900 Macroeconometric Modeling for Policy
analysis there is a clear gain from using the congruent model, and avoiding basing
policy recommendations on a misspecified model of the supply side. In forecast-
ing, the link between model misspecification and forecast failure is not always as
straightforward as one would first believe. The complicating factor is again non-
stationarity, regime shifts and structural change. For example, a time series model
formulated in terms of the change in the rate of inflation adapts quickly to changes
in equilibria – a location shift – and is therefore robust to before-forecast structural
breaks of this type, even though it is clearly a misspecified model of the DGP
over the historical data period (see Clements and Hendry, 1999, Ch. 5). There-
fore a double-differenced device (DDD) can deliver near-unbiased forecasts when
a location shift has occurred prior to the preparation of the forecast. Conversely, a
forecasting model of the equilibrium correction type is less adaptable. Indeed, fol-
lowing an equilibrium shift, EqCM forecasts tend to move in the opposite direction
to the data, thereby causing forecast failure (cf. Hendry, 2006). Eitrheim, Husebø
and Nymoen (1999) have shown that this new theory of forecasting has practical
relevance for understanding the properties of forecasts from a medium-sized fore-
casting model of the Norwegian economy, in particular the more adaptive nature
of DDD to several historical examples of structural breaks.
The new theory of forecasting that we build on does not deliver carte blanche
for using non-congruent models for prediction, though. DDD and near-equivalent
forecasting devices are robust for one particular reason: they do not equilib-
rium correct, and are therefore insulated from changes in the parameters that are
most pernicious for forecasting. Replacing a congruent and adequate EqCM with
another, less adequate, EqCM model for forecasting is not a good idea. The non-
congruent EqCM is also subject to forecast failure on its own premises and, without
location shifts, it will forecast worse than the congruent EqCM. In this respect there
is a cost to compromising model adequacy also in forecasting. In terms of the con-
testing supply-side models of section 17.4.3, this is illustrated by Bårdsen, Jansen
and Nymoen (2002), who show that, although the PCM is robust to some of the
location shifts that can damage forecasting from the ICM, the cost of high forecast
variance and bias due to misspecified equilibrium correction dominates.
17.4.5.2 Real-time forecast performance
As mentioned above, NAM is part of the Normetrics system of models which was
initiated in 2005. Model-based forecasts for the Norwegian economy are produced
in January, March, June and September each year and published on the web. So
far, these model-based forecasts have performed relatively well compared to com-
peting forecasts. As an example, Figure 17.9 shows forecasts for core inflation
in 2006. Because of inflation targeting, this variable is among the most thor-
oughly analyzed variables by professional forecasters both in the private and public
sectors.
Figure 17.9 shows Norges Bank’s projections together with the average of other
professional Norwegian forecasters. The third line in the graph shows the sequence
of forecasts from the two Normetrics forecasting models, automatized inflation
forecasts (AIF) and NAM.^17 The figure shows that Normetrics forecasts were never