Evaluating forecasting models in volatile exchange rate environment: Insights from Sierra Leone

This paper evaluates the predictive performance of several econometric models for exchange rate volatility, focusing on the Sierra Leonean Leone (SLE) to US Dollar (USD) exchange rate. Given Sierra Leone's dependency on imports and remittances, accurate forecasting of exchange rates is critical, particularly during periods of heightened political and economic uncertainty, such as the covid19 pandemic and the 2023 general elections. The analysis compares baseline models (mean, drift, naïve), traditional regression models (linear, exponential, stepwise), and advanced time series techniques (SARIMA, ETS) using monthly data from 2014 to 2024. Empirical results indicate that traditional models, while effective in capturing short-term trends, are insufficient during periods of structural breaks and volatility. Advanced models, particularly ETS, outperform others by dynamically adjusting to irregular seasonality and external shocks. ETS exhibits superior performance in volatile conditions, offering the lowest forecast errors (RMSE and MAE) and greater adaptability. Traditional models like ARIMA, SARIMA, and ETS are preferred over Prophet and GARCH in this context due to their superior interpretability, ability to handle small datasets and seasonality, and suitability for providing actionable insights in economic policy-making