Can We Improve the Perceived Quality of Economic Forecasts

SUMMARY
A number of topics are discussed concerning how economic forecasts can be improved in quality or at
least in presentation. These include the following: using 50% uncertainty intervals rather than 95%; noting
that even though forecasters use many different techniques, they are all occasionally incorrect in the same
direction; that there is a tendency to underestimate changes; that some expectations and recently available
data are used insufficiently; lagged forecasts errors can help compensate for structural breaks; series that
are more forecastable could be emphasized and that present methods of evaluating forecasts do not capture
the useful properties of some methods compared to alternatives.
'Economic predictions are fallible and advice is highly variable',
Resident Jimmy Carter, Camp David Summit, Summer 1979,
quoted by Klein (1981)
1. CRITICISMS OF FORECASTS
It is easy to find criticisms of economic forecasts, both of their perceived quality and of the
methods used in their construction. Some of the criticism is from academics, such as by Keating
(1985), and more appear in the business press. Examples of the latter are Robert Chote (New
Scientist, 31 October 1992), 'Why the chancellor is always wrong', with sub-head 'No one takes
the governments forecasts of economic upturns seriously. The problem lies with the Treasury's
computer model', and by Robert Samuelson (Newsweek, 13 February 1995), 'Soothsayers on
the decline', 'Economists know less than they-or most Americans-think.' A great deal of
this criticism is probably deserved. Much of the purely academic writing on forecasts, for
example, seems to be simplistic, inflexible, too conservative and insufficiently concerned with
presentation. The initial objectives of this paper are to ask if some of the points made by the
critics are helpful, if a substantial rethinking of standard practices is worth attempting and if this
will produce a noticeable improvement in forecasts.
No forecast can be properly evaluated in isolation and so it is worth noting that 1995 is the
silver anniversary of the publication of the famous book by George Box and Gwilym Jenkins
(1970) whose univariate models, at least, provide substantial opponents in any forecasting
competition. This paper is dedicated to their path-breaking work.
So that attention is paid to realistic situations, it is worth starting by looking at some actual
forecasts. The 3 December 1994 issue of The Economist included British Treasury forecasts for
1995 and 1996 (first half) as shown in Table I.