COMPARISON OF ARIMA, GARCH AND NNAR MODELS FOR MODELLING EXCHANGE RATE IN NIGERIA

Abstract

In the pursuit to understand the importance of exchange rates to any economy, it becomes very expedient to build reliable models for the prediction of the volatility of exchange rates of home currency vis-à-vis the currencies of the developed nations, especially the nations with whom the home country have bilateral economic relationship; such as the United States of America (USA), China, Japan, to mention but few. Thus, this study investigates the characteristics or features of Nigeria exchange rate (Naira/USD), as well as the conventional facts of the exchange rate using Neural Network Autoregressive (NNAR) model and popular BJ-type models such as Autoregressive Integrated Moving Average (ARIMA) and Generalized Autoregressive Conditional Heteroscedastic (GARCH) models. The study utilized secondarily sourced daily time series data from Central Bank of Nigeria websites that covers the period between January 2021 and December 2022. The return series was computed and the Box-Jenkins, GARCH and NNAR modelling methodologies were implemented in R environment. The study empirical results revealed that among this thirteen (13) candidate ARIMA models estimated,  returned as the most parsimonious ARIMA model with the lowest Akaike Information Criterion (AIC). Also,  returned as the most parsimonious GARCH-type models for the series. Lastly,  model returned as the appropriate fitted NNAR models for the series. Furthermore, the three (3) utilized accuracy functions i.e., Root Mean Square Error (RMSE), Mean Square Error (MSE) and Mean Square Error (MAE) criteria established  with the minimum accuracy values across the three (3) evaluation criteria. Thus, this study concludes that  is the optimal model for the examined exchange rate returns series and it outperformed the and  time series models.