Effects of Exchange Rate Policy on Balance of Payments in Nigeria: A Machine Learning Approach

Abstract

The effects of exchange rate policy on Balance of payments in Nigeria is a sensitive and significant issue because it examines how Nigeria’s exchange rate policy influences external sector stability especially trade, capital flows, foreign reserves and overall balance of payments (BoP) outcomes and uses Machine Learning (ML) to capture effects that may be non-linear, delayed and regime dependent. It can provide timely, data-driven evidence for the Central Bank of Nigeria (CBN) and fiscal policy decisions than a purely conventional linear econometric approach. This study develops a hybrid-macro financial model that integrates the Mundell-Flemming open economy framework with machine learning techniques to analyze exchange-rate policy and balance-of-payments (BoP) dynamics in Nigeria. The theoretical structure follows the IS-LM-BP specification where the goods market (IS), money market (LM) and external balance (BP) jointly determine output, interest rates and the nominal exchange rate under alternative regimes (fixed, managed float, free float). Within this structure, key behavioral equations for net exports, capital flows and money demand are estimated using Nigerian data on output, inflation, interest rates, oil prices, foreign reserves and BoP components. To capture non-linearities, regime shifts and high-frequency dynamics that standard linear specifications miss, we employed Machine learning algorithms such as CatBoost, Ridge regression and SVR (RBF Kernel). Policy simulations examine how monetary policy shocks, fiscal expansions, oil price shocks and changes in exchange rate regime propagate through the model to affect output, inflation, reserves and the current and capital accounts. Results indicate that the hybrid Mundell-Flemming ML model outperforms purely linear IS-LM-BoP balances particularly during periods of policy regime changes and external shocks. The findings reveal that a managed-float regime complemented by rules-based interventions informed by ML-enhanced early-warning signals can improve external stability while preserving monetary policy effectiveness in Nigeria. This research demonstrates that Nigeria's current account balance can be predicted with meaningful accuracy from a small set of economically fundamental, same-year indicators, without resorting to time-series engineering. Ridge regression's outperformance over both a tree-based ensemble (CatBoost) and a kernel method (SVR) reflect the constraints of a genuinely small annual macro dataset — regularized linear models remain the most reliable choice at this scale.

Keywords: Machine learning, balance of payment (BoP), Exchange rate, inflation, output, model, current account

DOI: 10.7176/JESD/17-6-04

Publication date: September 30^th 2026