## 1rwaea2023005

## Source details

**Canonical URL:** [1rwaea2023005](https://www.imf.org/-/media/files/publications/cr/2023/english/1rwaea2023005.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/cr/2023/english/1rwaea2023005.pdf.md)
- [Structured JSON version](/-/media/files/publications/cr/2023/english/1rwaea2023005.pdf.json)

---

### Mission overview
- Virtual TA mission on “Expanding the nowcasting toolbox at the National Bank of Rwanda” (NBR) during December 12–21, 2022.
- Mission team: Messrs. Eilert Husabø and Pål Bergset Ulvedal (STXs, the Norges Bank), and Ms. Jianping Zhou (Mission Chief, MCM).
- Daily technical sessions attended by senior managers and staff from the Monetary Policy & Research Directorate forecasting teams; separate sessions for GDP and CPI nowcasting teams. Opening and concluding sessions included Mr. Thierry Mihigo Kalisa (Chief Economist of the NBR), Mr. Mathias Karangwa (Director of Research Department), and Mr. Bruno Mwenese (Head of Forecast Division).
- Report sent to NBR for review in January; all factual comments included.

### Executive summary — key completed tasks and findings
- Mission objectives:
  - (i) further strengthen the nowcasting framework for GDP and CPI inflation;
  - (ii) initiate development of a simple, efficient, and robust process for forecasting external variables for the Quarterly Projection Model (QPM).
- Completed tasks:
  - Improved nowcasting frameworks for GDP growth and CPI inflation by expanding the nowcasting toolbox and building presentation templates and internal toolboxes for new data releases.
  - Implemented a system for forecast error analysis and applied it in historical forecasts for GDP and inflation.
  - Unified and simplified external data by replacing multiple ad-hoc databases with one database for historical external data and one database for external forecasts.
  - Set up Excel programs to produce higher-quality charts for monthly Monetary Policy Reports.
  - Shared detailed presentations for concluding sessions with the authorities.
- Operational recommendations (high-level):
  - Use new CPI and GDP Near-Term Forecast (NTF) tools monthly as part of forecasting process and FPAS work.
  - CPI NTF includes monthly forecasts of ten subgroups of core CPI inflation and two subgroups of food inflation to assess drivers and monitor monthly outcomes relative to the forecast.
  - GDP NTF focuses on forecasting production across sectors and showed better performance than the current Dynamic Factor Model (DFM); authorities should consider phasing out the DFM.
  - Prioritize improving forecasts of foreign variables and external assumptions for commodity prices; use external providers given resource constraints.
  - Integrate FPAS more fully into monetary policy decision-making; increase permanent FPAS staffing and consider additional IMF TA.

### Key recommendations (priorities and timeframes)
- Establish a simple, robust, and efficient system for producing external assumption forecasts managed by at least two persons; use forecasts from outside providers and enable efficient monitoring of external developments.
  - Priority: High; Timeframe: Near-term
- Replace charts in Monetary Policy Reports with charts produced at this mission.
  - Priority: High; Timeframe: Near-term
- Consider phasing out the current DFM for GDP nowcasting and replace it with the new model system developed during this mission.
  - Priority: High; Timeframe: Near-term
- Continue refining NTF models for the twelve main CPI COICOP groups to move toward bottom-up forecasting of core, energy, and food groups.
  - Priority: Medium; Timeframe: Medium-term
- Note: Near-term defined as < 12 months; Medium-term defined as 12 to 24 months.

### Nowcasting framework — structure, role, and operations
- FPAS technical elements: frameworks for analyzing and forecasting foreign developments, the nowcasting framework, and the QPM.
- Nowcasting team structure and responsibilities:
  - Managed by sector experts (1–3 staff per key variable/sector such as inflation, real sector/GDP, fiscal, labor market).
  - Tasks: monitoring variables; developing/running analytical tools and models (aggregate and disaggregate); performing forecast error analysis; providing judgements; contributing to MPR text and internal memos for new data releases.
  - Nowcasting specialists typically differ from core model team and may belong to separate divisions within Research or within the Monetary Policy and Research Directorate (MPRD).

### CPI nowcasting developments, tools, and data management
- Historical TA progression:
  - September 2021: tools for disaggregate inflation analysis and measures of underlying inflation; outputs used in MPRs.
  - March 2022: nowcasting models based on different indicators; model averaging and out-of-sample forecast error calculations; covered headline, core, food, and energy; outperformed benchmark ARMA and existing VAR up to six months.
  - December 2022 (this mission): extended NTF-system to forecast headline and core CPI from the 12 main COICOP groups, trained staff, developed toolbox bridging NTF and policy model, and built framework for regular forecast error analysis.
- Inflation Dashboard (Excel-based):
  - Fixed and coherent bottom-up CPI forecasting system.
  - Starts with monthly seasonally adjusted level-index forecasts for core, food, and energy from NTF-models; monthly judgement can be applied to these three groups.
  - Aggregates to headline CPI and constructs quarterly indices and growth rates for the policy model.
  - Highly automated and easy to update between forecast rounds; enables real-time monitoring of monthly inflation outcomes relative to forecasts.
  - NTF-forecasts for the twelve COICOP groups inherit judgement applied to core, food, or energy, allowing detailed disaggregate contributions to overall CPI forecasts.
- Data management improvement:
  - Import function amended to improve daily frequency data handling (exchange rates and commodity prices); inflation team should always update high-frequency data before running NTFs.
- Operational recommendation:
  - Use all tools developed during this TA mission and previous missions on a monthly basis to assess new inflation outcomes.

### Forecast error analysis and uncertainty communication
- Forecast error framework developed and applied to evaluation period 2018 Q3 to 2022 Q3.
- Findings:
  - Forecasts for headline CPI and sub-groups have been unbiased (average error = 0).
  - Forecast accuracy has been good (RMSE lower than standard deviation of data).
- Confidence bands:
  - Output used to produce measures of forecast uncertainty (confidence bands) with tools for charting.
  - Charts useful for communicating uncertainty to the Monetary Policy Committee (MPC).
  - NBR concern: small sample for bands and potential confusion if MPC sees bands not aligned with QPM model confidence bands.

### GDP nowcasting — design, evaluation, and model recommendation
- Prior framework (pre-March 2022): bridge equations and a DFM to backcast estimated monthly GDP; lacked out-of-sample evaluations and relied on annual changes.
- March 2022 system: E-views-based GDP nowcasting using simple bivariate bridge equations, model averaging, producing backcast and nowcast; outperformed a simple benchmark autoregressive model for one and two quarter horizons.
- December 2022 refinements:
  - Extended NTF-system to provide disaggregate forecasts for Agriculture, Industry, Services and Net Taxes.
  - New GDP NTF system components:
    1) Aggregate approach: several bridge equations forecast aggregate GDP directly; forecasts combined via weighted average with weights reflecting historical forecasting performance; different sets for two horizons.
    2) Two-sector approach: Agricultural and Non-Agricultural sectors forecast separately using same framework.
    3) Four-sector approach: Agriculture, Industry, Services and Net Taxes forecasted separately using same framework.
  - GDP forecast spreadsheet in Excel summarizes nowcasts and allows sector experts to add judgement and compare to historical outcomes and model nowcasts.
- Forecast error analysis (evaluation sample: 2019Q1-2022Q2; results hold if extended to 2015Q1):
  - RMSE: Horizon 1 / RMSE: Horizon 2
    - Four-sector approach: 2.7 / 4.1
    - Two-sector approach: 2.6 / 3.1
    - Aggragate: 2.7 / 3.3
    - DFM (old system): 6.2 /
    - AR1: 6.4 / 7.4
  - Conclusion: The new system outperforms both the DFM model and a simple autoregressive benchmark (AR1).
- Outputs used to produce confidence bands around GDP forecasts, analogous to CPI bands.
- Analytical guidance: Focus on seasonally adjusted quarterly growth rates rather than annual changes when analyzing real sector outcomes.
- Recommendation: Use GDP outcome analysis tools regularly and replace the DFM with the new NTF-system.

### External assumptions — identified problems and new setup
- Identified problems with existing system:
  - Automatic linking of historical data to out-of-date annual forecasts produced unrealistic quarterly paths (example: US PCE in November 2022 MPR where up-to-date US PCE data were linked to CBO year-over-year inflation forecasts from May 2022).
  - MATLAB driver backing out quarterly growth rates consistent with historical data and y-o-y forecasts led to unrealistically low forward inflation and strange q-o-q patterns.
  - Similar issues for GDP and commodity price forecasts when annual-frequency forecasts linked to monthly-frequency historical data.
- New external database setup:
  - Consolidated from three main databases (US, euro area, world food prices) and several ad hoc databases into:
    - One database for historical external data (mirroring the domestic database).
    - One database for external forecasts.
  - Benefits: easier navigation, better overview, simpler and more robust update procedures, easier access for CPI and GDP NTF-systems.
- Proposed operational approach:
  - At least two persons should oversee linking historical data to forecasts and monitor external developments.
  - Use forecasts directly from external providers rather than in-house forecasts given human resource constraints.
  - Mission conclusion: using outside providers for external assumptions is more efficient at the current stage; NBR forecast team should focus on domestic forecasting.
  - Specific proposal: forecasts for GDP and inflation (US and euro area) should be based on up-to-date forecasts from Bloomberg Consensus (in principle updated daily).
  - External assumptions team should be able to add judgement to provider forecasts when needed; workload should be light but robust staffing (at least two persons) is required.
  - For remaining variables, forecasts should be based on financial market variables as indicated in mission materials.
  - Caution: ensure new systems do not produce unintended effects on policy model functioning.

### Enhancing communications and operational templates
- Developed presentation templates for nowcasting meetings (CPI inflation and GDP) that:
  - Analyze recent outcomes with focus on sectoral contributions or CPI subgroups.
  - Compare recent outcomes with the previous nowcast and explain deviations.
  - Leverage disaggregated nowcasts to provide a clearer story behind forecast errors.
- Templates present recent developments in key indicators, updated nowcasts, model nowcasts, and sector-expert judgement.
- Recommendation: Develop templates for internal memos presenting recent outcomes for GDP and CPI; memos should include outputs from tools developed during missions and indicate how forecasts fare against new outcomes. Such memos are standard at leading central banks and typically circulated a few hours after NBS releases.
- NBR staff later indicated that analysis on outcomes relative to forecasts can be added to existing inflation briefs without additional assistance.

### Human resources, sustainability, and next steps
- FPAS progress swift, but human resources are a binding constraint.
- Current situation:
  - FPAS tasks include monitoring, model development, forecast error analysis, updating forecasts between official rounds, and communications — workload increasing.
  - Both nowcasting and QPM teams are smaller than in most inflation-targeting central banks and often dedicate only part of their time to FPAS.
  - The nowcasting team should be freed to continue FPAS development to maintain sustainability and credibility.
  - When report prepared, only one NBR staff operated the CPI nowcasting; according to the authorities, the team has been expanded to include three staffs and one young economist.
- Next steps emphasized by mission:
  - Continue development and monthly use of CPI and GDP NTF tools as core parts of the forecasting process.
  - Phase in the new GDP nowcasting model system and refine CPI NTF models toward bottom-up forecasting of core, energy, and food groups.
  - Establish and staff a simple, robust external-assumption forecasting system (managed by at least two persons, leveraging external providers).
  - Fully integrate FPAS outputs into monetary policy decision-making and expand permanent FPAS staffing; consider further IMF TA support.
  - Staff should prepare a manual for the new tools and new measures on underlying inflation.
  - Preferred modality for next TA mission: in person.

### FPAS logframe highlights and status (selected entries as of 12/21/2022)
- Objective: Improve the analytical and forecasting capabilities for monetary policy decision-making – MAF
  - Outcome: FPAS fully integrated into the monetary policy decision-making process.
  - Annual Assessment Rating: 2 Partially Achieved
  - Outcome Rating Date: 12/21/2022
- Selected outcome indicators and assessments (as of 12/21/2022):
  - "A structured forecasting process and schedule is in place with specified responsibilities.": Partially achieved as of 12/21/2022
  - "Analytical reports are published routinely on the same frequency. A monetary policy report is published regularly": Largely achieved as of 12/21/2022
  - "Forecasting team presents macroeconomic developments and forecast report to the monetary policy committee (MPC).": Largely achieved as of 12/21/2022
  - "MPC understands the structure of the core QPM and the underlying FPAS mechanisms.": Partially achieved as of 12/21/2022
  - "MPC uses staff projections in their policy decision making.": Partially achieved as of 12/21/2022
  - "Regular interactions between MPC and FT take place according to the forecast calendar.": Partially achieved as of 12/21/2022
- Selected milestone ratings:
  - Forecast schedule adjusted for important data releases: Target Completion Date 4/30/2022 — Milestone Rating 2 Partially Achieved as of 12/21/2022
  - Monetary policy meeting schedule published half a year in advance: Target Completion Date 4/28/2023 — Milestone Rating 1 Not Achieved as of 12/21/2022
  - Forecast process includes regular interaction between policy makers and staff: Target Completion Date 4/30/2023 — Milestone Rating 2 Partially Achieved as of 12/21/2022
- Outcome: Improved economic analysis and forecasting capabilities at the CB.
  - Annual Assessment Rating: 2 Partially Achieved
  - Outcome Rating Date: 11/22/2021
  - Selected indicators and assessments (as of 12/21/2022):
    - "A forecasting team is set up and integrated in the decision-making process.": Largely Achieved as of 12/21/2022
    - "A near term forecasting toolkit is operational, used in the regular forecasting exercise rounds, and regularly updated.": Largely achieved as of 12/21/2022
    - "A quarterly projection model is operational...": Largely achieved as of 12/21/2022
  - Selected milestone ratings (examples):
    - Sector Expert capacity being developed and integrated: Target Completion Date 4/30/2022 — Milestone Rating 2 Partially Achieved as of 12/21/2022
    - A suite of near-term forecasting tools and models developed: Target Completion Date 4/30/2023 — Milestone Rating 3 Largely Achieved as of 12/21/2022
    - Benchmark QPM extended with relevant additional blocks: Target Completion Date 4/30/2024 — Milestone Rating 3 Largely Achieved as of 12/21/2022

*Source: IMF TA mission report (1rwaea2023005) — mission findings, system descriptions, evaluation metrics, and recommendations as provided in the chapter text.*

### Preface____________________________________________________________________5

### Preface____________________________________________________________________5

### Mission overview
- A virtual TA mission on “Expanding the nowcasting toolbox at the National Bank of Rwanda” (NBR) took place during December 12–21, 2022.
- Mission team: Messrs. Eilert Husabø and Pål Bergset Ulvedal (STXs, the Norges Bank), and Ms. Jianping Zhou (Mission Chief, MCM).
- Daily technical sessions attended by senior managers and staff from the Monetary Policy & Research Directorate forecasting teams; separate sessions for GDP and CPI nowcasting teams. Opening and concluding sessions included Mr. Thierry Mihigo Kalisa (Chief Economist of the NBR), Mr. Mathias Karangwa (Director of Research Department), and Mr. Bruno Mwenese (Head of Forecast Division).
- The report was sent to the NBR for review in January and all factual comments have been included.

### Executive Summary — key completed tasks and findings
- Mission objectives:
  - (i) further strengthen the nowcasting framework for GDP and CPI inflation;
  - (ii) initiate development of a simple, efficient, and robust process for forecasting external variables for the Quarterly Projection Model (QPM).
- Completed tasks:
  - Improved nowcasting frameworks for GDP growth and CPI inflation by expanding the nowcasting toolbox and building presentation templates and internal toolboxes for new data releases.
  - Implemented a system for forecast error analysis and applied it in historical forecasts for GDP and inflation.
  - Unified and simplified external data by replacing multiple ad-hoc databases with one database for historical external data and one database for external forecasts.
  - Set up Excel programs to produce higher-quality charts for monthly Monetary Policy Reports (replacing current copy-paste of PDF charts).
  - Shared detailed presentations for concluding sessions with the authorities.
- Operational recommendations from the mission:
  - The new CPI and GDP Near-Term Forecast (NTF) tools should be used monthly as part of the forecasting process and FPAS work at the NBR.
  - The CPI NTF system now includes monthly forecasts of ten subgroups of core CPI inflation and two subgroups of food inflation, enabling assessment of key drivers and nature of shocks and allowing “real time” monitoring of monthly inflation outcomes relative to the forecast.
  - The GDP NTF system focuses on forecasting production across sectors and significantly improves forecasting results; the authorities should consider phasing out the current Dynamic Factor Model (DFM) in favor of the new model system developed during the mission, which demonstrated better forecasting performance.
  - Improving forecasts of foreign variables and external assumptions for commodity prices should be prioritized; given resource constraints, using external providers is currently more efficient than relying solely on in-house forecasting.
  - FPAS needs fuller integration into the monetary policy decision-making process; increase permanent FPAS staffing and consider additional IMF TA to strengthen the NBR’s monetary policy process.

### Key recommendations (excerpted priorities and timeframes)
- NBR should continue establishing a simple, robust, and efficient system for producing external assumption forecasts for the policy model; system should be managed by at least two persons, be based on forecasts from outside providers (such as Macro Bonds/Bloomberg consensus), and allow efficient monitoring of external developments.
  - Priority: High; Timeframe: Near-term
- NBR should replace charts in Monetary Policy Reports with charts produced at this mission.
  - Priority: High; Timeframe: Near-term
- NBR should consider phasing out the current DFM for GDP nowcasting and replace it with the new model system developed during this mission that has demonstrated far better forecasting performance.
  - Priority: High; Timeframe: Near-term
- NBR should continue refining the newly developed models in the NTF system for the twelve main CPI COICOP groups, aiming to move from forecasting core, energy, and food inflation directly to forecasting these groups bottom-up.
  - Priority: Medium; Timeframe: Medium-term
- Note: Near-term defined as < 12 months; Medium-term defined as 12 to 24 months.

### Nowcasting framework — structure and role
- FPAS technical elements: frameworks for analyzing and forecasting foreign developments, the nowcasting framework, and the QPM.
- Nowcasting team structure and responsibilities:
  - Managed by sector experts (1–3 staff per key variable/sector such as inflation, real sector/GDP, fiscal, labor market).
  - Tasks include monitoring variables, developing/running analytical tools and models (aggregate and disaggregate), performing forecast error analysis, providing judgements, contributing to MPR text and internal memos for new data releases.
  - Nowcasting specialists typically differ from core model team and may belong to separate divisions within Research or within the Monetary Policy and Research Directorate (MPRD).

### CPI nowcasting developments and tools
- Historical TA progression:
  - September 2021: tools developed for deeper disaggregate analysis of inflation and measures of underlying inflation; outputs used in MPRs.
  - March 2022: system of nowcasting models based on different indicators/explanatory variables developed; system used out-of-sample forecast error calculations and model averaging; covered headline, core, food, and energy inflation; outperformed benchmark ARMA and existing VAR up to six months.
  - December 2022 (this mission): refined outputs, extended NTF-system to forecast headline and core CPI from the 12 main COICOP groups, trained staff, developed toolbox bridging NTF and policy model, and built framework for regular forecast error analysis.
- Inflation Dashboard (Excel-based):
  - Provides fixed and coherent bottom-up CPI forecasting system.
  - Starts with monthly seasonally adjusted level-index forecasts for core, food, and energy from NTF-models; judgement can be applied monthly to these three groups.
  - Aggregates to headline CPI and constructs quarterly indices and growth rates for the policy model.
  - Highly automated and easy to update between forecast rounds; enables real-time monitoring of monthly inflation outcomes relative to forecasts.
  - NTF-forecasts for the twelve COICOP groups inherit judgement applied to core, food, or energy, allowing detailed disaggregate contributions to overall CPI forecasts.
- Data management improvement:
  - Import function of CPI-NTF-system amended to improve daily frequency data handling (exchange rates and commodity prices); inflation team should always update high-frequency data before running NTFs.

### Forecast error analysis and uncertainty communication
- Forecast error framework developed and applied to evaluation period 2018 Q3 to 2022 Q3.
- Findings from forecast error analysis:
  - Forecasts for headline CPI and sub-groups have been unbiased (average error = 0).
  - Forecast accuracy has been good (RMSE lower than standard deviation of data).
- Confidence bands:
  - Output used to produce measures of forecast uncertainty (confidence bands) with easy-to-use tools for charting.
  - Charts useful for communicating uncertainty to the Monetary Policy Committee (MPC).
  - NBR concern: small sample for bands and potential confusion if MPC sees bands not aligned with QPM model confidence bands (QPM bands reflect model uncertainty from that system).

### GDP nowcasting and model recommendation
- GDP NTF system emphasizes forecasting production by sector and significantly improves forecasting results relative to previous methods.
- Recommendation: Consider phasing out the current Dynamic Factor Model (DFM) and replace it with the new model system developed during the mission, given its demonstrated better forecasting performance.

### Human resources and sustainability
- FPAS progress has been swift, but human resources are a binding constraint for further progress.
- Current situation:
  - FPAS tasks include monitoring, model development, forecast error analysis, updating forecasts between official rounds, and communications — workload is increasing.
  - Both nowcasting and QPM teams are smaller than in most inflation-targeting central banks and often dedicate only part of their time to FPAS.
  - The nowcasting team should be freed to continue FPAS development to maintain sustainability and credibility.
- Staffing note: When the report was prepared, there was only one NBR staff operating the CPI nowcasting; according to the authorities, the team has been expanded to include three staffs and one young economist.

### Next steps emphasized by the mission
- Continue development and monthly use of CPI and GDP NTF tools as core parts of the forecasting process.
- Phase in the new GDP nowcasting model system and refine CPI NTF models toward bottom-up forecasting of core, energy, and food groups.
- Establish and staff a simple, robust external-assumption forecasting system (managed by at least two persons, leveraging external providers).
- Fully integrate FPAS outputs into monetary policy decision-making and expand permanent FPAS staffing; consider further IMF TA support.

*Source: 1rwaea2023005 - Preface____________________________________________________________________5*

### 17.      The mission also expanded the variables analyzed in the inflation outcome

### 1rwaea2023005 - 17.      The mission also expanded the variables analyzed in the inflation outcome

### Inflation outcome system and operational use
- The mission expanded the variables analyzed in the inflation outcome system and trained NBR staff on how to use the system.
- Output from the inflation outcome system was refined to produce higher quality graphs used for external and internal communication.
- Previously, graphs were copied from the system and pasted directly into the MPRs, causing layout and image quality conflicts with the MPRs.
- The mission resolved issues with the domestic database, including duplicate data and incorrect weights.
- Outputs from all tools developed in this mission were added to a template of nowcasting presentations for the forecast meetings.
- Recommendation: All tools developed during this TA mission as well as the previous missions should be used on a monthly basis to assess the new outcomes of inflation.

### The Nowcasting System for GDP — design and evaluation
- Prior to March 2022, the NBR framework included bridge equations based on domestic explanatory variables and a Dynamic Factor Model (DFM) used to "backcast" an estimated monthly GDP variable (quarterly GDP divided by 3); the framework lacked out-of-sample evaluations and relied on annual changes.
- March FPAS mission: built an E-views-based GDP nowcasting system similar to CPI, emphasizing transparency, judgement, and storytelling.
  - System uses simple bivariate bridge equations with indicators shown to perform well in out-of-sample forecasting.
  - Produces both a backcast (previous quarter) and a nowcast (current quarter) using model averaging.
  - System outperforms a simple benchmark autoregressive model for both the one and two quarter horizons.
- Implementation challenge: The March system had not yet been used at the NBR due to difficulties updating external variables; this highlights the need for systematic organization and updating of external data. The new external database eases this process.
- December 2022 mission refinements:
  - Extended the NTF-system to provide disaggregate forecasts for Agriculture, Industry, Services and Net Taxes to complement aggregate nowcasts and provide a story behind the nowcast.
  - The new NTF system for GDP now consists of:
    1) Aggregate approach: several bridge equations forecast aggregate GDP directly; forecasts are combined via weighted average where weights reflect historical forecasting performance; different bridge equation sets used for the two horizons.
    2) Two-sector approach: Agricultural and Non-Agricultural sectors forecast separately using the same bridge-equation and model-averaging framework.
    3) Four-sector approach: Agriculture, Industry, Services and Net Taxes forecasted separately using the same framework.
  - A GDP forecast spreadsheet in Excel summarizes nowcasts and allows sector experts to add judgement and compare to historical outcomes and model nowcasts.
- Forecast error analysis (evaluation sample: 2019Q1-2022Q2; results hold if extended to 2015Q1):
  - RMSE: Horizon 1 / RMSE: Horizon 2
    - Four-sector approach: 2.7 / 4.1
    - Two-sector approach: 2.6 / 3.1
    - Aggragate: 2.7 / 3.3
    - DFM (old system): 6.2 / 
    - AR1: 6.4 / 7.4
  - Conclusion: The new system outperforms both the DFM model and a simple autoregressive benchmark (AR1).
- Outputs used to produce confidence bands around GDP forecasts, analogous to CPI inflation uncertainty bands.
- Analytical guidance: When analyzing real sector outcomes, focus on seasonally adjusted quarterly growth rates rather than annual changes to avoid missing current developments.
- Recommendation: Tools developed for GDP outcome analysis and nowcasting should be used regularly as new data becomes available and the NTF-system should replace the current DFM.

### External assumptions — problems identified and new setup
- Identified problems: the existing system for constructing external assumptions sometimes produced significant model dynamics distortions and policy recommendation effects.
- Existing system details:
  - Uses MATLAB drivers to automatically link historical data to forecasts and transform data to quarterly frequency for policy models.
  - Forecasts are collected from external sources such as the US Congressional Budget Office (CBO).
- Problem example:
  - Out-of-date external projections linked to up-to-date historical data produced unrealistic quarterly paths (illustrated for US PCE in the November 2022 MPR, where up-to-date US PCE data were linked to CBO year-over-year inflation forecasts from May 2022, when actual US inflation was much higher).
  - The MATLAB driver backs out a quarterly growth rate consistent with historical data and the y-o-y forecasts, causing unrealistically low forward inflation and strange q-o-q inflation patterns.
  - Similar issues affect GDP and commodity price forecasts when annual-frequency forecasts are linked to monthly-frequency historical data.
- New external database setup:
  - Consolidated from three main databases (US, euro area, world food prices) and several ad hoc databases into:
    - One database for historical external data (mirroring the domestic database).
    - One database for external forecasts.
  - Benefits: easier navigation, better overview, simpler and more robust update procedures, easier access for CPI and GDP NTF-systems.
- Proposed new system for external assumptions:
  - At least two persons should oversee linking historical data to forecasts and monitor external developments and impacts on forecasts.
  - Due to human resource constraints, recommend using forecasts directly from external providers rather than in-house forecasts.
  - Mission conclusion: using outside providers for external assumptions is more efficient at the current stage; NBR forecast team should focus on domestic forecasting.
  - Specific proposal: forecasts for GDP and inflation (US and euro area) should be based on up-to-date forecasts from Bloomberg Consensus (in principle updated daily).
  - External assumptions team should be able to add judgement to provider forecasts when needed; workload should be light but robust staffing (at least two persons) is required.
  - For remaining variables, forecasts should be based on financial market variables as indicated in Figure 9.
  - Caution: when building new systems, ensure no unintended effects on policy model functioning.

### Enhancing communication tools
- Developed presentation templates for nowcasting meetings (CPI inflation and GDP) that:
  - Analyze recent outcomes with focus on sectoral contributions or CPI subgroups.
  - Compare recent outcomes with the previous nowcast and explain deviations.
  - Leverage disaggregated nowcasts to provide a clearer story behind forecast errors.
- Templates present recent developments in key indicators, updated nowcasts, model nowcasts, and sector-expert judgement.
- Recommendation: Future missions should assist NBR in continued development of more informative nowcasting presentations based on outputs from the new nowcasting frameworks.
- Recommendation: Develop templates for internal memos presenting recent outcomes for GDP and CPI; memos should include outputs from tools developed during missions and indicate how forecasts fare against new outcomes. Such memos are standard at leading central banks and typically circulated a few hours after NBS releases.
- Note: NBR staff later indicated analysis on outcomes relative to forecasts can be added to existing inflation briefs and they would not require assistance developing that product at the time.

### Next steps and institutional recommendations
- Continue refining and improving the nowcasting framework to support deeper monetary policy analysis; disaggregated frameworks improve understanding and policy communication.
- Preferred modality for next TA mission: in person.
- Operationalization and usage:
  - New CPI and GDP NTF tools should be used monthly as part of the forecasting process and FPAS work at the NBR.
  - The new CPI NTF system includes monthly forecasts of ten subgroups of the core CPI inflation and two subgroups of food inflation, enabling assessment of key drivers and nature of inflation shocks and real-time monitoring of monthly outcomes relative to forecast.
  - The new GDP NTF-system focuses on production forecasting in different sectors and significantly improves forecasting results; it should replace the DFM.
  - Staff should prepare a manual for the new tools and new measures on underlying inflation.
- External assumptions system:
  - Continue establishing a simple, robust, efficient system managed by at least two persons, based on external providers (such as Macro Bonds/Bloomberg consensus), and enabling efficient monitoring of external developments.
- Human resources:
  - More permanent staff needed to work on FPAS.
- Additional IMF TA could help strengthen the monetary policy process and better align forecast/decision-making with communication; communications should be part of broader TA on communications.
- Reference: NBR’s “Explanatory Note on the Indicators of Underlying Inflation in Rwanda” received after the mission describes different ways of measuring inflation in Rwanda (headline, core, trimmed mean, etc.).

### FPAS logframe highlights and status (selected entries)
- Objective: Improve the analytical and forecasting capabilities for monetary policy decision-making – MAF
- Outcome: FPAS fully integrated into the monetary policy decision-making process.
- Annual Assessment Rating: 2 Partially Achieved
- Outcome Rating Date: 12/21/2022
- Selected outcome indicators and current assessments (as of 12/21/2022):
  - "A structured forecasting process and schedule is in place with specified responsibilities.": Partially achieved as of 12/21/2022
  - "Analytical reports are published routinely on the same frequency. A monetary policy report is published regularly": Largely achieved as of 12/21/2022
  - "Forecasting team presents macroeconomic developments and forecast report to the monetary policy committee (MPC).": Largely achieved as of 12/21/2022
  - "MPC understands the structure of the core QPM and the underlying FPAS mechanisms.": Partially achieved as of 12/21/2022
  - "MPC uses staff projections in their policy decision making.": Partially achieved as of 12/21/2022
  - "Regular interactions between MPC and FT take place according to the forecast calendar.": Partially achieved as of 12/21/2022
- Selected milestone ratings:
  - Forecast schedule adjusted for important data releases: Target Completion Date 4/30/2022 — Milestone Rating 2 Partially Achieved as of 12/21/2022
  - Monetary policy meeting schedule published half a year in advance: Target Completion Date 4/28/2023 — Milestone Rating 1 Not Achieved as of 12/21/2022
  - Forecast process includes regular interaction between policy makers and staff: Target Completion Date 4/30/2023 — Milestone Rating 2 Partially Achieved as of 12/21/2022
- Outcome: Improved economic analysis and forecasting capabilities at the CB.
  - Annual Assessment Rating: 2 Partially Achieved
  - Outcome Rating Date: 11/22/2021
  - Selected indicators and assessments (as of 12/21/2022):
    - "A forecasting team is set up and integrated in the decision-making process.": Largely Achieved as of 12/21/2022
    - "A near term forecasting toolkit is operational, used in the regular forecasting exercise rounds, and regularly updated.": Largely achieved as of 12/21/2022
    - "A quarterly projection model is operational...": Largely achieved as of 12/21/2022
  - Selected milestone ratings (examples):
    - Sector Expert capacity being developed and integrated: Target Completion Date 4/30/2022 — Milestone Rating 2 Partially Achieved as of 12/21/2022
    - A suite of near-term forecasting tools and models developed: Target Completion Date 4/30/2023 — Milestone Rating 3 Largely Achieved as of 12/21/2022
    - Benchmark QPM extended with relevant additional blocks: Target Completion Date 4/30/2024 — Milestone Rating 3 Largely Achieved as of 12/21/2022

*Source: IMF TA mission report (1rwaea2023005) — mission findings, system descriptions, evaluation metrics, and recommendations as provided in the chapter text.*

---


_Source: https://www.imf.org/-/media/files/publications/cr/2023/english/1rwaea2023005.pdf_