FORECASTING CONSUMER PRICE INDEX USING MACROECONOMIC VARIABLES: A COMPARATIVE ANALYSIS OF MACHINE LEARNING AND DEEP LEARNING APPROACHES

Abstract

The Turkish economy has faced many economic difficulties throughout it's history. At this point, predicting inflation accurately is very important for policy makers, businesses, investors and consumers. This study aims to estimate the Turkish Consumer Price Index. Producer price index, M1 money supply, gold price, dollar price, natural gas price and interest rate variables were used to estimate the CPI for Turkey. The variables used in the research were obtained through EVDS, the Central Bank's Electronic Data Management System. Monthly data from January 2003 to August 2023 was used in the study. The obtained data were estimated using DDPG, XGBoost, SVR, KNN and CNN-BiLSTM methods. Model performances were compared using RMSE, MSE, MAE, MAPE and R2 statistical coefficients. When model performances were evaluated, the best CPI prediction for Turkey was obtained by the SVR method.