uct categories enumerated in the Classification of Individual Consumption by
Purpose (COICOP). The COICOP serves as a guideline for the disaggregation of
private consumption expenditures in the national accounts.1 For some product
categories (e.g. food) prices for quite a number of goods are required to represent
the heterogeneous basket. For other categories, e.g. electricity or heating gas, a
small number of prices suffices to characterize price changes. We argue that, if the
prices of many products (in a category) rise, consumers will perceive inflationary
trends more intensely.
We use simple transformations of the data taken from the CPI statistics to con-
struct several indicators of inflation sentiment. First, we calculate the unweighted
median. By comparing it to the weighted mean of the price changes of the indi-
vidual goods and services (the current CPI), we get an impression of the skewness
of their distribution. If the median is larger than the weighted mean, the overall
price trend reflects a relatively large number of similar price changes for individ-
ual products. In that situation we would expect consumers to perceive inflation
more strongly. Secondly, we test a diffusion index measuring the share of prices
which grow faster than current CPI. Finally, a momentum index is defined as the
difference between the share of prices which grow faster than in the previous
period, and the share of prices which grow more slowly. These indicators are
calculated for quarterly data for the US and Germany.
In analysing the forecasting power of these indicators we follow the technique
proposed in their seminal paper by Stock and Watson (1999). Specifically, we
estimate several variants of the Phillips curve in which our indicators and, as in
the standard procedure, the CPI itself are each used as the relevant price term,
respectively. Starting with an initial sample length we employ all indicators for an
out-of-sample forecast up to a maximum horizon of two years. Then, the sample is
expanded by one quarter and another set of forecasts is made. By continuing this
procedure, we generate a series of out-of-sample forecasts which can then be used
to evaluate the forecasting power of all candidate approaches.
The remainder of this paper is organized as follows: In the next section we discuss
the data, and the indicators that are used to characterize the inflation sentiment.
Section 3 presents the econometric approach to forecast inflation and the methods
which are employed to evaluate the estimates. In the fourth section we discuss our
results. Section 5 concludes.
See e.g. http://unstats.un.org/unsd/cr/registry/regcst.asp?Cl=5&Top=2&Lg=1.