A non-linear autoregressive neural network model for forecasting Indian index of industrial production

Abstract

For a developing country such as India, to have the best usage of resources, public planning requires good forecasts of future trends. India's Index of Industrial Production (IIIP) is an index which conveys the status of production in the industrial sector of the economy. In this study, an artificial neural network (ANN) was applied to forecast IIIP. Accordingly, the inputs to the ANN consisted of data spanning from F.Y. 2004-05 to F.Y. 2013-14 of Gross Domestic Product (GDP), Consumer Price Index (CPI), Wholesale Price Index (WPI) and Index of the Eight Core Industries (Electricity, Steel, Refinery Products, Crude Oil, Coal, Cement, Natural Gas and Fertilizers) to forecast IIIP. Therefore, a methodology for forecasting was developed using Nonlinear Autoregressive (NAR) and Nonlinear Autoregressive neural network with exogenous inputs (NARX) models. Several structures of the neural networks were tested for forecasting, and then the results were compared in terms of forecasting error. The NARX network with 11 hidden layers and 1 delay line provided the best results with a Mean Square Error (MSE) of 2.168. Thus, ANNs can be used for accurate forecasting of Industrial Production.