## Executive Summary

## Source details

**Canonical URL:** [Executive Summary](https://www.imf.org/-/media/files/publications/tar/2025/english/tarea2025012-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/tar/2025/english/tarea2025012-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/tar/2025/english/tarea2025012-print-pdf.pdf.json)

---

### Background
- In September 2020, the Governor of the Central Bank of Bosnia and Herzegovina (CBBH), Mr. Senad Softić, requested technical assistance from the Institute for Capacity Development (ICD) to build a Financial Programming and Policies (FPP) framework.
- Bosnia and Herzegovina (BiH) institutional structure: two largely autonomous entities (Federation of Bosnia and Herzegovina (FBiH) and Republic of Srpska (RS)) and the Brčko District; FBiH includes 10 cantons with significant autonomy.
- Since 1997 there is a currency board with a fixed exchange rate with the Euro.
- December 2022: European Council granted European Union candidacy status to BiH.
- March 2019: one-week customized FPP course for BiH delivered; attendance: 35 participants (10 from the Central Bank of Bosnia and Herzegovina, 10 from the Ministry of Finance and Treasury of BiH, 5 from the Department of Economic Planning of BiH, and 12 from ministries of finance of the Federation of Bosnia and Herzegovina and Republika Srpska (6 participants each)).
- CBBH requested a country-specific, in-house model to address institutional fragmentation and improve forecasting; IMF European Department supported an ICD TA project with CBBH as main recipient.
- Project timeline: Phase I across four TA missions over April 2021 - April 2022 (first three missions remote in 2021; last one in-person in April 2022).
- For 2024, GDP per capita is estimated at "$8,320 (and $20,730 in purchasing power parity terms)".

### Outputs — The Quarterly Macroeconomic Forecasting Framework (QMFF)
- QMFF characteristics:
  - Gap-based modeling: behavioral equations quantified as deviations of real variables from long-run trends.
  - Semi-structural simultaneous system of calibrated equations covering all key FPP sectors and accounting restrictions within and across sectors.
  - Flexible to adjust for different countries and exchange rate regimes.
  - Expectations: baseline QMFF uses adaptive expectations; can be adjusted to forward-looking, model-consistent expectations.
  - Implementation environment: Matlab with automated export to FPP-style tables in Excel.
  - Automation: automatic reports on diagnostics (e.g., impulse-response functions) and forecast output; automatized database system for data transformations.
- Methodology highlights:
  - Combines gap-based modeling advantages (business cycle analysis, tractability, fast automated reports) with FPP granularity and full accounting consistency.
  - Calibration and empirical approaches used to bring model to BiH data.
- Deliverables from Phase I:
  - QMFF prototype, data processing, and technical/reporting infrastructure.
  - Knowledge transfer activities and capacity building with CBBH staff.

### Participation, Capacity Building, and Use
- CBBH Office of the Chief Economist (OCE) team actively participated in all development stages and built understanding of framework transmission channels.
- OCE team requires additional time and sustained practice to fully exploit QMFF potential.
- Recommendations to facilitate full utilization:
  - Formalize an internal forecast calendar to facilitate forecasting rounds.
  - Give OCE team members regular tasks and opportunities to work with the model to gain full ownership and operate it efficiently.
  - Integrate QMFF results into policymaking discussions with senior CBBH management and communicate results to the public on a regular basis.
  - Sustain capacity building to strengthen analytical skills and institutional use.

### Achievements and Assessment
- TA project success: met key objective of developing a medium-term macroeconomic framework to strengthen macroeconomic forecasting and analysis at the CBBH.
- QMFF serves as a pilot combining gap-based modeling and FPP.
- Authorities and IMF country team highly satisfied with contributions to CBBH forecasting and analytical capabilities.
- Tool equips authorities with an instrument for policy analysis, deepens Article IV mission discussions, and facilitates information sharing.

### Key Implementation and Operational Notes
- Operational environment: Matlab with automated export to Excel.
- Adaptive-expectations specification developed for CBBH.
- Framework adjustable across exchange rate regimes and country specifics.
- Continued regular use, internal forecasting routines, and integration into decision making necessary for full benefits.

### Project phases and key results
- Phase I (virtual missions and one in-person mission):
  - Compilation of database for FPP framework.
  - Initial code infrastructure for data processing and reporting using a Matlab-based toolbox.
  - Prototype model covering real, external, fiscal, and monetary sectors.
  - Calibration of behavioral relationships in the real economy.
  - Initial infrastructure for baseline projections and scenario analysis.
  - Transfer of QMFF results to Excel and support for CBBH hands-on Matlab experience.
- Phase II (four in-person missions between November 2022 and December 2023):
  - Introduced and linked trends in relative prices and quantities; adjusted calibration of steady-state values.
  - Introduced cross-restrictions within and across sectors for full accounting consistency.
  - Extended fiscal block (state level) by linking discretionary fiscal policy and GDP.
  - Reviewed and finetuned model calibration and dynamic properties.
  - Streamlined and finalized model and reporting infrastructure.
  - Populated ICD’s Debt Dynamics Tool (DDT) with QMFF projections for state-level public debt tracking.
  - Continued knowledge transfer via shadow forecast rounds and worked towards institutionalization.

### Key features of the QMFF
- Modeling approach:
  - Gaps measured as percentage deviations from trends using the Hodrick-Prescott filter.
  - Semi-structural simultaneous system with calibrated behavioral equations.
  - Baseline adaptive expectations; forward-looking option available.
  - Matlab-based implementation with Excel export for FPP presentation.
- Automation and outputs:
  - Automatic diagnostics (impulse-response functions), baseline and scenario reporting, and standard FPP Excel export.
  - Excel template contains 12 quarters of historic data and 12 quarters of forecasts.
- Off-model information:
  - Incorporated via transparent residual adjustments in respective equations.
- Scenario example:
  - "Higher Spending on Goods and Services for 1 p.p. of GDP between 2023Q3 and 2024Q4" included as an illustrative scenario.

### Calibration, model properties, and empirical methods
- Calibration objectives:
  - Iterative parameterization emphasizing plausible dynamic properties and satisfactory in-sample forecasting performance.
  - Calibrated to replicate growth rates of nominal and real variables, ensure nominal expenditures relative to nominal GDP consistency, and explain key cyclical movements.
  - Steady-state shares: initial 2017-19 averages recalibrated to 2021-2022 averages. "All steady-state shares were eventually calibrated to 2021-22 averages."
- Equation treatment:
  - Most equations log-linear; level model variables (except interest rates) expressed in natural logarithms and multiplied by 100.
  - Nominal variables from fiscal, external, and monetary sectors specified as ratios to nominal GDP and forecast via autoregressive convergence toward HP-filtered trend ratios.
  - Includes trends in relative prices and real quantities so projections align with balanced growth path.
- Model diagnostics:
  - Impulse response functions verify dynamic plausibility and country-specific transmission (e.g., dominance of external demand, strong response of private transfers and FDI, minor role of domestic real interest rates under currency board).
  - In-sample simulations: iterative automated "forecasts" starting in Q1 2010 and continued successively eight quarters ahead, assuming known exogenous variables and future trends.

### Forecasting, reporting, and scenario analysis
- Forecasting steps:
  - Update historical database and exogenous variables, solve model, discuss initial conditions, incorporate nowcasts and judgment, produce baseline medium-term forecast.
- Reports:
  - Automated reports cover historic data, initial conditions, baseline forecast, alternative scenarios, and scenario comparisons.
  - Decomposition graphs show projected gaps and driving factors; tables present key variables in annual or quarterly frequency in standard FPP format.
- Excel use:
  - Excel used only for presentation; Matlab performs modeling and report generation.
  - Built-in Excel macro copies Matlab-generated data into a single data sheet to populate templates.

### Linkages to debt analysis and documentation
- DDT linkage:
  - QMFF projections exported to ICD’s DDT to produce public debt projections as a share of GDP and identify debt-stabilizing primary balances and fiscal adjustment paths under alternative consolidation strategies (constant primary balance; constant adjustment in primary balance; constant debt changes; customized combinations).
  - Annual data and forecasts from QMFF exported as inputs into the DDT Excel file.
- Documentation:
  - Core CBBH team produced an internal user manual with step-by-step guidance on model operation and code structure.
  - User manual presented to ICD team; mission team recommended further improvement of documentation of model structure and properties.

### Knowledge transfer, training, and technology use
- Knowledge transfer modalities:
  - Mix of virtual and in-person sessions emphasizing theoretical underpinnings, hands-on analytical work, and shadow forecasting rounds.
  - Virtual missions recorded in 2022 to facilitate transfer of technical knowledge to the OCE.
- Capacity building:
  - CBBH staff received hands-on Matlab coding training and participated in specification, calibration, and shadow forecast rounds.
  - Mission team encouraged OCE staff to attend JVI courses: FPP, Macroeconomic Diagnostics, Macroeconometric Forecasting and Analysis, Monetary Policy Analysis and Forecasting, and Vulnerability Diagnostics.
- Statistical integrity:
  - Automatic checks embedded in data processing codes to compare database vintages, flag renaming, missing variables, and large data revisions.

### Challenges, data issues, and risks
- Structural/methodological:
  - Integrating gap-based methodology with classic FPP levels required consistency of trends in relative prices and quantities and cross-sector restrictions.
  - Retrospective: developing a smaller functional model first might have been more efficient.
- Resource and staffing constraints:
  - OCE core team of seven staff; constrained resources and high workload slowed absorption and proficiency.
  - Staff turnover occurred, including a core member who left in final stages.
- Data quality and accounting inconsistencies:
  - Discrepancies between CPI inflation and private consumption deflator and sizable revisions in GDP expenditure item series.
  - Fiscal data limitations: CBBH GFS provides total taxes only; indirect taxes from Indirect Taxation Authority used to infer direct taxes by residual, occasionally producing counter-intuitive values (e.g., negative direct taxes).
  - Reconciling fiscal statistics with government consumption from national accounts: example discrepancy around 5% of GDP between sum of compensation of employees and spending on goods and services versus national accounts government consumption.
  - Net acquisition of nonfinancial assets appears underestimated (around 1% of GDP); investment not split between public and private in national accounts.
  - Many data series published with delays and without a fixed schedule; in mid-December 2023 last available BoP and fiscal data were for Q2 2023, producing a two-quarter lag in forecasts.
- Suggested approach to discrepancies:
  - Model discrepancies separately as a mean reverting process and adjust assumed steady state if discrepancies shrink over time.
- Main risks:
  - Availability of adequate resources and staff time given core team size.
  - Staff retention and turnover risk, with specialized skills costly and time-consuming to replace.

### Expected benefits and operational efficiencies
- Automation gains:
  - Highly automatized data-processing and sectoral report generation including frequency transformation, X13-ARIMA seasonal adjustment, HP filtering, growth rates and ratios.
  - Efficient quarterly projections solved in Matlab with automatic reports (impulse responses, decompositions, in‑sample simulations) and automatic export to Excel.
- Analytical gains:
  - Transparent incorporation of expert judgment and off-model information via residual adjustments.
  - Seamless generation of alternative and stress-test scenarios and easy comparison of forecast vintages.

### Recommendations for the OCE / CBBH
- Adopt the QMFF as core forecasting and policy analysis tool; two OCE members should operate the tool regularly with sectoral analysts providing judgments, and other OCE members should become familiar with QMFF basics to enable rotation.
- Integrate forecasting and policy analysis results into policymaking discussions and public communication:
  - Share the TA report and finalized tool with CBBH senior management.
  - After 2-3 shadow forecast rounds, discuss baseline and alternative scenarios with policymakers to foster buy-in and institutionalization.
  - Evaluate starting publication of medium-term macroeconomic forecasts produced with the QMFF in line with CBBH mandate.
- Formalize an internal forecast calendar with timelines, roles, and deadlines while embedding flexibility to accommodate data publication delays.
- Finalize and extend the QMFF user manual and technical documentation to strengthen business continuity and knowledge sharing.
- Address staff retention and capacity:
  - Consider differentiated compensation and recognition for staff performing complex analytical tasks.
  - Increase OCE team capacity in terms of number of staff with relevant educational background and experience.

### Appendix B — Main economic relationships underlying the QMFF (model specification highlights)
- Modeling in gaps: percentage deviations of real variables from trends; gaps denoted X̃_t.
- Baseline assumptions:
  - Adaptive expectations (adjustable to forward-looking).
  - Well-defined balanced growth path and stable long-run nominal expenditure shares.
- Monetary regime and exchange rate:
  - Model consistent with BiH currency board and de jure open capital account: no independent monetary policy.
  - Domestic policy rates i_t equal eurozone policy rates i*_t.
  - Nominal exchange rate s_t fixed at 1.95 BAM per euro.
- Price block:
  - Phillips curve: π_t depends on lagged and expected inflation (expected approximated by foreign effective inflation π*_t), output gap ỹ_t, real exchange rate q_t, and exogenous relative prices rpx_t^.
  - Expenditure deflators p_t^N are weighted averages of GDP deflator p_t^D and import deflator p_t^M using domestic value-added share β^N.

### Behavioral equations (selected forms and calibrated coefficients for BiH)
- Private consumption:
  - ĉ_t = β1^c ĉ_{t−1} + β2^c ỹ_t + β3^c p̃t^tr + β4^c r̂^h_t + β5^c capb̃_t + ε_t^c
  - Consumption calibration: β1^c = 0.35; β2^c = 0.10; β3^c = 0.10; β4^c = 0.05; β5^c = 0.25.
- Government consumption:
  - ĝ_t = β1^g ĝ_{t−1} + β2^g pex̃_t + ε_t^g
  - Calibration: β1^g = 0.60; β2^g = 0.17.
- Investment:
  - î_t = β1^i î_{t−1} + β2^i ỹ_t + β3^i ỹ*_t + β4^i fdĩ_t + β5^i r̂^c_t + β6^i cẽ_t + ε_t^i
  - Calibration: β1^i = 0.45; β2^i = 0.40; β3^i = 0.90; β4^i = 0.02; β5^i = 0.30; β6^i = 0.10.
- Exports:
  - x̂_t = β1^x x̂_{t−1} + β2^x q̂_t + β3^x ỹ*_t + β4^x fdĩ_t + ε_t^x
  - Calibration: β1^x = 0.20; β2^x = 0.30; β3^x = 3.00; β4^x = 0.02.
- Imports:
  - m̂_t = β1^m ĉ_t + β2^m ĝ_t + β3^m î_t + β4^m x̂_t + β5^m q̂_t + ε_t^m
  - Calibration: β1^m = 0.23; β2^m = 0.02; β3^m = 0.12; β4^m = 0.63; β5^m = 0.30.
- Domestic demand and national accounts identity:
  - d̂_t = β1^d ĉ_t + β2^d ĝ_t + β3^d î_t + ε_t^d
  - ỹ_t = β1^y ĉ_t + β2^y ĝ_t + β3^y î_t + β4^y x̂_t + β5^y m̂_t + ε_t^y
  - Domestic demand shares: β1^d = 0.60; β2^d = 0.17; β3^d = 0.23.
  - National accounts identity shares: β1^y = 0.67; β2^y = 0.19; β3^y = 0.26; β4^y = 0.45; β5^y = 0.57.
- Private transfers:
  - pt̃_t = β1^{pt} pt̃_{t−1} + β2^{pt} ỹ*_t + β3^{pt} d̂_t + ε_t^{pt}
  - Calibration: β1^{pt} = 0.30; β2^{pt} = 2.20; β3^{pt} = 0.30.
- FDI:
  - fdĩ_t = β1^{fdi} fdĩ_{t−1} + β2^{fdi} ỹ*_t + ε_t^{fdi}
  - Calibration: β1^{fdi} = 0.20; β2^{fdi} = 30.0.
- Cyclically adjusted fiscal balance:
  - capb̃_t = β1^{rev} car̃_t − β2^{exp} capẽ_t
  - Calibration: β1^{rev} = 0.44; β2^{exp} = 0.34.
- Inflation (Phillips curve):
  - π_t = β1^π π_{t−1} + (1−β1^π) π*_t + β2^π ỹ_t + β3^π q̂_t + β4^π rpx_t^. + ε_t^π
  - Calibration: β1^π = 0.40; β2^π = 0.25; β3^π = 0.25; β4^π = 0.25.
- Deflator and import-price decompositions:
  - Consumption deflator shares and import deflator shares: β1^{C,dom} = 0.20; β1^{C,other} = 0.05; β1^{C,food} = 0.27; β1^{C,oil} = 0.8.
  - Import deflator shares: β1^M = 0.55; β2^M = 0.35; β3^M = 0.05.
- Nominal exchange rate calibration:
  - β1^s = 67.3 (Model variables are log transformed and multiplied by 100; 67.3 = 100*log(1.95).)

### Trends, steady states, and import-deflator calibration
- Trends of real variables are functions of potential GDP and real exchange rate trend so nominal expenditure ratios to nominal potential GDP are constant in the long run.
- National accounts deflators: P_C = (P_D)^{α_C}*(P_M)^{1−α_C} * exp(ε_C); relative-price trend definitions and log-linear trend equations implemented as in Appendix C.
- Import deflator modeled as function of foreign CPI (effective), world food prices, world metal prices, and world oil prices; calibration via OLS and constrained optimization with theoretical homogeneity and sign restrictions.
- Condition for total import share:
  - sℎ_M = sℎ_C*(1−α_C) + sℎ_G*(1−α_G) + sℎ_J*(1−α_J) + sℎ_X*(1−α_X).
- Practical calibration note: smoothed data (HP-filtered series with low lambda = 1) often used to avoid excessive volatility.

_Italic: IMF Technical Assistance Report – Bosnia and Herzegovina — Executive Summary (content unit provided)._

### Executive Summary ......................................................................................................

### Executive Summary

### Background
- In September 2020, the Governor of the Central Bank of Bosnia and Herzegovina (CBBH), Mr. Senad Softić, requested technical assistance from the Institute for Capacity Development (ICD) to build a Financial Programming and Policies (FPP) framework.
- Bosnia and Herzegovina (BiH) has a complex, decentralized institutional structure with two largely autonomous entities (Federation of Bosnia and Herzegovina (FBiH) and Republic of Srpska (RS)) and the Brčko District; FBiH includes 10 cantons with significant autonomy.
- Since 1997, there is a currency board which sets a fixed exchange rate with the Euro.
- In December 2022, the European Council granted European Union candidacy status to BiH.
- In March 2019, a one-week customized FPP course for BiH was delivered; attendance: 35 participants (10 from the Central Bank of Bosnia and Herzegovina, 10 from the Ministry of Finance and Treasury of BiH, 5 from the Department of Economic Planning of BiH, and 12 from ministries of finance of the Federation of Bosnia and Herzegovina and Republika Srpska (6 participants each)).
- The CBBH requested a country-specific, in-house model to address institutional fragmentation and improve forecasting; the IMF’s European Department supported an ICD TA project with the CBBH as the main recipient.
- Project timeline: Phase I organized across four TA missions over the period of April 2021 - April 2022 (first three missions remote in 2021; last one in-person in April 2022).
- For 2024, GDP per capita is estimated at "$8,320 (and $20,730 in purchasing power parity terms)" (source noted in the original content).

### Outputs of the Project — The Quarterly Macro Forecasting Framework (QMFF)
- The QMFF is a novel quarterly framework that quantifies its main behavioral equations in terms of deviations of real variables from their long-run trends (gap-based modeling).
- Model structure and features:
  - Simultaneous system of calibrated equations covering all key FPP sectors and key accounting restrictions within and across sectors.
  - Flexibility to be adjusted for different countries and exchange rate regimes.
  - Ability to incorporate backward- or forward-looking expectation formation; this project developed a QMFF with adaptive expectations.
  - Implemented and operated in Matlab, with capacity to automatically export output to FPP-style tables in Excel format.
- Methodology highlights:
  - Combines gap-based modeling advantages (business cycle analysis, tractability, fast automated reports) with FPP granularity and full accounting consistency.
  - Custom-built and gradually brought to BiH data via a combination of empirical approaches.
- Documentation and delivery:
  - Phase I produced a QMFF prototype, data processing, and technical and reporting infrastructure.
  - The project included knowledge transfer activities and capacity building with CBBH staff.

### Participation, Capacity Building, and Use
- CBBH counterparts (Office of the Chief Economist (OCE) team) actively participated in all development stages and gradually built understanding of the framework’s transmission channels.
- Despite considerable progress, the OCE team will need more time to fully exploit the QMFF’s potential.
- Recommendations to facilitate full utilization:
  - Formalize an internal forecast calendar to facilitate forecasting rounds.
  - Give OCE team members regular tasks and opportunities to work with the model to gain full ownership and operate it efficiently.
  - Integrate QMFF results into policymaking discussions with senior CBBH management and communicate results to the public on a regular basis.
  - Capacity building should be sustained to strengthen analytical skills and institutional use.

### Achievements and Assessment
- The TA project was judged successful in meeting its key objective: development of a medium-term macroeconomic framework to strengthen macroeconomic forecasting and analysis at the CBBH.
- The QMFF serves as a pilot version of a new approach to financial programming combining gap-based modeling and FPP.
- Both the authorities and the IMF country team were highly satisfied with the project’s contribution to CBBH forecasting and analytical capabilities.
- The tool equips authorities with a vital instrument for policy analysis and can deepen discussions with the Fund in Article IV missions and facilitate information sharing.

### Key Implementation and Operational Notes
- QMFF operational environment: Matlab with automated export to Excel.
- The model’s adaptive-expectations specification was developed for the CBBH.
- The framework is designed to be adjustable across exchange rate regimes and country specifics.
- Continued regular use, internal forecasting routines, and integration into decision making are necessary for full benefits.

_Italic: IMF Technical Assistance Report – Bosnia and Herzegovina — Executive Summary (content unit provided)._

### 2022. The virtual missions were recorded, thus facilitating the transfer of technical knowledge to the OCE

### 2022. The virtual missions were recorded, thus facilitating the transfer of technical knowledge to the OCE

### Project phases and key results
- Phase I (virtual missions and one in-person mission):
  - Compilation of a database required for setting up the FPP framework.
  - Development of an initial code infrastructure for data processing and reporting using a Matlab-based toolbox.
  - Development of a prototype model covering key economic sectors: real, external, fiscal, and monetary.
  - Calibration of behavioral relationships in the real economy.
  - Building the initial infrastructure for baseline projections and scenario analysis.
  - Transfer of the quarterly macroeconomic framework in Excel using the data and forecasts generated in Matlab.
  - Supporting the CBBH team in acquiring hands-on experience with using Matlab and operating the framework.
- Phase II (four in-person missions between November 2022 and December 2023):
  - Ensuring internal consistency of the QMFF by explicit introduction and linking of trends in relative prices and quantities and adjusting the calibration of steady-state values.
  - Ensuring full accounting consistency of the framework by introducing cross-restrictions within and across sectors.
  - Review and extension of the fiscal block (at the state level) by introducing a link between discretionary fiscal policy and GDP.
  - Review and finetuning of model calibration and verification of the dynamic model properties.
  - Streamlining and finalizing model and reporting infrastructure.
  - Populating the IMF’s Debt Dynamics Tool (DDT) with QMFF projections to allow authorities to track public debt dynamics at the state level within the CBBH mandate.
  - Continuing transfer of knowledge to the OCE team via shadow forecast rounds.
  - Working towards institutionalization of the framework, ensuring its integration and sustainable use within the organization's processes and systems.

### Key features of the Quarterly Macroeconomic Forecasting Framework (QMFF)
- Modeling approach and implementation:
  - The QMFF quantifies behavioral equations in terms of deviations from their medium-term trends ("gap model"); gaps are measured as percentage deviations of levels of macroeconomic variables from their trends.
  - The QMFF uses the Hodrick-Prescott filter to calculate gaps.
  - The backbone is a semi-structural simultaneous system of calibrated behavioral equations covering all key FPP sectors and key accounting restrictions within and across sectors.
  - The baseline QMFF assumes adaptive expectations but can be adjusted to forward-looking, model-consistent expectations.
  - Implementation environment: Matlab (model and automation) with export to Excel for FPP-style presentation.
- Automation and outputs:
  - Automatic reports on model diagnostics (e.g., impulse-response functions) and forecast output across all FPP sectors.
  - Results exported to Excel in a standard FPP format.
  - An automatized database system executes data transformations for real, monetary, external, and fiscal sectors.
  - The QMFF infrastructure includes a program that populates an Excel-based FPP macroframework and updates ICD’s Debt Dynamics Tool (DDT) template.

### Calibration, model properties, and empirical methods
- Calibration objectives and strategy:
  - Iterative parameterization to produce plausible dynamic properties and satisfactory in-sample forecasting performance, emphasizing overall model behavior over individual equation econometric precision.
  - Calibrated to replicate growth rates of nominal and real variables, ensure consistency with observed nominal expenditures relative to nominal GDP, and explain key cyclical movements.
  - Steady-state and shares: initial use of 2017-19 average shares, recalibrated to 2021-2022 averages; "All steady-state shares were eventually calibrated to 2021-22 averages."
- Equation types and treatment of variables:
  - Most equations are log-linear; level model variables (except interest rates) are expressed in natural logarithms and multiplied by 100.
  - Nominal variables from fiscal, external, and monetary sectors are specified as ratios to nominal GDP and forecast via autoregressive convergence toward HP-filtered trend ratios.
  - QMFF includes trends in relative prices and real quantities so projections are consistent with a balanced growth path (nominal expenditure shares converge to stable shares; real expenditure gaps close; real and nominal growth rates are constant in steady state).
- Model diagnostics and empirical verification:
  - Impulse response functions used to verify dynamic plausibility and country-specific transmission (e.g., dominance of external demand, strong response of private transfers and FDI, minor role of domestic real interest rates under a currency board regime).
  - Example shock: a 1 p.p. shock to foreign demand (Figure caption: "1 %. Foreign Demand Gap Shock") shows growth in real GDP, inflationary pressures, initial negative real interest rates, and temporary external position improvements.
  - Decomposition graphs used to verify behavioral equations; example: foreign demand and FDI dominate the exports gap equation.
  - In-sample simulations: iterative automated "forecasts" starting in Q1 2010 and continued successively in each period eight quarters ahead; noted as mechanical simulations assuming known exogenous variables and future trends.

### Forecasting, reporting, and scenario analysis
- Forecasting process:
  - Steps: update historical database and exogenous variables, solve model, discuss initial conditions, incorporate nowcasts and judgment, produce baseline medium-term forecast.
  - Detailed automated reports cover historic data, initial conditions, baseline forecast, alternative scenarios, and scenario comparisons.
  - Decomposition graphs show projected gaps and key driving factors; tables present key variables in annual or quarterly frequency in a standard FPP representation.
- Excel presentation and templates:
  - Excel used solely for presentation of results; Matlab performs modeling, linking, and report generation.
  - Built-in Excel macro copies Matlab-generated data into a single data sheet to populate templates.
  - Excel template currently set up to contain 12 quarters of historic data and 12 quarters of forecasts.
- Off-model information and scenarios:
  - Off-model information incorporated transparently via residual adjustments in respective equations.
  - Scenario reports provide transmission analysis and forecast vintage comparisons.
  - Example scenario: "Higher Spending on Goods and Services for 1 p.p. of GDP between 2023Q3 and 2024Q4" with comparative panels and tables.

### Linkages to debt analysis and documentation
- Debt Dynamics Tool (DDT) linkage:
  - QMFF projections exported to ICD’s DDT to produce public debt projections as a share of GDP and identify debt-stabilizing primary balances and fiscal adjustment paths under alternative consolidation strategies (constant primary balance; constant adjustment in primary balance; constant debt changes; customized combinations).
  - Annual data and forecasts from QMFF are exported as inputs into the DDT Excel file.
- Documentation:
  - The core CBBH team produced an internal user manual providing step-by-step guidance on model operation and code structure.
  - The user manual was presented to the ICD team during the final mission; mission team recommended further improvement of documentation of model structure and properties.

### Knowledge transfer, training, and use of technology
- Knowledge transfer modalities:
  - Mix of virtual and in-person sessions emphasizing theoretical underpinnings, hands-on analytical work, and shadow forecasting rounds.
  - Recordings of virtual missions and most face-to-face engagements were shared with the CBBH core team to facilitate drafting the user manual and ongoing learning.
- Capacity building:
  - CBBH staff received hands-on Matlab coding training and participated in specification, calibration, and shadow forecast rounds.
  - Mission team encouraged OCE staff to attend JVI courses: FPP, Macroeconomic Diagnostics, Macroeconometric Forecasting and Analysis, Monetary Policy Analysis and Forecasting, and Vulnerability Diagnostics.
- Statistical integrity:
  - Automatic checks embedded within data processing codes; example script compares updated database with previous vintage, flags variable renaming, missing variables, and data revisions exceeding thresholds.

### Challenges and data issues
- Structural and methodological challenges:
  - Integrating gap-based methodology with classic FPP levels required ensuring consistency of trends in relative prices and quantities and imposing cross-sector restrictions.
  - Retrospective observation: developing a smaller functional model first might have been more efficient before introducing full granularity.
- Resource and staffing constraints:
  - The OCE core team consists of seven staff; constrained resources and high workload slowed absorption and proficiency.
  - Staff turnover occurred, including an additional core member who left in final stages.
- Data quality and accounting inconsistencies:
  - Discrepancies between CPI inflation and private consumption deflator and sizable revisions in GDP expenditure item series.
  - Fiscal data limitations: CBBH GFS provides total taxes only; indirect taxes from Indirect Taxation Authority used to infer direct taxes by residual, occasionally producing counter-intuitive values (e.g., negative direct taxes).
  - Reconciling fiscal statistics with government consumption from national accounts: example discrepancy around 5% of GDP between sum of compensation of employees and spending on goods and services versus national accounts government consumption.
  - Net acquisition of nonfinancial assets appears underestimated (around 1% of GDP); investment not split between public and private in national accounts.
  - Many data series published with delays and without a fixed schedule; in mid-December 2023 last available BoP and fiscal data were for Q2 2023, producing a two-quarter lag in forecasts.
- Suggested approach for data discrepancies:
  - Model discrepancies separately as a mean reverting process and adjust assumed steady state if discrepancies shrink over time.

### Achievements, expected benefits, and risks
- Achievement assessment:
  - The TA project met its key objective: development of a medium-term macroeconomic framework to strengthen forecasting and analysis at the CBBH.
  - CBBH counterparts actively participated in data population, calibration, cross-sector linkages, and testing dynamic and in-sample properties; further time needed to realize full potential.
- Expected medium-term benefits (operational efficiencies and analytical gains):
  - Highly automatized data-processing and generation of sectoral reports including frequency transformation, X13-ARIMA based seasonal adjustment, HP filtering, growth rates and ratios.
  - Efficient generation of quarterly projections solved in Matlab with automatic reports (impulse responses, decompositions, in-sample simulations) and automatic export to Excel.
  - Transparent incorporation of expert judgment and off-model information via residual adjustments.
  - Seamless generation of alternative and stress-test scenarios and easy comparison of forecast vintages.
- Main risks:
  - Availability of adequate resources and staff time given the core team of seven people.
  - Staff retention and turnover risk, with specialized skills costly and time-consuming to replace.

### Recommendations for the OCE / CBBH
- Adopt the QMFF as the core forecasting and policy analysis tool; two OCE members should operate the tool regularly with sectoral analysts providing judgments, and other OCE members should become familiar with QMFF basics to enable rotation.
- Integrate forecasting and policy analysis results into policymaking discussions and public communication:
  - Share the TA report and finalized tool with CBBH senior management.
  - After 2-3 shadow forecast rounds, discuss baseline and alternative scenarios with policymakers to foster buy-in and institutionalization.
  - Evaluate starting publication of medium-term macroeconomic forecasts produced with the QMFF in line with CBBH mandate.
- Formalize an internal forecast calendar with timelines, roles, and deadlines while embedding flexibility to accommodate data publication delays.
- Finalize and extend the QMFF user manual and technical documentation to strengthen business continuity and knowledge sharing.
- Address staff retention and capacity:
  - Consider differentiated compensation and recognition for staff performing complex analytical tasks.
  - Increase OCE team capacity in terms of number of staff with relevant educational background and experience.

*IMF Technical Assistance Report – Bosnia and Herzegovina.*

### APPENDIX B. THE MAIN ECONOMIC RELATIONSHIPS UNDERLYING THE QMFF FOR THE CBBH

### APPENDIX B. THE MAIN ECONOMIC RELATIONSHIPS UNDERLYING THE QMFF FOR THE CBBH

### Model structure and main behavioral relationships
- The QMFF is specified in terms of gaps: percentage deviations of any real variable from its trend (for a variable X_t the gap is X̃_t).  
- Baseline assumptions:
  - Adaptive expectations (framework can be adjusted to forward‑looking, model‑consistent expectations).  
  - Well‑defined balanced growth path and stable “great ratios” in steady state.  
- Real economy block includes all key real expenditure variables of national accounts (private consumption, public consumption, total investment, exports, imports, domestic demand, national accounts identity).  
- Monetary regime and exchange rate:
  - Model consistent with BiH currency board and de jure open capital account: no room for independent monetary policy.
  - Domestic policy rates i_t equal eurozone policy rates i*_t.
  - Nominal exchange rate s_t is fixed at 1.95 BAM per euro.  
- Price block:
  - Phillips curve: inflation π_t (quarterly annualized change in consumption deflator) depends on lagged and expected inflation (expected approximated by foreign effective inflation π*_t in the backward‑looking BiH version), output gap, real exchange rate q_t, and exogenous relative prices rpx_t^.  
  - National accounts deflators p_t^N (N = C, G, J, X) are weighted averages of GDP deflator p_t^D and import deflator p_t^M (domestic value‑added share β^N, with 1−β^N import share).  
- Supply side:
  - Potential output approximated by Hodrick‑Prescott filter (framework adaptable to Kalman filter or alternatives).

### Behavioral equations (summary of Table 4)
- Private consumption:
  - ĉ_t = β1^c ĉ_{t−1} + β2^c ỹ_t + β3^c p̃t^tr + β4^c r̂^h_t + β5^c capb̃_t + ε_t^c  
  - Real households’ lending rate r̂^h_t negatively affects consumption.
  - Cyclically‑adjusted primary balance capb̃_t (restrictive fiscal policy) reduces consumption.
- Government consumption:
  - ĝ_t = β1^g ĝ_{t−1} + β2^g pex̃_t + ε_t^g
- Total investment:
  - î_t = β1^i î_{t−1} + β2^i ỹ_t + β3^i ỹ*_t + β4^i fdĩ_t + β5^i r̂^c_t + β6^i cẽ_t + ε_t^i
  - Real corporate lending rate r̂^c_t negatively related to investment; FDI and capital expenditures raise investment demand; inertia via lagged investment.
- Exports:
  - x̂_t = β1^x x̂_{t−1} + β2^x q̂_t + β3^x ỹ*_t + β4^x fdĩ_t + ε_t^x
  - Depend on foreign demand ỹ*_t, real exchange rate q̂_t (price competitiveness), and FDI.
- Imports:
  - m̂_t = β1^m ĉ_t + β2^m ĝ_t + β3^m î_t + β4^m x̂_t + β5^m q̂_t + ε_t^m
  - Real exchange rate depreciation (q̂_t) induces import substitution (imports fall).
- Domestic demand and national accounts identity:
  - d̂_t = β1^d ĉ_t + β2^d ĝ_t + β3^d î_t + ε_t^d
  - ỹ_t = β1^y ĉ_t + β2^y ĝ_t + β3^y î_t + β4^y x̂_t + β5^y m̂_t + ε_t^y
  - Coefficients β are normalized shares of nominal components to nominal potential GDP.
- Private transfers:
  - pt̃_t = β1^{pt} pt̃_{t−1} + β2^{pt} ỹ*_t + β3^{pt} d̂_t + ε_t^{pt}
  - Positive relation with foreign demand, negative with domestic demand; lagged transfers capture inertia.
- FDI:
  - fdĩ_t = β1^{fdi} fdĩ_{t−1} + β2^{fdi} ỹ*_t + ε_t^{fdi}
- Cyclically adjusted fiscal balance:
  - capb̃_t = β1^{rev} car̃_t − β2^{exp} capẽ_t
- Inflation (Phillips curve form):
  - π_t = β1^π π_{t−1} + (1−β1^π) π*_t + β2^π ỹ_t + β3^π q̂_t + β4^π rpx_t^. + ε_t^π
- Expenditure deflators:
  - p_t^N = β1^N p_t^D + (1−β1^N) p_t^M + ε_t^N
- Import deflator:
  - p_t^M = β1^M p_t^{M,eff*} + β2^M p_t^{food*} + β3^M p_t^{metal*} + (1−β1^M−β2^M−β3^M) p_t^{oil*} + ε_t^{M}

### Calibration and numerical parameterization (Bosnia and Herzegovina)
- Consumption equation parameters:
  - β1^c = 0.35; β2^c = 0.10; β3^c = 0.10; β4^c = 0.05; β5^c = 0.25.
- Government consumption:
  - β1^g = 0.60; β2^g = 0.17.
- Investment:
  - β1^i = 0.45; β2^i = 0.40; β3^i = 0.90; β4^i = 0.02; β5^i = 0.30; β6^i = 0.10.
- Exports:
  - β1^x = 0.20; β2^x = 0.30; β3^x = 3.00; β4^x = 0.02.
- Imports:
  - β1^m = 0.23; β2^m = 0.02; β3^m = 0.12; β4^m = 0.63; β5^m = 0.30.
- Domestic demand shares:
  - β1^d = 0.60; β2^d = 0.17; β3^d = 0.23.
- National accounts identity shares (nominal expenditure shares relative to nominal potential GDP):
  - β1^y = 0.67; β2^y = 0.19; β3^y = 0.26; β4^y = 0.45; β5^y = 0.57.
- Private transfers:
  - β1^{pt} = 0.30; β2^{pt} = 2.20; β3^{pt} = 0.30.
- Nominal exchange rate calibration:
  - β1^s = 67.3. (Model variables are log transformed and multiplied by 100; 67.3 = 100*log(1.95).)
- FDI:
  - β1^{fdi} = 0.20; β2^{fdi} = 30.0.
- Cyclically adjusted fiscal balance:
  - β1^{rev} = 0.44; β2^{exp} = 0.34.
- Phillips curve coefficients:
  - β1^π = 0.40; β2^π = 0.25; β3^π = 0.25; β4^π = 0.25.
- Deflators — domestic/import value‑added shares for national accounts deflators and import price decomposition:
  - β1^{C,dom} = 0.20; β1^{C,other} = 0.05; β1^{C,food} = 0.27; β1^{C,oil} = 0.8.  
  - Import deflator shares: β1^M = 0.55; β2^M = 0.35; β3^M = 0.05.

### Trends, steady state relationships, and identity derivations (Appendix C)
- Trends of real variables are functions of potential GDP and the real exchange rate trend; chosen so nominal national accounts expenditure ratios to nominal potential GDP are constant in the long run.  
- Notation and relationships:
  - P_D, P_C, P_M = GDP, households’ consumption, and import deflators.
  - Y, C = real GDP, real households’ consumption.
  - T_D, T_C, T_Q = trends in real GDP, real households’ consumption, and trend in the relative price Q (consumption deflator relative to GDP deflator).  
- Consumption deflator assumed: P_C = (P_D)^{α_C}*(P_M)^{1−α_C} * exp(ε_C). Hence relative price P_C/P_D = (P_M/P_D)^{1−α_C} * exp(ε_C). Define Q = (P_M/P_D) and trend T_Q.  
- Log relationships yield trend equations used to derive trends for any real variable Ω_t consistent with constant nominal shares sℎ_Ω in the long run:
  - log(T_{Ω,t}) = log(T_{D,t}) − s_{Ω}^{M}·log(T_Q) + log(sℎ_Ω) + ε_{Ω,t}^T  (summary form — model implements the detailed log linear derivations in Appendix C).  
- National accounts identity in gaps form:
  - Ỹ_t = sum over expenditure components of sℎ_component · ( (1−α_component)·Q̃_t + component̃_t )  with total import share sℎ_M equal to sum of expenditure import shares weighted by nominal shares.

### Import demand and national accounts deflators calibration (Appendix D)
- Calibration approach:
  - Impose theoretical restrictions and match observed data; estimate log‑linear deflator equations via OLS.
  - Deflators modeled as functions of import price and domestic value‑added price; α_N is estimated domestic value‑added share (1−α_N is import share).
  - Additional constraints imposed so expenditure‑side import shares add up to total import share and homogeneity conditions hold.
- Import deflator specification:
  - Import deflator modeled as function of current and lagged foreign CPI (effective), world food prices, world metal prices, and world oil prices. Fit is reasonably good though data subject to revisions.
- Condition for total import share:
  - sℎ_M = sℎ_C*(1−α_C) + sℎ_G*(1−α_G) + sℎ_J*(1−α_J) + sℎ_X*(1−α_X).  
- Practical calibration notes:
  - Estimated equations are often set up in Excel and coefficients quantified via constrained optimization (minimize squared residuals subject to homogeneity and sign/theoretical constraints). Expert judgment frequently used.
  - To avoid excessive volatility, smoothed data (HP‑filtered series with low lambda = 1) were often used during calibration/estimation.

*Source: APPENDIX B. THE MAIN ECONOMIC RELATIONSHIPS UNDERLYING THE QMFF FOR THE CBBH (Technical Report – Bosnia and Herzegovina)*

---


_Source: https://www.imf.org/-/media/files/publications/tar/2025/english/tarea2025012-print-pdf.pdf_