Economic Models - Convex Optimization

Inflation Control in Central and Eastern European Countries 187
test (further details about all the tests and a summary of the estimates are
in the tables at the end).
The inflation dynamics, on the other hand, was not stable when the
whole sample was taken into account. The instability, though was clearly
restricted to the first part of the sample, and could indeed be removed dropping
the years 1991 and 1992. It is interesting to observe that it is exactly
the period of the recovery of the inflationary shock due to the German
re-unification, a phenomenon that was usually regarded as temporary and
exceptional. Diagnostic tests confirmed that the equation estimated using
data from 1993 onwards was correctly specified and stable; notice anyway,
the distribution of the residuals did not appear to be normal, so the interpretation
of the other tests can only be justified asymptotically. Point estimates
of ρ 1 , ρ 2 , and ρ 3 were all very close to −1, suggesting a very high value of
α 4π , or a strong seasonal component in inflation; we also found that lagged
values of yt − 1 had a more significant effect, albeit even in this case the P-
value of the test H 0 : {αy = 0} vs. H 0 : {α y > 0} was between 5% and 10%.
Considering the model as a whole, anyway, we find these estimates
satisfactory, although not as much as those presented by Rudebush and
Svensson for the United States: we conjecture that the lower precision of
the estimates for Germany depended on the smaller number of observations
available.
3.2. Poland
Opposite to Germany, we consider Poland (and the other CEECs later)
as a country small enough to be affected by the international trade, so
we augmented the model with the real exchange rate depreciation with
respect to Germany and with the German output gap. Albeit our dataset
included 1993 too, we found that both the equations suffered from residual
autocorrelation, resulting in inconsistent estimates. We interpreted this as
evidence of instability, because when 1994 was taken as a starting point,
the residual autocorrelation was largely removed and the other diagnostic
tests were broadly compatible with a stable, correctly specified model. The
estimated model was
⎧
ŷ ⎪⎨ t = 0.020 + 0.467y t−1 − 0.0027y t−1
(0.008) (0.142) (0.001)
⎪⎩
̂π t =−0.64π t−1
(0.12)
− 0.67π t−2
(0.101)
− 0.51π t−3 + 21.46(q t−1 − q t−5 )
(0.10)
(7.88)
over the period 1994Q1–2004Q1 for the AD, and 1994Q1–2004Q1 for
the PC.