WO R K I N G PA P E R S E R I E S
N O 1 3 4 9 / J U N E 2 011
CONSUMER
CONFIDENCE AS
A PREDICTOR OF
CONSUMPTION
SPENDING
EVIDENCE FOR THE
UNITED STATES AND
THE EURO AREA
by Stéphane Dées
and Pedro Soares Brinca
WO R K I N G PA P E R S E R I E S
N O 13 49 / J U N E 2011
CONSUMER CONFIDENCE AS
A PREDICTOR OF CONSUMPTION
SPENDING: EVIDENCE FOR
THE UNITED STATES
AND THE EURO AREA 1
by Stéphane Dées 2
and Pedro Soares Brinca 3
NOTE: This Working Paper should not be reported as representing
the views of the European Central Bank (ECB).
The views expressed are those of the authors
and do not necessarily reflect those of the ECB.
In 2011 all ECB
publications
feature a motif
taken from
the €100 banknote.
This paper can be downloaded without charge from http://www.ecb.europa.eu or from the Social Science
Research Network electronic library at http://ssrn.com/abstract_id=1852208.
1 We are grateful to Annika Alexius, Robert Anderton and Filippo di Mauro for helpful comments. Any views expressed
represent those of the authors and not necessarily those of the European Central Bank or the Eurosystem.
2 Corresponding author: European Central Bank, Kaiserstrasse 29, D-60311 Frankfurt am Main,
Germany; phone: (+49) (0)69 1344 8784; e-mail: [email protected]
3 Stockholm University, Sweden.
© European Central Bank, 2011
Address
Kaiserstrasse 29
60311 Frankfurt am Main, Germany
Postal address
Postfach 16 03 19
60066 Frankfurt am Main, Germany
Telephone
+49 69 1344 0
Internet
http://www.ecb.europa.eu
Fax
+49 69 1344 6000
All rights reserved.
Any reproduction, publication and
reprint in the form of a different
publication, whether printed or produced
electronically, in whole or in part, is
permitted only with the explicit written
authorisation of the ECB or the authors.
Information on all of the papers published
in the ECB Working Paper Series can be
found on the ECB’s website, http://www.
ecb.europa.eu/pub/scientific/wps/date/
html/index.en.html
ISSN 1725-2806 (online)
CONTENTS
Abstract
4
Non technical summary
5
1 Introduction
7
2 Data
10
3 Do confidence indicators bring additional
information beyond economic fundamantals?
3.1 Unit root and Granger causality tests
3.2 Estimation of a simple model
for consumption
3.3 VAR analysis
12
12
13
15
4 Isolating periods in which confidence matters
4.1 A threshold model
4.2 In-sample evidence
4.3 Out-of-sample evidence
19
20
21
23
5 Conclusions
25
References
26
Tables and charts
28
ECB
Working Paper Series No 1349
June 2011
3
Abstract
For most academics and policy makers, the depth of the 2007-09 …nancial crisis, its longevity
and its impacts on the real economy resulted from an erosion of con…dence. This paper proposes
to assess empirically the link between consumer sentiment and consumption expenditures for
the United States and the euro area. It shows under which circumstances con…dence indicators
can be a good predictor of household consumption even after controlling for information in
economic fundamentals. Overall, the results show that the consumer con…dence index can be in
certain circumstances a good predictor of consumption. In particular, out-of-sample evidence
shows that the contribution of con…dence in explaining consumption expenditures increases
when household survey indicators feature large changes, so that con…dence indicators can have
some increasing predictive power during such episodes. Moreover, there is some evidence of a
“con…dence channel”in the international transmission of shocks, as U.S. con…dence indices lead
consumer sentiment in the euro area.
Keywords: Consumer Con…dence, Consumption, International Linkages, Non-linear modeling.
JEL Classi…cation : C32, E17, F37, F42.
4
ECB
Working Paper Series No 1349
June 2011
Non-Technical Summary
The 2007-09 …nancial crisis led to the most severe global economic recession
since the Great Depression. Most academics and policy makers agree to say
that it is an erosion of con…dence that has ensured the depth and the longevity
of the crisis, especially as regards its impacts on the real economy.
The link between con…dence and economic decisions has been widely covered in the literature, which has focused on two main aspects. First, from a
theoretical viewpoint, the literature has concentrated on the conceptualisation
of con…dence and its role in modern theories of consumption. In particular, to
explain deviations from the Permanent Income Hypothesis, con…dence has been
justi…ed for "precautionary savings" arguments in the context of perceived uncertainty by households. Second, for an empirical viewpoint, the literature has
been concerned on whether the con…dence indicators contain any information
beyond economic fundamentals. The concern is whether con…dence can be explained by current and past values of variables such as income, unemployment,
in‡ation or consumption or, in other way, whether con…dence measures have any
statistical signi…cance in predicting economic outcomes once information from
the variables cited above is used. While the evidence is overall rather mixed,
most authors seem to, at least, …nd a signi…cant statistical relationship between
con…dence measures and economic variables, current and future. In particular,
some stress the special importance of con…dence indicators in predicting periods
of strong ‡uctuations in the economy, such as recessions and recoveries or during
periods of major economic or political shocks. Such periods are usually associated with high volatility of consumer con…dence, suggesting that large swings
in con…dence could be useful indicators of consumption.
The purpose of the paper is …rst to empirically assess the role of con…dence
in explaining household consumption in the United States and the euro area
and show to what extent con…dence indicators bring additional information
beyond variables usually found to have some explanatory power for household
real consumption expenditures (e.g. income, wealth or interest rates). Second, it
will check the existence of a con…dence channel in international transmission of
shocks. Finally, it will identify under which circumstances con…dence indicators
can be a good predictor of household consumption. In particular, it will measure
the contribution of con…dence during periods associated with large movements
in household survey indicators.
Although not providing any methodological novelty to the empirical literature, the value added of this paper concerns …rst the use of a relatively long and
up-to-date database that in particular includes the recent …nancial crisis. It also
provides a comparison between the United States and the euro area. Finally, it
allows for international linkages in con…dence.
ECB
Working Paper Series No 1349
June 2011
5
41 $,/(1(" + $5(#$-"$ (2 ! 2$# .- 3'1$$ #(>$1$-3 ,$3'.#.+.&(" + //1. "'$2
(123 - //1. "' ! 2$# .- 2(,/+$ ".-24,/3(.- $04 3(.- (-#(" 3$2 3' 3 6'(+$
%.1 3'$ ".-9#$-"$ (-#(" 3.12 #. -.3 2$$, 3. '$+/ /1$#("3 ".-24,/3(.$7/$-#(341$2 6'$- 3 *(-& %4-# ,$-3 +2 (-3. "".4-3 3'$8 , 8 ".-3 (- 2.,$
5 +4 !+$ (-%.1, 3(.- (- 3'$ $41. 1$ " 2$ .1$.5$1 3'$1$ (2 2.,$ $5(#$-"$ .%
3'$ /1$2$-"$ .% .-$6 8 ".-9#$-"$ "' --$+ !$36$$- 3'$ -(3$# 3 3$2 -#
3'$ $41. 1$ - .3'$1 6.1#2 / 23 "' -&$2 (- 3'$ ".-9#$-"$ (-#(" 3.12
6.4+# ".-3 (- (-%.1, 3(.- !.43 "411$-3 "' -&$2 (- $41. 1$ ".-24,/3(.-
'$ .//.2(3$ 6.4+# -.3 !$ 314$
$".-# 3'$2$ 1$24+32 1$ ".-91,$# !8 //1. "' 6'("' +2. 2'.62 3'1.4&' '(23.1(" + #$".,/.2(3(.- $7$1"(2$ 3' 3 ".-9#$-"$ 2$$,2 3. , 33$1 (2.,$ 2/$"(9" $/(2.#$2 6'("' (- ,.23 " 2$2 ".11$2/.-#2 3. /$1(.#2 6'$1$ 3'$1$
1$ + 1&$ "' -&$2 (- '.42$'.+# 2415$8 (-#(" 3.12 +(*$ #41(-& 9- -"( + "1(2$2 .1
&$./.+(3(" + 3$-2(.-2 %.1 (-23 -"$
. " /341$ !$33$1 3'$ % "3 3' 3 ".-9#$-"$ ,(&'3 /+ 8 1.+$ .-+8 #41(-&
4-"$13 (- 3(,$2 3'(1# $,/(1(" + //1. "' 1$+($2 .- -.-+(-$ 1 $23(, 3(..% .41 ".-24,/3(.- $04 3(.- 42(-& 3'1$2'.+# ,.#$+ '(2 ,.#$++(-& /
/1. "' 224,$2 3' 3 3'$1$ (2 "$13 (- 4-*-.6- 3'1$2'.+# (- ".-9#$-"$ (-#$7
"' -&$2 !$8.-# 6'("' ".-9#$-"$ 23 132 (,/ "3(-& ".-24,/3(.- !$' 5(.412 -
2 ,/+$ -# .43.%2 ,/+$ $7$1"(2$2 2'.6 3' 3 3'$ ".-31(!43(.- .% ".-9#$-"$ ($7/+ (-(-& ".-24,/3(.- $7/$-#(341$2 (-"1$ 2$2 6'$- '.42$'.+# 2415$8 (-#(" 3.12 %$ 341$ + 1&$ "' -&$2 2. 3' 3 ".-9#$-"$ (-#(" 3.12 " - ' 5$ 2.,$ (-"1$ 2(-&
/1$#("3(5$ /.6$1 #41(-& 24"' $/(2.#$2
5$1 ++ 3'$ 1$24+32 2'.6 3' 3 3'$ ".-24,$1 ".-9#$-"$ (-#$7 " - !$ (- "$1
3 (- "(1"4,23 -"$2 &..# /1$#("3.1 .% ".-24,/3(.- .1$.5$1 3'$1$ (2 2.,$
$5(#$-"$ .% <".-9#$-"$ "' --$+= (- 3'$ (-3$1- 3(.- + 31 -2,(22(.- .% 2'."*
2 ".-9#$-"$ (-#("$2 +$ # ".-24,$1 2$-3(,$-3 (- 3'$ $41. 1$ (- ++8
3'$ 1$+ 3(.-2'(/ !$36$$- 2$-3(,$-3 -# ".-24,/3(.- ' 2 !$$- 2'.6- 3. !$ -.+(-$ 1 6$ " - -.3("$ 3' 3 ,.#$+2 6(3' ".-9#$-"$ 2$$, 3. .43/$1%.1, #41(-&
/$1(.#2 .% &$./.+(3(" + 3$-2(.-2 +(*$ (- .1 #41(-& 3'$ 9- -"( + "1(2(2
4341$ 1$2$ 1"' (-"+4#$2 $73$-2(.-2 3. .3'$1 ".4-31($2 -# .3'$1 5 1( !+$2
2 2415$82 +2. 1$/.13 !42(-$22 2$-3(,$-3 2(,(+ 1 1$2$ 1"' ".4+# !$ #.-$ .3'$ /1$#("3(5$ /.6$1 .% ".-9#$-"$ .- (-5$23,$-3 #$"(2(.-2 8 $73$-2(.- ".,
!(-(-& ".-24,$1 -# !42(-$22 2415$8 # 3 ".4+# '$+/ (,/1.5$ 3'$ %.1$" 23 .%
:4"34 3(.-2 (- ++8 6$ ' 5$ .-+8 42$# 3'$ &&1$& 3$ (-#$7 .% ".-9#$-"$
2415$82 1(.42 24!".,/.-$-32 ".4+# !$ 42$# 2 +3$1- 3(5$2 2 3'$8 ,(&'3
/1.5(#$ ,.1$ /1$"(2$ (-%.1, 3(.- !.43 &$-32; /$1"$/3(.-2 !.43 3'$ %4341$
6
ECB
Working Paper Series No 1349
June 2011
1
Introduction
The 2007-09 …nancial crisis led to the most severe global economic recession since the Great Depression. Just after the bankrupcy of Lehman Brothers, Nobel Prize Joseph Stiglitz said that this
"…nancial crisis springs from a catastrophic collapse in con…dence”1 . While it remains di¢ cult to
assert whether the collapse of con…dence was the cause or the consequence of the …nancial crisis,
most academics and policy makers agree to say that it is this erosion of con…dence that has ensured
the depth and the longevity of the crisis, especially as regards its impacts on the real economy.
The link between con…dence and economic decisions has been widely covered in the literature,
which has focused on two main aspects. First, from a theoretical viewpoint, the literature has concentrated on the conceptualisation of con…dence and its role in modern theories of consumption. If
consumers were to behave according to the Permanent Income Hypothesis (PIH), no information
known to the consumer when the consumption choice was made could have any predictive power
for how consumption will change in future periods (Hall, 1978). Any deviation from the PIH can
be theoretically justi…ed by liquidity constraints or uncertainty relative to future income. This
uncertainty can then lead households to decrease their current consumption and build precautionary savings to face a possible drop in their income. Against this background, consumer con…dence
indices could be helpful as they might capture information about expected income. Another approach to consumption relates to the existence of "animal spirits" (Katona, 1975 or Eppright et
al., 1998). Consumer expenditures could be in‡uenced by non-economic factors, such as political
tensions or wars, that would a¤ect the willingness of households to consume by increasing perceived
uncertainty (Acemoglu and Scott 1994).
Second, for an empirical viewpoint, the literature has been concerned on whether the con…dence indicators contain any information beyond economic fundamentals. The concern is whether
con…dence can be explained by current and past values of variables such as income, unemployment, in‡ation or consumption or, in other way, whether con…dence measures have any statistical
signi…cance in predicting economic outcomes once information from the variables cited above is
used. While the evidence is overall rather mixed, most authors seem to, at least, …nd a signi…cant
statistical relationship between con…dence measures and economic variables, current and future.2
In particular, some stress the special importance of con…dence indicators in predicting periods of
strong ‡uctuations in the economy, such as recessions and recoveries (Howrey, 2001; Haugh, 2005)
1
See
J.E.
Stiglitz,
"The
fruit
of
hypocrisy",
The
Guardian,
16
September
2008,
http://www.guardian.co.uk/commentisfree/2008/sep/16/economics.wallstreet
2
Many authors …nd that consumer con…dence can reduce forecast errors made by models including traditional
macroeconomic variables: Ludvigson (2004) or Wilcox (2007) for the U.S.; Kwan and Cotsomitis (2006) for Canada;
Easaw and Heravi (2004) for the UK. However, Smith (2009) using real-time data for the UK shows that including
consumer sentiment in a VAR model to predict consumer expenditures does not improve its forecast accuracy.
Similarly, Al-Eyd et al. (2008) show that con…dence e¤ects on consumption are weak when other key determinants
of consumption are taken into account across …ve major OECD countries. Claveria et al. (2007) show that despite
survey indicators provide useful information for improving forecasts of many euro area macroeconomic variables, such
improvements are signi…cant in a limited number of cases.
ECB
Working Paper Series No 1349
June 2011
7
5B8 CH<9FG !5FB9F 5B8 .<FCCD GI;;9GH H<5H H<9M 7CI@8 69 <9@D:I@ 8IF=B; D9F=C8G C:
A5>CF 97CBCA=7 CF DC@=H=75@ G<C7?G
-I7< D9F=C8G 5F9 IGI5@@M 5GGC7=5H98 K=H< <=;< JC@5H=@=HM C:
7CBGIA9F 7CBO89B79 GI;;9GH=B; H<5H @5F;9 GK=B;G =B 7CBO89B79 7CI@8 69 IG9:I@ =B8=75HCFG C: 7CB
GIADH=CB
5FFC@ 9H 5@
:CF =BGH5B79 7=H9 7CBGIA9F 7CBO89B79 5G H<9 @958=B; 75IG9 :CF H<9
/- F979GG=CB
.<9 DIFDCG9 C: H<9 D5D9F =G OFGH HC 9AD=F=75@@M 5GG9GG H<9 FC@9 C: 7CBO89B79 =B 9LD@5=B=B;
<CIG9<C@8 7CBGIADH=CB =B H<9 /B=H98 -H5H9G 5B8 H<9 9IFC 5F95 5B8 G<CK HC K<5H 9LH9BH 7CBO89B79
=B8=75HCFG 6F=B; 588=H=CB5@ =B:CFA5H=CB 69MCB8 J5F=56@9G IGI5@@M :CIB8 HC <5J9 GCA9 9LD@5B5HCFM
DCK9F :CF <CIG9<C@8 F95@ 7CBGIADH=CB 9LD9B8=HIF9G 9
;
=B7CA9 K95@H< CF =BH9F9GH F5H9G
-97CB8
=H K=@@ 7<97? H<9 9L=GH9B79 C: 5 7CBO89B79 7<5BB9@ =B =BH9FB5H=CB5@ HF5BGA=GG=CB C: G<C7?G
=B5@@M =H
K=@@ =89BH=:M IB89F K<=7< 7=F7IAGH5B79G 7CBO89B79 =B8=75HCFG 75B 69 5 ;CC8 DF98=7HCF C: <CIG9<C@8
7CBGIADH=CB
#B D5FH=7I@5F =H K=@@ A95GIF9 H<9 7CBHF=6IH=CB C: 7CBO89B79 8IF=B; D9F=C8G 5GGC7=5H98
K=H< @5F;9 ACJ9A9BHG =B <CIG9<C@8 GIFJ9M =B8=75HCFG
@H<CI;< BCH DFCJ=8=B; 5BM A9H<C8C@C;=75@
BCJ9@HM HC H<9 9AD=F=75@ @=H9F5HIF9 H<9 J5@I9 58898 C: H<=G D5D9F 7CB79FBG OFGH H<9 IG9 C: 5 F9@5H=J9@M
@CB; 5B8 IDHC85H9 85H565G9 H<5H =B D5FH=7I@5F =B7@I89G H<9 F979BH OB5B7=5@ 7F=G=G
#H 5@GC DFCJ=89G
5 7CAD5F=GCB 69HK99B H<9 /B=H98 -H5H9G 5B8 H<9 9IFC 5F95
=B5@@M =H 5@@CKG :CF =BH9FB5H=CB5@
@=B?5;9G =B 7CBO89B79
.<9F9 =G 5B =ADCFH5BH 75J95H HC GHF9GG ID:FCBH 5G F9;5F8G H<9 A95GIF9A9BH C: 7CBO89B79
"CIG9
<C@8 G9BH=A9BH =G 5 D9FGCB5@ 5B8 GI6>97H=J9 5GG9GGA9BH C: H<9 9BJ=FCBA9BH 7IFF9BH 5B8 :IHIF9 =B
K<=7< 5;9BHG H5?9 97CBCA=7 897=G=CBG
'CF9CJ9F 5G G<CKB :CF =BGH5B79 6M CA=B=HN 5B8 '5BG?=
H<9 7CBGIA9F G9BH=A9BH =B8=79G A=;<H GIV9F :FCA A95GIF9A9BH 9FFCFG 5G GIFJ9M EI9GH=CBG
5F9 J9FM C:H9B 5A6=;ICIG :CF H<9 F9GDCB89BH 5B8 HCC AI7< EI5@=H5H=J9 HC 69 IG98 :CF EI5BH=H5H=J9
5GG9GGA9BH
"9F9 @=?9 =B ACGH DF9J=CIG F9G95F7< K9 5GGIA9 H<5H 7CBO89B79 =B8=79G 89F=J98 :FCA
GIFJ9MG 5F9 5 F9@5H=J9@M ;CC8 DFCLM C: <CIG9<C@8GQ D9F79DH=CBG 56CIH H<9=F 97CBCA=7 9BJ=FCBA9BH
5B8 7CI@8 69 IG98 5G 9LD@5B5HCFM J5F=56@9G C: H<9=F 7CBGIADH=CB 9LD9B8=HIF9G
AD=F=75@@M 5G G<CKB :CF =BGH5B79 6M 5FFC@ 9H 5@
A95GIF9G C: 7CBGIA9F 7CBO89B79
5F9 <=;<@M 7CFF9@5H98 K=H< F95@ 7CBGIADH=CB
D@CH C: H<9 G9F=9G :CF H<9 /B=H98 -H5H9G =;IF9
5B8 H<9 9IFC 5F95 =;IF9 G<CKG =B8998 GCA9 7CACJ9A9BH 69HK99B H<9 @C; 7<5B;9 =B F95@
7CBGIADH=CB 5B8 H<9 56GC@IH9 7<5B;9 =B 7CBGIA9F 7CBO89B79 =B89L
#B H<9 /B=H98 -H5H9G H<9
7CFF9@5H=CB 69HK99B F95@ 7CBGIADH=CB ;FCKH< 5B8 H<9 7<5B;9 =B 7CBGIA9F G9BH=A9BH A95GIF98
6M H<9 /B=J9FG=HM C: '=7<=;5B CBGIA9F -9BH=A9BH #B89L =G F5H<9F <=;<
1<=@9 =H =G H<9 <=;<9GH
K<9B 7CADIH98 7CBH9ADCF5B9CIG@M H<9 7CFF9@5H=CB 69HK99B @5;;98 7CBGIA9F 7CBO89B79
5B8 7CBGIADH=CB F9A5=BG <CK9J9F F9@5H=J9@M 9@9J5H98 K=H< 5 CB9D9F=C8 @5; 5B8 K=H< 5
HKCD9F=C8 @5; =B8=75H=B; GCA9 DCH9BH=5@ @958=B; DFCD9FH=9G :CF 7CBGIA9F G9BH=A9BH
#B H<9 9IFC
5F95 H<9 7CFF9@5H=CB 69HK99B 7CBO89B79 5B8 7CBGIADH=CB =G H<9 <=;<9GH K<9B 7CBO89B79 =G @5;;98
6M CB9 D9F=C8 .<9 7CFF9@5H=CB F9A5=BG 5@GC @5F;9 :CF <=;<9F @5;G :CF 5 D9F=C8 @5; 5B8
:CF 5 D9F=C8 @5;
8
ECB
Working Paper Series No 1349
June 2011
-+63) #0-5)( !5%5)4 1046/25-10 +3185, 3,4 %0( ',%0+) -0 '10:()0') .,4
-+63) 631 %3)% 1046/25-10 +3185, 3,4 %0( ',%0+) -0 '10:()0') .,4
BCH<9F 5GD97H H<5H <5G 699B B9;@97H98 =B H<9 @=H9F5HIF9 =G H<9 7FCGG7CIBHFM 7CFF9@5H=CB C:
7CBO89B79 =B8=75HCFG =;IF9 "9F9 5;5=B H<9 7CFF9@5H=CB 69HK99B H<9 7<5B;9G =B /- 5B8 9IFC
5F95 7CBO89B79 =G F9@5H=J9@M @5F;9 7CBH9ADCF5B9CIG@M
<5B;9G =B /- 7CBO89B79 G99A HC
@958 =B GCA9 D9F=C8G 7<5B;9G =B H<9 9IFC 5F95 7CBO89B79
.<9 7CFF9@5H=CB 69HK99B H<9 HKC J5F=56@9G
=G H<9 <=;<9GH K<9B H<9 9IFC 5F95 7CBO89B79 =G @5;;98 6M HKC D9F=C8G ECB
Working Paper Series No 1349
June 2011
9
Figure 3: Change in con…dence in the United States (rhs) and the euro area (lhs)
These comovements do not preclude however any causal link between con…dence indicators
across countries and between con…dence and consumption. Con…dence indicators could also be just
a good proxy for other fundamental variables.
Overall, the results show that the consumer con…dence index can be in certain circumstances a
good predictor of consumption. In particular, out-of-sample evidence shows that the contribution
of con…dence in explaining consumption expenditures increases when household survey indicators
feature large changes, so that con…dence indicators can have some increasing predictive power during
such episodes. Moreover, there is some evidence of a “con…dence channel” in the international
transmission of shock, as U.S. con…dence indices lead consumer sentiment in the euro area.
The rest of the paper is organised as follows. Section 2 gives a brief description of the data
used in this paper. Section 3 describes the empirical framework used for our research and reports
results from both univariate and multivariate analyses. Section 4 extends the previous analysis by
proposing a non-linear approach in order to isolate periods where con…dence explains signi…cantly
more consumption developments. Section 5 contains some concluding remarks.
2
Data
The dataset used covers the period from the …rst quarter of 1985 to the second quarter of 2010. The
observations are seasonally adjusted though no detrending or business cycle adjustments are made.
Since we are agnostic with respect to which are the mechanisms driving any statistical relation, we
abstain from applying any economic theory motivated technique in order to …lter the series from
phenomena such as business cycles or trends. Real personal consumption series are taken from the
national account sources (BEA for the United States - NIPA table 1.1.6 (millions of chained 2005
dollars) and Eurostat for the euro area). For the euro area, national account series are available
from 1995 only. To backdate the series, we use the rates of growth of the series in the Area Wide
Model database (AWM, see Fagan et al., 2001 for details).
10
ECB
Working Paper Series No 1349
June 2011
CB79FB=B; H<9 7CBO89B79 =B8=75HCFG 5 7CADF9<9BG=J9 89G7F=DH=CB :CF H<9 /
-
75G9 75B 69
:CIB8 =B F5A 5B8 &I8J=;GCB G H<9 5IH<CFG BCH9 H<9 /B=J9FG=HM C: '=7<=;5BQG CBGIA9F
-9BH=A9BH #B89L /'-# 5B8 H<9 CB:9F9B79 C5F8QG CBGIA9F CBO89B79 #B89L # 5F9
H<9 ACGH K=89@M :C@@CK98 A95GIF9G C: /
-
7CBGIA9F 7CBO89B79
.<9M 6CH< 7CBG=GH C: OJ9 GIFJ9M
EI9GH=CBG K<9F9 F9GDCB89BHG 5F9 5G?98 HC 5GG9GG 7IFF9BH 7CB8=H=CBG 5B8 HC F9J95@ 9LD97H5H=CBG 56CIH
H<9 :IHIF9
1=H< F9GD97H HC H<9 7IFF9BH 7CB8=H=CBG H<9 GIFJ9MG 8=V9F GCA9K<5H K<=@9 H<9 /'-#
GIFJ9MG 7CBGIA9FG 56CIH A5>CF 9LD9B8=HIF9G 5B8 7<5B;9G =B H<9 F9GDCB85BHQG OB5B7=5@ G=HI5H=CB
H<9 # =BEI=F=9G :C7IG CB >C6 5J5=@56=@=HM 5B8 7IFF9BH 6IG=B9GG 7CB8=H=CBG
#H =G H<9B DCGG=6@9
H<5H 6CH< =B89L9G A5M 5H H=A9G G9B8 A=L98 G=;B5@G 56CIH 7CBGIA9F 7CBO89B79 89D9B8=B; CB H<9
F9@5H=J9 D9F:CFA5B79 C: H<9 J5F=56@9G H<5H A5M 69 ACF9 F9@9J5BH HCK5F8G 9LD9B8=HIF9G 897=G=CBG
5B8 7IFF9BH 65@5B79G K=H< F9GD97H HC H<9 :57HCFG H<5H 89H9FA=B9 IB9AD@CMA9BH 5B8 CH<9F @56CF
A5F?9H F9@5H98 =GGI9G
G H<9 A5=B F9GI@HG 5F9 FC6IGH K<5H9J9F =B89L IG98 K9 DF9G9BH <9F9 CB@M
H<CG9 65G98 CB H<9 /'-#
,9GI@HG 65G98 CB H<9 # 5F9 5J5=@56@9 IDCB F9EI9GH
CF H<9 9IFC 5F95 K9 IG9 H<9 =B89L 7CBGHFI7H98 6M H<9 =F97HCF5H9 !9B9F5@ :CF 7CBCA=7 5B8
=B5B7=5@ V5=FG C: H<9 IFCD95B CAA=GG=CB K<=7< 7CB8I7HG F9;I@5F <5FACB=G98 GIFJ9MG :CF
8=V9F9BH G97HCFG C: H<9 97CBCA=9G =B H<9 IFCD95B /B=CB / 5B8 =B H<9 5DD@=75BH 7CIBHF=9G
.<9M 5F9 588F9GG98 HC F9DF9G9BH5H=J9G C: H<9 =B8IGHFM A5BI:57HIF=B; H<9 G9FJ=79G F9H5=@ HF589
5B8 7CBGHFI7H=CB G97HCFG 5G K9@@ 5G HC 7CBGIA9FG #
.<9 9LD@5B5HCFM J5F=56@9G H<5H K9 K=@@ HF95H @5H9F 5G 97CBCA=7 :IB85A9BH5@G 5F9 J5F=56@9G
IGI5@@M :CIB8 HC <5J9 GCA9 DF98=7H=J9 DCK9F HC 9LD@5=B 7<5B;9G =B 7CBGIADH=CB
.<9M =B7@I89
F95@ 8=GDCG56@9 =B7CA9 OB5B7=5@ 5B8 <CIG=B; K95@H< F95@ GHC7? DF=79G G<CFHH9FA =BH9F9GH F5H9G
IB9AD@CMA9BH F5H9 5B8 F95@ C=@ DF=79G
CF H<9 /B=H98 -H5H9G 5G F5A 5B8 &I8J=;GCB K9 IG9
85H5 :FCA H<9 (5H=CB5@ *FC8I7H #B7CA9 77CIBHG DFC8I798 6M H<9 IF95I C: 7CBCA=7 B5@MG=G
C: H<9 9D5FHA9BH C: CAA9F79
.C DFCLM F95@ =B7CA9 K9 IG9 F95@ K5;9G 5B8 GIDD@9A9BHG B9H C:
HF5BG:9FG IG=B; H<9 (#* H56@9 =H9AG 15;9G 5B8 G5@5FM 8=G6IFG9A9BHG A=BIG *9FGCB5@ 7IFF9BH
HF5BG:9F F979=DHG 5B8 H<9B 58>IGH98 6M H<9 D9FGCB5@ 7CBGIADH=CB 9LD9B8=HIF9G 89P5HCF H5?9B :FCA
(#* H56@9 CF =BH9F9GH F5H9G K9 IG9 H<9 EI5FH9F@M 5J9F5;9G C: H<9 ACBH<
HF95GIF9 6=@@ F5H9 5J5=@56@9 :FCA H<9 C5F8 C: !CJ9FBCFG C: H<9 989F5@ ,9G9FJ9 -MGH9A
CAD5F98
K=H< F5A 5B8 &I8J=;GCB K9 <5J9 5 F9@5H=J9@M F=7< G9H C: K95@H< J5F=56@9G
9G=89G H<9
HF58=H=CB5@ -H5B85F8 5B8 *CCFQG 7CADCG=H9 =B89L K9 5@GC A5?9 IG9 C: =B5B7=5@ 5B8 ,95@ GH5H9
195@H< GHC7?G
=B5@@M H<9 IB9AD@CMA9BH F5H9 =G C6H5=B98 :FCA H<9 IF95I C: H<9 &56CF GH5H=GH=7G
5B8 K9 H5?9 H<9 5J9F5;9 C: F9BH I65= 5B8 1.# 5G C=@ DF=79G
@@ BCA=B5@ J5F=56@9G 5F9 89P5H98
A5?=B; IG9 C: H<9 56CJ9 F9:9FF98 7CBGIADH=CB 89P5HCF
1=H< F9GD97H HC H<9 85H5 F9@5H98 HC H<9 9IFC 5F95 J5F=56@9G H<9M 5F9 H5?9B :FCA H<9 F951=89
'C89@ 5H565G9
.<9 <=GHCF=75@ 85H5 5F9 65G98 CB H<9 5;;F9;5H=CB C: 5J5=@56@9 7CIBHFM =B:CFA5H=CB
K<9B H<9 CF=;=B5@ 1' 85H565G9 K5G 7CAD=@98
.<9 A5=B GCIF79 :CF H<9 7CIBHFM =B:CFA5H=CB =G
IFCGH5H 7CAD@9A9BH98 6M H<9 ) (5H=CB5@ 77CIBHG H<9 ) '5=B 97CBCA=7 =B8=75HCFG
H<9 #- 5B8 H<9 ') 85H565G9G
.<9 85H5 5G K9@@ 5G A9H<C8C@C;M BCH9G 5F9 5J5=@56@9 5H
<HHDKKK
9567B
CF;85H55KA=B89L
<HA
@GC :CF H<9 GHC7? A5F?9H K9 IG9 H<9 IFCF95
ECB
Working Paper Series No 1349
June 2011
11
Share Real Price Index.
3
Do con…dence indicators bring additional information beyond
economic fundamantals?
We start our empirical analysis by running unit root ans casality test. Thereafter, we estimate a
simple consumption equation where a con…dence indicator is considered as an explanatory variable
for consumption together with standard variables used in the empirical literature on consumption
expenditures. Finally, we estimate a VAR model to derive impulse response functions and historical
decomposition.
3.1
Unit root and Granger causality tests
We …rst perform Augmented Dickey-Fuller tests in order to determine the order of integration of the
variables. For the United States, most variables are integrated of order one or I(1). Unexpectedly,
the natural logarithm of real personal consumption expenditure appears be I(2). However, graphical
inspection and theoretical insights should put some caution in this pure statistical analysis, as there
is no theory that may suggest such an outcome. In the empirical part, we will therefore ignore this
odd outcome, as trend stationarity or even integrated of order one processes seem to be the most
theoretical sound data generator process for variables as income and consumption. For the euro
area, all variables, including consumption, are found to be I(1).
We then study Granger causality among the various variables of our dataset. Table 1 presents
the results of the Granger causality analysis with 5 lags. P -values reported for the probability of
Row NOT Granger-causing Column. Rejections of the null hypothesis for 5% signi…cance level are
in bold. For the United States, consumption is Granger-caused by wealth and equity prices. For
the euro area, consumption is Granger-caused by con…dence and interest rates only.
We also perform the Granger causality tests for con…dence. In the United States, con…dence is
Granger-caused by …nancial wealth and equity prices, while in the euro area, unemployment rate,
interest rates and foreign con…dence are the only variables that Granger cause domestic con…dence.
The results of the Granger causality tests are somewhat informative. They con…rm that wealth
e¤ects might be stronger in the United States than in the euro area. In the U.S. case, wealth
seems not only to cause consumption but also consumer con…dence. Also, con…dence seems to
in‡uence consumption in the euro area while no causal link can be found in the U.S. case. Finally,
the causality found between U.S. con…dence and euro area con…dence indicates possible con…dence
linkages between the two economic areas. These results remain however very partial. The estimation
of consumption models together with a dynamic analysis are necessary to understand better the
role of the various possible determinants of consumption expenditures and their complex dynamic
relationships.
12
ECB
Working Paper Series No 1349
June 2011
3.2
Estimation of a simple model for consumption
We extend our causality analysis with the estimation of a very simple model where the change in
(ln) consumption (
ln Ct ) only depends on the change in con…dence indicators ( conft ),
ln Ct =
+
q
X
i
conft
i
+
(1)
i
i=1
where
i
is the error term.
Table 2 presents the results for both the United States and the euro area. The lag order (q) is
determined using standard information criteria and 2 lags are found to be optimal for all models
estimated. Using only the past changes in consumer con…dence indicators could explain 8% and
10% of the variation of consumption changes in the U.S. and the euro area respectively.
We then compare this simple model with alternatives that include a set of fundamental variables. We study three di¤erent sets of fundamentals. The …rst (Zt1 ) only includes past changes in
consumption together with past changes in (ln) real disposable income ( ln Yt ). The second one
(Zt2 ) also includes changes in wealth (both …nancial and housing wealth
of fundamentals
(Zt3 )
ln Wt ). The third set
includes variables that might in‡uence consumption behaviours even though
no theory includes them directly as fundamentals. These variables are the log-changes in real equity
prices (
ln q), the changes in short-term interest rates ( i), in unemployment rate ( u) and in
(ln) real oil prices (
2
ln roil). For each set of fundamentals, we compare the R of the model (Eq.
2) with that of an alternative version that also includes the changes in con…dence indicators (Eq.
3).
ln Ct =
+
q
X
k
i Zt i
+ i ; for k = 1; 2; 3:
(2)
i=1
ln Ct =
+
q
X
i=1
i
conft
i
+
q
X
k
i Zt i
+ i ; for k = 1; 2; 3:
(3)
i=1
2
As shown by Table 2, expanding the set of fundamentals increases the R . While Zt1 already
2
explains 18% of the linear variation in U.S. consumption expenditure changes, the R increases to
2
35% with Zt2 and 42% with Zt3 . For the euro area, the R are surprisingly lower with Zt1 and Zt2
2
(9% and 7% respectively). The R increases however to 16% when using Zt3 .
Di¤erences between the two economic areas are even more striking when adding con…dence
indicators to the models. While for the United States, including con…dence indicators does not
increase the goodness of …t of the model, it improves that of the euro area. Indeed, in the U.S. case,
2
the R are similar for models (3) and (2), except for the case when we compare the Zt1 model with
one with just con…dence. For the euro area the best model is the one including con…dence indicators
together with Zt3 , explaining almost 18% of the linear variation in consumption expenditure changes.
2
While the R might appear low given the number of variables used, it is important to notice that
ECB
Working Paper Series No 1349
June 2011
13
BC 7CBH9ADCF5B9CIG J5F=56@9G 5F9 IG98 =B H<9G9 9GH=A5H=CBG G=B79 K9 :C7IG CB H<9 DF98=7H=J9 DCK9F
C: J5F=56@9G 5H 5 CB9D9F=C8 5<958 <CF=NCB
=B5@@M K9 7CBG=89F 5 @5GH AC89@ H<5H =B7@I89 CB HCD C: H<9 @5F;9GH G9H C: :IB85A9BH5@G 5B8
H<9 8CA9GH=7 7CBO89B79 =B8=75HCFG H<9 7CFF9GDCB8=B; :CF9=;B :IB85A9BH5@G G 69:CF9 K9
H9GH =: 588=B; H<9 7CBO89B79 =B8=75HCFG 56FC58 ('
=
9
9IFC 5F95 7CBO89B79 =B H<9 /
-
AC89@
5B8 /
-
7CBO89B79 =B H<9 9IFC 5F95 AC89@ =ADFCJ9G H<9 OH 6M 7CAD5F=B; H<9 C: 5B8 ('
('
('
"9F9 5;5=B K<=@9 588=B; 9IFC 5F95 J5F=56@9G K=H< CF K=H<CIH 9IFC 5F95 7CBO89B79 8C9G BCH
=ADFCJ9 H<9 /
-
AC89@QG ;CC8B9GGC:OH H<9 9IFC 5F95 CB9 =ADFCJ9G K<9B 588=B; /
-
:IB85A9B
H5@G H<9 F=G9G :FCA HC 5B8 =ADFCJ9G 9J9B :IFH<9F K<9B 588=B; /
-
7CBO89B79 H<9
F=G9G HC .<=G F9GI@H IB89@=B9G H<9 FC@9 C: /
-
89J9@CDA9BHG HC DF98=7H 9IFC 5F95 CB9G
G
G<CKB 6M !=5BBCB9 9H 5@
H<9 /
-
6IG=B9GG 7M7@9 =G @958=B; H<9 9IFC 5F95 6IG=B9GG 7M7@9
.<9
=B7F95G9 =B H<9 DF98=7H=J9 DCK9F =B H<9 9IFC 5F95 AC89@G K<9B =B7@I8=B; /
-
J5F=56@9G H<9B F9P97HG
H<9 :57H H<9 /
-
7CBGIA9FG D9F79=J9 95F@=9F H<5B H<9 9IFC 5F95 7CBGIA9FG 5 7<5B;9 =B H<9 6IG=B9GG
7M7@9
'CF9CJ9F H<9 :IFH<9F =B7F95G9 =B H<9 OH 69HK99B 5B8 GI;;9GHG H<9 9L=GH9B79 C: 5
7CBO89B79 7<5BB9@ H<5H F9P97HG H<9 :57H H<5H B9KG GDF958G 57FCGG H<9 ;@C69 EI=7?@M
7<5B;9 =B
H<9 /
-
7CBO89B79 5BH=7=D5H=B; 5 :IHIF9 7<5B;9 =B H<9 6IG=B9GG 7M7@9 A=;<H H<9B 5V97H 8CA9GH=7
7CBO89B79 K<=7< =B HIFB <5G 5B =AD57H CB 8CA9GH=7 6IG=B9GG 7M7@9
.<=G F9GI@H =G =B @=B9 K=H< 5 F9
79BH GHI8M 6M 9= K<C OB8G 9AD=F=75@ 9J=89B79 C: H<9 9L=GH9B79 C: 5 7CBO89B79 7<5BB9@ :FCA
R@5F;9 7CIBHF=9GS HC RGA5@@9F 7CIBHF=9GS
J9B 5:H9F <5J=B; 7CBHFC@@98 :CF 8CA9GH=7 A57FC97CBCA=7
75IG9G C: 7CBO89B79 @9J9@ J5F=5H=CBG H<9 @9J9@ C: 7CBO89B79 C: 5;9BHG =B @5F;9 7CIBHF=9G 8C9G <5J9
5B =BPI9B79 CB H<9 @9J9@ C: 7CBO89B79 C: 5;9BHG =B GA5@@9F 7CIBHF=9G
@H<CI;< H<=G 5B5@MG=G F9A5=BG F9@5H=J9@M G=AD@9 =H <9@DG HC 8F5K GCA9 H9BH5H=J9 7CB7@IG=CBG
=FGH K<=@9 :CF H<9 /
-
7CBO89B79 =B8=75HCFG 8C BCH G99A HC <9@D DF98=7H=B; 7CBGIADH=CB 9LD9B8=
HIF9G K<9B H5?=B; :IB85A9BH5@G =BHC 577CIBH H<9M A5M 7CBH5=B GCA9 J5@I56@9 =B:CFA5H=CB =B H<9
9IFC 5F95 75G9
-97CB8 H<9 F9GI@HG C: AC89@ =B8=75H9G H<9 DF9G9B79 C: 5 CB9K5M 7CBO89B79
@=B?5;9 69HK99B H<9 /B=H98 -H5H9G 5B8 H<9 9IFC 5F95
#B CH<9F KCF8G D5GH 7<5B;9G =B H<9 /
-
7CBO89B79 =B8=75HCFG KCI@8 7CBH5=B =B:CFA5H=CB 56CIH 7IFF9BH 7<5B;9G =B 9IFC 5F95 7CBGIADH=CB
.<9 CDDCG=H9 KCI@8 BCH 69 HFI9
.<=G F9GI@H =G BCH B979GG5F=@M GIFDF=G=B; ;=J9B H<9 =ADCFH5B79 H<9
F9@95G9 C: /
-
7CBO89B79 =B8=75HCFG <5G =B H<9 97CBCA=7 DF9GG 5B8 CB OB5B7=5@ A5F?9HG
.<=G =G
@9GG GC K<9B 9IFC 5F95 7CBO89B79 =B8=75HCFG 5F9 F9@95G98
14
ECB
Working Paper Series No 1349
June 2011
3.3
VAR analysis
We now set up a VAR modelling framework to help us analyse the dynamics of the impacts of
a shock to con…dence on consumption expenditures through impulse response functions. In a
…rst step, we estimate country-speci…c VAR models using the same variables as in the univariate
estimations above. This allows us to test the statistical signi…cance of the con…dence indicators
through the error bounds of the impulse response functions. We also estimate models where a
foreign con…dence indicator is added to the set of domestic fundamentals, in order to …nd evidence
of any "con…dence linkage" between the United States and the euro area. In a second step, we
compute historical forecast error variance decomposition to graphically see how the contribution of
con…dence shocks has changed over time.
We estimate …rst the preferred model for U.S. consumption (Eq. (3)) in a VAR setting, using
the largest set of fundamentals (Zt3 ).
We therefore have the following VAR model to estimate:
yt =
q
X
Ai y t
i
i=1
0
B
where y = @
and
i
ln Ct
+
1
C
conft A
i;
(6)
(7)
Zt3
is a vector of orthogonalised shocks. The orthogonolasition is done via a Choleski ap-
proach using the following ordering (as Bram and Ludvigson, 1998): con…dence, …nancial variables,
interest rates, wealth, consumption and income. Here again, the optimal lag order (q) is found to
be equal to 2 according to standard information criteria.3
Figure 4 shows the impulse response functions of a shock to con…dence on real consumption
in the United States. The impulse responses of the con…dence shocks appear signi…cant for two
periods ahead but insigni…cant at the 95% level for future periods. The variance decomposition
(not shown here) also shows some e¤ect of con…dence on consumption (slightly more than 10% of
the variance would be due to con…dence), though the con…dence bounds4 shows that the degree of
signi…cance remains borderline.
3
The lag order was selected by using various criteria, including the Sequential modi…ed LR test statistic, Final
prediction error, Akaike information criterion, Schwarz information criterion and Hannan-Quinn information criterion.
Although the …nal selection has remained subjective - since the various criteria often give con‡icting results -, we
decided to keep the same lag order whatever model considered for the sake of comparability and consistency.
4
The 95% con…dence bounds are boostrapped, 1000 draws.
ECB
Working Paper Series No 1349
June 2011
15
-+63) #0-5)( !5%5)4
)42104)4 51 % 4( -0017%5-10 -0 '10:()0') 10
'1046/25-10 +3185, 8-5, )3313 &160(4
CF H<9 9IFC 5F95 =;IF9 5 G<C7? HC 7CBO89B79 <5G GCA9 G<CFHH9FA G=;B=O75BH =AD57H CB
7CBGIADH=CB
(CB9H<9@9GG 5G :CF H<9 /B=H98 -H5H9G H<9F9 =G BC @CB;FIB G=;B=O75BH 7CFF9@5H=CB
69HK99B 7CBO89B79 5B8 7CBGIADH=CB ;FCKH<
05F=5B79 897CADCG=H=CB BCH G<CKB <9F9 5@GC G<CKG
H<5H 5FCIB8 C: H<9 J5F=5B79 KCI@8 69 9LD@5=B98 6M 7CBO89B79 G<C7?G
-+63) 631 %3)%
)42104)4 51 % 4( -0017%5-10 -0 '10:()0') 10 '1046/25-10
+3185, 8-5, )3313 &160(4
16
ECB
Working Paper Series No 1349
June 2011
G K=H< H<9 G=AD@9 AC89@G K9 5@GC 588 HC CIF G9H C: :IB85A9BH5@ J5F=56@9G H<9 :CF9=;B J5F=56@9G
7CFF9GDCB8=B; HC 5G K9@@ 5G H<9 7<5B;9 =B :CF9=;B 7CBO89B79 ('
.<=G 5@@CKG IG HC 7<97? HC
K<5H 9LH9BH :CF9=;B 7CBO89B79 7CI@8 <5J9 GCA9 =AD57H CB 8CA9GH=7 7CBO89B79 K<=@9 7CBHFC@=B; :CF
H<9 :CF9=;B =BPI9B79G ;C=B; H<FCI;< H<9 @=B?5;9G 69HK99B :CF9=;B 5B8 8CA9GH=7 :IB85A9BH5@G
1<=@9
:CF H<9 /B=H98 -H5H9G 5 G<C7? HC 9IFC 5F95 7CBO89B79 <5G BC =AD57H CB B9=H<9F /
-
7CBO89B79 BCF
/
-
7CBGIADH=CB H<9 =AD57H C: 5 /
-
7CBO89B79 G<C7? <5G GCA9 G=;B=O75BH =AD57H CB H<9 9IFC
5F95 7CBO89B79 =B H<9 G<CFH FIB =;IF9 H<IG 7CBOFA=B; H<9 9L=GH9B79 C: 5 7CBO89B79 7<5BB9@
69HK99B H<9 /
-
5B8 9IFC 5F95 97CBCA=9G
-+63) 631 %3)%
)42104)4 51 % 4( -0017%5-10 -0 #! '10:()0') 10 '10:()0') 8-5, )3313 &160(4
19 IG9 B9LH H<9 <=GHCF=75@ 897CADCG=H=CB H97<B=EI9 HC 89G7F=69 H<9 F9@5H=J9 =ADCFH5B79 C: G<C7?G
HC 7CBO89B79 5B8 G<C7?G HC H<9 CH<9F :IB85A9BH5@ J5F=56@9G
G OFGH 9LD@5=B98 6M IF6=8;9 5B8
"5FF=GCB H<9 <=GHCF=75@ 897CADCG=H=CB =G 65G98 CB H<9 F9CF;5B=N5H=CB C: H<9 ACJ=B; 5J9F5;9
F9DF9G9BH5H=CB C: 1 K<9F9 1 =G 5 7C@IAB J97HCF C: H<9 9B8C;9BCIG J5F=56@9G =G H<9 ACJ=B; 5J9F5;9 A5HF=L =AD@=98
6M H<9 9GH=A5H98 7C9U7=9BHG 5B8 5 J97HCF C: =BBCJ5H=CBG H<9 BCB:CF975GH56@9 7CADCB9BHG C: 1
CF 5 D5FH=7I@5F D9F=C8 - # EI5H=CB 75B 69 KF=HH9B 5G
1 ECB
Working Paper Series No 1349
June 2011
17
K<=7< F9DF9G9BHG H<9 <=GHCF=75@ 897CADCG=H=CB
.<=G 897CADCG=H=CB <5G HKC HMD9G C: H9FAG
.<9
:5F F=;<H G=89 H9FA F9DF9G9BHG H<9 9LD97H5H=CB C: 1 ;=J9B =B:CFA5H=CB 5J5=@56@9 5H H=A9 - H<9 65G9
DFC>97H=CB C: H<9 J97HCF 1
.<9 OFGH H9FA CB H<9 F=;<H<5B8 G=89 G<CKG H<9 8=V9F9B79 69HK99B H<9
57HI5@ G9F=9G 5B8 H<9 65G9 DFC>97H=CB 8I9 HC H<9 GHFI7HIF5@ =BBCJ5H=CBG =B H<9 J5F=56@9G GI6G9EI9BH
HC D9F=C8 - H<5H =G =H G<CKG H<5H H<9 ;5D 69HK99B 5B 57HI5@ G9F=9G 5B8 =HG 65G9 DFC>97H=CBG H<9 GIA
C: H<9 K9=;<H98 7CBHF=6IH=CBG C: H<9 GHFI7HIF5@ =BBCJ5H=CBG HC H<9 =B8=J=8I5@ G9F=9G =B H<9 5B5@MG=G
.<IG H<9 57HI5@ 85H5 5H D9F=C8 - # 5F9 H<9 GIA C: H<9 65G9 DFC>97H=CB 5B8 H<9 K9=;<H98 GHFI7HIF5@
=BBCJ5H=CBG HC H<9 GMGH9A J5F=56@9G
G =B8=75H98 56CJ9 H<9G9 GHFI7HIF5@ =BBCJ5H=CBG 5F9 5GGIA98
CFH<C;CB5@ HC CB9 5BCH<9F
)J9F5@@ H<9 <=GHCF=75@ 897CADCG=H=CB 5@@CKG IG HC EI5BH=:M H<9 F9@5H=J9
=ADCFH5B79 C: GD97=O7 G<C7?G HC 957< J5F=56@9
#B CIF 75G9 H<9 <=GHCF=75@ 897CADCG=H=CB 5@@CKG
IG HC 89H9FA=B9 H<9 F9@5H=J9 =ADCFH5B79 :CF F95@ 7CBGIADH=CB 9LD9B8=HIF9G 9=H<9F =B 5 D5FH=7I@5F
EI5FH9F CF CJ9F GCA9 @CB;9F H=A9 D9F=C8 C: 5 G<C7? HC 7CBO89B79 CF HC 5 :IB85A9BH5@ J5F=56@9
=;IF9G 5B8 G<CK F9GD97H=J9@M :CF H<9 /- 5B8 H<9 9IFC 5F95 H<9 7CBHF=6IH=CBG HC H<9 89J=5H=CBG
69HK99B 57HI5@ 7CBGIADH=CB 5B8 =HG 0,65G98 :CF975GHG C: G<C7?G HC 7CBO89B79 5B8 H<9 G9H C:
:IB85A9BH5@ J5F=56@9G G 9LD97H98 H<9 7CBO89B79 G<C7?G D@5M 5 F9@5H=J9@M GA5@@ FC@9 CB
5J9F5;9 F9@5H=J9 HC G<C7?G HC :IB85A9BH5@G
"CK9J9F H<9F9 5F9 D9F=C8G 8IF=B; K<=7< 7CBO89B79
G99AG HC D@5M 5B =ADCFH5BH FC@9
-+63) -4513-'%. *13)'%45 )3313 ()'1/214-5-10 ; #0-5)( !5%5)4
&CC?=B; 5H H<9 /
-
F9GI@HG OFGH =;IF9 K9 75B BCH=79 H<5H K<=@9 H<9 7CBHF=6IH=CB C: 7CBO
89B79 G<C7?G H9B8G HC CG7=@@5H9 5FCIB8 N9FC GI7< G<C7?G <58 @5F;9F B9;5H=J9 =BPI9B79 HC :CF975GH
9FFCFG 8IF=B; J9FM GD97=O7 9D=GC89G
19 75B HFM HC A5D GI7< B9;5H=J9 7CBHF=6IH=CBG HC D5FH=7I@5F
9J9BHG H<5H 5F9 @=?9@M HC <5J9 5V97H98 7CBGIA9F 7CBO89B79
CF =BGH5B79 7CBO89B79 G<C7?G 7CB
HF=6IH9 B9;5H=J9@M HC :CF975GH 9FFCFG 8IF=B; ;9CDC@=H=75@ H9BG=CBG CF /- A=@=H5FM =BH9FJ9BH=CBG =B H<9
18
ECB
Working Paper Series No 1349
June 2011
Middle East in early 1990s and in 2001-02. Negatively contributions also appear clearly during the
2007-09 …nancial crisis (2008Q2 following Bear Stearns and 2008Q4 following Lehman Brothers).
Figure 8: Historical forecast error decomposition –Euro area
Concerning the euro area (Figure 8), con…dence shock contributions are clearly negative during
recessions and …nancial crises (like in 1992-93 or 2008-09), going therefore in the same direction as
shocks to fundamentals and even anticipating them sometimes (like in the early 1990s). Negative
con…dence shock contributions are however sometimes absorbed by positive contributions of other
shocks, like during the Asian crisis in 1997 or at the turn of the millenium. On the contrary, strong
positive contributions can also be found in 1994 at the time where contributions to shocks to
fundamentals were clearly negative. Similarly, con…dence shocks contributed positively in 2006-07,
counterbalancing negative contributions by shocks to fundamentals.
Overall, the historical contribution exercise shows that con…dence seems to matter in some
speci…c episodes, which in most cases corresponds to periods where there are large changes in
household survey indicators, like during …nancial crises or geopolitical tensions for instance.
4
Isolating periods in which con…dence matters
To capture better the fact that con…dence might play a role only in some particular circumstances,
we perform a non-linear estimation of our consumption equation, assuming that there is a certain
(unknown) threshold in con…dence index changes beyond which con…dence starts impacting consumption behaviours. We present …rst the threshold estimation approach before showing evidence
for the U.S. and the euro area consumption, both in sample and out of sample.
ECB
Working Paper Series No 1349
June 2011
19
4.1
A threshold model
In this exercise we try to test to what extent removing low frequency observations from the con…dence indicators can improve the goodness of …t of the forecasting equations. Following Desroches
and Gosselin (2002), we use the following criterion to censor observations:
conftC
=
(
0 if j conft j <
(10)
conft otherwise
This produces a series of con…dence vectors (conftC ), each of them that will have their observations censored i.e. set to zero, if their absolute change is below a determined threshold. By letting
go from zero to its highest possible value i.e. max j conft j, we produce a number of di¤erent
con…dence vectors equal to the total sample size. We then estimate our models using each one of
this vectors and look for the one with the best performance, thus being able to …nd the optimum
i.e. the threshold for which by setting all changes in con…dence to zero below this value, we
maximize the …t of the model (as described below in the text).
By following the above threshold methodology, we test the assumption that small changes in
con…dence do not matter much to explain future consumption expenditures but that large con…-
dence shocks is likely to bring some extra information beyond economic fundamentals. This can
be understood as a study of a particular type of non-linear behavior of con…dence with respect to
consumption growth or as a structural break test to the standard linear relation estimated in the
previous sections.
As before, we will estimate a series of models, …rst with con…dence alone, and then adding the
set of controls. If we use lagged consumption in the estimation, then autocorrelation of the residuals
will render the OLS estimates biased and inconsistent. As Bram and Ludvigson (1998) point out,
due to the nature of consumption and its composition between durables and non-durables, we
should expect a consumption regression to include moving average components which then render
OLS as an inadequate estimation procedure in the presence of lagged dependent variables. Instead
of adding more lags to remove the residual autocorrelation (as usually done), we prefer adding a MA
component under the argument that this is the correct speci…cation to assume once consumption
expenditures includes durables. In general, all the speci…cations we use in our estimations will be
particular cases of what are known in the literature as ARMAX, auto-regressive moving average
with external regressors models:
X t = "t +
p
X
i=1
'i X t
i
+
q
X
i=1
i "t i
+
b
X
i dt i
i=1
As said before, as long as 'i = 0; 8 i, an estimation procedure using ordinary least squares
will produce consistent estimates. The Newey-West procedure will assymptotically correct any
heteroscedastic structure in the residuals. However, as long as 'i 6= 0 , we need to use an alternative estimation method and we use an iterative procedure that minimizes a robusti…ed quadratic
20
ECB
Working Paper Series No 1349
June 2011
prediction error criterion (see Ljung, 1999). Initial parameter values are constructed resorting to a
special-four stage LS-IV algorithm.
However, regardless of the adequate estimation procedure to each of the cases, the approach
is still the same. We aim at …nding, for each model, the con…dence vector that will maximize a
measure of …t (which would be to maximize R2 for the OLS case, and minimize the Final Prediction
Error (FPE) when minimizing the robusti…ed quadratic prediction error criterion). We then report
some summary statistics concerning the estimated model and have a focus both on the timing of
the censored con…dence observations for the optimal con…dence vector and the magnitude of the
threshold below which it was optimal to set to zero all observations from the con…dence vector.
4.2
In-sample evidence
As above, we …rst estimate models relating consumption and con…dence before estimating models
that also include other fundamentals.
The …rst threshold model to be tested is similar to Eq. (1) except that we limit the lag order
to one, include a moving average component and two lags of the dependent variable. We limit the
lag order of the con…dence measure to one as this variable is subject to censoring Although having
two thresholds (one for each lag of the con…dence measure) would not be technically impossible, it
would complicate the analysis.
ln Ct =
+
2
X
ln Ct
i
+
conftC 1 +
t
+
t 1
(11)
i=i
where the upperscript C stands for "censored" i.e. some observations were set to zero. This forecast
performance of this model is tested against two alternatives: a model without con…dence and a
model where con…dence is not censored. Results are reported in Table 3 (Column A) for the United
States and Table 4 (Column A) for the euro area.
For the United States, the FPE is about 3% smaller when censoring observations according to
our threshold algorithm than when we do not censor observations at all. Both point estimates seem
to be similar and their statistical signi…cance is well above 99%. From an optimization perspective,
43 out of 101 observations are censored. The optimal threshold found is of 2:4 which from a model
…tting perspective means that it is optimal to set all con…dence changes to zero if they are found
to be below that value.
For the euro area, the number of observations that are censored in the optimal regression is of
41 (Table 4 - Column A). the FPE is about 12% smaller when censoring observations according
to our threshold algorithm than when we do not censor observations at all. The point estimate
of the coe¢ cient of the censored con…dence variable is close to its uncensored counterpart, and
statistically signi…cance is well above 99% for both, though the coe¢ cient of con…dence in the
threshold model was found to have slightly higher statistical signi…cance. The optimal threshold
was found to be of 1.4, well below the standard deviation for con…dence of 2.9204.
ECB
Working Paper Series No 1349
June 2011
21
.<9 G97CB8 H<F9G<C@8 AC89@ =BHFC8I79G H<9 G9F=9G C: 97CBCA=7 :IB85A9BH5@G H<5H <5J9 699B
IG98 69:CF9 .<9 G5A9 @5; @9B;H< =G IG98 :CF 5@@ H<9 :IB85A9BH5@ J5F=56@9G
.C F9A5=B
7CBG=GH9BH K=H< H<9 DF9J=CIG 5B5@MG=G K9 IG9 HKC @5;G
.<=G @5; CF89F =G 5@GC 7CBG=GH9BH K=H<
GH5B85F8 =B:CFA5H=CB 7F=H9F=5
G 69:CF9 7CBO89B79 A95GIF9 =G @5;;98 CB@M 6M CB9 D9F=C8
.<9
GD97=O75H=CB =G H<9 CB9 69@CK
(CH=79 H<5H 5@GC =B7@I89 HKC @5;G C: H<9 89D9B89BH J5F=56@9
@GC
K9 =B7@I89 5 ACJ=B; 5J9F5;9 7CADCB9BH 5G DF9J=CIG@M A9BH=CB98
('
,9GI@HG 5F9 F9DCFH98 =B .56@9 C@IAB :CF H<9 /B=H98 -H5H9G 5B8 .56@9 C@IAB :CF
H<9 9IFC 5F95
CF H<9 /
-
75G9 K9 OB8 H<5H H<9 H<F9G<C@8 AC89@ <5G 5B =ADFCJ9A9BH C: OH F9@5H=J9 HC H<9
AC89@ =B7@I8=B; IB79BGCF98 7CBO89B79 =B8=75HCFG
.<9 F9G=8I5@G CB9GH9D 5<958 DF98=7H=CB 9FFCFG
5F9 GA5@@9F
.<9 DC=BH 9GH=A5H9 C: H<9 79BGCF98 7CBO89B79 J97HCF 7C9U7=9BH =G ACF9 H<5B
H=A9G @5F;9F H<5B H<9 CB9 :CF H<9 IB79BGCF98 AC89@ 5B8 G=;B=O75BH 5H H<9 @9J9@
'5BM
ACF9 7CBO89B79 C6G9FJ5H=CBG 5F9 BCK 79BGCF98 5G CB@M 5F9 @9:H IB79BGCF98 K<=7< =G =B @=B9 K=H<
CIF DF9J=CIG OB8=B;G =
9
7CBO89B79 =G =B8998 BCH FC6IGH HC H<9 =B7@IG=CB C: :IB85A9BH5@G
.<9
CDH=A5@ H<F9G<C@8 =G :CIB8 HC 69 9EI5@ HC =
9
@5F;9F H<5B HKC GH5B85F8 89J=5H=CBG C: 7CBO89B79
.<9 DC=BH H<5H =B H<9 7CBH9LH C: H<=G AC89@ 9LHF9A9 7CBO89B79 7<5B;9G 5F9 H<9 CB9G H<5H A5HH9F
H<9 ACGH =G H<9F9:CF9 F9=B:CF798
G G<CKB 6M =;IF9 H<9 D9F=C8G K<9F9 7CBO89B79 =G IB79BGCF98
7CFF9GDCB8G HC H<9 69;=BB=B;G C: F979GG=CB
-+63) #0-5)( !5%5)4 1046/25-10 +3185, 3,4 %0( ',%0+) -0 '10:()0') .,4
CF H<9 9IFC 5F95 H<9 F9GI@HG 8C BCH 7<5B;9 AI7< =B @=B9 HC CIF DF9J=CIG OB8=B;G H<5H H<9
=B:CFA5H=CB 7CBH5=B98 =B 7CBO89B79 =G 7CAD5F56@M ACF9 FC6IGH HC H<9 =B7@IG=CB C: 97CBCA=7 :IB85
A9BH5@G H<5B =B H<9 /
-
75G9
.<9F9 =G CB@M 5 AC89GH =ADFCJ9A9BH =B H<9 OH C: H<9 *G F9@5H=J9
22
ECB
Working Paper Series No 1349
June 2011
to the model with uncensored con…dence indicators. The point estimate of the censored con…dence
variable slightly exceeds the uncensored one and they are signi…cant at 99%. Too many observations were left uncensored for us to try to map them into some historical events, but it is worth
noting that the time where con…dence is uncensored for the larger consecutive number of periods
is precisely the 2007-09 years.5
Figure 11: Euro area: Consumption growth (rhs) and change in con…dence (lhs).
(Grey areas show periods where con…dence is uncensored)
4.3
Out-of-sample evidence
Finally, we perform some out-of-sample analysis to check to what extent and in which circumstances
the threshold models outperform models where the con…dence vector is subject to no censoring or
not included at all. Each exercise is performed both for the United States and the euro area and
consists of one-step ahead forecasts over the period 2002Q1-2010Q2. These forecasts are obtained
using models, whose parameters are estimated using data up to t-1. Root mean square errors are
computed and used as comparison between models.
5
We have also estimated a third threshold model, which includes a foreign con…dence indicator among the explanatory variables. In the U.S. case, adding euro area con…dence as a regressor does not a¤ect the number of
censored regressions. With respect to the euro area, by adding two lags of the U.S. con…dence measure, 7 more euro
area con…dence observations are censored. Moreover, the model …t increases when compared with their counterpart
without U.S. con…dence, thus reinforcing previous …ndings of some linkages between U.S. con…dence and euro area
variables. Results are available upon request.
ECB
Working Paper Series No 1349
June 2011
23
-+63) #0-5)( !5%5)4 &41.65) *13)'%45 )33134
(CH9 'C89@ K=H< 7CB: 7CFF9GDCB8G HC H<9 H<F9G<C@8 AC89@ K=H< 79BGCF98 7CBO89B79 'C89@ K=H< 7CB:
7CFF9GDCB8G HC H<9 AC89@ K=H< IB79BGCF98 7CBO89B79 'C89@ K=H<CIH 7CB: 7CFF9GDCB8G HC H<9 AC89@ K=H<CIH
7CBO89B79
-+63) 631 %3)% &41.65) *13)'%45 )33134
(CH9 'C89@ K=H< 7CB: 7CFF9GDCB8G HC H<9 H<F9G<C@8 AC89@ K=H< 79BGCF98 7CBO89B79 'C89@ K=H< 7CB:
7CFF9GDCB8G HC H<9 AC89@ K=H< IB79BGCF98 7CBO89B79 'C89@ K=H<CIH 7CB: 7CFF9GDCB8G HC H<9 AC89@ K=H<CIH
7CBO89B79
@H<CI;< =H =G 8=U7I@H HC ;F5D<=75@@M 8=G79FB 5 AC89@ H<5H GMGH9A5H=75@@M DFC8I79G GA5@@9F
:CF975GH 9FFCFG =;IF9 :CF H<9 /B=H98 -H5H9G 5B8 =;IF9 :CF H<9 9IFC 5F95 K9 75B BCH=79 H<5H
AC89@G K=H< 7CBO89B79 G99A HC CIHD9F:CFA 8IF=B; D9F=C8G C: ;9CDC@=H=75@ H9BG=CBG @=?9 =B CF 8IF=B; H<9 OB5B7=5@ 7F=G=G
1<9B 7CADIH=B; H<9 ,'-G C: H<9 H<F99 AC89@G .56@9 H<9M
5F9 5@@ GA5@@9F H<5B H<9 GH5B85F8 89J=5H=CB C: H<9 /- 7CBGIADH=CB A95GIF9 :CF H<9 D9F=C8 IB89F
5B5@MG=G
.<9 G5A9 5DD@=9G :CF H<9 9IFC 5F95
24
ECB
Working Paper Series No 1349
June 2011
In relative terms, the threshold models result in a smaller relative RMSE (0.897 for the United
States and 0.892 for the euro area) than the model with uncensored con…dence (slightly above
1) or the simple model with no con…dence (normalized to 1). Forecast accuracy tests (based on
Clark and West, 2007) show that the threshold models provide signi…cant improvements compared
with a model without con…dence for both the U.S. and the euro area. These results show that the
improvement in forecast accuracy due to nonlinearity is statistically signi…cant.
5
Conclusions
This paper has proposed an empirical assessment of the link between consumer sentiment and
consumption expenditures for the United States and the euro area. Overall, the results show that
the consumer con…dence index can be in certain circumstances a good predictor of consumption.
In particular, out-of-sample evidence shows that the contribution of con…dence in explaining consumption expenditures increases when household survey indicators feature large changes, so that
con…dence indicators can have some increasing predictive power during such episodes. Moreover,
there is some evidence of a “con…dence channel”in the international transmission of shock, as U.S.
con…dence indices lead consumer sentiment in the euro area.
Future research includes extensions to other countries. In the euro area case, in particular, it
would be interesting to verify whether the conclusions at the area level are con…rmed at the level
of the di¤erent countries. For instance, our results contrast with those found by Al-Eyd et al.
(2008), who do not …nd consumer con…dence to be a good predictor for consumption for the three
largest euro area countries (Germany, Italy and France). It would then be worth understanding
whether aggregation could be responsible for these di¤erences. Moreover, extensions could concern
other variables. As surveys also report business sentiment, similar research could be done on
the predictive power of con…dence on investment decisions. By extension, combining consumer
and business survey data could help improve the forecast of GDP ‡uctuations.6 Finally, we have
only used the aggregate index of con…dence surveys. Various subcomponents could be used as
alternatives, as they might provide more precise information about agents’perceptions about the
future.
6
Taylor and McNabb (2007) provde some evidence for Europe that both consumer and business con…dence indicators are procyclical and generally play a signi…cant role in predicting business cycle downturns.
ECB
Working Paper Series No 1349
June 2011
25
References
Acem oglu, D, Scott, A. (1994). Consumer Con…dence and Rational Expectations: Are Agents
Beliefs Consistent With the Theory? The Economic Journal 104: 1–19.
Al-Eyd, A., Barrell, R. and Philip, E. (2008). Consumer con…dence indices and short-term forecasting of consumption. Manchester School 77: 96-111.
Bram, J, Ludvigson, SC. (1998). Does Consumer Con…dence Forecast Household Expenditure? A
Sentiment Index Horse Race. Federal Reserve Bank of New York, Economic Policy Review: 59–78.
Burbidge, J, Harrison, A. (1985). A Historical Decomposition of the Great Depression to Determine
the Role of Money. Journal of Monetary Economics 16: 45-54.
Carroll, C, Fuhrer, J, Wilcox, D. (1994). Does Consumer Sentiment Forecast Household Spending?
If So, Why? American Economic Review 84: 1397–1408.
Claveria, O, Pons, E, Ramos, R. (2007). Business and consumer expectations and macroeconomic
forecasts. International Journal of Forecasting 23: 47–69.
Clark, T.E., West, K. (2007). Approximately normal tests for equal predictive accuracy in nested
models. Journal of Econometrics 138: 291-311.
Desroches, B, Gosselin, M-A. (2002). The Usefulness of Consumer Con…dence Indexes in the United
States. Bank of Canada Working Paper No. 2002-22.
Dominitz, J, Manski, CF. (2004). How Should We Measure Consumer Con…dence? The Journal of
Economic Perspectives 18: 51-66.
Easaw, JZ, Heravi, SM. (2004). Evaluating consumer sentiments as predictors of UK household
consumption behavior: Are they accurate and useful? International Journal of Forecasting 20:
671-681.
Eppright, DR, Arguea, NM, Huth, WL. (1998). Aggregate consumer expectation indexes as indicators of future consumer expenditures. Journal of Economic Psychology 19: 215-235.
Fagan, G, Henry, J, Mestre, R. (2001). An area-wide model (AWM) for the euro area. ECB
Working Paper No. 42.
Fei, S. (2011). The con…dence channel for the transmission of shocks. Banque de France Working
Paper No. 314.
Fuhrer, JC. (1988). On the Information Content of Consumer Survey Expectations. The Review
of Economics and Statistics 70: 140-144.
Giannone, D.m Lenza, M. and Reichlin, L. (2009). Business cycles in the euro area. In Alesina, A.
and Giavazzi, F. (eds), Europe and the euro. University of Chicago Press.
Hall, RE. (1978). Stochastic Implications of the Life-Cycle/Permanent Income Hypothesis: Theory
and Evidence. Journal of Political Economy 96: 971–87.
26
ECB
Working Paper Series No 1349
June 2011
"5I;< &
.<9 #BPI9B79 C: CBGIA9F CBO89B79 5B8 -HC7? *F=79G CB H<9 /B=H98 -H5H9G
IG=B9GG M7@9
(/ /B=J9FG=HM
' 1CF?=B; *5D9FG (C
"CKF9M *
.<9 *F98=7H=J9 *CK9F C: H<9 #B89L C: CBGIA9F -9BH=A9BH
FCC?=B;G*5D9FG
CB 7CBCA=7 7H=J=HM T
%5HCB5 !
*GM7<C@C;=75@ 7CBCA=7G
(9K 3CF? @G9J=9F -7=9BH=O7 *I6@=G<=B; CAD5BM
&>IB; &
-MGH9A #89BH=O75H=CB .<9CFM :CF H<9 /G9F *F9BH=79 "5@@
&I8J=;GCB -
CBGIA9F CBO89B79 5B8 CBGIA9F -D9B8=B;
$CIFB5@ C: 7CBCA=7
*9FGD97H=J9G T
%K5B CHGCA=H=G $
.<9 IG9:I@B9GG C: 7CBGIA9F 7CBO89B79 =B :CF975GH=B; <CIG9
<C@8 GD9B8=B; =B 5B585 5 B5H=CB5@ 5B8 F9;=CB5@ 5B5@MG=G
7CBCA=7 #BEI=FM T
-A=H< '
CF975GH=B; /H=@=HM C: /% CBGIA9F -9BH=A9BH #B89L9G =B ,95@ .=A9 C CB
GIA9F -9BH=A9BH -IFJ9MG #ADFCJ9 CBGIADH=CB CF975GHG =B ,95@ .=A9 /B=J9FG=HM C: ,=7<ACB8
9D5FHA9BH C: 7CBCA=7G "CBCFG *5D9FG
.5M@CF % '7(566 ,
IG=B9GG M7@9G 5B8 H<9 ,C@9 C: CBO89B79 J=89B79 :CF IFCD9
)L:CF8 I@@9H=B C: 7CBCA=7G 5B8 -H5H=GH=7G .<FCCD 1
CBGIA9F -9BH=A9BH #HG 5IG9G 5B8 V97HG
989F5@ ,9G9FJ9 5B? C: -5B
F5B7=G7C 7CBCA=7 ,9J=9K 1=@7CL $ CF975GH=B; CADCB9BHG C: CBGIADH=CB K=H< CADCB9BHG C: CBGIA9F -9BH=
A9BH IG=B9GG 7CBCA=7G ECB
Working Paper Series No 1349
June 2011
27
Tables and charts
Table 1: Granger causality tests
United States
Euro area
conf
Notes:
ln C
conf
ln C
conf
ln C
0.27
0.06
-
0.37
0.00
-
ln Y
0.27
0.64
0.10
0.63
ln W
ln W f
ln W h
0.01
0.37
0.02
0.00
0.25
-
0.25
-
ln q
i
u
roil
0.07
0.77
0.33
0.26
0.01
0.65
0.56
0.10
0.66
0.01
0.01
0.29
0.22
0.00
0.97
0.42
conf
0.19
0.39
0.00
0.21
conf : con…dence index; C : real consumption expenditures; Y : real disposable income; W : wealth; q : real
equity prices; i: short-term interest rates;
u: unemployment rate; roil: real oil price; conf : foreign con…dence.
indicate …rst di¤erences. P-values reported for the probability of Row NOT Granger-causing Column. Analysis
performed using 4 lags. For the euro area, only total wealth is available, whereas for the United States, both
…nancial wealth (W f ) and housing wealth (W h ) are used as fundamentals.
28
ECB
Working Paper Series No 1349
June 2011
.56@9 /B=J5F=5H9 GD97=O75H=CBG
EI5H=CB
/B=H98 -H5H9G
(CH9G
('
('
('
('
('
('
('
IFC
* " . +("%
:CF :CF9=;B J5F=56@9G
CF H<9 89OB=H=CB C: H<9
J5F=56@9G G99 .56@9 BCH9G
ECB
Working Paper Series No 1349
June 2011
29
.56@9 /B=H98 -H5H9G #BG5AD@9 :CF975GH=B; 9L9F7=G9
'C89@G
1=H<CIH 1=H< =B5@ *F98=7H=CB FFCF *
1=H<CIH 7CBO89B79 9
1=H< (' 9
1=H< (' 9
9
9
9
1=H< ('
1=H< ('
C9U7=9BHG ',(+ (, (+ (' &,.+ -(-% (,+/-"(', (CH9G
* " . +("%
CF H<9 89OB=H=CB C: H<9 J5F=56@9G G99
.56@9 BCH9G
'C89@ K=H< 7CB: 7CFF9GDCB8G HC H<9 AC89@ K=H< IB79BGCF98 7CBO89B79 'C89@ K=H< 7CB:
7CFF9GDCB8G HC H<9 H<F9G<C@8 AC89@ K=H< 79BGCF98 7CBO89B79
=G H<9 J5@I9 C: H<9 H<F9G<C@8
(' &,.+
F9DCFHG H<9 GH5B85F8 89J=5H=CB C: H<9 7CBO89B79 =B89L CJ9F H<9 9GH=A5H=CB D9F=C8
30
ECB
Working Paper Series No 1349
June 2011
.56@9 IFC 5F95 #BG5AD@9 :CF975GH=B; 9L9F7=G9
'C89@G
1=H<CIH 1=H< =B5@ *F98=7H=CB FFCF *
1=H<CIH 7CBO89B79 9
1=H< (' 9
1=H< (' 9
9
9
9
1=H< ('
1=H< ('
C9U7=9BHG ',(+ (, (+ (' &,.+ -(-% (,+/-"(', (CH9G
* " . +("%
CF H<9 89OB=H=CB C: H<9 J5F=56@9G G99
.56@9 BCH9G
'C89@ K=H< 7CB: 7CFF9GDCB8G HC H<9 AC89@ K=H< IB79BGCF98 7CBO89B79 'C89@ K=H< 7CB:
7CFF9GDCB8G HC H<9 H<F9G<C@8 AC89@ K=H< 79BGCF98 7CBO89B79
=G H<9 J5@I9 C: H<9 H<F9G<C@8
(' &,.+
F9DCFHG H<9 GH5B85F8 89J=5H=CB C: H<9 7CBO89B79 =B89L CJ9F H<9 9GH=A5H=CB D9F=C8
.56@9 )IHC:G5AD@9 :CF975GH=B; 9L9F7=G9 CJ9F ++
/B=H98 -H5H9G IFC 5F95
EI5H=CB
1=H<CIH 7CBO89B79 1=H< (' 1=H< (' (',.&)-"('
,CCH '95B -EI5F9 FFCF
(CH9G 'C89@ K=H< 7CB: 7CFF9GDCB8G HC H<9 AC89@ K=H< IB79BGCF98 7CBO89B79 'C89@ K=H< 7CB:
7CFF9GDCB8G HC
H<9 H<F9G<C@8 AC89@ K=H< 79BGCF98 7CBO89B79
(',.&)-"(' F9DCFHG H<9 GH5B85F8 89J=5H=CB C: H<9
@C; 7<5B;9 =B 7CBGIADH=CB CJ9F H<9 9GH=A5H=CB D9F=C8
EI5@ DF98=7H=J9 DCK9F H9GH F9>97H98 5H F9>97H=CB =:
HGH5H <=;<9F H<5B -99 @5F? 5B8 19GH ECB
Working Paper Series No 1349
June 2011
31
Wo r k i n g Pa p e r S e r i e s
N o 1 1 1 8 / n ov e m b e r 2 0 0 9
Discretionary
Fiscal Policies
over the Cycle
New Evidence
based on the ESCB
Disaggregated Approach
by Luca Agnello
and Jacopo Cimadomo