Palgrave Handbook of Econometrics: Applied Econometrics

930 Monetary Policy, Beliefs, Unemployment and Inflation

as in Nickell and Bell (1995) and Nickell (1998); and estimates of a cointegrating
equation using Johansen’s maximum likelihood (ML) method, as in Nickell and
Bell (1995). Henry and Kirby argue that neither satisfactorily explains long-run
trends in unemployment in the UK as its proponents claim. Reasons for this are
summarised below. First, however, ARDL estimates of the long-run equation are
shown in Table 18.1.^21 For this test, the full sample period is 1964Q4 to 1992Q4 as
in Nickell (1998). The variables are described in the appendix to this chapter: full
definitions are found in Nickell and Bell (1995) and Henry and Kirby (2007).

Table 18.1 ARDL estimates of long-run unemployment

equations

Const. IT RR∗ TT Skill UP T r

1964Q4–1992Q4

−35.5 0.11 0.049 9.99 0.09 1.90 0.035 0.021

∗∗ (0.6) (1.4) (2.2) (3.0) (2.0) (2.7) (2.1)

1964Q4–1992Q4

−45.5 0.17 2.19 11.2 0.06 1.9 0.039 0.01

(4.0) (1.2) (1.5) (2.89) (2.1) (1.82) (2.7) (1.5)

1964Q4–1989Q4

−71.9 0.34 2.5 6.9 0.05 −0.6 0.067 0.01

(4.5) (2.4) (1.87) (1.6) (2.2) (0.5) (3.55) (1.2)

1964Q4–1984Q4

−65.1 0.14 0.4 −13.1 0.08 2.0 0.069 0.02

(5.6) (1.6) (0.4) (1.8) (2.6) (2.2) (4.5) (2.6)

Notes: Each uses a maximum lag of 4 on each variable.t-statistics

are in (.).

∗Nickell and Bell (1995) report estimates of the parameter for this

variable of 8.95 and 4.88 for the ARDL and Johansen estimates

respectively and appear to use the log of the replacement ratio.

The estimate above from Nickell (1998) is apparently its level. We

use the former throughout.

∗∗Not-statistic is given in the original.

The first equation is taken from Nickell (1998). It is clear that the remaining
estimates reveal substantial variation in the estimated parameters and, in crucial
cases, a complete turnaround in the significance of the estimates. This feature of
parameter instability confirms what a number of critics have pointed out, namely
that such wage-pressure variables are unlikely to account for long-run movements
in unemployment since they were no worse in the mid 1980s than in the 1960s,
yet unemployment still rose.
On a more methodological note, the use of ARDL to estimate long-run equations
is not generally acceptable as it requires that the right-hand-side variables do not