Abstract:
The purpose of this work is to study and compare existing models and
methods, and build an optimal model for forecasting inflation in Kazakhstan,
taking into account seasonal adjustment of CPI.
This paper compares the performance of two seasonal adjustment methods,
TRAMO-SEATS and X-13-ARIMA-SEATS. Based on the results of seasonal
adjustment, the X-13-ARIMA-SEATS method detects more outliers than the
TRAMO-SEATS method; however, the difference in the seasonally adjusted time
series is negligible. Thus, further comparison of these two methods is conducted
based on the result analysis of the forecast models. To forecast the inflation rate,
two models are used - the ARIMA-GARCH model, and the VAR model.
The out-of-sample forecast is made for a short-term period of six months from
June 2019 to December 2019. Based on the error measures in the validation period
the most adequate and accurate model is the VAR(2) model with CPI seasonally
adjusted by the X-13-ARIMA-SEATS method.