13
encompasses a given alternative out of these four candidates is rejected for all
forecast horizons at a 95% or 99% level. However, according to the encompassing
tests we can confidently consider the standard model to encompass that using the
sMom indicator.
[Table 4: about here]
So far, the inflation sentiment indicators we propose seem to be a useful forecast-
ing tool. However, the question is whether this is true for all periods in our sam-
ple. This question is addressed by the forecast breakdown test. Of course, for all
forecast horizons the fit is better in-sample than the out-of-sample. However, as
shown in Table 5, forecast breakdowns occur in rare cases only. Models featuring
inflation sentiment indicators are quite stable at commonly used significance lev-
els. Further calculations show that the forecasts based on the sentiment indicators
seem to be more stable than our benchmark, for which forecast breakdowns occur
more frequently.
[Table 5: about here]
4.2 Germany
As it can be seen in Figure 1, inflation rates in Germany are less volatile compared
to those of the US. This may be the main reason why the out-of-sample accuracy
of our benchmark Phillips curve is better in the German case. This holds for both
samples considered and for both specifications of the real side of the economy.10
Nevertheless, taking point estimates of relative RMSFE at face value, also for
West Germany most alternative Phillips curve specifications outperform our
benchmark in the period from 1985 to 1998 (Table 6). When GDP growth is cho-
sen as real economy variable, the alternative inflation indicators even improve
forecast accuracy for all forecast horizons and all information criteria. As an ex-
ample, consider a one-year-ahead-forecast: Here, the RMSFE can be reduced up
to one half when replacing inflation by the s Diff -indicator in the Phillips curve.
However, the accuracy gains are only significant for all indicators in the case of
three-quarter-ahead-forecasts, and in many cases when one-year-ahead predictions
are constructed.
[Table 6: about here]
As in the case of the US, the sMed -indicator and the s Diff -indicator apparently
tend to produce better forecasts than the sMom -indicator. Indicators which are
10 On the other hand, the low volatility in inflation rates makes it much harder to beat a na-
ïve forecast.