Analytical framework
ACD uses two linked but distinct analytical systems. The climate system is country-specific and Bayesian. The trade system is a pooled Arab-country transmission layer estimated separately and connected to country-specific debt baselines and debt dynamics. Keeping the two systems separate avoids presenting pooled trade coefficients as if they had been estimated inside each country Dynare model.
Historical and physical climate conditions are translated into country-specific standardized climate stress.
Terms of trade, world demand, energy-trade conditions and export-demand innovations are translated into standardized economic shocks.
Climate or trade responses are applied to institutional debt baselines through a transparent annual debt identity.
The climate system is a country-specific linear macro-fiscal transmission model estimated in Dynare using Bayesian methods. The trade system is a semi-structural pooled fixed-effects transmission model with current and lagged shocks. Neither system should be described as a deterministic forecast of sovereign debt.
Country model library
The 13-country published model library uses a common debt architecture while retaining country-specific samples, financing blocks, baselines and validation caveats. Production specifications identify the model form and governance treatment; they are not rankings of economic importance.
| Country | ISO | Specification | Estimation period | N | Production specification | Validation note |
|---|---|---|---|---|---|---|
| Egypt | EGY | Flexible + spread | 1999–2024 | 26 | Standard final | Output-robust; convergence and prior-sensitivity disclosures retained |
| Saudi Arabia | SAU | Peg + global financing | 1995–2024 | 30 | Standard final | MCMC-clean; wider-prior sensitivity governance note retained |
| Lebanon | LBN | Flexible + spread | 2001–2019 | 19 | Standard final | Pre-crisis estimation sample; post-2025 baseline is portal status-quo, not IMF forecast |
| United Arab Emirates | ARE | Peg + spread | 2007–2024 | 18 | Standard final | MCMC-clean final specification |
| Qatar | QAT | Peg + spread | 1995–2024 | 30 | Standard final | Output-robust; convergence disclosure retained |
| Bahrain | BHR | Peg + spread | 1995–2024 | 30 | Standard final | Output-robust; convergence disclosure retained |
| Kuwait | KWT | Reduced · peg · no spread | 1995–2024 | 30 | Reduced-data | Separately validated reduced-data specification with narrower climate channels |
| Oman | OMN | Peg + global financing | 1995–2024 | 30 | Standard final | MCMC-clean final specification |
| Jordan | JOR | Peg + spread | 1995–2024 | 30 | Standard final | Output-robust; convergence disclosure retained |
| Morocco | MAR | Flexible + spread | 1995–2024 | 30 | Standard final | Output robustness reviewed; convergence disclosure retained |
| Tunisia | TUN | Flexible + spread | 1995–2024 | 30 | Multimodal final | Two-mode posterior envelope retained; no arbitrary mode selection |
| Algeria | DZA | Flexible + spread | 1995–2024 | 30 | Standard final | Output and prior-sensitivity governance notes retained |
| Iraq | IRQ | Peg + spread | 2006–2024 | 19 | Standard final | Output-robust; convergence disclosure retained |
Kuwait is available through a reduced-data specification rather than the standard v14.2 Dynare model. It uses Kuwait-specific climate, output and fiscal persistence plus a transparent median calibration from the published Saudi Arabia and Oman peg/no-spread climate models; inflation and sovereign-spread climate channels remain disabled.
Climate intelligence
The climate measure combines heat stress, hydrological drought and drought persistence. Monthly observations are assessed against each country's own calendar-month distribution so unusual climate conditions retain a country-specific meaning, including in highly arid environments.
Temperature observations are converted into empirical distribution scores relative to the country and calendar month.
Precipitation, soil moisture and runoff information are combined into the hydrological stress component.
Dry-month frequency and consecutive dry spells capture persistence rather than only contemporaneous drought intensity.
Annual components are standardized over the estimation-era climate history and combined into the country climate-stress state. Historical reconstruction tests for the physical translator reproduce the annual climate index to numerical precision across the currently modelled countries.
Physical climate pathway translation
Physical mode accepts annual temperature anomalies and precipitation changes. The same empirical transformations used in the historical climate index translate those assumptions into incremental standardized climate stress.
Apply the anomaly to the calendar-month temperature climatology and translate it through the historical empirical distribution.
Apply the percentage change to monthly climatological precipitation and convert the resulting level into a country-specific wetness/drought score.
Map the physical precipitation disturbance through the historical soil-moisture and runoff relationships used in the climate index.
Reconstruct dry-month frequency and consecutive drought duration under the physical pathway.
Physical scenarios are evaluated against the country's historical climate support. Results can therefore be labelled within historical support, moderate extrapolation or high extrapolation rather than treating every physical assumption as equally well supported by the observed record.
Climate econometric system
The country climate models are linear systems around macroeconomic gaps and standardized external states. Terms enter according to the relevant exchange-rate and financing specification.
Within the country Bayesian climate model, climate, energy and global activity are represented as country-relevant external states.
Output responds to persistence, financing conditions, exchange-rate movements where relevant, climate stress, energy prices and world activity.
Inflation combines persistence, lagged activity and external cost pressures.
The fiscal block captures debt feedback, cyclical conditions and climate-related pressure.
Where the country specification contains a spread block, financing conditions respond to debt, fiscal performance and external pressures.
The estimated debt state is a structural consistency measure. The public portal separately reconstructs the level debt path around the institutional baseline.
x output gap · π inflation gap · r global financing benchmark · pb primary balance · s financing spread · d debt state · Δe exchange-rate change · c climate stress · q energy-price state · w world-growth state
The q and w states above are controls inside the country Bayesian climate models. The operational trade-to-debt system described below is estimated separately and does not overwrite or re-estimate the v14.2 country Dynare models.
Bayesian estimation
Country climate systems are estimated offline in Dynare using Bayesian methods. Preliminary country regressions inform signs, persistence, initialization and economically defensible parameter ranges. The climate-output coefficient χc, climate-fiscal coefficient ψc and, where applicable, climate-financing coefficient ηc are estimated jointly with domestic shock scales.
The final analytical freeze is complete. Original MCMC convergence diagnostics remain visible rather than being silently converted to passes. Production governance therefore separates chain diagnostics, policy-output robustness, local prior sensitivity and, for Tunisia, explicit multimodal posterior uncertainty. Kuwait is governed separately as a reduced-data specification.
Trade data and shock construction
The trade layer begins with a 13-country analytical panel. It combines the portal macro-fiscal dataset with internationally comparable trade indicators and the external-shock series already used by the project. Kuwait now enters the debt-scenario layer through its separately governed reduced-data specification.
Net barter terms of trade are transformed into annual log changes and standardized using each country's historical volatility.
World real-GDP growth deviations provide a common external demand shock across the Arab-country panel.
Oil and LNG price movements are oriented by the importer/exporter role so a positive shock is favourable and a negative shock is adverse for the country.
Export-demand stress is based on innovations in export-volume growth rather than an arbitrary percentage change in trade value.
Energy importers receive the opposite orientation from energy exporters. Qatar is treated as a gas exporter using a 35-percent oil and 65-percent LNG price mix before re-standardization. The construction is designed so the sign has a consistent economic interpretation across countries.
ACD now reports genuine UNCTAD–Eora GVC indicators as 2015–2018 structural averages for backward participation, forward participation, total GVC participation and GVC position. These measures are treated as slow-moving production-network characteristics rather than as 2024 observations. They are available for descriptive country and regional intelligence but are not used as scenario multipliers.
Trade × GVC interactions were evaluated in 03G and then subjected to the 03H production gate: explicit hydrocarbon- exporter shock heterogeneity, restricted wild-cluster bootstrap-t inference, leave-one-country-out sign stability and Benjamini–Hochberg multiple-testing adjustment. Eleven interactions entered the robustness stage and zero satisfied the full production-eligibility criteria. The validated 03C debt-scenario coefficients therefore remain unchanged.
Trade-to-macro transmission estimation
The operational trade layer is estimated using pooled country fixed effects with standard errors clustered by country. Each equation contains the contemporaneous shock and one annual lag. Country-specific and exporter/importer-group regressions are retained as diagnostics; they are not used as the operational coefficients in the scenario engine.
Operational channels pass a reliability gate based on the cumulative current-plus-lag effect, statistical support at the 10-percent level and the expected economic sign where the sign is theoretically unambiguous. Channels that fail the gate are exported with zero operational coefficients while their raw diagnostic estimates remain preserved.
| Shock | Response | Cumulative 1-SD effect | Operational status |
|---|---|---|---|
| Terms of trade | Real GDP growth | +0.695 pp | Operational |
| Terms of trade | Primary balance | +1.803 pp GDP | Operational |
| Terms of trade | Current account | +2.617 pp GDP | Operational |
| World growth | Real GDP growth | +1.894 pp | Operational |
| World growth | Primary balance | +1.890 pp GDP | Operational |
| World growth | Sovereign spread | −66.15 bp | Operational |
| World growth | Current account | +0.860 pp GDP | Operational |
| Favourable energy trade | Primary balance | +2.427 pp GDP | Operational |
| Favourable energy trade | Current account | +2.737 pp GDP | Operational |
| Export-demand innovation | Real GDP growth | +1.071 pp | Augmented |
| Export-demand innovation | Current account | +1.778 pp GDP | Augmented |
No direct inflation channel passes the current operational reliability gate. Energy-trade shocks therefore affect debt primarily through the primary balance and external-account response in the present specification, while world-demand shocks additionally transmit through growth and, where available, the sovereign spread.
Cross-country differences in trade scenario debt results should not be interpreted as 12 separately estimated structural trade elasticities. The operational transmission coefficients are pooled. Country differences arise from the economic-to-SD translation, country debt baselines, financing structure and debt dynamics.
Economic trade-shock translation
The user enters an economically interpretable assumption. A dedicated translation layer converts that assumption into the exact standardized shock units used by the trade-transmission estimates. Exposure variables are descriptive and are not used as arbitrary multipliers on pooled coefficients.
A user-specified percentage change is therefore scaled by the country's observed historical terms-of-trade volatility.
The same global-growth disturbance is expressed relative to the historical volatility of the common world-growth series.
Importer/exporter sign orientation and, where relevant, energy composition are applied before standardization.
This is an export-volume innovation, not a direct percentage change in export value.
A common economic shock does not necessarily imply the same standardized shock across countries. For example, the same energy-price movement maps differently when historical volatility and energy-trade role differ.
Validation, stability and evidence tiers
Validation is not treated as a single pass/fail statistic. Climate models are reviewed for dynamic stability, MCMC behaviour, posterior identification, parameter-bound sensitivity, persistence, climate reconstruction, debt-baseline reconciliation and economic interpretation. The trade layer is separately validated through data coverage checks, sign and significance gates, exact zero-shock reproduction and end-to-end economic-scenario replication.
Published systems passed the required structural model checks after documented country regularization. Posterior convergence is reported separately and is not relabelled as a pass where the original diagnostic thresholds were not met.
The physical translator reproduces the annual climate-stress index to numerical precision across the modelled countries.
The trade engine reproduces the stored baseline exactly when the standardized trade shock is set to zero.
Economic assumptions translated by the trade layer reproduce the same debt paths as the corresponding standardized 03C scenario.
Examples include Lebanon's pre-2019 structural climate sample, Bahrain's stability regularization, Tunisia's persistent debt dynamics, shorter climate-model samples for the UAE and Iraq, pooled trade-transmission coefficients and the absence of operational GVC amplification.
Scenario engines
The Scenario Lab exposes three user-facing analytical modes: standardized climate stress, physical climate pathways and trade or external-sector stress. Climate and trade scenarios are executed by separate engines but converge on a common debt accounting framework.
A standardized climate-state disturbance is propagated through the country Bayesian response system.
Physical assumptions are translated to the same standardized climate state before macro-fiscal propagation.
An economic trade assumption is translated to SD units, transmitted through the operational trade equations and then passed to the sovereign debt identity.
Climate shock propagation
Trade shock propagation
Standardized trade stress uses positive values for favourable or expansionary shocks and negative values for adverse shocks. The economic translator applies this sign convention before the debt engine is called.
Public-debt reconstruction
The portal distinguishes model-state responses from the projected level debt path. Both climate and trade scenarios ultimately operate around the institutional baseline stored in the country bundle.
For climate scenarios, Cₜ can represent an explicit direct climate fiscal-cost assumption when the user activates that feature. The trade engine does not add a direct trade fiscal cost; its effects enter through estimated macro-fiscal and financing responses. Current-account responses are reported as an external-sector diagnostic and are not mechanically inserted into the debt identity.
Output, inflation, fiscal and financing responses are applied recursively to the institutional debt baseline.
Trade-induced growth, primary-balance and applicable spread responses are applied to the same country debt accounting structure.
Climate v14.2 treats x as an output-gap level, so the annual real-GDP-growth response used in the climate debt identity is approximated by Δxₜ = xₜ − xₜ₋₁. The trade system instead estimates the real-GDP-growth response directly.
Baselines, source treatment and data lineage
Country scenarios are anchored to the baseline stored in each portal model bundle. Where a complete institutional projection is available, the debt identity is reconciled through an annual stock-flow adjustment so the model reproduces the published baseline before stress is introduced.
Physical and standardized climate scenarios preserve the country model version and baseline metadata.
Trade scenario outputs preserve the economic assumption, standardized shock, operational channels and baseline source.
AI analytical layer
The AI assistant is designed as an orchestration and explanation layer over the analytical platform. It does not replace the climate or trade engines and should not generate sovereign-risk numbers from language-model memory.
Extract country, shock family, economic magnitude, duration, start year and requested output from natural language.
Invoke climate, trade, comparison, data-catalogue and reporting endpoints rather than inventing results.
Translate engine output into decision-ready language while retaining assumptions, caveats and source information.
The assistant may formulate and explain a scenario, but every quantitative result presented as a portal estimate must come from an analytical endpoint or stored validated output.
Model governance and versioning
Every published country is associated with a model tier, publication status, model version, data vintage, baseline vintage and explicit caveats in the country registry. Climate and trade engines are versioned separately so additions to the trade layer do not silently alter validated climate-model bundles.
- Climate engine. Code 02 v14.2 remains the current country climate-debt scenario engine for the published final bundles.
- Trade data. 03A builds the Arab trade-debt analytical panel without changing the climate models.
- Trade transmission. 03B v2 estimates the pooled operational current-and-lag shock responses and exports only channels that pass the reliability gate.
- Trade translation. 03D converts economic assumptions into standardized shock units without arbitrary exposure multipliers.
- Trade debt engine. 03C v1 applies the operational responses to the standard v14.2 baselines; the Kuwait-compatible v1.1 availability layer uses the same equations and coefficients once the validated reduced-data Kuwait bundle exists. 03E provides an end-to-end economic scenario runner.
- Kuwait. Published as a reduced-data specification. Climate-output and climate-primary-balance loadings are regional calibrations checked against Kuwait country diagnostics; the model must not be presented as a standard country-specific Dynare estimate.
Interpretation, scope and disclaimer
Portal outputs are conditional scenario estimates. Climate results describe model-implied macro-fiscal responses to selected climate stress. Trade results are semi-structural stress-test estimates based on pooled operational transmission coefficients and country-specific debt baselines and dynamics. Neither should be presented as an official forecast.
- Conditional interpretation. Results depend on the selected baseline, shock magnitude, start year, duration, persistence and the relevant model configuration.
- Climate interpretation. Temperature and precipitation enter the macro-fiscal system through the composite climate-stress state rather than as independent structural coefficients.
- Trade interpretation. The current operational trade coefficients are pooled across the Arab-country panel; cross-country rankings are scenario sensitivities rather than structural rankings of exposure.
- Current-account channel. Trade-induced current account responses are disclosed but are not mechanically added to the public-debt identity.
- Financing spread. A spread response is only applied where the country debt bundle contains the corresponding spread block.
- GVC. Structural GVC intelligence is operational as a descriptive layer using genuine UNCTAD–Eora 2015–2018 averages. No synthetic or post-2018 extrapolated GVC series are used. GVC amplification is inactive because zero interaction candidates passed the full 03H small-cluster production gate.
- Uncertainty reporting. Trade scenarios expose conditional parameter-uncertainty envelopes from the validated bootstrap layer; these are not forecast probabilities. Climate scenarios emphasize central paths and deterministic sensitivity, while original posterior convergence and multimodality disclosures remain visible rather than being converted into probability claims.