Inflation persistence – a disaggregated approach

Inflation persistence and its role for efficient monetary policy has been given a lot of attention in empirical literature. The main contribution has been made by the Inflation Persistence Network, seeking to assess the level of persistence in the euro zone countries, the influence of the Monetary Union for the inflation persistence heterogeneity within countries and categories of products (see: Altissimo, Mojon, Zaffaroni, 2009). The research concerning New Member States started to appear later on and concentrates mostly on the comparison of the level of inflation persistence among NMS (or individual country analysis), eventually correlating it to the level of inflation (see eg. Franta, Saxa, Smidkova, 2010). Our contribution to the subject is the evaluation of persistence not only in the inflation aggregate in Poland, but also in the disaggregated components of Polish CPI in the aim of discerning sectors which contribute to the largest extent to persistence of headline inflation. We are also interested in assessing whether the adoption of direct inflation targeting in Poland (and the decrease in the targeted value of inflation) contributed to the decrease in persistence and, given the sustained high level of inflation in recent years in Poland, whether it was due to the local increase in the level of persistence (which could make the monetary policy reaction less effective) or to other factors, unrelated to the response of inflation to shocks. The measures of inflation persistence used in the exercise rely solely on time series methods (see: Marques, 2004, Pivetta, Reis, 2007 and Baillie, 1996) and are applied to Polish CPI and its 11 components (CPI ex. food and energy, food and non-alcoholic beverages, processed food, unprocessed food, energy, goods, services, CPI ex. administered prices, administered prices, administered energy, administered services, m-o-m, SA). We avoid structural methods such as New Keynesian Phillips Curve (Gali, Gertler, 1999), which, on one hand, provide information on the sources of persistence (intrinsic or inherited from the developments of economic activity), but on the other are highly dependent on estimating disaggregated output gap. In the first part of the research, to check if the inflation persistence in some sub-categories of CPI is not infinite, we conduct the unit root/stationarity tests. The results of the ADF/PP tests suggest that all of the series are stationary, but the KPSS test casts some uncertainty to those findings in the case of some series, leaving room for the fractional integration analysis, presented in the second part of the research. Meanwhile, still in the framework of linear univariate AR modelling, several persistence measures are calculated in the 9-year rolling windows for all of the series in the aim of assessing and comparing the inflation persistence level and dynamics in the defined sectors. The measures are as follows: sum of autoregressive coefficients, largest autoregressive root, half-life. In this context, the sectors with the highest persistence are core inflation and its components – goods and services as well as administered prices and processed food. There is also evidence of the decrease in persistence, the most apparent in the case of core inflation and its components as well as in the aggregate CPI. The only component where an increase in persistence can be observed are administered services. To check the robustness of the results above (the uncertain results of unit root/stationarity tests inter alia) and to verify the hypothesis that structural changes may have influence on the level of persistence in the context of changing monetary policy environment, the unit root test assuming a potential structural break has been carried on (Zivot-Andrews). The break date was defined endogenously, allowing us to confirm or reject our a priori statements about the possible impact of a shift in monetary policy strategy on the inflation developments. The significance of the break in mean was checked with the Chow stability test. The following results emerged: firstly, in average the level of persistence in the models with break in mean is lower than in models not accounting for the break. However, in case of some of the series the difference is negligible, especially in the second part of the sample. Secondly, in most cases the decrease in persistence is negligible. This is an argument confirming the hypothesis that higher level of inflation in the first part of the sample is contributing to artificially oversizing the persistence measured in the simple AR framework. The largest difference in comparison to the results without the break can be observed once again in the case of core inflation and its components (goods as well as services) and in administered prices. The highest level of persistence can now be stated in the processed food and administered services (in the second part of the sample), the lowest still in the energy sector (also in its administered component) and unprocessed food, which is a result often quoted in the literature (see eg. Bilke, 2005). The second part of the research refers to the discussion and to some empirical evidence (see Gadea, Mayoral, 2006) that inflation process may also be characterised by fractional integration. This assumption leads to a different set of persistence measures than the univariate AR analysis. Firstly, the value of integration (or “memory”) parameter bears information about the duration of a shock, its influence on the series level in the medium and long run and, as such, has been used in our research as one of persistence measures. Both Geweke-Porter-Hudak and local Whittle estimator have been used to evaluate the memory parameter. Secondly, the values of impulse-response function for 3 and 12 months, have been used to illustrate the differences between two alternative specifications and to compare different model specifications. The conclusions stemming from this analysis are the following: firstly, the estimates of the level of persistence confirm in general the results found in the univariate AR analysis. The highest medium and long term persistence can be observed in core inflation and its both components, as well as in the administered prices, the lowest – in the unprocessed food prices, food & beverages and energy (in the case of Whittle estimator). This finding confirms the usefulness of the core inflation aggregate (CPI ex. food and energy) as an inflation measure concentrating mostly on the most persistent CPI components (see eg. Walsh, 2011). Secondly, the rolling sample estimation showed that in case of almost all the series a decrease of level of integration parameter can be observed. However, its scale depends on the series and the estimator of d. The largest changes are detected in case of the overall CPI, CPI ex. administered prices and the prices of goods . The last step consisted of discriminating between the alternative specifications, AR including the break in mean and the fractional integration process, using appropriate statistical tests. In this purpose a test of I(d) vs. I(0) with structural breaks, proposed by Dolado, Gonzalo, Mayoral (2006), has been applied. In the case of series that did not display a statistically significant break in mean an appropriate version of the test above and the EFDF test (Dolado, Gonzalo, Mayoral, 2009) has been used to check whether the hypothesis of fractional integration is acceptable or not. It appears that fractional integration can be a better specification for the series of: CPI, food, energy, CPI ex. administered prices and administered services. Other series are better described as I(0) processes with break in mean (except from administered energy which seems to be an I(0) process) and as such can be modeled within the AR framework. See above See above