Gunnar Bårdsen and Ragnar Nymoen 855shocks to the economic system, as implied by (4) above.^1 However, structure is
partial in two respects. First, autonomy is a relative concept, since an econometric
model cannot be invariant to every imaginable shock. Second, all parameters of an
econometric model are unlikely to be equally invariant, and only the parameters
with the highest degree of autonomy represent structure. Since elements of struc-
ture typically will be grafted into equations that also contain parameters with a
lower degree of autonomy, forecast breakdown may frequently be caused by shifts
in these non-structural parameters.^2
A strategy for model evaluation that puts emphasis on forecast behavior, without
a careful evaluation of the causes of forecast failureex post, runs a risk of discarding
models that actually contain important elements of structure. Hence, e.g., Doornik
and Hendry (1997) and Clements and Hendry (1999, Ch. 3) show that the main
source of forecast failure is location shifts (shifts in means of levels, changes, etc.),
and not shifts in such coefficients that are of primary concern in policy analysis, i.e.,
the derivative coefficients of behavioral equations. Therefore, a rough spell in terms
of forecasting performance does not, by itself, disqualify the model’s relevance for
policy analysis. If the cause of the forecast failure is location shifts, they can be
attenuatedex postby intercept correction or additional differencing “within” the
model (Hendry, 2004). With these add-ons, and once the break period is in the
information set, the model forecast will adapt to the new regime and improve
again. Failure to adapt to the new regime may then be a sign of a deeper source
of forecast failure, in the form of non-constant derivative coefficients, which also
undermines the models relevance for policy analysis.^3 In general, without adap-
tive measures, models with high structural content will lose regularly to simple
forecasting rules (see, e.g., Clements and Hendry, 1999; Eitrheim, Husebø and
Nymoen, 1999). Hence different models may be optimal for forecasting and for
policy analysis, which fits well with the often heard recommendations of a suite
of monetary policy models.
Structural breaks are always a main concern in econometric modeling but, like
any hypothesis or theory, the only way to judge the significance of a hypothesized
break is by confrontation with the evidence in the data. Moreover, given that an
encompassing approach is followed, a forecast failure is not only destructive but
represents potential for improvement, if successful respecification follows in its
wake (cf. Eitrheim, Jansen and Nymoen, 2002). In the same vein, one important
intellectual rationale for DSGE models is the Lucas critique. If the Lucas critique
holds, any “reduced-form” equation in a model is liable to be unstable over the
historical sample, due to regime shifts and policy changes that have taken place
in the economy. Hence, according to the Lucas critique, parameter instability may
be endemic in any model that fails to obey the rational expectations hypothesis
(REH), with the possible consequence that without integration of the REH, the
model is unsuited for policy analysis. However, as stated by Ericsson and Irons
(1995), the Lucas critique is only a possibility theorem, not a truism, and the
implications of the Lucas critique can be tested (see also, e.g., Hendry, 1988; Engle
and Hendry, 1993; Ericsson and Hendry, 1999). In Bårdsen, Jansen and Nymoen