Modelling Producer Prices Using Artificial Neural Network and ARIMA Methods for Türkiye

Abstract

The increase in the producer prices limits the activities of the companies and leads to the deterioration of the national product, employment and consumer prices. In this study, the relations between the oil price, exchange rate, interest rate, wages and the producer prices for the period of 2002Q01-2022Q03 in Türkiye were examined using autoregressive moving averages (ARIMA) and artificial neural network (ANN) methods. The employed ANN structure consists of an input layer, a hidden layer with 100 neurons and an output layer. The ANN is trained and then the modelling and forecasting performances of the traditional ARIMA and nonlinear ANN methods are compared. RMSE, MAE, MAPE and R2 criteria were used to evaluate the predictive power of the ARIMA and ANN models. As a result of automatic ARIMA model estimation, it has been determined that the producer prices can be modelled using an ARMA(4.4) model, which is a subset of the ARIMA modelling. MAE, RMSE, MAPE and R2 values of ARIMA and ANN models show that the ARMA(4,4) model has slightly better accuracy compared to the ANN model. In addition, according to the ARMA(4.4) model, it is shown that the interest rate, exchange rate, oil prices and wages affect producer prices. In this context, our policy recommendations are to follow a low interest policy and to encourage the use and production of electric vehicles to reduce the use of fossil fuels in order to reduce producer prices. Keywords: Producer price index, artificial neural network ARIMA, economic modelling.