Le texte de l’ouvrage est en anglais, tel qu’il a été écrit ; seule l’interface du site est en français.

Annex J — International extension protocol (comparative panel and the 44 LDCs)

Annexe 10 sur 17 · ≈ 24 min de lecture · 104 paragraphes · 32ᵉ des 40 sections de lecture · rechercher dans l’ouvrage

of the accounting-incidence and partial-attribution framework for monetary externalities¶ 1

Global comparative panel and application to the 44 least developed countries (LDCs)¶ 2

Proposed methodology — Version 1.0¶ 3

Status of sources and of the LDC list: 23 July 2026¶ 4

Status: proposed research protocol. It does not yet contain causal results for the countries studied.¶ 5

Executive summary — Annex J

As at 23 July 2026, the United Nations list comprises 44 LDCs: 32 in Africa, 8 in Asia, 1 in the Caribbean and 3 in the Pacific.[1]¶ 6

Bangladesh, the Lao People’s Democratic Republic and Nepal remain included in the reference panel, even though their graduation from the category is scheduled for 24 November 2026. The panel must be frozen according to status at the start of the study period in order to avoid a break in comparability.[2]¶ 7

The preliminary audit of the international data infrastructures leads to the following findings:¶ 8

• 41 of the 44 LDCs have a country page in the World Bank’s International Debt Statistics (IDS); Kiribati, South Sudan and Tuvalu do not appear in the country list displayed by IDS, which does not imply the absence of any data but requires alternative sources.[3]
• 18 LDCs appear in the QEDS-GDDS availability file updated on 30 April 2026; 13 of them have reported data at least up to the second quarter of 2024.[4]
• 23 of the 44 LDCs appear on the World Bank’s FY26 list of fragile and conflict-affected situations: 14 in the “conflict” category and 9 in the “institutional and social fragility” category.[6]
• General macroeconomic data are more widely accessible than the contractual clauses required to measure floating-rate debt, reset dates, hedges and the ultimate identity of those bearing losses and gains.[5]

The recommended strategy is therefore a three-ring architecture:¶ 9

1. Calibration and control ring: Tunisia, Morocco, Jordan, Kenya, South Africa, Chile and Indonesia.
2. Stress ring: Ghana, Pakistan, Sri Lanka, Nigeria and Egypt.
3. LDC census: standardised application to the 44 countries, with depth varying according to data quality.

Structure of the protocol

1. Purpose, scope and levels of evidence¶ 10

2. Research architecture: calibration, stress tests and LDC census¶ 11

3. Unit of analysis, period and common taxonomy¶ 12

4. Data architecture and provenance register¶ 13

5. Identification of source-jurisdiction monetary shocks¶ 14

6. Counterfactuals and predetermined exposure¶ 15

7. Measurement by mechanism and elimination of double counting¶ 16

8. Partial attribution, panel econometrics and heterogeneity¶ 17

9. Distribution across agents, offsets and consolidation¶ 18

10. Handling of missing data and stopping rules¶ 19

11. Feasibility for the 44 LDCs and deployment in waves¶ 20

12. Governance, replication and publication¶ 21

13. Project organisation, schedule and deliverables¶ 22

14. Limits, antithesis and rejection criteria¶ 23

Annex J.A. Minimum data dictionary¶ 24

Annex J.B. Preliminary feasibility matrix for the 44 LDCs¶ 25

Annex J.C. Country passport template¶ 26

J.1 Purpose, scope and levels of evidence

The purpose of the protocol is to extend the method of the new Annex F to a set of countries sufficiently diverse to test its external validity, and then to the whole set of LDCs. The extension must preserve the distinction between four objects that cannot be conflated:¶ 27

• prior exposure to the shock;
• the accounting incidence observed relative to a counterfactual;
• the share plausibly attributable to an identified monetary shock;
• the qualification in terms of welfare, equity or legal responsibility.

The project normally stops at the first two levels for the universal census and crosses the third level only for countries satisfying the identification conditions. The fourth level falls to S0–S6 qualification and to the corresponding institutions of this report; it is never mechanically deduced from an economic estimate.¶ 28

J.1.1 Scale of permitted results

J.1.2 Comparative research question

The central question is the following: to what extent does one and the same global monetary tightening cycle produce different incidences according to the debt structure, the exchange-rate regime, financial openness, fiscal room, institutional credibility and the adaptive capacity of recipient countries?¶ 29

The objective is therefore not merely to measure costs. It is to distinguish:¶ 30

• the common component of the global shock;
• the exposure structure inherited before the shock;
• domestic amplifications;
• economic policy responses;
• contemporaneous gains and their beneficiaries;
• the elements that remain impossible to attribute.

J.2 Research architecture: calibration, stress tests and LDC census

J.2.1 Ring 1 — calibration and control panel

The first ring must comprise countries with relatively rich data and contrasting financial structures. It serves to validate the measurement functions, the non-double-counting rules, the counterfactuals and the econometric tests before any generalisation.¶ 31

J.2.2 Ring 2 — stress tests

The second ring brings together countries where external shocks combine with domestic crises, restructurings or regime changes. It verifies that the method is capable of refusing excessive attribution.¶ 32

• Ghana: restructuring and strong interaction between fiscal policy, the exchange rate and debt;
• Pakistan: external constraints, energy, the IMF and multilateral debt;
• Sri Lanka: default and restructuring, a case in which the simple counterfactual breaks down;
• Nigeria: oil, exchange-rate reforms and domestic monetary financing;
• Egypt: devaluations, large financing needs and a multiplicity of simultaneous shocks.

J.2.3 Ring 3 — census of the 44 LDCs

The third ring aims at universal coverage, but with differentiated products. All countries receive a data and exposure passport. Only countries crossing the quality thresholds produce accounting-incidence ranges; only those with sufficiently granular data and an identifiable shock reach in-depth causal analysis.¶ 33

J.3 Unit of analysis, period and common taxonomy

J.3.1 Indices and unit of observation

The protocol takes up the architecture of Annex F, correcting the consolidation of transfers and the currency conversion. The indices are: recipient country i, source monetary jurisdiction j, affected agent h, mechanism m, primitive economic component k, period t and counterfactual c.¶ 34

The primitive component is the unit of aggregation. One and the same determinant — for instance the exchange-rate movement — may feed several reporting mechanisms, but its monetary value is counted only once.¶ 35

J.3.2 Study period

The recommended main window is 2019–2025:¶ 36

• 2019–2021 to measure prior exposures and trends;
• 2022–2024 for the core of the tightening cycle;
• 2025 for delayed effects, refinancings and first normalisations;
• longer windows are used to estimate elasticities and to check pre-trends.

For countries in conflict, default or restructuring, the window must be segmented into regimes. An average over the whole period would be misleading if the economic mechanism changes in nature.¶ 37

J.3.3 Common taxonomy of mechanisms

The fifteen mechanisms of this report remain the reporting channels: A1–A7 for the monetary and financial channels; B1–B2 for imported prices; C1–C3 for the real economy; O1–O3 for contemporaneous offsets. The S item for the revaluation of the debt stock remains outside cash flows.¶ 38

J.4 Data architecture and provenance register

J.4.1 Hierarchy of sources

J.4.2 Mandatory provenance register

Each data item must have a stable identifier and the following fields:¶ 39

• country, source institution, URL or reference, date of extraction and version;
• unit, currency, price base, frequency, coverage and revision method;
• status: observed, reported, reconstructed, imputed or simulated;
• start and end date, series break and metadata quality;
• economic agent initially affected;
• primitive component, reporting mechanism and allocation rule;
• licence, confidentiality and possibility of publication.

J.4.3 Data readiness score

A data readiness score (DRS) is computed before any estimation. It measures neither the quality of economic policy nor the governance of the country; it measures only the capacity of the case file to support a given level of evidence.¶ 40

Proposed thresholds: DRS < 40: L0–L1; 40–59: cautious L2; 60–74: robust L2 / L3 preparation; 75 and above: candidate for L3, subject to shock identification and contractual cooperation.¶ 41

J.5 Identification of source-jurisdiction monetary shocks

J.5.1 From the scalar shock to the multidimensional vector

To represent a tightening cycle by a scalar in basis points amounts to assuming that all the impulses of a central bank are transmitted through the same channel, with the same horizon and the same signature. The assumption is false: a policy-rate surprise, a revision of the expected path, new information revealed about the economy, a balance-sheet reduction and a withdrawal of refinancing operations reach neither the same segments of the curve, nor the same agents, nor the same lags.¶ 42

S(j,t) = [ S_level , S_path , S_information , S_balance_sheet , S_liquidity , S_operational , S_backstop ]¶ 43

Why enrich the upstream rather than the downstream. Adding transmission mechanisms would make the register of primitive components, the overlaps to be audited and the risks of double counting grow combinatorially. Enriching the shock vector leaves the taxonomy intact and turns the addition into a gain in identification: each additional impulse supplies a distinct prediction across mechanisms already defined. That is the difference between making a model more complex and constraining it further.¶ 44

J.5.2 The seven components

The backstop component calls for a remark: it is the only one whose expected effect is of opposite sign to the others. An active swap line attenuates transmission; its absence or its withdrawal aggravates it. It therefore enters the vector in the role of a modulator, and not of an additional adverse shock — to conflate it with the other six would invert its interpretation.¶ 45

J.5.3 Signature matrix: the identification gain

The main benefit of the enrichment is not to measure more things. It is that each component possesses a predictable signature across the transmission mechanisms already defined, so that the whole forms an overdetermined system of restrictions. A seven-component vector applied to some fifteen mechanisms produces far more signed predictions than free parameters — which makes the construction refutable rather than merely richer. This multidimensional identification strategy extends the decomposition of monetary policy shocks (Gürkaynak, Sack and Swanson, 2005; Swanson, 2021).¶ 46

J.5.4 Rank conditions and fallback scale

Moving from one impulse to seven over a single episode multiplies sevenfold the problem that the 2022–2025 window already poses. The protocol must therefore state, before estimation, under what conditions the decomposition is identifiable and what it does when it is not.¶ 47

Pre-registered fallback scale. If the conditions fail, the protocol does not relax its thresholds: it steps down one rung on a scale declared in advance. Seven components when all the conditions are satisfied; four in the event of partial failure, grouping level and path on the one hand, balance sheet and liquidity on the other, information and backstop remaining separate; two in the event of broad failure, separating only the price of money and the quantity of money; a single composite impulse as a last resort, with an explicit declaration that the decomposition is not identifiable for the episode. Each step down is published together with the diagnostics that motivated it. Choosing the rung after inspecting the estimates is expressly prohibited.¶ 48

J.5.5 Information effect and factor rotation

The information component is the most important addition to the vector and the most demanding to identify. Its logic is the following: when a central bank tightens, it also reveals what it knows about the economy. Part of the market reaction — and therefore of the incidence in the recipient countries — reflects not policy but information about the state of the world. To attribute that part to policy would overstate the attributable effect; to ignore it would understate it.¶ 49

Identifying restriction. The separation rests on the co-movement, around the announcement, between interest rates and the prices of risky assets. A pure policy tightening raises rates and depresses assets; a favourable information revelation raises both together. The sign restriction on this co-movement is what identifies the component; without it, a factor rotation produces orthogonal axes devoid of structural meaning. The protocol requires that the restriction be declared before estimation and that the sensitivity of the result to its relaxation be published. The corresponding high-frequency identification methods — the Jarociński & Karadi (2020) decomposition for the Federal Reserve, the EA-MPD database (Altavilla et al., 2019) for the ECB — are those of Annex G (§G.3).¶ 50

The same requirement applies to the separation between level surprise and path surprise: the statistical orthogonality of two factors confers upon them no structural interpretation. The rotation adopted, its identifying restriction and its alternative must be documented. A decomposition whose interpretation depends on an untested choice of rotation is not identified.¶ 51

J.5.6 Multi-source-jurisdiction model and exposure matrix

The LDCs are exposed to several monetary jurisdictions. A separate estimation for the Fed, the ECB, the Bank of England or others, then summed, may count a common global component twice. The protocol adopts a joint shock vector.¶ 52

The shocks are identified by a combination of high-frequency surprises around announcements, narrative instruments, revisions of expected paths and global risk factors. Each method produces a separate series; their convergence and their divergences are published.¶ 53

The relevance of a source jurisdiction varies according to: the currency of the debt, the invoicing currency of trade, the exchange-rate regime, the share of reserves, the creditors, the banking structure and the exposure of investors. The protocol does not use a single arbitrary weight; it produces a pre-shock exposure matrix.¶ 54

J.5.7 Source-jurisdiction instruments and recipient responses

The instruments of the source-jurisdiction central bank belong to the shock vector; the responses of the recipient central bank never do. An increase in the domestic policy rate, foreign-exchange intervention, a drawdown on reserves, capital controls and macroprudential measures are response variables, to be treated as endogenous mediators, with instrumentation or sensitivity bounds, and not as additional impulses. To conflate them would amount to attributing to the source jurisdiction a decision taken by the recipient.¶ 55

Conditional activation by regime. Some components have no channel in certain recipient countries: the balance-sheet component has no purchase on an economy without a liquid domestic bond market; the exchange-rate channel is structurally neutralised under a hard peg or dollarisation. Such components are explicitly deactivated in the passport of the country concerned, together with the reason, rather than estimated and reported as non-significant. Non-significance through the absence of a channel and non-significance through the absence of an effect are two different results and must be distinguished.¶ 56

J.5.8 Prevention of double counting

Each component of the vector feeds primitive components allocated only once, with weights summing to one. Three overlaps specific to this enrichment must be audited explicitly: between path and balance sheet, which both act on intermediate maturities; between liquidity and operational, which both affect banking conditions; and between information and path, whose separation depends entirely on the sign restriction of §5.¶ 57

Principal allocation and robustness. The principal allocation follows the structural ordering of the impulses — level, path, information, then execution through the balance sheet and liquidity — each component being measured conditionally on the preceding ones. The Shapley decomposition is retained as a robustness test and not as the principal allocation: it assumes a symmetry between contributions that structurally ordered shocks do not satisfy, and its combinatorial cost grows exponentially with the number of components. A large discrepancy between the structural allocation and the Shapley allocation is information to be published, not a reason to prefer the latter.¶ 58

J.5.9 Composite index: strictly limited use

An aggregate index of effective tightening, gross or net of offsets, is useful for communication and for comparing episodes. It cannot serve as an input to an attribution estimate, for a simple reason: it glues back into a scalar what the vector has just separated, and its weightings then carry the whole normative content of the result. The protocol authorises its publication provided that the weightings are declared, that the sensitivity of the ranking to alternative weightings is published, and that the index appears in no regression whose output feeds a qualification or a calibration. The wording “instrument of communication, not of attribution” accompanies every publication of the index.¶ 59

J.5.10 Priorities and graduated compliance

The vector is deployed by priorities and follows the graduated compliance of Annex F (§F.15): a mandatory core, an advanced module and jurisdictional modules, each activated according to the availability of the series and the identification conditions.¶ 60

It changes the nature of the object measured: tightening ceases to be a number in basis points and becomes a set of impulses announced, anticipated, executed, offset and transmitted differently across recipients. It also changes the evidentiary regime, substituting for a free parameter a matrix of signed predictions that the data can contradict.¶ 61

It changes nothing in what follows. The taxonomy of transmission mechanisms remains unchanged. The attainable level of evidence remains that of accounting incidence and bounded partial attribution: a better-specified vector does not cross the bridge to welfare. And the most likely conclusion for the 2022–2025 episode remains that some of the components will not be separately identifiable — which is precisely what the fallback scale is designed to make publishable rather than concealable.¶ 62

J.6 Counterfactuals and predetermined exposure

J.6.1 Minimal family of counterfactuals

• c0 — rates held at their end-2021 level: a demanding scenario, but not an upper bound by definition;
• cN — gradual normalisation towards a neutral rate;
• cR — path based on a reaction rule or on market expectations;
• cS — scenario with an identical supply shock but different monetary tightening, in order to isolate monetary policy from energy and food.

Each counterfactual is a coherent vector of rates, risk-free curves, exchange rates, global activity, risk aversion and commodity prices. It is not acceptable to alter the policy rate while mechanically holding all the other observed variables.¶ 63

J.6.2 Fixed exposure and dynamic exposure

The fixed measure is the principal attribution result. The dynamic measure describes the incidence realised after adaptations. The gap between the two is informative about the cost or the benefit of national responses, without presuming that they were avoidable.¶ 64

J.7 Measurement by mechanism and elimination of double counting

J.7.1 Specific functions

No single “base × price gap” formula is suitable for the fifteen mechanisms. The contractual debt channels may use flow identities; the real-economy channels require elasticities, local projections or structural models.¶ 65

J.7.2 Register of primitive components

Each primitive component k receives an identifier, an amount, an agent, a source, a sign and an allocation. The sum of the allocation weights across mechanisms must be equal to 1.¶ 66

Example: the rise in the local cost of an energy bill due to the exchange rate cannot be counted at once as an import bill, as household inflation and as a loss of purchasing power. It is a primitive component, broken down for reporting but aggregated only once.¶ 67

J.8 Partial attribution, panel econometrics and heterogeneity

J.8.1 Bounded attribution

For each country, the attribution coefficients belong to a joint set of admissible values. The bounds are not chosen independently when the same exchange-rate, global-risk or interest-rate paths link several mechanisms.¶ 68

J.8.2 Panel local projections

The common empirical protocol may use local projections with country and time fixed effects, interacted with the predetermined exposures:¶ 69

Rᵢ represents the structural characteristics: exchange-rate regime, capital controls, share of foreign-currency debt, credibility, level of reserves, market access and import dependence. Average coefficients are transferred to an LDC only if the transportability assumptions are explicit and tested.¶ 70

J.8.3 Partial pooling

Countries with weak data may benefit from partial pooling of the parameters, but never from a mechanical transfer of results. A hierarchical structure can reduce instability while allowing the effects to vary by region, regime and debt structure.¶ 71

The national result must present separately the component estimated from the country’s own data and the component derived from pooling. An estimate dominated by the group is labelled as such.¶ 72

J.8.4 Mandatory tests

• pre-trends and temporal placebos;
• fictitious shocks over periods with no announcement;
• sensitivity to counterfactuals and to windows;
• interactions of rates × exchange rate × refinancing;
• regime breaks, defaults and restructurings;
• successive exclusion of influential countries;
• fixed-exposure / dynamic-exposure comparison;
• replication by a second team and a log of all decisions.

J.9 Distribution across agents, offsets and consolidation

J.9.1 Partition of agents

The categories must be mutually exclusive: the State, the central bank, banks, non-financial corporations, households and local authorities. Vulnerability, income, gender, region or generation are dimensions of breakdown, not additional agents that can be added together.¶ 73

J.9.2 Corrected transfer matrix

A gain accruing to one agent offsets another agent only if an observed or legally committed transfer is documented. The formula must deduct the transferred share from the initial beneficiary so as not to create resources:¶ 74

Without transfer data, T = 0. The return on reserves, the valuation of remittances and the fall in certain global prices are therefore published separately by beneficiary.¶ 75

J.9.3 Common unit and time

Two coherent methods are authorised, but only one must be applied within a given result:¶ 76

• conversion into the reference currency at the rate of each period, then deflation in the reference currency;
• deflation in local currency, then conversion at the base-year exchange rate.

The protocol prohibits mixing the price index of the recipient country in the denominator with that of the reference currency in the numerator without an explicit parity model.¶ 77

J.9.4 Three non-additive registers

• A — market and budgetary costs: cash flows, margins, credit and output losses;
• B — welfare losses: health, employment, poverty and public services, measured separately;
• C — transfers between agents or jurisdictions: distributive losses whose global total may be nil.

An LDC total must never add A, B and C together. The global protocol publishes three distinct aggregates and retains the national distributive vector as the primary result.¶ 78

J.10 Handling of missing data and stopping rules

J.10.1 Hierarchy for the treatment of gaps

1. search for the primary national source;¶ 79

2. cross-checking with IDS, QEDS, the IMF, the BIS and creditor sources;¶ 80

3. construction of a physical or accounting bound;¶ 81

4. regional proxy only if transportability is demonstrated;¶ 82

5. partial pooling with the country’s own contribution made explicit;¶ 83

6. abstention from quantifying if the interval remains uninformative.¶ 84

J.10.2 Stopping rules

J.11 Feasibility for the 44 LDCs and deployment in waves

J.11.1 Feasibility verdict

International coverage is more favourable than a pessimistic reading might suggest: IDS provides complete annual data on the external debt stocks and flows of reporting countries, while QEDS provides, for participants, quarterly breakdowns by sector, maturity, instrument and currency.[3][4]¶ 85

However, participation in QEDS is voluntary, and the e-GDDS remains less demanding than the SDDS. The IMF stresses that fragile States are over-represented among the countries not yet publishing through a National Summary Data Page.[5]¶ 86

J.11.2 Proposed waves

This prioritisation is a preliminary methodological triage, not a judgement on national institutions. It will have to be replaced by the DRS computed from an actual audit.¶ 87

J.11.3 Special cases

• LDCs approaching graduation: retain the status and the composition of the panel at the starting point; produce a transition note after graduation.
• Small island States: give priority to data on public debt, the balance of payments, disasters and transfers; small numbers may increase the volatility of ratios.
• Countries in conflict: segment the regimes, document the areas not covered and avoid any continuous interpolation across an institutional break.
• Monetary unions: distinguish the common central bank’s shock, regional policy and national vulnerabilities; do not treat the national rate as an autonomous instrument.
• Countries under restructuring: separate liquidity shock, default, relief, arrears, creditor losses and the effects of global monetary policy.

J.12 Governance, replication and publication

J.12.1 Pre-registration

Before any computation, the project publishes: the list of countries, the mechanisms, the counterfactuals, the gₘ functions, the allocation rules, the DRS thresholds, the exclusions, the tests and the authorised products. Any modification is dated and justified.¶ 88

J.12.2 Double coding and audit

• independent double coding of the contractual data and of the agents;
• inter-coder agreement and documented arbitration;
• automatic audit of units, signs, duplicates and dates;
• a replication team distinct from the production team;
• publication of null, negative and non-identifiable results.

J.12.3 Publication package

• codebook and data dictionary;
• raw sources or persistent links;
• extraction, cleaning and estimation scripts;
• register of primitive components;
• matrix of agents and transfers;
• scenarios and joint constraints;
• log of versions and decisions;
• replication report and limits by country.

J.13 Project organisation, schedule and deliverables

J.13.1 Minimum team

The staffing figures are planning assumptions, not standards. Country teams may be pooled by region, but replication must remain institutionally separate.¶ 89

J.13.2 Indicative schedule

A public L0–L2 atlas can reasonably be targeted within 18 months. In-depth contractual and causal attribution for several countries requires rather more: 24 to 30 months, subject to access to national data.¶ 90

J.13.3 Deliverables

• Global atlas of monetary exposures and incidences;
• 44 LDC passports with data score and stopping rules;
• 12 comparative calibration and stress studies;
• 8 to 12 L3 monographs with national partners;
• open database, code and documentation;
• methodological paper, panel paper and regional papers;
• institutional note for HuIB/GIRA specifying the authorised passage towards S0–S6.

J.14 Limits, antithesis and rejection criteria

J.14.1 Main antithesis

The adverse hypothesis is that the exercise would measure above all the internal vulnerability of the LDCs and the weakness of their data, rather than the effect of foreign monetary policies. This objection is serious. The protocol answers it through pre-shock exposures, control countries, multiple counterfactuals, domestic interactions and the publication of non-attributable results. It does not eliminate it by the sophistication of the equations alone.¶ 91

J.14.2 Possible biases

• transparency bias: the best-documented countries may appear more exposed;
• crisis bias: countries in difficulty accumulate domestic and external causes;
• survivorship bias: restructured debt or debt in arrears alters the observable flow;
• transportability bias: an elasticity estimated in Chile or South Africa is not automatically valid in Malawi or Tuvalu;
• currency bias: actual invoicing and hedging may differ from the currencies reported;
• distributive bias: an aggregate national gain may conceal uncompensated losers;
• political bias: the selection of shocks, counterfactuals or countries may be interpreted as a search for culpability.

J.14.3 Criteria for rejection or downgrading

• reject a national estimate if the bounds are so wide that they distinguish no scenario;
• downgrade a country to L1 if the contractual exposures cannot be reconstructed;
• do not consolidate the LDCs if the units, periods or components differ;
• abandon a common coefficient if panel heterogeneity dominates the average effect;
• refuse any pricing if the case file has not separately passed the S0–S6 analysis and the legal basis.

Annex J.A — Minimum data dictionary

Annex J.A.1 — Minimum automatic checks

Before any national computation or panel aggregation, the pipeline must run and archive at least the following checks:¶ 92

• consistency of units, currencies, dates, signs and price years;
• uniqueness of each primitive-component identifier and absence of cross-mechanism duplication;
• allocation weights summing to 1 for each component distributed across several channels;
• transferred shares less than or equal to the available gain and exact conservation of the total after transfer;
• automatic exclusion of stocks and memorandum items from all flow totals;
• presence of a versioned counterfactual for each quantified exposure;
• reconciliation of subtotals with the aggregates published by national and international sources;
• version fingerprint of the data, the code and the parameters, accompanied by a change log.

Any blocking anomaly produces a “not computable” status rather than an automatic imputation or a silent replacement by a proxy.¶ 93

Annex J.B — Preliminary feasibility matrix for the 44 LDCs

How to read: IDS indicates the existence of a country page in the list displayed by the World Bank; QEDS indicates the last period reported in the availability file as at 30 April 2026; FCV reproduces the FY26 list. The wave is a provisional methodological prioritisation. “Potential L2” does not mean that an estimate is already valid.¶ 94

* Kiribati does not appear in the displayed IDS country list, but has a recent QEDS history. South Sudan and Tuvalu require further alternative sources or direct cooperation. The absence of an IDS page must not be interpreted as an absolute absence of data.¶ 95

Annex J.C — Country passport template

Verified sources and reference documents

[1] United Nations — UN-OHRLLS, “List of LDCs”. List of 44 countries consulted in July 2026. ¶ 96

[2] United Nations — LDC Portal, “Countries approaching graduation and already graduated”. Graduation timetable. ¶ 97

[3] World Bank, International Debt Statistics (IDS). Annual data on external debt stocks and flows for countries reporting to the Debtor Reporting System. ¶ 98

[4] World Bank, Quarterly External Debt Statistics (QEDS), GDDS and SDDS availability files, updated 30 April 2026. ¶ 99

[5] International Monetary Fund, “Standards for Data Dissemination”: e-GDDS, SDDS and SDDS Plus. ¶ 100

[6] World Bank, FY26 List of Fragile and Conflict-affected Situations. ¶ 101

[7] Yaiche, N. (2026), The Humanity Compact, Annex F — Formal accounting decomposition and bounded partial attribution framework, enhanced version. ¶ 102

[8] Yaiche, N. (2026), The Humanity Compact — The Global Accountability Chain, the present volume, version 2.2. ¶ 103

Note of caution. Country statuses, graduation timetables, data coverage and fragility classifications may change. The project register will have to time-stamp each extraction and retain earlier versions for reproducibility.¶ 104

@incollection{yaiche_hc_annexe_j,
  author    = {Yaïche, Nizar},
  title     = {Annex J — International extension protocol (comparative panel and the 44 LDCs)},
  booktitle = {The Humanity Compact},
  year      = {2026},
  version   = {2.3},
  doi       = {10.5281/zenodo.22746533},
  publisher = {Zenodo},
  url       = {https://humanitycompact.org/ouvrage/annexe-j/}
}

Ouvrage de Nizar Yaïche, ancien ministre des Finances de la République tunisienne. Qui est l’auteur ?