4.4 Robustness checks
To evaluate the robustness of our results, we estimated several alternative versions
of our model. First, we changed the Cholesky ordering of the variables and, ad-
ditionally, used generalized impulse responses.18 For instance, the interest rate is
often ordered behind the money variable in similar VAR models, so that we also
tried this option with nearly no consequences for the results. The same is true for
generalized impulse response analysis. Second, additional variables were added to
the model, namely a commodity price index (like already mentioned earlier), the oil
price (as an alternative for the commodity price index) and a long-term interest rate
(specified by 10-year government bond yields). Both former variables were involved
in only very few significant impulse responses with the most interesting of them
being a short-term rise of the interest rate to a commodity price shock. The other
findings of our model again proved to be stable. As the commodity price index and
the oil price did not solve the ”price puzzle” and did not show significant effects on
the price level, we dropped them in the analysis illustrated above not least in order
to save degrees of freedom.
The long-term interest rate was added as a substitute for the short-term rate and
as a complement of our system as well. In the former case, results were very similar
to the use of the short-term rate. In particular, no evidence was found that global
liquidity fuels bond markets. When using both rates signs of duplications were
found. For instance, shocks to both rates caused a decline of the GDP deflator and
the house price index. Notwithstanding the fact that the long-term interest rate
might contain additional information, the relationship to the short-term interest
rate seems to be close enough such that the more parsimonious model may be more
adequate in order to diminish overparameterization. As a third aspect, different
lag lengths were used. Particularly, 4 lags were tried, but no contradicting results
occurred.
18See Pesaran and Shin (1998) for theoretical derivations of generalized impulse response analysis.
23