2. Measuring inflation sentiment
In our empirical work we rely on three time series of quarterly inflation rates
(Figure 1) that we construct as simple averages of seasonally adjusted monthly
data. Our series for the US covers the period 1978:1 to 2006:4. It uses the CPI-U
research data that were calculated by the Bureau of Labour Statistics to account
for methodological changes. The number of goods and services included in this
data set varies from 60 at the beginning to 212 at the end of the sample period.
[Figure 1: about here]
For Germany, we investigate two CPI series. The first covers the years 1985 to
1998 and relates to West Germany, the second provides data for unified Germany
for the years 1993 to 20072. These periods comprise quite different inflationary
regimes. The first sub-period is characterized by large fluctuations in inflation
rates; they stick very closely to the inflation objective in the more recent sub-
period. The German CPI covers about 750 single items, which are aggregated in
several steps to gain inflation rates for product categories.
[Table 1: about here]
To be able to derive comparable results for Germany and the US, we utilize these
data on a four digit COICOP-level, which allows us to include about 100 product
categories in our indices (Table 1). We employ three indicators to measure infla-
tion sentiment.3 Firstly, for each quarter t we construct the difference, stMed , be-
tween (unweighted) median inflation, πtMed , and the headline inflation rate, πt .
stMed =πtMed-πt. (2.1)
This indicator measures the skewness of the distribution of the price increases of
the individual CPI components. When a high proportion of price components is
rising, the unweighted median tends to be above the weighted average. In that
case, we expect consumers and price setters to perceive inflation more strongly.
The second indicator of inflation sentiment, stDiff , captures the difference between
the share of prices which grow faster than current CPI, and the share of prices
2 Price data for unified Germany are available since 1991. However, we omitted the years 1991
and 1992 because they were strongly influenced by irregularities in the follow-up of unification.
3 Note that all inflation sentiment indicators should be stationary by construction. However,
augmented Dickey-Fuller tests for the inflation sentiment indicators partly suggest that these
variables may not be stationary and, therefore, should be included in differences. We include
them in levels for two further reasons: Augmented Dickey-Fuller tests are not valid in small
samples under 100 observations. Moreover, KPSS tests, which can also be found in the Appen-
dix, largely confirm the stationarity assumption.