AIR-Automatized Inflation Reports
15 February 2007, Ragnar Nymoen
An evaluation of AIR forecast for Norwegian core
inflation for the period 2004q2-2006q4.
Automatized econometric inflation forecasts have been published twize a year, starting in
July 2004. The forecasts are automatized, with a minimum of intervention after the
econometric specification of the forecasting mechanism is completed. Therefore these
econometrically based forecasts have been dubbed Automatized Inflation Reports
forecasts (AIR for short).
Nymoen (2005) gives an evaluation of how the forecasting model underlying AIR
performs relative to Norges Bank’s inflation forecasts in 2002 and 2003. Here, we
evaluate the true ex-ante forecasts based on forecast errors for the period 2004q2-2006q4.
The forecasted variable is the annual rate of inflation.
The graphs in the upper panel in the figure show the mean forecast errors, MFE.
A negative MFE means that the inflation forecasts are on average higher than the actual
inflation rates in the period. The biases of the AIR forecasts are small for forecast
horizons 1, 2, 3 and 4: the MFEs are less than 0.25 percentage points. Norges Bank’s IR
forecasts are markedly biased over these horizons. The 4-quarter-ahead forecasted rates
of inflation in the Inflation Reports are on average 0.8 percentage points higher than
actual rates inflation.
0.0
Econometric AIR forecasts
MFE
-0.5
-1.0
Norges Bank's Inflation Reports
1
2
3
4
5
6
7
8
9
10
11
10
11
Length of forecast horizon (quarters)
2.5
2.0
1.5
Norges Bank's Inflation Reports
1.0
Econometric AIR forecasts
0.5
1
2
3
4
5
6
Length of forecast horizon (quarters)
7
8
9
Figure 1: First panel: Mean forecast errors, MFE, for core inflation (annual rate of change in CPIATE). Inflation report forecast in red, and AIR forecast in blue. Second panel: Mean squared
forecast errors, MSFE. Source: Inflation Reports 2/04 - 3/06 and AIRs published at
http://folk.uio.no/rnymoen/forecast_air_index.html
1
AIR-Automatized Inflation Reports
15 February 2007, Ragnar Nymoen
The biases of the AIR forecasts become markedly bigger for forecasts of length 5 and 6
quarters, but they are still less than 0.5 percentage points higher compared to the actual
annual rates of inflation. The graph shows forecast biases for all forecast horizons form
one quarters to 10 quarters i.e. one and a half year. The biases in Norges Bank’s forecasts
are consistently larger than the AIR biases, with the exception of the 1 quarter horizon
where the two biases are almost identical.
The second panel of the figure shows the mean squared forecast errors, MSFE, to
which large forecast errors contribute relatively more than small errors. Based on this
measure, the AIR is practically as good as the IR for one quarter forecasts. For the other
horizons, the AIR forecasts are better, so the picture is more or less the same as in the
first panel.
This evaluation of AIR forecasts will be updated in April 2007.
Previous evaluation of AIR: November 2006.
2