CBN's Sentiment-Augmented Inflation Model — Nigeria's Central Bank Adds Machine-Learned Text Mining to Its Forecasting System

The Central Bank of Nigeria (CBN) Research Department has since 2022 explored augmenting its existing Short-Term Inflation Forecasting (STIF) model with a machine-learning-derived sentiment index built by text-mining public sources, after COVID-19 and the Russia-Ukraine war destabilised the reliability of its established econometric forecasts.
CBN researchers Adebiyi, Adenuga, Olusegun and Mbutor computed a sentiment index using machine-learning text-mining methods and tested whether adding it as an input improved the STIF model's predictions of headline, core and food inflation, publishing the results in the Bank's peer-reviewed Economic and Financial Review (Vol. 60, No. 1, 2022).
The study found that STIF specifications including the computed sentiment index out-performed those without it in both in-sample and out-of-sample tests across all inflation components, leading the authors to recommend sentiment-based forecasting alongside "open mouth operations" forward guidance. A follow-on, methodologically independent 2025 study by researchers at Federal University Dutsin-Ma, using the same CBN/National Bureau of Statistics monthly inflation series (2000-2024), quantified just how large the gap between traditional and ML-informed approaches can be: standalone ARIMA produced a Mean Absolute Percentage Error (MAPE) of 37.74%, a standalone Artificial Neural Network cut that to 0.059%, and a hybrid ARIMA-ANN model achieved the lowest error, at 0.054%.
Neither study is an independently audited, production-grade accuracy report of CBN's live forecasting operations, and the 2025 university benchmark is a separate academic exercise rather than a validation of CBN's specific in-house model; the improvement figures should be read as evidence that machine-learning-augmented approaches suit Nigeria's inflation data well, not as CBN's own published live-model accuracy.

Read the full analysis: https://dc.cbn.gov.ng/efr/vol60/iss1/1/