Abstract:
This study is to investigate forecasts of Bangladesh’s inflation by the linear forecasting method like Autoregressive Integrated Moving Average (ARIMA) and Neural Network (NN) model as Nonlinear Autoregressive network with exogenous inputs (NARX). Inflation forecast is used as guide in the formulation of the monetary policy by the money policy makers worldwide. Monetary policy decisions are based on inflation forecast extracted from the information from different models and other indicators, which influence the macroeconomics conditions of the economy.
ARIMA method is an extrapolation method for forecasting, based on probability theory and statistical analysis with a certainty of distributions assumed in advance and like any other such methods, it requires only the historical time series data for the variable under forecasting. Five different plausible ARIMA estimated models are selected by various diagnostic and selection & evaluation criteria. On the basis of in sample and out of sample forecast and forecast evaluation statistics two candid models among the five models which have sufficient predictive powers and the findings are well compared to the other models are proposed. Artificial neural network (ANN) models are data-driven self- adaptive methods in that there are few a priori assumptions about the models for problems under study. An ANN model is developed to forecast the inflation of Bangladesh as a function of its own previous value. The model selects a feed-forward back-propagation ANN with the input of previous inflation and an exogenous variable of exchange rate, five hidden neurons and one output as the optimum network. The model is tested with actual time series data of inflation in case of Bangladesh and forecast evaluation criteria. The forecast performance of the ANN model is compared with ARIMA based model and observed that RMSE of ANN based forecasts is much less than the RMSE of forecasts based on ARIMA models. So it can be said that forecasting of inflation with ANN offers better performance in comparison with ARIMA methods.
