FORECASTING EXCHANGE RATES IN PAKISTAN: A COMPARATIVE STUDY OF HYBRID ARIMA AND ARTIFICIAL NEURAL NETWORK APPROACHES

Abstract

In the sphere of research, ARIMA and ANN offer a strong approach for time series data prediction.  Both linear and nonlinear patterns are frequently seen in time series data.  Consequently, when it comes to modeling and forecasting time series data, neither ARIMA nor neural networks are suitable.  The majority of current research appears to employ the same specification for forecasting and estimating when using linear models, but the dynamic influence of the relevant variables is disregarded.  To capture both the linear and nonlinear data, this study merged the ARIMA and artificial neural network model using both an equal weighted technique and a profit weighted strategy Elements of the exchange rate, as well as creating a hybrid approach utilizing autoregressive integrated moving average (ARIMA) and artificial neural network (ANN) models. The results were compared to those of the ANN and hybrid ARIMA models.  The future dollar exchange rate as well as data imports and exports are predicted using hybrid models.  The results shown that the benefits of both linear and nonlinear modeling may be obtained by combining the ARIMA and ANN models Standard statistical measurements like mean absolute error (MAE), root mean square error (RMSE), and mean squared error (MSE) are used to evaluate the two models' capabilities. Models effectiveness the effectiveness of the models are analyzed for the foreign exchange rate, imports and exports of the data and concluded that hybrid techniques provided the best forecasting results