Investigation of Inflation Forecasting

Forecasting methods of the neural network, ARIMA, ARIMA-GARCH, exponential smoothing and others are introduced. Then using U.S. inflation data, based on the out-of-sample forecasting test, the paper studies the advantages and disadvantages of these methods by the empirical comparisons. The empirical results show that, firstly, from the superior to the inferior, the ranking order of the six methods are, the ARIMA-GARCH, ARIMA, neural networks, median method of autoregressive model, least squares method of autoregressive model, exponential smoothing, no matter based on sample mean absolute error or absolute error for one-step forecasting, or absolute error for two-steps forecasting. Secondly, the ARIMA-GARCH method is suitable most to forecast the inflation level in the USA and sometimes sophisticated methods such as neural networks can not improve the forecasting results. Thirdly, according to the out-of-sample forecasting, directions of forecasting errors of these methods are almost the same, indicating that these forecasts have underestimated the inflation level in the USA.