EVALUATING THE EFFECT OF EXCHANGE RATE AND PRESIDENTIAL ELECTION YEAR INDICATORS ON INFLATION FORECAST ACCURACY IN NIGERIA: A MACHINE LEARNING APPROACH

ABSTRACT

Accurate inflation forecasting matters for central banks and policymakers, particularly in developing economies. In Nigeria, the Dollar-Naira exchange rate and recurring political uncertainty around presidential elections both influence inflation. Yet, the combination of these two factors in a machine learning forecasting framework has received little attention. This research tests the effects of exchange rate data and a presidential election-year counter on four models, LASSO, Ridge, Random Forest, and XGBoost, forecasting Nigeria's inflation at 1, 3, and 6-month horizons using Central Bank of Nigeria data. The study tests four cases: using inflation data alone as the baseline, adding exchange rate data, adding the election counter, and adding both together. Of all the combinations tested, Random Forest with the election counter alone gave the lowest forecasting error at every horizon, with RMSE values of 0.68, 1.58, and 2.78 for the 1, 3, and 6 month forecasts. Ridge achieved the lowest 1-month RMSE of 0.58 in the baseline case, but fell behind at longer horizons. LASSO gave the same result in all four cases, regardless of which variables were added. At the same time, Ridge changed across cases, showing that the extra variables do carry a useful signal that LASSO throws away. Because election dates are known years in advance, the counter requires no real-time data, making it a cheap and practical input for inflation forecasting in Nigeria and other countries with a fixed election cycle.

Keywords: machine learning, inflation forecasting, presidential election cycle, exchange rate, Nigeria, LASSO, Ridge, Random Forest, XGBoost