PERSISTENCE IN INFLATION: LONG MEMORY, AGGREGATION, OR LEVEL SHIFTS?

Mehmet BALCILAR (Kırgız-Turkish Manas University)

A number of countries have experienced very long periods of inflation. It is argued that inflationary processes in these countries are determined by their own inflationary experience and in the absence of new shocks the inflation reproduces itself. This implies that time series of inflation rates are highly persistent. Turkey is one of the very typical among these countries with a very long period of high inflation experience since late 1970s. The inflationary process in Turkey is believed to be highly persistent.

Long memory models are very commonly used to embody highly persistent time series data. In this paper, we examine persistence in inflation rates using several CPI and WPI based inflation measures of the Turkish Economy. In the first part of the paper, using several parametric and nonparametric estimators of the long memory models we show that aggregate inflation rates show moderate to high persistence. The estimators used include Geweke and Porter?Hudak's log periodogram regression estimator, Robinson's Gaussian semiparametric estimator, Whittle's approximate maximum likelihood estimator, Sowell's exact maximum likelihood estimator, and a wavelet based estimator. ARIMA models are also estimated for comparison. For parametric models persistence is evaluated using the impulse response function of the model. Standard errors of the impulse response functions are obtained via bootstrap.

The second part of the paper examines sources of the long memory in inflation rates. Although the long memory in inflation rates may be inertial, level shifts and aggregation can also create apparent long memory. Aggregation over a large number of sectors each subject to white noise shocks may lead to long memory in aggregate inflation rates. In order to examine this possibility we estimate ARFIMA models for disaggregated inflation rates. Estimates show that disaggregated data display very weak memory. The hypothesis of no long memory cannot be rejected for sector specific inflation rates. Apparent long memory can also be caused by neglected occasional level shifts. The paper extends the STOPBREAK model of Engle and Smith (1999) to ARFIMA case and shows that when an ARFIMA model with occasional level shifts is estimated the long memory in inflation rates weakens further.