14
based on one of the two core inflation measures improve forecast accuracy, too,
but (in terms of point estimates) they seem to be inferior to the sMed - and the
s Diff -indicators in most cases. Table 6 also documents that, issues of statistical
significance aside, all indicators perform better in combination with GDP growth
than with changes in unemployment.
The second sub-period relates to unified Germany. We should not be too sur-
prised, if results turn out to be different. On the one hand, it is quite an open ques-
tion, whether the inflation sentiment among East Germans and those among West
Germans, respectively, coincide, and what might happen as they become entan-
gled with one another. On the other hand, much of the sampling period was rela-
tively stable with respect to inflation and, thus, intuitively it should be difficult to
outperform the standard Phillips curve approach. Indeed, none of our candidate
forecast models is performing significantly better than the standard model. By
contrast, for several forecast horizons some of the candidates, in particular that
employing sMom perform significantly worse. This holds especially for models
involving the unemployment rate as a predictor.
Again, we also employ encompassing tests to assess whether our candidate vari-
ables make a contribution to improved inflation forecasts. Since according to our
point estimates, all indicators perform better than the inflation rate in the first sub-
sample, the tests might state that these variables add significant information. Table
7 documents that the null hypothesis is rejected for a large share of the candidates
at all forecast horizons. The associated significance levels are particularly high
with regard to three-quarter-ahead forecasts. Less uniform are the results for the
more recent sub-period, in which inflation sentiment indicators do not outperform
the standard Phillips curve approach in most cases.
Yet, the encompassing tests show that in some cases inflation sentiment indicators
can add forecast-relevant information, in particular the sMed - and the s Diff -
indicator. Again, the alternative indicators lose their appeal when they are com-
bined with unemployment instead of GDP growth. In line with the forecast results,
the sMom -indicator and the core measures do not add significant forecast-relevant
information to the inflation rate, the combination of the str20 -indicator and GDP
being an exception. All in all, also in periods where inflation is quite stable, sim-
ple inflation sentiment indicators might improve forecast accuracy, whereas this
does not seem to be the case for familiar core measures of inflation.
[Table 7: about here]
Since the inflationary regime in Germany seems to differ between the (partly
overlapping) sub-periods, the forecast breakdown can be expected to provide