Common Biases in OECD and IMF Forecasts:Who Dares to be Different?

Heinz GlückOesterreichische Nationalbank

Stefan P. SchleicherUniversity of Graz

Project LINK Meeting in Mexico CityMay 16-20, 2005

Motivation for this paper

We are comparing forecasts and ex-post data revisions for G-7 countries

Real-time estimatesGDP estimates by OECD and IMF

Agnostic data analysis We do not investigate the reasons for differences

We investigate the hypothesis of common biasesCommon exogenous assumptionsInteraction of forecastersUncertainty decreases over time as forecasts evolve

Dated OECD estimates for USAOECD Estimates GDP USA

Estimation errors

-2,00

0,00

2,00

Year -

2 Sec.

Year -

1 Firs

tYear

-1 Sec.

Year 0

First

Year 0

Sec

.Yea

r +1 F

irst

Year +

1 Sec

.Yea

r +2 F

irst

Year +2 S

ec.La

st Rep

orted

Per

cent

poi

nts

diff

eren

ce

to la

st re

port

ed 1990

2003

1995

2000

Target yearBefore target year After target year

Dated IMF estimates for USAIMF Estimates GDP USA

Estimation errors

-2,00

0,00

2,00

Year -

2 Sec.

Year -

1 Firs

tYea

r -1 S

ec.Yea

r 0 Firs

tYea

r 0 Sec.

Year +

1 Firs

tYea

r +1 S

ec.

Year +

2 Firs

tYear

+2 Sec

.Las

t Rep

orted

Perc

ent p

oint

s di

ffer

ence

to

last

repo

rted 1990

2003

1995

2000

Target yearBefore target year After target year

Comparison of OECD / IMF estimates for USA

IMF Estimates GDP USAEstim ation errors

-2,00

0,00

2,00Yea

r -2 S

ec.

Year

-1 Fi

rstYe

ar -1

Sec.

Year 0

Firs

tYea

r 0 Sec

.Yea

r +1 F

irst

Year

+1 S

ec.

Year +

2 Firs

tYea

r +2 S

ec.

Last R

epor

ted

Per

cent

poi

nts

diff

eren

ce

to la

st r

epor

ted 1990

2003

1995

2000

Target yearBefore target year After target year

OECD Estimates GDP USAEstimation errors

-2,00

0,00

2,00

Year

-2 Se

c .Yea

r -1 Firs

tYea

r -1 S

ec.

Year 0

First

Year 0

Sec.

Year +

1 Firs

tYe

ar +1 S

ec.

Year +

2 Firs

tYea

r +2 S

ec.

Last R

eport

ed

Perc

ent p

oint

s di

ffere

nce

to la

st re

port

ed 1990

2003

1995

2000

Target yearBefore target year After target year

IMF Estimates GDP JapanEstimation errors

-2,00

0,00

2,00

Year -2

Sec.

Year -

1 Firs

tYea

r -1 Sec

.Yea

r 0 Firs

tYea

r 0 S

ec.

Year +1 F

irst

Year +

1 Sec.

Year +2 F

irst

Year +

2 Sec.

Last

Reporte

d

Perc

ent p

oint

s di

ffer

ence

to

last

repo

rted

1990

200319952000

Target yearBefore target year After target year

Comparison of OECD / IMF estimates for Japan

OECD Estimates GDP JapanEstimation errors

-2,00

0,00

2,00

Year -2 Sec.

Year -1 Firs

tYea

r -1 Sec.

Year 0 F irs

tYear 0

Sec

.Year +

1 First

Year +1 Sec.

Year +2 Firs

tYea

r +2 S

ec.Last R

eported

Perc

ent p

oint

s di

ffer

ence

to

last

repo

rted

1990

20031995

2000Target yearBefore target year After target year

We investigate two types of potential biases

Step 1: Common biases among OECD / IMF estimates for the G7

Evidence of strong common biases for OECD / IMF estimates for each G7 country

Step 2: Common biases among G7 country estimatesfor IMF

But only a few country groups exhibit common biases within the set of OECD / IMF estimates

This is a puzzling resultDespite strong common biased estimates between OECD and IMF estimates for each G7 country, this is in general not true among countries

Step 1:Common biases among OECD/IMF

We investigate common biases among OECD / IMF estimates countries for each G7 country

Methodology:Sequentially estimated correlations and biases with fading memory

Common biases among OECD/IMF USA

GDP USACorrelation and bias with last reported values

-1

0

1

1985 1990 1995 2000

Cor

rela

tion

-1

0

1

Bia

sPe

rcen

tage

poi

nts

OECD

IMF

Common biases among OECD/IMF Japan

GDP JapanCorrelation and bias with last reported values

-1

0

1

1985 1990 1995 2000

Cor

rela

tion

-1

0

1

Bia

sPe

rcen

tage

poi

ntsOECD

IMF

Common biases among OECD/IMF Germany

GDP GermanyCorrelation and bias with last reported values

-1

0

1

1985 1990 1995 2000

Cor

rela

tion

-1

0

1

Bia

sP

erce

ntag

e po

ints

OECD IMF

Common biases among OECD/IMF France

GDP FranceCorrelation and bias with last reported values

-1

0

1

1985 1990 1995 2000

Cor

rela

tion

-1

0

1

Bia

sPe

rcen

tage

poi

nts

OECD

IMF

Common biases among OECD/IMF Italy

GDP ItalyCorrelation and bias with last reported values

-1

0

1

1985 1990 1995 2000

Corr

elat

ion

-1

0

1

Bias

OECD

IMF

Common biases among OECD/IMF United Kingdom

GDP United KingdomCorrelation and bias with last reported values

-1

0

1

1985 1990 1995 2000

Corr

elat

ion

-1

0

1

Bias

Perc

enta

ge p

oint

s

OECD

IMF

Common biases among OECD/IMF Canada

GDP CanadaCorrelation and bias with last reported values

-1

0

1

1985 1990 1995 2000

Cor

rela

tion

-1

0

1

Bia

sPe

rcen

tage

poi

nts

OECD

IMF

Common biases among OECD/IMFConclusions

Very strong common biases between OECD and IMF estimatesOECD estimate is a good predictor for IMF estimate and vice versa

Accuracy of estimates varies over timeEstimates improve since 1990 (except USA)Biases always positive (underestimation)but become lower

OECD and IMF don’t dare to be different

Step 2:Common biases among countries

We investigate common biases among G7 countries within the set of IMF estimates

Methodology:Sequentially estimated covariances with fading memory

Common biases among countriesUSA / *

-1

0

1

1985 1990 1995 2000

Corr

elat

ion

USA/CAN

-1

0

1

1985 1990 1995 2000

Corr

elat

ion

USA/GB

-1

0

1

1985 1990 1995 2000

Corre

latio

n

USA/ITA

-1

0

1

1985 1990 1995 2000

Corre

latio

n

USA/FRA

-1

0

1

1985 1990 1995 2000

Cor

rela

tion

USA/DEU

-1

0

1

1985 1990 1995 2000

Cor

rela

tion

USA/JPN

Common biases among countriesJapan / *

-1

0

1

1985 1990 1995 2000

Cor

rela

tion

JPN/CAN

-1

0

1

1985 1990 1995 2000

Corre

latio

n

JPN/GBR

-1

0

1

1985 1990 1995 2000

Corr

elat

ion

JPN/ITA

-1

0

1

1985 1990 1995 2000

Cor

rela

tion

JPN/FRA

-1

0

1

1985 1990 1995 2000

Corr

elat

ion

JPN/DEU

Common biases among countries Germany / *

-1

0

1

1985 1990 1995 2000

Corr

elat

ion

DEU/CAN

-1

0

1

1985 1990 1995 2000

Corr

elat

ion

DEU/GB

-1

0

1

1985 1990 1995 2000

Corre

latio

n

DEU/ITA

-1

0

1

1985 1990 1995 2000

Corre

latio

n

DEU/FRA

Common biases among countries France / *

-1

0

1

1985 1990 1995 2000

Corr

elat

ion

FRA/CAN

-1

0

1

1985 1990 1995 2000

Cor

rela

tion

FRA/GB

-1

0

1

1985 1990 1995 2000

Cor

rela

tion

FRA/ITA

Common biases among countries Italy / *

-1

0

1

1985 1990 1995 2000

Corre

latio

n

ITA/CAN

-1

0

1

1985 1990 1995 2000

Corr

elat

ion

ITA/GBR

Common biases among countries United Kingdom / *

-1

0

1

1985 1990 1995 2000

Corre

latio

n

GBR/CA

Common biases among countriesConclusions

Only among a very few country groups we can observe common biases

USA – CanadaGermany – France – ItalyUnited Kingdom – Canada

Japan is “isolated”Decoupled from USA after 1990

This is a puzzling result

Despite significant biases in all OECD / IMF estimates only a few show common features

Final remarks

We only reported the results for the first forecast for the target year.

The estimates show systematic and significant biases

The estimates can be improved for policy analysis by correcting for these biases