Implementation of the ARIMA-GARCH Model on USD/JPY Exchange Rate Forecasting

Exchange rates exhibit unique market behavior with influences on many macroeconomic factors, making it an interesting subject for time series analysis. The Autoregressive Integrated Moving Average (ARIMA) and Generalized Autoregressive Conditional Heteroskedasticity (GARCH) models have been employed to address the topic, however, the liquidity and non-linearity of the foreign exchange market as a financial time series makes exchange rate forecasting complex. This study examines the application of ARIMA-GARCH models in forecasting the USD/JPY exchange rate, focusing on capturing linear trends and time-varying volatility. The study employs data transformations, including logarithmic scaling and differencing, to achieve stationarity and optimize model performance. The ARIMA model identifies and forecasts the conditional mean, while the GARCH model captures volatility clustering in residuals, and rolling window back testing is implemented. Results indicate that the remaining residuals of the ARIMA-GARCH model are white noise and stationery but depart from normality, characteristic of financial time series. The findings demonstrate the utility of ARIMA-GARCH models in modeling the dynamics of foreign exchange rates, including volatility clustering and short-term shocks, while predicting a pattern of depreciation and diminishing volatility, or market stabilization, in the USD/JPY exchange rate. The study offers practical insights for policymakers and market participants managing exchange rate risks in volatile financial environments.