Forecasting Stock Price Using Macroeconomic Variables: A Hybrid ARDL, ARIMA and Artificial Neural Network

Forecasting accuracy is the most important factor selecting any forecasting methods. Research for improving the accuracy of forecasting models has never been stopped. The idea in this paper is simple and old while the practice is straightforward using the software technology in use. We intend to filter out the residuals from a multivariate time series causality model by a univariate (residual term) time series model, then to remove any possible systematic component if left at all, by using an artificial neural networks. Doing so, we believe the hybrid method will take the advantages of each and all model in use. In this practice we have compared the ultimate residuals left out of ARDL, ARIMA and ANNs linked, with that of ARDL, ARIMA and ANN, individually. The data set in our experiments consists of few macroeconomic variables such as consumer price index, interest rate, exchange rate and money volume, used to forecast the time series Tehran stock index, a very small and volatile market. Experimental results as expected, has indicated that the hybrid model is a better predictor for stock price compared to each of the ARDL, ARIMA and the Artificial Neural Networks.