Because of the previously mentioned advantages, the exercise is conducted
only on HD SVM and not on HD SO. The results are shown in Table 7. The
coefficients of the estimated Taylor equations are significant for almost all vari-
ables with the exception of inflation expectations which is only significant (at
the 10%-level) in the standard Taylor rule equation for the LIBOR-rate. When
HD is included, inflation expectations suffer from a further loss of significance in
many equations, in particular for the MRO rate, while HD terms tend to be sig-
nificant for all specifications. The signs of the estimated coefficients are positive,
indicating that a rise in both economic slack and inflation expectations tend to
be associated with rises in interest rates, which is consistent with the literature.
As said, the lagged HD-index is statistically significant in all equations. Fur-
thermore, a rise in R
2
is observed generally, indicating an improvement in the fit
for all specifications. Such evidence tends to be consistent with similar results
in the literature, in the sense that the HD-index is able to capture changes in
the tone of communication anticipating movements in key interest rates. This
is also moderately confirmed by the correlation between the residuals from the
Taylor rule (with no lagged HD-index term) and the HD-index, as shown in
Figure 12. For the overall HD index, correlations show a peak of close to 0.4 for
the MRO rate and slightly above 0.3 for the Libor rate, in both cases at zero
lags. Correlations tend to be slightly higher for the HD index computed after
the press conference, though with a broadly unchanged pattern.
While the Taylor rule formulation as presented above is generally used in
literature and in practise (see e.g. Lucca & Trebbi (2009); Sturm & De Haan
(2011), non-stationarity of the variables could lead to spurious regressions and
inconsistent parameter estimates
¨
Osterholm (2005). An Augmented Dickey
Fuller test shows that all variables used in the Taylor rule are integrated of
order one. All pairwise Johansen cointegration tests between the independent
variables and MRO/LIBOR accept the hypothesis of no cointegration. To en-
sure that the previously found significant relationship between communication
and the future path of the interest rate is valid, we supplement our Taylor rule
exercise with a set of Taylor rule regressions conducted using the first differences
of all the included variables. The results are reported in Table 8. When taking
first differences, the coefficient of the inflation expectations becomes significant
for all equations. The coefficients of inflation expectations and economic slack
remain positive, indicating that increasing output gap and increasing inflation
expectations are associated with rises in the interest rates. R
2
-values are lower
than those reported in Table 7, especially for the MRO rate, yet a large pro-
portion of the variance can still be explained by the models and model fit still
increases when the HD-indicator is introduced in the equations. Both the over-
all ∆HD SVM and ∆HD SVM after the press conference are significant at the
5%-level for MRO and LIBOR.
Overall, these results tend to confirm the conclusion that communication
plays a significant role in characterizing the ECB’s reaction function. Further-
more, given how the HD-index is constructed, the empirical evidence tends to
ECB Working Paper 2085, July 2017