The Gate Before the Scales: Choosing an SPV Jurisdiction for a Deal That Has to Close

I. Why this post exists now

My September post looked at Ukraine’s closed capital account from the inside, through what it does to the risk-free rate. A Ukrainian exporter meets the same wall facing the other way: it needs investors beyond it to buy paper backed by its future export proceeds, and one way to arrange that is a cross-border securitisation through a vehicle registered abroad. The first practical question such a deal raises is where to register the special-purpose vehicle (SPV), and the usual answer is where the trouble starts.

Sit in the adviser’s chair for a moment. The exporter’s team brings a page from a jurisdiction brochure: a colour-coded grid with a score in every cell, a total in the last column and the highest total circled. It looks rigorous because it is arithmetic, and it is perfectly capable of ranking a jurisdiction first for a deal that cannot close. What it cannot show is whether the winner survives a change in the investor base or the settlement system.

My answer to that page is now in print. The article “Choosing an SPV jurisdiction for cross-border securitisation” was published on 30 September in The Ukrainian Economic Journal, No. 14, 2026, pp. 54–60 (DOI 10.32782/2786-8273/2026-14-7); the full text is in English and open access. It builds a two-stage model and applies it to an illustrative cross-border securitisation of future iron-ore export receivables of a Ukrainian producer under two issuance routes: an EU-facing issue within Regulation (EU) 2017/2402 and a UK-facing issue listed on the London Stock Exchange. The question sits close to my dissertation, which examines securitisation in the capital formation of extractive-sector companies.

The country of registration decides more than the brochure suggests: the listing venue, the investor base, the settlement system and the securities-issuance rules. Yet foreign incorporation alone does not remove assets from the originator’s insolvency estate. Take the address from the brochure, and the deal can fail on a condition no score in the matrix ever measured. What could stop the deal has to be checked before anything is weighed.

II. What a weighted matrix cannot see

The weighted matrix is candid about one thing: how much of one criterion the decision-maker would give up for a unit of another. Addition is the whole trick and the whole flaw, because once everything ends up in one total, a high score on one line can pay for a low score on another. Cinelli and colleagues make the point in a paper on the uses of multi-criteria decision analysis (MCDA) in energy systems analysis: MCDA treats weights as trade-off coefficients between scales, and threshold conditions are a different object from compensatory preferences. Nothing in the arithmetic is mistaken; the trouble is the premise that every defect can be bought off with points. For a defect that stops the transaction, that premise makes no sense: no amount of tax neutrality turns an ineffective assignment into an effective one.

Matrices of this kind mislead in four ways. They can rank a jurisdiction even when the transaction cannot be completed there. They confuse a failed condition with missing data. A missing file is not a failed test, and a matrix that gives both the same low score cannot say which one it is looking at. They hide the source and date behind each score, so a statute and a sales pitch look the same. And they report a single ranking without testing whether it survives a change in the assumptions or in the conditions of placement.

Of the four, the last is the quietest: a brochure ranking reads as universal, while for a Ukrainian exporter the answer depends on the issuance route and, within a UK route, on the settlement mechanism. Lahdelma and Salminen supplied the tool for testing a fragile ranking: stochastic rank-acceptability analysis, which reports the share of scenarios in which an alternative attains each rank over a defined uncertainty space. None of the studies I build on makes the choice of an SPV jurisdiction conditional on non-compensatory conditions or on an explicit issuance route. The model’s answer is to put the gate before the scales, and to write the route into both – one condition at the gate, one criterion on the scales – instead of assuming it away.

III. Stage 1: the gate, and the two conditions no address clears

A gate carries no weights, and Stage 1 is one. It screens the seven jurisdictions examined – Luxembourg, Ireland, Jersey, the Netherlands, the Cayman Islands, Bermuda and the British Virgin Islands (BVI) – against six non-compensatory admissibility conditions. They are asset and transaction fit (G1), effectiveness under Ukrainian law (G2), legal and organisational separateness of the SPV (G3), lawful cash movement (G4), operational continuity (G5) and the issuance perimeter (G6). A pass requires direct confirmation. A conditional pass identifies a precondition to satisfy before the transaction completes, and a fail records an incompatibility with the route that no favourable score elsewhere can offset.

Two of the six conditions are set by Ukrainian law and currency rules, and a weighted matrix has no column for either. Effectiveness under Ukrainian law starts from Articles 512–519 of the Civil Code, which permit the assignment of receivables subject to the contract and applicable law. The Law on Private International Law preserves mandatory Ukrainian rules despite a foreign-law clause. Under Article 42 of the Code of Ukraine on Bankruptcy Procedures, transactions concluded after bankruptcy proceedings open, or in the three years before, can be challenged on grounds the law enumerates. None of this by itself establishes a true sale or excludes future receivables from the insolvency estate; identifiability, notice, priority, set-off, avoidance and recharacterisation are for Ukrainian counsel to assess. As at 22 July 2026, the date at which the article’s legal sources were verified, Law of Ukraine No. 4466-IX on factoring was due to take effect on 30 July 2026, and the securitisation bill, Draft Law No. 15172, had not been adopted. An address abroad changes none of these tests.

Lawful cash movement (G4) is where the closed circuit from the September post comes back, as an admissibility condition: the Law on Currency and Currency Transactions and the wartime restrictions of the National Bank of Ukraine require a bank-confirmed payment path compatible with export-proceeds supervision, so directing buyers to pay into a foreign account is not sufficient. A foreign address changes where the notes are sold, not where the money starts. I do not read this as a fault in the regime: the requirement is the rules working as written, and a structure built to step around it simply fails the condition.

The issuance perimeter is the only condition that moves with the route. Under the EU route, Article 4 of Regulation (EU) 2017/2402 bars a securitisation special-purpose entity (SSPE) established in a third country on the EU list of high-risk anti-money-laundering jurisdictions, and Commission Delegated Regulation (EU) 2026/83 places the BVI on that list. Under the UK route, regulation 8A of the Securitisation Regulations 2024 bars an SSPE only in jurisdictions under a FATF call for action – Iran, North Korea and Myanmar. The article reads both lists as they stood on 22 July 2026. So the BVI fail G6 under the first route and receive a conditional pass under the second: the one jurisdiction the gate stops, and on a single route.

Picture 1. Stage 1 as Table 1 of the article records it: the seven jurisdictions examined against six admissibility conditions, with the issuance perimeter (G6) shown separately for the EU route and the UK route. The two Ukrainian conditions, effectiveness under Ukrainian law (G2) and lawful cash movement (G4), show a conditional pass for every jurisdiction, as do operational continuity (G5) and the issuance perimeter under the UK route. No address clears the Ukrainian conditions, and the single fail sits on one route.

The decision column is the part a single ranking would flatten. Luxembourg, Ireland and Jersey are admitted conditionally: they proceed to Stage 2, although some preconditions for closing are still open. The Netherlands, the Cayman Islands and Bermuda, and the BVI under the UK route, are not ranked: comparable Stage-2 data are missing, and “not ranked” says nothing about legal unsuitability; “excluded” applies only to the BVI under the EU route. In Picture 1 the Cayman Islands’ gate row is identical to Luxembourg’s, so the gate is not what keeps them off the scales.

IV. Stage 2: the scales, for the three with complete data

Three jurisdictions reach the scales: Luxembourg, Ireland and Jersey. Picture 2 shows the whole model in one scene, with the gate on the left and the scales on the right. The criteria are maturity of the legal regime (weight 0.25), creditor control beyond the gate minimum (0.25), potential for domicile-level tax neutrality (0.20), execution simplicity (0.15) and market and settlement functionality (0.15), each scored from 1 to 5 points. Tax neutrality carries 0.20 because tax leakage eats into debt-service coverage. The last two carry 0.15 each because, once the conditions are verified, they mainly affect cost and timing. The weights reflect the priorities of this transaction, and no expert consensus stands behind them.

Picture 2. The two-stage model as one scene: a gate of six non-compensatory admissibility conditions, then scales with five weighted criteria. Three of the seven jurisdictions examined reach the scales, and none of them without conditions. The Netherlands, the Cayman Islands and Bermuda fail no condition but wait in the grey “not ranked” zone for want of comparable data; the BVI stop at the issuance perimeter under the EU route. No weight placed on the scales opens the gate.

Each score arrives with its reason. On the legal regime, Luxembourg receives 5 points for the maturity of its regime and current legislation, Ireland 5 points for the officially recorded number of Irish financial vehicles, and Jersey 4 points for a specialist regime without a comparable official body of data. On creditor control, Luxembourg receives 5 points for statutory creditor control, Jersey 4 points because its 2005 Bankruptcy (Netting, Contractual Subordination and Non-Petition Provisions) Law makes non-petition provisions enforceable as written, and Ireland 3 points on general company-law tools. Tax neutrality scores an equal 4 points for all three: each regime is materially favourable but conditional, so the equal score records comparable potential, not identical outcomes. On execution, an SPV in Luxembourg or Ireland remains subject to the risk-retention and reporting requirements of Regulation (EU) 2017/2402 regardless of the listing venue, so both score 2 points. Jersey’s consent-based regime is lighter at 4 points, provided that no party to the securitisation is established in the EU. Before settlement enters, Luxembourg’s lead is built on law and creditor control; tax does not separate the three.

Market and settlement functionality is the one criterion the route can move, and the one that can change the leader – provided the notes on the UK route settle directly in CREST. Under the EU route the notes are issued as a global note and settle through an international central securities depository (ICSD), which is domicile-neutral, so the criterion scores venue depth: 4 points for Luxembourg and Ireland, 3 points for Jersey. Multiply each score by its weight and add, and the EU-route baseline is Luxembourg 4.20 points, Jersey 3.85 points and Ireland 3.70 points. Cost, the column the exporter will ask about first, is deliberately left out of the model: its level is not comparable across jurisdictions and varies with the parameters of the specific transaction, and published exchange fees do not reflect all-in cost. Leaving it out keeps such a figure from deciding the order.

Now change the route. Under the assumed direct-CREST configuration of the UK route (the article’s “U-CREST”), Jersey notes settle directly in CREST while Luxembourg and Irish notes use a CREST depositary interest, so settlement scores 5 points for Jersey and 3 points for the other two. The totals become Jersey 4.15 points, Luxembourg 4.05 points and Ireland 3.55 points. The top two swap places (Picture 3), and the swap needs checking before anyone relies on it.

Picture 3. Baseline scores of the three jurisdictions that reach the scales, under the EU route and under the assumed direct-CREST configuration of the UK route. Four of the five criteria score the same on both routes, so the reversal at the top comes from the settlement criterion alone. The right-hand panel holds only if the CREST operator accepts the specific issue for direct settlement.

V. One hundred thousand scenarios, and what the shares are not

A baseline is one point in a space of plausible inputs, and showing a decision-maker only the point means showing one draw and calling it the answer. The model therefore follows the rank-acceptability approach: vary everything uncertain at once, then count how often each jurisdiction comes first. Each weight moves within ±20 %, and the medium- and low-confidence scores move by integer values within one point of their baselines. Each of the two samples contains 100,000 scenarios. The first covers the admissible values uniformly; in the second, the centre-weighted sample, the baseline score occurs twice as often.

Under the EU route Luxembourg ranks first in 61.8 % of scenarios in the uniform sample and 70.5 % in the centre-weighted one – 62–71 % in round numbers – with Jersey second. Under the assumed direct-CREST configuration Jersey ranks first in 61.6 % and 62.6 % of scenarios (Picture 4). The two leads look alike but are not of the same kind: Jersey’s stands on settlement scores coded as low-confidence assumptions, while Luxembourg’s needs no assumption about CREST.

Picture 4. First-place shares of the leading jurisdiction across 100,000 scenarios in each of two samples, uniform and centre-weighted: Luxembourg under the EU route, Jersey under the assumed direct-CREST configuration of the UK route. Bar height measures how firmly the lead holds across the declared ranges, not the quality of the assumption underneath. The right-hand pair would not apply if the operator declined direct settlement for the specific issue.

A first-place share of 62–71 % is not a probability that Luxembourg is the right answer. The shares describe the declared scenario conditions; they show sensitivity to assumptions, not how the transaction will perform. The narrower reading holds: within the declared ranges for weights and scores, Luxembourg stays first in most joint scenarios, and its lead does not hang on the baseline point alone. One limit sits inside the arithmetic: the additive model assumes approximately interval scales and preferential independence of the criteria – a one-point step means the same anywhere on the scale, and the gain from a better creditor-control score does not depend on where the tax score sits.

VI. The route, the originator, and where the model stops

The most tempting shortcut is to treat the listing venue as a way of choosing the rules. It does not work like that. A UK route does not lighten the risk-retention and reporting duties of an EU-established SSPE: the rules follow the vehicle, not the venue, so execution scores the same on both routes and only settlement depends on the route.

Settlement is where the route bites, and even there the answer splits. Under an ICSD global note, settlement is domicile-neutral, so the UK route reproduces the EU result and Luxembourg leads. Jersey moves ahead only if its notes settle directly in CREST, which depends on the operator accepting the particular issue under Article 17 of the Companies (Uncertificated Securities) (Jersey) Order 1999. Hence the low-confidence coding. Until the operator has confirmed acceptance of the specific issue, the article treats settlement as a condition of completing the transaction rather than a driver of the baseline ranking, so for this illustrative deal the article’s baseline choice is Luxembourg.

One check comes before the choice of country. The address does not make the house: a vehicle registered anywhere can pass on only the receivables that the originator still generates, and the article makes the point with Ferrexpo’s published disclosures. On 15 April 2026 Ferrexpo reported first-quarter production of 592,751 tonnes – 72 % below the prior-year quarter – after a large part of operations was suspended following strikes on energy infrastructure. On 22 April the company reported access to only limited funds and warned that, without additional funding, it faced insolvency. These statements do not establish that insolvency occurred, but they require confirmation of going-concern status, eligible-receivable generation and backup servicing before a jurisdiction is chosen. The article uses the case only to show what originator-level verification looks like; it describes no Ferrexpo transaction, and nothing here is a view on the company’s credit.

The model structures due diligence; it is not a legal opinion on a specific transaction, and tax, accounting, regulatory-capital, sanctions and conflict-of-laws effects must be analysed separately. It is bounded by the seven jurisdictions examined, the two routes, the 22 July 2026 date of its sources and uneven source-data availability. The scores are my own coding. CIRA publishes my CEO column, but this post is not the agency’s judgement on any jurisdiction, company or transaction: the article is my academic work, and the conclusions here are mine.

VII. What I’d say now

If I were in that adviser’s chair, I would follow the order the article’s model sets out. I would turn the brochure face down and start with the originator: whether it still generates the receivables, and whether a backup servicer could take over if it stopped. No choice of jurisdiction repairs a negative answer there. Then I would run the gate, with Ukrainian counsel on the assignment and a bank on the payment path, before a single score is written down. I would fix the route before weighing anything, treat settlement as a condition of closing and keep cost on a separate page. Only then would the addresses that passed, and that have comparable data, go on the scales.

A brochure is quicker to read than a gate, and an exporter raising money in wartime, or later for the rebuilding, will always be tempted by the quicker read. Benjamin Graham asked for a margin of safety before he asked about the upside; the gate is the same habit applied to a structure, and only what has passed it is worth putting on the scales.

Communicating Vessels with the Tap Shut: What a Closed Capital Account Does to the First Number in Every Valuation

I. Why this post exists now

The last post here ended on a claim I still hold: the market sets the price of money, and the regulator mostly accompanies it. This one is the uncomfortable sequel, because the market only does that where it is let in.

The premise I have grown tired of arguing – in seminars, in valuation reviews, in conversations with people who build discount rates for a living – is that the risk-free rate is a lookup. Find the currency, take the number, move on. The canon appears to license exactly that: under Aswath Damodaran, the local-currency risk-free rate is the government bond yield minus the sovereign’s default spread, while the base for the dollar and the euro comes from US and German government bonds. The whole construction hangs on one unstated word – can. It assumes an investor who can actually buy the instrument being quoted. For a Ukrainian institution today, neither US Treasuries nor German Bunds are within reach, so the international table describes somebody else’s choice rather than theirs.

What made me write this up rather than grumble about it again was the September issue of the risk-free rates CIRA publishes monthly. Two definitions sit on that page for the hryvnia, and this month they moved apart – 8.7 % against 9.2 %. The temptation is to treat the gap as noise and average it away. It is not noise. It is the shape of the problem.

The stake is not cosmetic. The risk-free rate is the first term of every other rate in the model: the cost of equity is that rate plus the equity risk premium and the country risk premium; the cost of debt is that rate plus the borrower’s credit spread. One percentage point of error at the base moves a company’s valuation by 5–20 %, depending on the level of rates and the growth rate. It is the single number in a DCF that people are least willing to argue about and least able to observe.

II. The identity everyone quotes is a result, not a definition

The theory itself is remarkably clean. A risk-free rate has only two components: a real reward for postponing consumption – the time premium – and compensation for the inflation of the currency the rate is quoted in. Inflation belongs to each currency alone; the time premium is roughly the same everywhere. Hence the textbook sentence that every valuation course repeats: the difference between the risk-free rates of two currencies is, in essence, the difference in their inflation rates, so a rate in one currency converts into another through the inflation differential.

The question usually skipped at that point is why the time premium should be roughly the same everywhere. It is not an axiom. It is an equilibrium, and the machine that produces it is arbitrage – a process that looks entirely mundane. A fund with dollar liquidity sees paper in another currency where the real reward over the same horizon is higher, and moves money there; its buying lifts the price of that paper and pushes the yield down. The reverse case is the mirror image. No single trade equalises anything. What equalises is the flow of such trades, and that flow compresses the difference in time premia for as long as the move still pays, leaving mostly the difference in inflation between the currencies.

So the identity that opens every textbook calculation is not a definition at all. It is an equilibrium result, and it holds for exactly as long as arbitrage can run. Communicating vessels equalise not because the vessels are alike but because liquid flows between them, and that flow is arbitrage. Shut the tap, and the levels diverge in every vessel, whatever the formula says.

III. Where the tap is shut

In Ukraine the tap is shut by regulation. The wartime currency regime rests on Resolution No. 18 of the Board of the National Bank of Ukraine of 24.02.2022, whose paragraph 14 permits cross-border transfers of currency values out of Ukraine only for a defined list of operations – everything is prohibited except what is permitted – and the purchase of foreign securities is not on that list now either.

The August easing is worth stating precisely, because it is easy to overstate and I have seen it overstated. Resolution No. 90 of 10.08.2026, in force since 11.08.2026, did not widen the paragraph 14 list. It added a separate paragraph 12, index 19, under which a bank may sell securities of foreign issuers to its own individual client, delivered inside Ukraine’s depository system, within a combined UAH 200,000 per calendar month at one bank – a limit shared with non-cash currency and bank metals. Under that construction no money crosses the border, and a transfer to an account with a broker abroad remains outside paragraph 14. The easing happens inside the circuit rather than through its wall.

The closure is older than the war, which is the part most commentary misses. Ukraine’s normalised Chinn–Ito index of capital-account openness – zero for a fully closed capital account, one for a fully free one – has read 0.000 at every annual observation since 2009, the latest available being 2023. The wartime regime sealed a circuit that was already closed; it did not create it.

The closure is also visible outside the regulatory text. The exit queue on restricted operations runs at around USD 47bn for 2026–2027 on the NBU’s own estimate, roughly USD 19bn of it already past due. Against that queue, USD 707m has actually been used out of the USD 1,271m of limits set for new money. This is no rebuke to the regulator: any protective regime has a price, and the price here has been counted – a CIRA study published on 27 July 2026 puts the contribution of the restrictions themselves over 2022–2025 at roughly USD 9–10bn. The arbitrageur who ought to equalise the time premia does not take the trade. Not because he cannot see the difference, but because a transfer of that size will not go through.

Picture 1. The closed circuit in three views: the non-resident share in domestic government bonds (OVDP) at three published observations, from 15.9 % at its February 2020 peak to 0.9 % at the end of June 2026; the exit queue set against the outflow the regime actually permits; and Ukraine at the closed end of the capital-account openness scale. Each indicator on its own is a detail of the regime; together they mark the boundary of a separate financial ecosystem, and inside that boundary the risk-free rate has to be built rather than imported.

With arbitrage switched off, the circuit becomes exactly that – a separate ecosystem. About a third of the hryvnia segment of domestic debt, UAH 656bn out of UAH 1,834bn, is held by the NBU itself. Neither Treasuries nor Bunds are within reach of capital locked inside, and a foreign-currency deposit at a Ukrainian bank or a foreign-currency government bond carries a different, Ukrainian risk. So in every currency of the circuit the rate stops being a single object.

Regimes like this are introduced as temporary, and the comparative record is short: Cyprus held one for 24 months, Greece for 50, Argentina for 67, Iceland and Nigeria for about 100. Ukraine’s has run for 53 months as at July 2026 – already longer than the whole Greek episode. As a researcher I find the case unusually interesting, because theory rarely gets to watch what happens to its most basic constant once the mechanism that keeps it constant is removed. But the tap here was shut by a war, not for an experiment: what interests me is the phenomenon, not its cause, and no experiment is worth that price. For those working inside the circuit the choice is not between studying it and ignoring it. It is between measuring and guessing.

IV. Two definitions, and why I would not pick one

The first definition is the international view, on a ten-year horizon: what the hryvnia rate would be if arbitrage were working. The dollar risk-free rate is the yield on ten-year Treasuries minus the US CDS spread (the market price of insurance against a US default), because the United States no longer carries an Aaa/AAA rating and Treasuries are therefore only quasi-risk-free. In the September issue, 4.73 % minus 0.33 p.p. gives 4.40 %, shown in the summary table as 4.4 %. The euro needs no subtraction, since ten-year Bunds are rated Aaa/AAA: 3.32 %, shown as 3.3 % – a rate in its own right for euro cash flows, not a second base for the hryvnia. The Fisher relation then carries the dollar rate into hryvnia through actual year-on-year inflation in both currencies for July 2026 – 7.7 % against 3.4 % – and lands us at 8.74 %, shown as 8.7 %. Actual rather than expected inflation is a deliberate choice: an observed number leaves no room for assumptions nobody can audit. This route openly borrows the identity that arbitrage produces beyond the circuit, which is precisely why it can anchor a long horizon.

The second definition never leaves the circuit. The nearest things to a risk-free instrument inside it are hryvnia government bonds from the Ministry of Finance’s primary auctions – but the sovereign is not default-free even in its own currency, so we deduct the credit spread implied by its local-currency rating from S&P and Fitch, taken from the default-spread ladder Damodaran publishes. That is a rating-based figure from outside CIRA, not a CIRA judgement on the sovereign’s creditworthiness, and the national rating scale plays no part in it. At the auction of 25.08.2026 the weighted average yield on the 329-day issue was 15.17 %; deduct a spread of 5.97 % and we have 9.20 %, shown as 9.2 %. It is a computed anchor, not a yield anyone can buy.

Two objects, then, not one rate computed two ways. This month they sit 0.5 p.p. apart. A month ago the distance was 0.1 p.p., and that near-coincidence confirmed nothing – current annual macro conditions simply happened to meet. What moved over the month is worth naming, because it was not the market: the deducted sovereign spread was reset from 6.37 % to 5.97 % in a new edition of the published ladder, while the yield on the one-year issue itself shifted by 0.02 p.p. A gap is more informative than a coincidence, but only over a longer series, and only if you remember which input moves it. The derivation travels with the inflation differential; the internal point travels with the government bond market.

Picture 2. Two routes to the hryvnia risk-free rate, each with its own inputs and its own horizon: on the left, from the dollar rate through the inflation differential over ten years; on the right, from the government bond yield at the auction of 25.08.2026 through the deducted sovereign credit spread, at about a year. 

Every step is a choice, and each one is worth showing rather than assuming. The benchmark is the one-year point because it is the shortest clean market point on the curve. And each month the result is checked against an independent bottom-up estimate, built the same way the theory is: inflation plus the real reward for waiting, approximated by real growth. On actual data that gives 7.7 % plus 0.6 %, or 8.3 %, against the market’s 9.20 % – a divergence of 0.9 p.p. Run the same estimate on household inflation expectations instead, 9.95 % over twelve months in the NBU’s July survey, and it rises to roughly 10.5 %. Showing only the check that agreed would pass a choice of input off as the absence of one. The gap itself measures how much future disinflation the government bond market has already priced in and the household survey has not. It is a check, not a standalone estimate.

V. Where the curve ends

Beyond the one-year point the internal curve offers two further clean points: 9.7 % at 658 days and 10.5 % at 1,274 days. There, at about 3.5 years, it ends. That second figure carries the same digits as the forward-looking cross-check two paragraphs above, and the two are different objects entirely: one is a point on a curve of accepted auction bids, the other a macro reconciliation. A line cannot be extended into territory where no clean market point exists, and stretching the internal rate to ten years would pass invention off as observation.

Picture 3. The internal hryvnia curve across the maturities that actually exist, and the emptiness beyond them, set against the ten-year international derivation. 

VI. The two objections, and the limits I keep

The first objection I get from an attentive reader is that country risk has been counted twice – deducted from the risk-free rate and then added back as a premium. It has not. The deducted spread returns as a separate term: the country risk premium in the cost of equity, the sovereign credit spread in the cost of debt. On the debt side the subtraction and the addition cancel exactly, reproducing the observed yield. Country risk is counted once; what changes is that it is now visible as its own line rather than buried in a base rate.

The second objection is sharper, and it is the one that makes people uncomfortable: the internal rate sits below the NBU’s key policy rate of 15.5 %, in force since 31.07.2026. There is no absurdity in that. It is the shadow yield of a hypothetical default-free hryvnia issuer, not a rate at which anyone places money. Even one-year government paper is placed below the key rate – 15.17 % against 15.5 % at the primary auction, a weighted average of accepted bids rather than a secondary-market print – which is itself another reading of a closed circuit carrying more hryvnia liquidity than it can lend out.

The limits belong in the open, because on a page like this they are part of the product rather than a footnote to it. The credit spread is rating-based rather than market-traded, from a ladder calibrated on long-dated dollar instruments, so for a one-year hryvnia point neither the maturity nor the currency matches, and the direction of the resulting bias is unknown. The international derivation rests on the current rather than the long-run inflation differential, and the dollar rate is overstated by roughly 0.1 p.p., because a monthly cycle uses the five-year US CDS where the canon would use the ten-year one. The most important limit is not methodological but temporal: the split into two definitions holds for exactly as long as the wartime regime does. It is a property of the regime, not of the hryvnia.

VII. What I’d say now

The conclusion is unsatisfying for anyone who wants one number, and more honest for that reason. For me, two numbers side by side are not a confession of uncertainty; they are an accurate description of a country where capital inside and capital outside live under different rules, and where reconstruction will be financed by both at once. Anyone discounting a cash flow is entitled to see what the rate doing the discounting is made of, and which of the two worlds it belongs to. A valuation that hides that choice inside a single imported figure is not more precise than one that shows it – it is only quieter about being wrong.

The vessels remain connected by design, and one day the tap will be opened. Arbitrage will flow again and pull the levels back together on its own, without any tables of ours. Until then, the level in each vessel has to be measured separately – and measuring it is still the cheaper of the two available options.

The Rooster Doesn't Make the Dawn: What Ukrainian Data Say About the NBU's Real Reach

 I. Why this post exists now

    I wrote this because I got tired. Tired of hearing the same unexamined premise in academic seminars and business conversations alike — that the National Bank of Ukraine is the institution doing the job: setting the rate, taming inflation, drawing the government yield curve by decree. It gets repeated as settled fact, and almost no one stops to test it against the data. The push to actually run that test came from Aswath Damodaran’s 2024 post, “Fed up with Fed Talk? Fact-checking Central Banking Fairy Tales!”, which makes the case for the United States: the Fed is far more a follower of markets than a setter of them — a rooster whose crowing the yard mistakes for the cause of sunrise.

    Damodaran ran it on the US. Ukraine is not the US — and here the dependency is less obvious, because our financial system is thinner and bank-centric rather than market-centric, and because since 2015 the NBU has operated under a far more rigorous, externally disciplined inflation-targeting regime. Those two features make the regulator look more powerful than its American counterpart, which is exactly why the myth is stickier here. But the fundamentals are always the same: a policy rate is a reaction to inflation and growth before it is a cause of anything, and the further you move from the overnight corridor, the less the central bank is setting and the more the market is.

II. The three myths in one sentence

    Ukrainian public discussion typically rolls three claims into one sentence: that by raising the key policy rate the NBU has administratively (1) set the rate of inflation, (2) set the yield on government bonds, and (3) set the cost of corporate bank credit. For journalism this is convenient. For practice it is dangerous — because it glues three different transmission mechanisms into a single administrative act. The paper tests each of them separately on Ukrainian data, and each of them fails.

III. Reading the policy rate as a reaction, not a decision

    If the central bank reacts to macro conditions rather than setting them, then the rate itself ought to be explainable from the simplest pair of those conditions: lagged inflation and lagged real growth. Damodaran’s intrinsically-implied risk-free rate is the arithmetic sum of those two; for our purposes a linear regression is more honest than the sum.

The empirical core of the paper is a two-variable regression of the quarterly average policy rate on lagged year-over-year CPI and lagged real GDP growth, calibrated separately on four institutional windows:

• 2002–2025, the long history: R² = 0.54. Useful as background but it mixes institutionally incompatible regimes — pre-Maidan and post-Maidan, peace and war.
• 2015–2025, the post-Maidan period: R² = 0.58. The NBU is operating under tighter external discipline from this point.
• 2015–2021, the cleanest peacetime approximation: R² = 0.62. Pre-war post-Maidan window.
• 2022–2025, wartime: R² = 0.73. The narrower we draw the window, the more disciplined the rate-setting looks.

    Picture 1. Actual NBU policy rate versus the implied value of the lagged CPI + GDP model. The fit tightens as the window narrows from the long history to wartime

    The climb in R² is the central reading. It is not consistent with an arbitrary administrative rate; it is consistent with a regulator that reacts more tightly the more dominant inflation becomes in the macro environment. Wartime also doubles the loading on lagged CPI: the coefficient moves from 0.33 in the peacetime window to 0.57 in wartime — each printed percentage point of inflation translates into roughly 0.57 p.p. of policy rate next quarter. That doubling is itself an argument against the administrative reading. If the rate were set arbitrarily, its sensitivity to inflation wouldn’t have to double in exactly the regime where inflation became the dominant macro stressor.

    A richer meeting-level reaction function (94 NBU rate decisions, February 2015 to March 2026) adding 12-month household inflation expectations, one-month hryvnia depreciation, and a war-regime dummy lifts R² to 0.74. The two heaviest coefficients are household expectations (β = 0.65, p < 0.001) and the war dummy (≈4.8 p.p., p < 0.001). Same reading: the policy rate looks like a reaction function, not a decision variable.

    The honest limit. A pseudo out-of-sample test inside each window confirms that the model does not beat the naive “rate stays unchanged” rule on the level of next quarter’s rate. Naive persistence is a hard benchmark on quarterly data — most variants of the Taylor rule on developed economies lose to it too. But on the direction of the next move — up, down, or hold — the model classifies correctly about twice as often as naive persistence in every window, including the wartime one. That is real structural information about where the rate is going; it just is not an instrument for forecasting the level.

IV. Where the signal actually passes — and where it doesn’t

    If the rate itself reads as reactive, the natural follow-up is how far its signal then propagates into market prices. The answer in Ukrainian data is “not nearly as far as the public discussion assumes.” The signal is tightest at the very short end of the money market and loosens steadily as we move outward.

    Picture 2. The hierarchy of direct rate-setting: how far the policy-rate signal reaches across UONIA, government-bond tenors, bank rates and consumer inflation.

    a.  UONIA — the overnight interbank rate. Operationally tied to the policy-rate corridor. Descriptive R² ≈ 0.96 between UONIA and the policy rate over 2020–2026; after the October 2023 corridor redesign it is closer to one. UONIA is the closest thing in the system to “the NBU sets this.” It is also the only thing the NBU literally sets, and it is not a price anyone outside the interbank market actually trades.

    b.  Domestic government bonds, by tenor. Pre-war the short end (<1 year) loads on the policy rate at β = 0.77 with R² = 0.89; the 1–3 year bucket at β = 0.53, R² = 0.81; the 3–5 year bucket at β = 0.31, R² = 0.53. The further out the curve, the less the policy rate explains, and the more the price is doing its own work — pricing inflation, liquidity, and fiscal risk. Post-24.02.2022 the short-end coupling collapses outright (β = 0.09, R² = 0.04): the NBU raised the rate to 25 % in June 2022 and government-bond yields did not follow immediately. The NBU’s own communication at the time acknowledged it was waiting for government-bond and deposit rates to reprice, which is the tell — administrators don’t wait for the price they set.

    Picture 3. NBU policy rate versus government-bond yields and spreads by tenor. The short end tracks the rate; the long end carries its own inflation, liquidity and fiscal premia.

    c.  Commercial bank lending and deposit rates. Pre-war these are the cleanest transmission story in the system: β = 0.57 for non-financial corporate credit, β = 0.62 for deposits, both with R² > 0.88, both with a one-month lag. Post-war the betas fall to ≈0.30 and R² to ≈0.66–0.68. Even in the tightest peacetime regime the pass-through is partial — credit risk, liquidity, bank competition, and a wartime premium all write the final corporate borrowing rate on top of the policy-rate signal.

    The combined reading: the NBU directly sets the corridor and indirectly anchors the very short end of the curve. Outside that narrow zone its signal arrives partially and with a lag — and increasingly so as the instrument lengthens or the regime breaks.

V. Inflation: the market does most of the work

    The third myth is the centrepiece — the claim that the central bank sets inflation. We test it in two steps.

    Step one: estimate a market-only CPI equation with lagged producer-price inflation, three-month hryvnia depreciation, and 12-month household expectations. The model explains 64.2 % of the variation in year-over-year CPI over August 2014 – March 2026, before any monetary variable enters. Household expectations carry most of the load (β ≈ 1.92); lagged PPI contributes the rest.

    Step two: add the 12-month-lagged policy rate to the same specification. R² rises from 64.2 % to 68.3 %, and the rate enters with a negative, statistically significant coefficient (β ≈ –0.47). On a robustness check with core CPI as the dependent variable, the rate coefficient remains negative and significant (β ≈ –0.41), with R² = 0.65.

    Picture 4. Actual CPI versus the market-only model and the model with the lagged policy rate added. The monetary variable adds a thin lagged layer on top of a market-driven base.

    That four-point lift is real, but it is small relative to the market layer that was already there before any monetary variable arrived. Inflation in Ukraine is born from the market first and gets a thin lagged disinflationary topcoat from the policy rate later. That is not “the NBU set the rate of inflation.” That is the NBU dampening, with a lag, an inflation print whose level was determined elsewhere.

VI. The cross-regime lesson

    The pattern across the three myths converges on one reading. The NBU’s signal is tightest where the rule is literally administrative — the corridor and the overnight interbank market — and loosens systematically as we move outward into prices that other players also write. By the time we reach consumer inflation, the policy-rate contribution is a thin lag on top of a market-driven layer that already explains most of the variation. The wartime window is the partial exception: in a crisis the NBU can move proactively, as it did in June 2022, but that proactive step does not translate automatically into government-bond yields, bank rates, or consumer prices either — the long tail of transmission stays partial and lagged even when the headline policy move is decisive.

VII. What I’d say now

    The conclusion I keep coming back to is deliberately less complimentary than the one that dominates the Ukrainian public discussion. The NBU is best read as an administrative, constrained, reactive institution. Its signal is tightest at the very short end of the money market and in wartime hryvnia repricing; outside those zones its impact on the broader structure of rates and on inflation is partial, lagged, and widely overstated. The policy-rate decision is not the cause of the macro picture; it is a structurally disciplined response to the macro picture, executed under tighter external discipline since 2015 and especially since 2022.

    Damodaran’s bird is the right metaphor — the rooster whose crowing the yard credits with the sunrise it merely accompanies. The practical implication for Ukrainian economic communication, in wartime and toward post-war reconstruction, is that we should stop building narratives of macro-stability on an overstated NBU power over prices. The market sets the rate; the regulator’s task is to be disciplined enough not to obstruct it.

Shell, Eleven Months Later: A Retrospective Check on the June 2025 Valuation

    I. Why this post exists

    In my previous post on Shell I valued the company at 13 June 2025, ran four methods in parallel, and ended with a Kennedy refrain handing the decision over to the reader. The post stopped at a buy-decision built on a 55.9 % probability of underpricing in the base run, with the explicit thesis that the next leg up would come from fundamentals rather than from a higher Brent print. It did not commit to anything beyond that. The honest thing to do – eleven and a half months later – is to come back and check.

The window is unusually clean. On 2 March 2026 a major geopolitical incident around the Strait of Hormuz spliced the observation period into two regimes – a quiet, fundamentals-driven stretch from 13 June 2025 to 27 February 2026, and a shock-driven stretch from 2 March 2026 through (so far) 26 May 2026 – twelve weeks long enough that the shock half is now meaningfully bigger than a single news cycle. That accidental split is the kind of natural experiment a valuation methodology rarely gets in clean form: a calm regime to test the fundamental layer of the model, a shock regime to test the option layer.

    II. What was on the table at 13 June 2025

    A quick recap, for readers who didn’t see the June post.

    At the valuation date Shell traded at $36.25 per share on the article basis. I applied four methods in parallel, each yielding a per-share number on a consistent share count:

    • Fundamentals normalisation – baseline (cycle-median revenue × 8 % operating margin): $26.50. The 8 % was the 2010–2024 cycle-median operating margin; the revenue side was the cycle-median top line over the same window. On its own logic, Shell looked overpriced.

    • Fundamentals normalisation – conservative (cycle-median revenue × 10 % operating margin): $42.16. I called the higher margin conservative precisely because the company was already earning above 11 % and management had given no sign of reversal. On its own logic, underpriced.

    • Oil-price normalisation: $41.68 at the spot of $73, $44.09 at the cycle-median $75.85, with a middle of $42.89. Underpriced.

    • Monte Carlo with correlated Brent and operating-margin draws (100 000 trials): median $40.50; 5th–95th percentile range $25.50–$64.30; 55.9 % of the mass above market. Underpriced on median.

    Four anchors, not one. The fundamentals baseline at $26.50 was deliberately the cheap stress test – it asks what Shell would be worth if its sustainable operating margin reverted from the 11–14 % range it had been printing back to the 2010–2024 cycle-median 8 %, i.e. if recent above-cycle profitability were treated as a temporary tailwind rather than a sustainable level. The other three methods clustered between $40 and $44. I bought at $36.25 with the explicit thesis that the next leg up would come from fundamentals – sustained margin above the cycle median, the buyback, the rising payout ratio – rather than from a higher Brent print.

    III. The window, the regimes, the gap that closed

    Between 13 June 2025 and 26 May 2026 the share price rose 18.3 % on the consistent article basis, from $36.25 to $42.88. Brent ran the opposite way through the calm half of that window – from $76.00 on the valuation date down to $71.32 on the eve of the shock, or –6.2 % and then broke regime on 2 March. The first half of the window is the cleaner test: the re-rating happened despite weak oil, not because of it.

Picture 1. Brent and Shell rebased to 13 Jun 2025 = 100. The dashed line marks 2 Mar 2026 – the Hormuz shock. Brent series ends 18 May 2026 (FRED 5-business-day publication lag).

    Splitting the window at the day before the Hormuz shock sharpens the point. In the pre-shock regime, from 13 June 2025 to 27 February 2026, Shell rose 15.1 % on basis – almost the entire long-run gap to the methods that had said underpriced closed in nine months on a falling Brent. The share price crossed the Monte Carlo median ($40.50) on 25 February 2026, the lower edge of the oil-price normalisation band ($41.68) on 27 February 2026 – literally the eve of the shock – and the conservative fundamental ($42.16) and the oil-price middle ($42.89) on 6 and 11 March 2026 respectively. None of those crossings coincided with Brent above $75. The market re-rated Shell to its intrinsic-value range on the strength of fundamentals, before there was any extra geopolitical premium to allocate.

Picture 2. Monte Carlo distribution of Shell’s intrinsic value (100 000 draws, log-normal calibrated to the June 2025 paper). Markers show where the realised price sat at three reference dates: the valuation date, the eve of the shock, and today.

    Then came 2 March 2026. Brent indexed to the valuation date jumped from 94 to a peak of 182 – close to a doubling off the week before. Shell on the same index moved from 115 to a post-shock peak of 130 on 7 April 2026, then drifted back to 118 by 26 May as Brent receded from its peak. That is a striking asymmetry: an almost doubling in spot Brent delivered, at peak, fifteen index points on top of a share that had already re-rated by fifteen – and most of that incremental fifteen has since unwound while Brent is still trading around 155. The interpretation is straightforward once we sit with the model. The bulk of Shell’s value sits in already-producing assets whose discounted cash flows are not very sensitive to a one-quarter spike in spot Brent. The option layer on undeveloped reserves – Whale, Bonga North, Manatee, the rest – is the only piece of the valuation that should react to a regime change in oil-price volatility, and it did, but in measured size, and only for as long as the volatility regime itself looked changed.

Picture 3. Shell daily close (article basis) against the three near-money valuation anchors – Monte Carlo median, conservative fundamental, oil-price middle – with the Hormuz shock dashed in. The stress-test anchor (Fundamentals 8 % at $26.50) sits below the plotted range.

IV. Reading out each method against eleven months of price tape

    Picture 4 below plots the mean absolute deviation of the realised price from each method’s central anchor, separately for the two regimes. The ranking is informative because it inverts.

Picture 4. Mean absolute deviation of realised price from each method’s central anchor – pre-shock regime (green) versus post-shock regime (orange). Computed from daily Shell article-basis prices, 13 Jun 2025 – 26 May 2026.

    a. Pre-shock regime (the fundamental period). Monte Carlo wins (MAD 9.4 %). The conservative fundamental at 10 % margin is a close second (12.9 %). The oil-price normalisation comes third (14.4 %). The fundamentals baseline at 8 % is last by a wide margin (38.6 %) – which confirms what was already obvious in June: an 8 % anchor is artificially low for a company already earning above 11 %.

    b. Post-shock regime (the geopolitical period). The ranking flips. Oil-price normalisation is now the most accurate (MAD 4.2 %) – intuitively correct, because that method is the one explicitly tied to where oil prices sit. The conservative fundamental holds up unexpectedly well (5.4 %), because the margin anchor turned out to be the right long-run anchor regardless of regime. Monte Carlo is third (9.5 %), barely worse than its pre-shock score; the reason it loses any accuracy at all is that the empirical 2005 – mid-2025 Brent distribution simply did not contain prints above $130, so the realised path strayed outside the bulk of the simulated distribution. The fundamentals baseline at 8 % is again last and by a much wider margin (67.4 %), because the share price kept walking away from a value anchor that was wrong to begin with.

    The cross-regime lesson is the central takeaway of this retrospective. No single method dominates in both states of the world. Monte Carlo is the right tool when the market is weighing fundamentals; oil-price normalisation is the right tool when the market is reacting to a price shock; the conservative fundamental is a useful sanity check across both; and the baseline at the cycle-median margin is mostly a stress test that tells us what Shell would be worth if everything we knew about its recent operating performance were wrong. The right output is the range the four methods jointly span, not any one anchor on its own.

    V. The option layer, an area to keep exploring

    In the June post I priced Shell’s undeveloped reserves as a portfolio of European call options on oil, using Black-Scholes with q = 1/n to penalise time-to-expiry. That layer added a non-trivial premium to the producing-asset DCF. A few fellow financiers reasonably asked whether the option layer was doing real work or just decorating the model with mathematics.

    Let me be precise about what the option-pricing model actually tells us. Black-Scholes is unambiguous on direction: higher realised volatility raises the value of a long-dated call, and lower volatility releases that premium back. Between 2 March and roughly the second week of April 2026 Brent’s realised volatility jumped sharply, then started normalising; over the same window Shell traced a path from $42 to a peak around $47 article basis on 7 April and back toward $43 by 26 May. The option layer must have contributed something to that round trip – the model leaves no room to claim otherwise. What I am not ready to tell you is how much. Disentangling the option-layer contribution from the producing-asset DCF response inside a single shock episode is its own piece of work; I treat it as an open area for further exploration, not a question I have answered here.

    VI. The Graham line, eleven months late

    Benjamin Graham’s observation about the market being a voting machine in the short run and a weighing machine in the long run gets quoted often enough that it has nearly worn itself out. The Shell window is a rare clean example of what he meant. In the nine months before the shock there was no news on oil that should have re-rated Shell by 15 % – Brent went down over that period. What changed was that the market progressively recognised what the four methods had already weighed in June: that Shell’s sustainable operating margin was closer to 14 % than to 8 %, that the buyback was real, that the reinvestment rate was settling at a level consistent with a mature commodity producer returning more cash than it deployed. The market weighed. Then geopolitics arrived and the market voted – briefly, on volatility – in a direction broadly consistent with what an option-pricing layer would predict, before giving most of the incremental premium back as volatility eased. The producing-asset DCF carried the underlying value through both halves.

    VII. And so, dear friends, you just have to carry on

    The cleanest evidence sits in the calm half of the window, not the loud half. By 27 February 2026, the eve of the Hormuz shock, Shell had already crossed the Monte Carlo median ($40.50), touched the lower edge of the oil-price normalisation band ($41.68), and was within striking distance of the conservative fundamental ($42.16) and the oil-price middle ($42.89). On a falling Brent. The market closed the gap between $36.25 and the band the four methods jointly defined while spot oil drifted from $76 to $71. That is the part of the test the Iran shock did not run for me: it had already finished, in my favour, before the first headline out of Hormuz.

    One footnote on hindsight. The post-shock numbers benefit from it; the pre-shock crossings do not – they happened in the calm regime, with Brent working against the trade, and they happened to all three near-money anchors before 2 March.

    Every time I sit down with a valuation I come back to professor Damodaran’s reminder: it is all about fundamentals. And this is another example of professor’s wisdom.