Session 12: Core inflation measurement methods UNECE Workshop on Consumer Price Indices Istanbul, Turkey,10-13 October 2011 Session 12: Core inflation measurement methods Presentation by Carsten Boldsen, UNECE
Overview What is core inflation Exclusion based methods Trend estimates Limited influence estimators Conclusion
What is core inflation? No general accepted definition of core inflation! It is usually associated with the idea of an underlying or long-term inflation, which is not disturbed by seasonal, erratic or one-time price variations. Normally used for inflation targeting by the monetary authorities High credibility of the measure is required
t : inflation measured by the CPI t* : core inflation What is core inflation? t = t* + vt t : inflation measured by the CPI t* : core inflation vt : temporary disturbances All information about t * should be utilised; only the noise (vt) should be excluded It is difficult to distinguish signals and noise Elements with noise may have a different trend than average, so that exclusion will bias the core inflation indicator
Different methods Exclusion based methods - Products - Indirect taxes - One-off shocks - Domestically generated inflation - Imputation methods Trend estimates Limited influence estimators - The weighted median - trimmed means
Exclusion based methods Exclusion of products Food and energy usually excluded Products can be excluded in two ways: Exclude always the same products Exclude products which are the most volatile Documentation of methods before publication is needed to ensure credibility
Exclusion based methods Exclusion of indirect taxes Exclude tax changes, or Exclude all indirect taxes – calculate a net price index One-off Schocks Exclude extreme price changes caused by exogenous schocks Domestically generated inflation Exclude imported products
Exclusion based methods Imputation methods Exclude volatile products and impute their price change by products likely to experience similar price development The idea is to keep the signal and exclude the noise
Trend estimates A commonly used multiplicative model is: CPIt = Tt * St * It Tt : Trend component St : Seasonal component It : Irregular component The trend can be estimated by moving average, for example 24 months (but latest period is always missing) The seasonal component can be estimated in regression models Both trend and seasonal component included in most seasonal adjustment programs
Limited influence estimators Weighted median Calculate the weighted median, which is less responsive to extreme price changes. This is a simple and transparent method Trimmed means: Exclude specified upper and lower tails of the distribution of the 12-months price changes. For example, exclude 10% of the weights at the top and the bottom In trimmed means indices large price changes are excluded, but smaller re-adjustments are included!
Some conclusions All methods have their advantages and disadvantages Some methods may work well in some period of time, but not in others The role of the SNO should be clear and communicated to the public Methods and sources should be transparent and documented and made publicly available