Since 2022, Rwanda's central bank has used an IMF-built machine-learning nowcasting toolbox — covering 10 core-CPI and 2 food-inflation subgroups — to feed its policy-analysis system with monthly, near-real-time inflation and GDP signals despite lagging official data.
Details
Maturity
Established
Promoter
National Bank of Rwanda (Banki Nkuru y'u Rwanda, BNR), with IMF Monetary and Capital Markets Department technical assistance
Period
March 2022 – ongoing (expanded October 2023 and January 2025)
Keywords
central banking, monetary policy, machine learning, inflation nowcasting, IMF technical assistance
Context
Rwanda's central bank, the National Bank of Rwanda (BNR), addresses a problem common to developing-economy central banks: official GDP and inflation data arrive with long lags, hampering real-time monetary policy decisions. Beginning with a March 2022 IMF technical assistance mission and expanded through further missions in October 2023 and January 2025, IMF experts and BNR staff built the CPI and GDP 'Near-Term Forecast' (NTF) toolbox.
Activities
The toolbox applies statistical and machine-learning techniques to high-frequency indicators to generate monthly forecasts disaggregated into ten core-CPI subgroups and two food-inflation subgroups, helping staff distinguish supply-shock-driven inflation from demand-driven inflation. Training also covered building uncertainty fan charts and a framework for assessing weather-shock effects on food prices.
Results
IMF technical assistance reports state the tools 'should be used on a monthly basis' within BNR's Forecasting and Policy Analysis System, and BNR's own Monetary Policy Reports, including May 2025, reference outputs from this nowcasting work.
Conclusions
Governance is relatively transparent by central-bank standards, with the IMF publishing full technical assistance reports detailing methodology and known limitations with the recipient country's consent. The available documentation does not fully quantify how directly monetary policy decisions incorporate the tool's outputs, so its impact is best read as operationally integrated into staff analysis rather than a proven driver of specific rate decisions.
Implementation
Indicative cost
Low (< €50k)
Time to results
Long (> 3 years)
Staffing & skills
National Bank of Rwanda (BNR) staff, IMF Monetary and Capital Markets Department technical assistance experts
Conditions for success
Multi-year, phased IMF technical assistance missions (2022, 2023, 2025)
Integration into BNR's existing Forecasting and Policy Analysis System (FPAS)
Building complementary capacity such as uncertainty fan charts and weather-shock frameworks
Common failure modes
No published ex-post accuracy statistics quantify forecast error reduction
Documentation does not clarify how directly monetary policy decisions incorporate the toolbox's outputs
Where it fits
Governance type
central bank with IMF technical assistance
Scale
national
Income level
low-income
Data sources
Where this practice's information was retrieved from, and when.
SARB researchers rigorously tested machine-learning models (LASSO, XGBoost, neural networks) against the bank's own published inflation forecasts: ML modestly beat …
In an April 2025 Staff Memo, Riksbank economists found random-forest and neural-network models measurably outperformed Sweden's traditional time-series forecasts for …