ENHANCING EXCHANGE RATE FORECASTING: THE IMPACT OF OUTLIER REMOVAL ON USD/KZT PREDICTIONS

Abstract

Exchange rate forecasting presents significant challenges due to the influence of market shocks. This study evaluates forecasting models (ARIMA, SARIMAX, and LSTM) for predicting the USD/KZT exchange rate following Kazakhstan’s 2015 transition to a floating exchange rate regime. Specifically, it investigates whether preprocessing through outlier removal via Changepoint Analysis and Isolation Forest improves forecasting accuracy. Using daily data from 2015 to 2024 and incorporating key exogenous variables, model performance is compared with and without outlier removal. The results show that outlier removal leads to narrower confidence intervals and improved model stability, particularly in ARIMA and SARIMAX forecasts. These findings contribute to understanding optimal preprocessing techniques for exchange rate forecasting in markets subject to external shocks.