Improving Business Cycle Forecasts’ Accuracy - What Can We Learn from Past Errors?

Improving Business Cycle Forecasts’ Accuracy

formational efficiency of forecasts of the expenditure side components of
GDP.

The paper is organised as follows: In section 2, the concept of informational ef-
ficiency of forecasts is discussed and the method employed in this paper is de-
scribed. Section 3 provides some details about the forecasts considered and
gives some measures of their accuracy. Furthermore the short term indicators
are described. In Section 4, the results of the tests for information accuracy are
presented. The final section offers some conclusions.

2. Methodology

2.1 Concepts of Informational Efficiency

Several measures to evaluate forecast efficiency are proposed in literature. A
simple test is the so called Mincer-Zarnowitz equation, which regresses the re-
alized rate of growth r of a variable on the predicted change f of the same
variable

(1)                               r_t =α 0 +α ₁ ∙ f_t +ε _t.

ε _t is an error term which is assumed to be normally distributed with mean zero.
In an efficient forecast (“Mincer-Zarnowitz-efficiency”), α₀ should be equal
to 0 andα₁ equal to 1, which can be tested by an F-test. In case α₁ is above 1,
the forecaster is “timid”, i.e. he underestimates high and overestimates low
growth rates: The opposite applies if α₁ is lower than 1. If α₀ differs from 0
while α₁ equals 1, the forecast is biased.

For a stronger test for informational efficiency, various extensions of the
Mincer-Zarnowitz equation are proposed. One approach is to enclose r_t-1 as
an additional explanatory variable

(2)                        r=α+α₁∙f+α∙r₁+ε.

t01t2t-1t

Ifα₂ differs from zero significantly, the forecast tends to covariate with the last
observation. If α₂ is negative, the forecast is systematically too optimistic in
years following a “good” year and over-pessimistic in those following “bad”
years. In the caseα₂ is positive the opposite applies. In a more direct test r_t-1 is
replaced by other data x_t which have been known by the forecaster when
making his prediction.

(3)                           r_t=α₀+α₁∙f_t+α₂∙x_i,t-1+ε_t.

Whether the factors included in (2) and (3) do really improve a forecast can be
tested in different ways. Stekler (2002: 224) proposes the null hypothesis

<<   4   5   6   7   8   9   10   >>

More intriguing information