expected FX loss, and the exposure carries the full tail: in a severe devaluation the public pays the Value-at-Risk. The risk lands on the sovereign, and — this is the operative word — implicitly. Nothing books it; it is the silent liability of §2, sitting unpriced on the public balance sheet until a currency moves.

4.2 The government FX guarantee — the same exposure, made visible

A government FX guarantee looks like the opposite of retaining: a private sponsor is the nominal payer, and the state merely stands behind the currency gap. But trace the cash. When the currency falls, the guarantee pays the sponsor the shortfall — the same shortfall the MoF would have funded directly under retention. The expected payout is the expected FX loss; the tail of the payout is the same Value-at-Risk. As an FX exposure, a government FX guarantee and retention are one position — identical expected payout and tail, holding the contracted obligation fixed. The engine prints them with identical expected cost and identical tail, and a unit test in the shipped module asserts exactly that equality, because it is the claim the rest of the paper rests on. Two scope conditions belong here, and both cut the guarantee's way. First, the equality is in expected cash flow under the modelled view, not in legal character: an enforceable, booked guarantee is a contractual claim a counterparty can pursue, where silent retention leaves the state the (sometimes valuable, sometimes costly) option to stretch or renegotiate a budget line — Pakistan's circular debt is precisely that deferral. Second, the equality holds for an uncapped, fee-free guarantee that pays the whole shortfall; a deductible, a payout cap, a trigger band, or a guarantee fee makes the guarantee strictly cheaper or thinner-tailed than retention, never worse — a refinement the engine prices and §8.5 demonstrates.

The decisive difference, and it is the case for the guarantee, is that it is explicit. The guarantee is a booked contingent liability — it can be provisioned, aggregated across the portfolio, reported to the legislature, and managed. Retention is the same exposure unbooked. An explicit, enforceable backstop can also earn a sponsor or lender de-risk discount that silent retention never does — a lower required return priced into the obligation the public ultimately pays — so in total contracted cost the guarantee is, if anything, the cheaper of the two. For that reason the guarantee dominates silent retention: unconditionally on visibility and manageability (the same FX exposure, now booked), and weakly on cost — equal in the fee-free, uncapped limit, strictly cheaper with any fee, cap, or de-risk discount. The lesson is not "never guarantee FX" — it is "if you are going to hold the risk, hold it explicitly, because a guarantee you have priced is a liability you can manage, and a mismatch you have not priced is a surprise you cannot."

The equivalence, formally. The "same exposure" claim is a one-line theorem, not a numerical coincidence. Let the FX-exposed obligation be \(O_t\) and, on FX path \(\omega\), the discounted open-position loss

\[L(\omega) \;=\; \sum_t \frac{O_t\,d_t(\omega)/(1+d_t(\omega))}{(1+r)^t},\]

where \(d_t\) is cumulative local depreciation and \(r\) the discount rate. Write each placement as the sovereign's cash-outflow random variable: retention \(C_{\mathrm{ret}}=L\); the guarantee with fee \(f\) and cap \(c\) (as shares of the open-position net present value, \(V\)), \(C_{\mathrm{guar}}=\min\{L,\,cV\}-fV\).

Proposition (placement equivalence). With \(f=0\) and \(c=\infty\), \(C_{\mathrm{guar}}(\omega)=L(\omega)=C_{\mathrm{ret}}(\omega)\) for every path \(\omega\). The two placements are the identical random variable, so their expected cost, their Value-at-Risk, and every law-invariant risk measure (any measure that depends only on the loss distribution) coincide. They differ only in booking.

Corollary (dominance). For \(f\ge0\) and \(c<\infty\), \(C_{\mathrm{ret}}-C_{\mathrm{guar}} = \max\{0,\,L-cV\}+fV \ge 0\) pathwise — strict on every path under a positive fee (by \(fV\)), and additionally where the cap binds (\(L>cV\)). The guarantee weakly dominates silent retention in every risk measure, strictly under any positive fee or finite cap, and it is booked.

Proof. \(\min\{L,\infty\}-0=L\) pathwise; distributional equality and the equality of every law-invariant measure follow because the two outcomes are the same measurable function of \(\omega\). The corollary is \(\min\{L,cV\}-fV\le L\). \(\square\)

The roots are standard and the increment is not. The guarantee-as-written-put isomorphism that makes the result intuitive is Merton's¹³; capping a guarantee to bound public exposure with a cost-benefit test is Brandão & Saraiva's¹⁴ — but for revenue, valuing a single instrument, and without identifying retention with an uncapped guarantee as the same random variable. The field's mental model in fact runs the other way: the fiscal-risk matrix files a firm payment and a guarantee in different cells⁸. The equivalence as a pathwise identity, and the fee/cap dominance corollary, are — as far as the literature we surveyed goes — new.

The content is economic, not a trick: FX risk inheres in the contracted obligation, not in who nominally holds the paper. Any placement that leaves the obligation's currency gap on the sovereign — directly (retain) or via a full backstop (guarantee) — is the same cash flow; only hedge (which makes the cost deterministic, \(C=h\)) and push-back (which moves it to the sponsor at the premium \(\pi\) it charges, \(C=\lambda(\mathbb{E}[L]+\pi)\)) change the sovereign's risk. The shipped engine embodies the Proposition exactly: a guarantee with fee = cap = 0 returns the retain expected cost and tail to the cent, and the module's unit test asserts that equality — because, with §2's recording asymmetry, it is the whole scandal: provably the same risk, demonstrably different visibility, and the rules reward the invisible one.

4.3 Hedge

A cross-currency swap is the only placement that genuinely removes the risk from both the sovereign and the sponsor: a counterparty takes the currency exposure in exchange for a fee, and the public's cost becomes certain — no FX tail, though it substitutes a smaller (and usually collateralised) swap-counterparty and convertibility risk in its place. In a deep, efficient market the swap's all-in cost is close to the expected depreciation already priced into the forward, so hedging buys away the tail at roughly the cost of the mean. The catch is structural: frontier and emerging markets frequently have no accessible long-dated cross-currency swap market at all, or one priced punitively for twenty-five-year tenors. The instrument therefore carries a feasibility flag, and §8 shows the difference it makes — a hedge that would be the cheapest option on paper is simply unavailable in two of the three worked cases.

4.4 Push back via tariff indexation

The fourth placement leaves the FX risk with the sponsor: the contract does not fully index the tariff to the exchange rate, so the sponsor must fund a hard-currency obligation from a tariff that will not keep pace, and bears the gap. (That the concessionaire should, as a default, carry currency risk "as is the case for firms in other sectors" is the Engel–Fischer–Galetovic position¹².) This moves the FX risk off the sovereign's direct account — though a sponsor who cannot fund a devalued obligation can default, returning part of it through a termination payment or a service interruption (the Ghana and halted-Egypt-2003 dynamic of §9). And it is not free. An undiversified sponsor, unable to lay the risk off cheaply, prices it into a higher required return, which the contract pays for through a higher tariff or availability payment. Following the pass-through logic of the Third-Party Guarantee paper — where transferring a risk raises the counterparty's hurdle by a multiple of the risk's value² — the cost of pushing back is the sponsor's risk-adjusted price of the FX exposure, typically above the expected loss, because the sponsor charges for bearing a tail it cannot diversify. Push-back is the honest transfer; the public simply pays the premium for it up front, in the tariff, rather than later, in a devaluation.

4.5 The shape of the four

The four placements line up on two axes the gate cares about — who holds the risk and is there still a tail — and the framework's contribution is to show that the menu is smaller than it looks:

Two of the four leave the tail with the sovereign and are the same FX exposure; the other two move the tail off it at a price. That is the menu a Ministry of Finance is actually choosing from.

4.6 A fifth form: place it with a third party

The hedge and the guarantee each have a multilateral form, and in a frontier market it is often the only form available. Where no commercial swap market exists (§8), the counterparty willing to take local-currency risk is usually a development institution: the TCX Fund writes local-currency hedges across more than 140 frontier currencies; MIGA sells currency-inconvertibility-and-transfer-restriction cover; and multilateral development banks offer partial-risk and partial-credit guarantees that backstop an FX or convertibility gap²¹. Each is a placement already in the menu — a third-party hedge or a third-party guarantee — distinguished only by who the counterparty is.

That turns the choice into the decision the companion Third-Party Guarantee paper frames: take the third-party cover if and only if it lowers the all-in cost to the public by more than its fee, valuing the sponsor's reduced required return through the same pass-through, Δhurdle ≈ λ × Δspread, used for push-back in §4.4². That paper prices the credit and political-risk guarantee and hands the currency exposure to the multi-currency engine; this paper prices the currency placement and routes its "place with a third party" branch back to that decision. Two of the six cascades are this option taken — Argentina's World Bank (IBRD) payment guarantee and Egypt's MIGA cover (§9) — and the practitioner case for contributing toward a hedge rather than writing a full FX guarantee, because it "strictly limits the budget liability"⁵, is itself an argument for pricing the placements side by side instead of reaching for the guarantee by reflex.

5. The decision rule

Why this matters: a comparison is not yet a decision. The gate needs a rule that turns four priced placements into a choice, and the rule has to respect the one thing a frontier Ministry of Finance cannot ignore — that the cheapest option in expectation may carry a tail its reserves cannot absorb.

5.1 The rule

Among the feasible placements, choose the one with the lowest expected cost to the public — but never retain silently when a guarantee is available, because the guarantee is the same cost made visible (§4.2), and read the recommendation against the tail: if the cheapest option still leaves a large Value-at-Risk on the sovereign, weigh that tail against the state's reserve adequacy before accepting it. Formally:

1. Drop any infeasible placement (a hedge with no market).
2. Among the rest, take the minimum expected cost; break ties in favour of the explicit guarantee over silent retention.
3. If the recommended placement still carries a tail and that tail is large relative to the open position, flag it — the choice is now a risk-tolerance judgment, not a cost minimisation.

5.2 The frontier-market trap

The rule has a sharp consequence that the worked examples make concrete. Where a swap market exists, the hedge usually wins: it is close to the cheapest option and it removes the tail, so a deeper-market sovereign can buy certainty cheaply. Where a swap market does not exist — the frontier case — the hedge drops out, and the Ministry is left choosing between two unattractive options: hold the risk (retain or guarantee) at the lowest expected cost but with the full devaluation tail, or push it to the sponsor at a premium to be rid of the tail. There is no free placement. The instrument's value in that setting is not to conjure one but to make the trade explicit: the price of certainty is the sponsor's premium, the price of the cheapest expected cost is the tail, and the gate must choose with both numbers in front of it.

5.3 Guarantee dominates silent retention — always

One element of the rule is unconditional and worth isolating. Because retention and the guarantee are the same FX exposure (§4.2), there is never a reason to retain silently. If the sovereign is going to hold the FX risk, it should hold it as a booked guarantee: the FX exposure is identical — and, once a de-risk discount, fee, or cap is allowed for, the booked version is weakly cheaper — and it can be provisioned and aggregated. The only thing silent retention buys is the comfort of not seeing the number — and that comfort is precisely what §9 shows to be expensive.

5.4 The placement decision as an optimization

The four-way comparison the instrument runs is the discrete face of a constrained optimization, and stating it that way turns the dominance of §5.3 from an observation into an optimality property. Let the planner's preference be a risk-adjusted cost \(\Phi(C)=\mathbb{E}[C]+\rho\,\big(\mathrm{CVaR}_q(C)-\mathbb{E}[C]\big)\) — the expected cost plus a charge on the tail excess over the mean, where \(\mathrm{CVaR}_q\) is the conditional Value-at-Risk (the average loss beyond the \(q\)-tail, i.e. expected shortfall) and \(\rho\ge0\) is the shadow price of the reserve capital that tail consumes (\(\rho=0\) is risk-neutral). A deterministic cost, whose \(\mathrm{CVaR}_q\) equals its mean, carries no penalty — so \(\Phi\) of a certain placement is just that certain cost, which is what makes the thresholds below exact. Allocating a fraction \(\alpha_P\) of the open position to each placement \(P\), the decision is

\[\min_{\{\alpha_P\}}\ \Phi\!\Big(\textstyle\sum_P\alpha_P\,C_P\Big)\quad\text{s.t.}\quad \sum_P\alpha_P=1,\ \ \alpha_P\ge0,\ \ \alpha_{\mathrm{hedge}}=0\ \text{if no swap market},\ \ (f,c)\in\text{policy bounds},\]

with a transparency constraint: any sovereign-borne portion must be booked, so retention enters only as its booked equivalent — which, by the Corollary of §4.2, dominates it. The solution has a clean structure:

1. Silent retention is never optimal. The program replaces any retained slice with the booked guarantee at no extra cost (the Corollary) — §5.3 is now a property of the optimization, not a claim laid alongside it.
2. The remaining choice is a threshold on \(\rho\) and the market costs. Hedge when the certain swap cost \(h<\Phi(\text{guarantee})\) — paying to remove the tail beats carrying it (favoured when tail-aversion \(\rho\) is high and the swap market is cheap). Push back when the sponsor's risk-adjusted premium \(\lambda(\mathbb{E}[L]+\pi)<\Phi(\text{guarantee})\) — moving the risk off-sovereign via the tariff is cheaper than backstopping it (favoured when the sponsor is diversified, \(\pi\) low, and pass-through \(\lambda\) is bounded). Hold it as a booked guarantee otherwise.

In the frontier-market case that defines this paper — no accessible swap market (\(\alpha_{\mathrm{hedge}}=0\)) and a large sponsor premium — the program returns the corner the worked examples reach by hand: hold the FX risk, but visibly. The shipped instrument implements the risk-neutral \(\rho=0\) case and reports the Value-at-Risk separately, leaving \(\rho\) — the price a ministry puts on tail capital — a policy parameter rather than a fitted constant; §5.4 is the formal generalization the four-way comparison instantiates, not a separate computation. The pieces here are individually standard — Monte-Carlo valuation of a guarantee as a distribution¹⁴, explicit pricing of the FX tail (the Lab's FX Hedging Facility prices it at a 99.7% threshold)¹⁷, and CVaR minimisation¹⁵; the novelty is their unification into one four-placement objective with retention and push-back priced in and the booking constraint that makes silent retention inadmissible. The same shape ties this paper to the series' other optimizers — the viability-gap-funding (VGF) sizing of the Feasibility Framework, the accept/counter/terminate argmin of The Renegotiation Triangle, the correlated portfolio-VaR of Stress-Testing PPP Portfolios — each a constrained minimisation of a risk-adjusted cost to government.

6. Wiring it into the fiscal envelope

Why this matters: pricing the placement is the analytical step; booking it is the fiscal one. The instrument's output is only useful to a Ministry of Finance if it lands inside the machinery that governs the country's contingent exposure — otherwise it is one more number that does not make it onto a report.

In the language of Paper 3, the chosen placement's exposure is a draw on the Stage-E fiscal envelope — the contingent commitment the Ministry of Finance underwrites at the gate¹. The placement-comparison output is designed to wire into the platform's fiscal-rules and FCCL machinery so that the FX exposure becomes a booked line, not a footnote: the expected loss as a provision, the Value-at-Risk as the tail the envelope must be able to absorb, and the chosen placement (retain / guarantee / hedge / push-back) recorded as the contract-level decision. The point of the wire is precisely the point of the paper — to convert the silent position of §2 into a visible, provisioned, audit-chained commitment that the gate sets before the currency moves, and that the medium-term fiscal framework carries forward thereafter.

This is the orchestrator's task, not the gatekeeper's. Paper 3 assigns the FX path, the guarantee menu, and the fiscal envelope to the Ministry of Finance by construction — they are country variables no contracting authority can price¹. A gatekeeper MoF, handed a one-page affordability note after the analysis is done, never sees the embedded FX leg; an orchestrator MoF, which already owns those variables, is the one actor positioned to recognise the exposure, place it, and book it. And because it sits at the portfolio level it can do what no single project can: aggregate the embedded-FX exposure across every contract alongside the explicit FX guarantees, and size the country's reserves and envelope to the combined currency tail rather than to one guarantee at a time. PFRAM already runs portfolio-level FX shocks; the missing step is to feed the embedded-indexation legs into that portfolio, not only the guarantees.

7. Architecture and reproduction in ten minutes

Why this matters: the series' discipline is that every framework is a running reference implementation, not a slide. The FX-placement instrument is a shipped platform module with the same audit and reproducibility guarantees as the rest.

7.1 The pipeline

A normalized FX-exposure record is submitted to the placement-comparison layer, which runs the seeded Monte-Carlo FX-loss distribution and prices the four placements; the result is written to the platform's tamper-evident SHA-256 audit chain and returned with an optional signed PDF. Nothing in the pipeline is bespoke to a country or a sector: the same schema and engine priced the power IPP, the hospital, and the toll road in §8.

7.2 One call

A Ministry-of-Finance gate integration is a single POST …/fm/fx-placement/compare carrying the obligation stream, the currencies, the spot rate, the depreciation view, and the placement parameters; the response carries the open-position present value, the expected FX loss, the Value-at-Risk, the four priced placements with their risk-holders and feasibility, and the recommendation. Appendix A gives the copy-pasteable call and a representative response.

7.3 Reproduce in ten minutes

The three worked examples of §8 are reproducible end to end from a single harness that calls the shipped engine; the obligation streams, depreciation views, and placement parameters are stated in the harness, and the outputs are written to a JSON the reader can diff. The engine is deterministic under its seed (the worked examples use n_sims = 20,000, seed = 42), so a reader re-running it sees the same numbers.

8. Three worked examples

Why this matters: the framework's claims are abstract until a real-shaped contract runs through it. We take three sectors that span the FX-mismatch landscape — a power IPP, an availability-payment hospital, and a demand-risk toll road — calibrate each to the shape of the §9 cascades, and run all three through the shipped engine. The numbers below are illustrative but realistic; every one traces to the harness output. All figures are in millions of the obligation's hard currency, present-valued at an 8% discount rate.

8.1 Power IPP — dollar capacity payments, no swap market

A roughly one-gigawatt thermal independent power producer sells under a take-or-pay power purchase agreement (PPA) with US$120 million per year of dollar-indexed capacity payments for twenty years, in a country with no accessible long-dated swap market — the shape of the Pakistan and Ghana cases (§9). The open position is US$1,178 million in present value. Against a depreciation view of 8% per year with 12% volatility, the engine prices:

The expected FX loss is US$433 million — 37% of the open position — and the 95th-percentile loss is US$661 million, 56%. Note the trap: the hedge, at 412, would be the cheapest option, but there is no market for it, so it is infeasible. Among the feasible placements the guarantee is cheapest, and the engine recommends it — but with a tail warning, because a 56% Value-at-Risk on a US$1.2 billion obligation is exactly the kind of exposure that became circular debt and arrears in Pakistan and a multi-billion-dollar sector shortfall in Ghana (§9). The recommendation is honest: hold it, but book it, and know the tail you are holding.

8.2 Availability-payment hospital — a euro correction factor, no swap market

A twenty-five-year design-build-finance-operate city hospital pays the sponsor €300 million per year of availability payments carrying a euro correction factor — the contract's own FX guarantee — funded from a local-currency health budget, again with no accessible swap market. This is the Turkey shape (§9). The open position is €3,202 million. Against a harsher depreciation view (15% per year, 20% volatility, reflecting the lira's path), the engine prices:

The expected FX loss is €1,725 million — 54% of the open position — with a Value-at-Risk of €2,349 million, 73%. The euro correction factor is the government guarantee in this taxonomy: the state has retained the FX risk explicitly, and the engine's recommendation (the guarantee) simply names what the contract already did. The lesson is the magnitude — over half the obligation's value is the expected currency loss, and nearly three-quarters in a bad scenario — and the fact that none of it is booked as a contingent liability under the standard frameworks (§2.2).

8.3 Toll road — a deeper market changes the answer

A demand-risk toll road services US$80 million per year of dollar debt for twenty-five years from a local-currency tariff, but in a country with an accessible cross-currency swap market and a calmer currency (4% per year depreciation, 8% volatility). The open position is US$854 million. Now the answer changes:

The expected FX loss is US$210 million (25%) with a Value-at-Risk of US$367 million (43%) — but here the hedge is feasible, costs US$154 million, and is both the cheapest option and the one that removes the tail. The engine recommends the hedge. This is the counter-case that proves the framework is not a one-note argument for guarantees: where a market exists to take the currency risk cheaply, the sovereign should use it. The frontier cases land on the guarantee not because guarantees are good but because hedging is unavailable.

8.4 What the three show together

Read across the three, the framework's two claims are visible at once. First, in every case, retain and the guarantee carry identical numbers — the same expected loss, the same tail — because they are the same FX exposure; the only difference is the word "implicit" versus "explicit." Second, the placement that wins flips on a single feature of the environment: the existence of a swap market. With one, hedge; without one, the sovereign is choosing between a booked guarantee with a tail and a sponsor premium without — and there is no free option. The instrument does not make the trade-off disappear. It makes it a number on the gate's desk instead of a surprise on the budget.

8.5 Sensitivity — the recommendation is a function of the inputs

Why this matters: the placement each worked example recommends is not an estimated fact about the world — it is the output of the gate's own inputs. A reader should see exactly which inputs move the verdict, so the recommendations are read as conditional, not as findings. We vary the §8.1 power IPP (open position ~US$1,178 million) one dimension at a time.

The swap market is the pivot. Whether a hedge exists, and at what price, decides the verdict: with no accessible market the frontier case lands on the guarantee; open a market and the hedge wins until its price exceeds the expected loss (~37% of the open position here), beyond which the guarantee returns.

The depreciation view scales the exposure. The expected loss and the tail rise with the gate's view of the currency; a gate that assumes 4%-per-year drift carries less than half the exposure of one that assumes 15%.

A fee or a payout cap makes the booked guarantee strictly better than silent retention — the computed form of §4.2's claim. The engine prices the guarantee as identical to retain only in the uncapped, fee-free limiting case; add a fee the sovereign collects or a cap on the payout, and the guarantee is cheaper, thinner-tailed, or both.

The three tables carry one lesson: the instrument does not hand the gate a verdict, it hands it the verdict's dependence — on the swap market, on the currency view, and on the guarantee's own terms. The recommendations in §8.1–§8.3 are what those inputs imply, not claims about how any real country should act.

8.6 The bankability break-point — how little it takes

Read the same power-IPP shape through the coverage lens of §3.4 — hard-currency debt service funded from a local tariff, designed at a 1.40× DSCR against a 1.20× lock-up covenant — and the picture is stark. Worst-year DSCR against a one-off devaluation, with no tariff pass-through (φ = 0):

The break-point is 16.7% — design ÷ covenant − 1, independent of the path — so any of §9's cascades (Argentina ~100%+, Egypt/Ghana/Sri Lanka 50–60%+) does not nick the covenant, it demolishes it. And under the §8.1 currency view itself (8%/yr depreciation, 12% volatility, 20 years) the breach is essentially certain: the probability the DSCR breaches 1.20× in some year is ~100% at zero pass-through, ~99% at half pass-through, and only falls to zero at full FX-indexation. The break-even is φ* ≈ 0.93 — the tariff must pass through about 90% of the currency move to keep the project bankable on retention. That is the bankability form of the paper's thesis: the un-indexed leg is not a slow leak but the line between a financeable project and a covenant breach, and the only placements that move it are the two that genuinely move the risk — a hedge, or near-full pass-through (§4.4). Retain and a fee-free guarantee leave the DSCR exactly where it was; whether the exposure is booked or silent, the project still breaches.

9. The evidence — six devaluation cascades

Why this matters: the framework predicts that a currency mismatch left on the sovereign — by retention, by guarantee, or by FX-indexation — becomes a fiscal liability when the currency falls. Six recent crises let us check that prediction against the record. We carry every figure with its verification status, and we do not print a number we could not source.

9.1 Two archetypes

The cases sort cleanly into two patterns the framework anticipates.

Archetype A — the risk was placed on the state by design, so the devaluation landed on the budget, not the project. Here the contract used FX-indexation or a guarantee to keep the lenders and the sponsor whole; where the currency fell hardest the cost crystallised as a fiscal bill (Turkey, Pakistan), and where the structure held the risk simply sat on the state's account (Egypt, Argentina).

• Turkey (2021–22). The clearest case. Twenty-five-year availability-payment city hospitals pay the Ministry of Health in lira, but the availability payment carries a euro-denominated correction factor — the contract states that the euro value of the payment cannot fall below the preceding period's — and the senior debt is 100% covered in all termination scenarios²⁷. The lira lost roughly 44% in 2021 and a further ~30% in 2022²⁷. The correction factor is an FX guarantee in all but name; the lira's collapse therefore raised the lira cost of the euro-pegged payments rather than impairing the lenders, exactly as §4.2 describes. (Reported aggregate figures for the hospital programme's lifetime cost circulate in the press but are single-source or unverified; we rely on the verified contract structure, not the headline totals.)
• Pakistan. Take-or-pay IPPs — the 1994/2002 policy plants and the CPEC coal plants — sell under PPAs with dollar-indexed capacity payments and sovereign guarantees²⁶. As the rupee fell from roughly 100 to 300 per dollar (a loss of over 60% across 2017–2025)²⁶, the rupee cost of the dollar-linked capacity charges climbed mechanically: capacity payments roughly doubled, from about Rs 1,082 billion to Rs 2,152 billion, and power-sector circular debt reached approximately PKR 2.6 trillion (~2.5% of GDP) by early 2024²⁶. The mismatch was on the sovereign by contractual design, and it crystallised as circular debt and arrears.
• Egypt (2016; 2022 onward). The renewable feed-in-tariff IPPs at Benban were deliberately insulated: Round 2 restructured the tariff as 70% indexed to the dollar and 30% fixed at EGP 8.88 per dollar, with DFI cover behind it²⁵. When the pound floated in November 2016 — from the ~8.8 peg to roughly 17.8–18.1 per dollar, a fall of about 50%²⁵ — the indexation kept the projects serviceable; the FX risk had been re-allocated to the state and its backstops, not removed. (Egypt also furnishes the historical version of the exact mismatch: its first IPP wave was halted in 2003 after an earlier devaluation doubled the local cost of dollar-denominated PPAs²⁵.)
• Argentina (2018). The RenovAr renewables programme priced PPAs in dollars and backstopped them with the FODER trust fund and an optional World Bank (IBRD) payment guarantee backstopping FODER's put-option / termination payments²². When the peso fell about 52% in 2018²², the structure largely held — making Argentina the instructive counter-example: an explicit, pre-funded backstop is the guarantee placement done deliberately. (We found no verified figure for a realised fiscal loss tied specifically to the 2018 RenovAr crisis, and do not assert one.)

Archetype B — the mismatch was simply unhedged, and crystallised as arrears and default. Here dollar costs met local-currency revenue with weaker contractual protection, and the gap became unpaid bills.

• Ghana (2022). Thermal IPPs sold to the distribution offtaker ECG under dollar-priced take-or-pay PPAs with government guarantees²³. As the cedi fell sharply in 2022 (commonly reported in the ~50–60% range, depending on the measurement window)²³, the dollar-denominated sector shortfall ballooned: the Ministry of Finance's Energy Sector Recovery Programme re-estimated the cumulative shortfall from about US$12.5 billion to roughly US$14.3 billion by 2023 — a ~US$1.8 billion (~14%) uplift — once a realistic ~7.4%-per-year depreciation path replaced the prior flat-rate assumption²³. Ghana suspended payments on most external debt in December 2022²³. The reform fix is the tell: new PPAs must be cedi-denominated, take-and-pay, with no government guarantees — the contract finally pushing the FX risk off the sovereign.
• Sri Lanka (2022). The state utility CEB bought dollar-priced power and imported fuel against rupee tariffs, with some private-producer obligations under public guarantees²⁴. The rupee fell from about 202 to over 300 per dollar, and Sri Lanka defaulted on its external debt in April 2022²⁴. The honest framing is that the sovereign default was driven mainly by international sovereign bonds, not by the power PPAs — so the power-sector mismatch was an amplifier, not the cause; but the unhedged dollar-versus-rupee position in the energy sector is the same pattern, and it crystallised as utility losses and unpaid producers.

9.2 The policy lesson the evidence supports

Across all six, one statement is consistent with the verified record: FX-indexation and government guarantees do not make currency risk disappear — they relocate it from the project's profit-and-loss to the sovereign balance sheet, where a devaluation converts a private financing risk into a public fiscal liability. Turkey's euro floor, Pakistan's dollar capacity payments, and Ghana's dollar shortfall are three forms of the same conversion. Relocating the risk to the sovereign is not in itself the error — a state is often the cheapest, most diversified holder of a project's currency risk, and doing so deliberately, with multilateral cover, can be efficient allocation (Argentina and Egypt are the deliberate version). The error is doing it silently: an exposure that is priced and booked can be managed; the same exposure unpriced is the one that becomes circular debt and arrears. The pattern is not anecdotal — the IMF's cross-country dataset of contingent-liability realizations puts the average fiscal cost near 6% of GDP, reaching some 40% in the worst episodes and clustering in exactly the crises that trigger devaluations¹⁸. The framework of §4 is, in effect, the contract-level mechanism behind these six country-level outcomes; the instrument exists so that a Ministry of Finance can see the conversion coming and price it at the gate, rather than discover it in a crisis.

10. Limitations

Why this matters: the framework is a pricing-and-placement aid, not a currency forecast, and the honesty about what it does not do is part of what makes it usable.

• It prices a view, it does not forecast the currency. The expected loss and the Value-at-Risk are only as good as the depreciation mean and volatility the gate supplies. The instrument's contribution is to turn a view into a placement decision and a booked liability — not to predict the exchange rate, which no one can.
• The depreciation model is deliberately simple. Annual depreciations drawn from a mean and a volatility capture the central feature — a one-sided, compounding exposure with a fat tail — but omit regime shifts, jumps, and the correlation between devaluation and the macro shock that usually accompanies it. A Ministry of Finance running this against a real programme should stress the view with explicit devaluation scenarios alongside the distribution, and the engine accepts a deterministic stress path for exactly that.
• The worked examples are illustrative, and their recommendations are conditional. The three sectors are calibrated to the shape of real cases, not drawn from a contract database; they demonstrate the instrument, they do not estimate a population, and §8.5 shows how the recommended placement moves with the swap market, the currency view, and the guarantee's own terms — so §8.1–§8.3 are conditional illustrations, not claims about how any country should act. The six cascades are post-mortems with their figures tagged by source, not a statistical sample.
• Hedge cost and sponsor premium are inputs. The framework prices the four placements given the swap cost and the sponsor's FX-risk premium; it does not derive them. In practice both require market soundings, and in frontier markets both are uncertain — which is itself part of the decision, and §8.5 maps the verdict's dependence on them.
• The guarantee defaults to its limiting case. The shipped engine prices a government FX guarantee as identical to retention unless a fee or a payout cap is supplied; with neither, the two are equal by construction, and the worked examples use that fee-free, uncapped case as the conservative bound. A real guarantee that carries a fee, deductible, or cap is strictly cheaper or thinner-tailed (§8.5), and the de-risk discount of §4.2 widens its advantage further. That de-risk discount is asserted on the same footing as the hedge cost and the sponsor premium — a per-deal input to be set, not a measured constant; the dominance does not depend on it, since the booking-and-visibility benefit holds on every path and a positive fee or finite cap already makes the guarantee strictly cheaper. So every refinement strengthens, never weakens, the dominance of an explicit guarantee over silent retention.
• The bankability lens is a stylised coverage identity, not a financing model. §3.4/§8.6 run one design ratio (1.40× DSCR, 1.20× covenant) and treat all debt as hard-currency and the tariff's FX-pass-through φ as the single mitigant; it isolates the break-point and the pass-through a project needs to stay bankable, not the full structure of any deal. A real financing carries an amortisation profile, a debt-service reserve, sculpting, and any local-currency or partially-hedged tranche (which raises the break-point); and local prices absorb some devaluation through inflation pass-through, so the effective φ is rarely zero. The point is the direction and the order of magnitude — coverage breaks on a modest devaluation unless the revenue is largely FX-indexed — not the precise 16.7%.
• Single-source figures are flagged, not laundered. Several headline numbers in §9 (notably the Turkish programme's aggregate totals and some Pakistani return figures) rest on single sources; the evidence base carries them as such and the paper relies on the verified structural facts rather than the contested totals.

11. The policy lead — make the silent liability visible

Why this matters: the instrument is useful to one Ministry of Finance at one gate; the policy point is for the doctrine. The recurring failure across the six cascades is not that countries chose badly among the placements — sometimes they had no good choice — but that they did not see the position they were holding until it came due.

The lead is therefore not a recommendation to hedge, or to guarantee, or to push back — the right placement depends on the swap market, the reserves, and the sponsor's price, as §5 and §8 show. The lead is to stop holding the FX mismatch silently. A currency position that is priced, booked as a contingent liability, provisioned against, and aggregated across the portfolio is a position a Ministry of Finance can manage — it can size its reserves to the tail, negotiate the placement deliberately, and defend the verdict to an audit institution. The same position unbooked is the thing that turned, in six countries in a decade, into circular debt, arrears, and default.

The instrument this paper ships is the route from one to the other: it converts the embedded FX-indexation clause that today reads as ordinary spending into a priced, placed, audit-chained line at the gate. That is the move from a passive fiscal gate that discovers its currency exposure in a devaluation to an intelligent one that prices it at signature — the same arc the rest of this series traces, applied to the one liability the fiscal machinery has been built to miss.

11.1 Roadmap

• The FCCL standard should book the indexation leg. The clearest doctrinal step is to bring FX-indexed availability payments and capacity charges inside the contingent-liability frameworks alongside explicit guarantees — because, as §4 shows, they are the same exposure. The companion Indexation Ceiling (Paper 7) generalises this from the FX special case to any indexed tariff, pricing the affordability ceiling as a written call of which the FX loss priced here is the ceiling-zero instance.
• Build the swap-market map. The decisive variable in the decision rule is whether an accessible long-dated cross-currency swap exists; a cross-country map of frontier hedging-market depth would tell Ministries of Finance which of them are in the trap of §5.2.
• Add countries to the cascade record. Six post-mortems motivate the framework; a structured, sourced database of multi-currency-PPP devaluation episodes would let the relocation thesis be tested rather than illustrated.

Appendix A — Reproducible API call

The placement-comparison layer is reached with a single authenticated POST to the financial-model module. The request is a normalized FX-exposure record.

Request:

POST /workspaces/{workspace_id}/fm/fx-placement/compare
Authorization: Bearer <token>
Content-Type: application/json

{
  "obligation_hard_ccy": [120, 120, 120, 120, 120, 120, 120, 120, 120, 120,
                          120, 120, 120, 120, 120, 120, 120, 120, 120, 120],
  "hard_currency": "USD",
  "local_currency": "LCY",
  "discount_rate_pct": 8.0,
  "scenario": {
    "spot_rate": 100.0,
    "annual_depreciation_mean_pct": 8.0,
    "annual_depreciation_vol_pct": 12.0,
    "n_sims": 20000, "seed": 42, "var_pct": 95.0
  },
  "params": {
    "hedge_all_in_cost_pct": 35.0, "hedge_available": false,
    "sponsor_risk_premium_pct": 20.0, "lambda_passthrough": 1.0
  }
}

Response (FXPlacementResponse, abridged):

{
  "open_position_npv": 1178.18,
  "expected_fx_loss_npv": 432.85,
  "var_fx_loss_npv": 661.25,
  "var_pct": 95.0,
  "placements": [
    {"placement": "Retain", "risk_holder": "Sovereign (implicit)",
     "expected_cost_npv": 432.85, "tail_cost_npv": 661.25, "has_tail": true, "feasible": true},
    {"placement": "Government FX guarantee", "risk_holder": "Sovereign (explicit)",
     "expected_cost_npv": 432.85, "tail_cost_npv": 661.25, "has_tail": true, "feasible": true},
    {"placement": "Hedge / cross-currency swap", "risk_holder": "Swap counterparty",
     "expected_cost_npv": 412.36, "tail_cost_npv": 412.36, "has_tail": false, "feasible": false},
    {"placement": "Push-back via tariff indexation", "risk_holder": "Sponsor (via tariff)",
     "expected_cost_npv": 668.48, "tail_cost_npv": 668.48, "has_tail": false, "feasible": true}
  ],
  "recommended": "Government FX guarantee",
  "warnings": ["Hedge marked infeasible — recommendation excludes it.",
               "Cheapest option carries a large devaluation tail ..."]
}

The expected cost and tail are present-valued in hard-currency-equivalent terms; the recommendation is the cheapest feasible placement, preferring the explicit guarantee over silent retention.

Appendix B — Placement and field reference

The four placements: Retain (sovereign, implicit), Government FX guarantee (sovereign, explicit — same cost/tail as Retain in the fee-free, uncapped default; weakly cheaper or thinner-tailed with a fee or cap), Hedge (counterparty, certain cost, feasibility-gated), Push-back (sponsor, via a higher tariff).

Appendix C — Glossary

• Open FX position — the unmatched exposure left when a hard-currency obligation is funded by local-currency revenue at a rate that will move.
• FX loss — the public's shortfall on the obligation when the currency depreciates, valued in hard currency at the prevailing rate; bounded by the obligation itself.
• Value-at-Risk (VaR) — the FX loss at a chosen tail percentile (here 95%): a bad-but-not-worst devaluation.
• Cross-currency swap — a contract that exchanges a stream of payments in one currency for another, converting a floating FX cost into a certain one.
• Availability payment — a fixed periodic payment a public offtaker makes for an available asset, regardless of usage.
• Take-or-pay / capacity payment — a contract (typically a PPA) under which the offtaker pays for capacity whether or not it offtakes the output.
• Tariff indexation — a clause adjusting the tariff to a reference (inflation, FX); full FX indexation moves the currency risk to the offtaker.
• Circular debt — the chain of arrears that builds when a utility cannot pass its costs (including FX-inflated ones) through to tariffs.
• FCCL — fiscal commitments and contingent liabilities; the framework for accounting a government's PPP exposure.
• PFRAM — the IMF–World Bank PPP Fiscal Risk Assessment Model.

Appendix D — Pseudocode for the placement comparison

open_npv      = NPV(obligation_hard_ccy, discount_rate)
# FX-loss distribution (seeded Monte Carlo)
for sim in 1..n_sims:
    dep_factor = 1
    for t, o in obligation:
        dep_factor *= max(0.01, 1 + Normal(dep_mean, dep_vol))   # cumulative depreciation
        d_t        = dep_factor - 1
        loss[t]    = o * d_t / (1 + d_t)        # hard-ccy-equiv, bounded by o
    loss_sim[sim] = NPV(loss, discount_rate)
E_loss = mean(loss_sim);   VaR = percentile(loss_sim, var_pct)

retain     = (cost=E_loss,                       tail=VaR,   holder=sovereign-implicit, tail=yes)
# guarantee: defaults fee=cap=0 -> identical to retain; a fee/cap make it weakly cheaper/thinner
fee    = fee_pct*open_npv
g_loss = [min(l, cap_pct*open_npv) for l in loss_sim] if cap_pct > 0 else loss_sim
guarantee  = (cost=mean(g_loss)-fee, tail=percentile(g_loss,var_pct)-fee, holder=sovereign-explicit, tail=yes)
hedge      = (cost=hedge_cost_pct*open_npv,      tail=cost,  holder=counterparty,       feasible=hedge_available)
pushback   = (cost=lambda*(E_loss + premium_pct*open_npv), tail=cost, holder=sponsor)

recommended = argmin(expected_cost over feasible placements,
                     tie-break: explicit guarantee over silent retain)

References & notes