Enhancing the Accuracy of the Seasonal Model for Forecasting CPI in the Short Run
This paper presents the third version of the model for predicting inflation up to one year ahead. The model can be used for operational monitoring, as it yields a forecast every month of the changes in the CPI in the next twelve months. The forecast is based on automatic identification of stochastic processes--trend and seasonality--in price changes in fourteen subgroups of the consumption basket. The parameters are estimated by means of the X-12-ARIMA program, and are updated with every update of the database.
The trend estimates in the subgroups of the general index enable us to obtain an estimate of core inflation. This estimate is needed for purposes of monetary projection, but it is difficult to obtain in real time and with high frequency because of price volatility in the components of the consumption basket. This paper proposes a smoothed estimate of core inflation in real time as a weighted trend, calculated simultaneously with the forecast, taking into account the interdependence of these estimates.
The new version of the forecast model has extended the exogenous factors of its equations, beyond the depreciation of the shekel against the dollar, and incorporated economic variables which hitherto were not included as continuous independent variables. The variables that have been added are fuel prices appropriate to the CPI components, and import price changes of consumer goods in dollar terms, estimated from the changes in the cross rates of the currencies of nine of Israel's trading partners, including China and Turkey. The new version of the model has been in use since the December 2005 forecast, and it has significantly reduced the ex ante forecast error for the following month by 0.13 percent, compared with 0.20 percent in the previous version. Some of the reduction in the ex ante error is explained by the lower price volatility in the last two years than that which prevailed in the previous three years, but most of it is explained by the additional explanatory power offered by the new version. One practical proposal put forward in the paper is to update the model's specifications at least once in three years.