934 Monetary Policy, Beliefs, Unemployment and Inflation
are measures of international price competition (COM) and the real price of imports
(RPM) (see Balakrishnan and Lopez-Salido, 2002; Herzberg, Kapetanios and Price,
2003; Batini, Jackson and Nickell, 2005, amongst others). Other, moread hocaddi-
tions which have figured in the empirical literature are also included, such as the
real oil price (OIL) (see Batini, Jackson and Nickell, 2005; Stock and Watson, 2002;
Henry and Nixon, 2000; Boivin and Giannini, 2003). The real exchange rate (RXR)
has also been used in the wage equation on the grounds of capturing real wage resis-
tance effects (Nickell, 1988), as well as being a way of extending the standard model
to allow for internal and external balance (see Layard, Nickell and Jackman, 2005).
Equating the real wage and markup again leads to a long-run unemployment
equation, this time of the form:
ut=β 1 +β 2 COMt+β 3 RPMt+β 4 z ̃wt+β 5 RXRt+β 6 OILt, (18.24)wherez ̃wtis now taken to be the price and tax wedge variable (T).^27
18.4.2.3 Cointegration results for unemployment
The strategy adopted here is to take the set of possible determinants of unemploy-
ment from equation (18.24) and derive a parsimonious version of the long-run
unemployment equation using cointegration methods.^28 From earlier findings it
appears that there is only one variable (T)which survives as a potential determi-
nant of long-run unemployment from the wage setting side and, in the light of
this, the rest of this section considers the empirical case for the use of variables
from the price-setting side reviewed above.
The complete set of variables used initially is given in (18.24). In this set there
is probably one cointegrating vector. Tests forr ≤1 give 47.4 (40.53) for the
Johansen eigenvalue test (λ)and 120.1 (102.6) for the Johansen trace test (95%
significance levels in brackets). The tests reject the hypothesis thatr≤2 with
λ=26.3(34.4)andTrace=72.7(76.0). Even so, it remains likely that sub-sets
of these six variables also cointegrate, and we build on this idea in following the
approach suggested in Davidson (1998) of selecting minimal sets of non-stationary
variables as contenders for the long-run equation. This explores the results of drop-
ping each variable (except unemployment) from the cointegrating vector, and the
results can be summarized as follows. It is found that a minimal set comprising
the two variablesCOMandOIL, together with unemployment, form a cointegrat-
ing vector (forr≤1;λ=40.3(22.0)andTrace=51.4(34.9)). Other contending
sub-sets of the six either do not cointegrate or have inappropriate signs and other
theoretical shortcomings (further details are given in Henry, Kirby and Riley, 2007).
The minimal cointegrating set selected is then:
ut=0.76+0.85OILt+10.7COMt. (18.25)(Tests for orders of integration of these variables are given in Table 18.4.) Although
this is simple by design, this long-run relation enters significantly in the error
correction model for changes in unemployment based on the just-identified VECM,
with a negative loading factor of 0.53 and at-statistic of 4.6. Tests also show that all