Cointegration of Consumption and Disposable Income: Evidence from Twelve OECD Countries

Author(s): Fuchun Jin

Source: Southern Economic Journal, Vol. 62, No. 1 (Jul., 1995), pp. 77-88
Published by: Southern Economic Association
Stable URL: http://www.jstor.org/stable/1061377
Accessed: 15-12-2015 17:54 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

Southern Economic Association is collaborating with JSTOR to digitize, preserve and extend access to Southern Economic
Journal.

http://www.jstor.org

This content downloaded from 142.51.1.212 on Tue, 15 Dec 2015 17:54:59 UTC
All use subject to JSTOR Terms and Conditions
Cointegrationof Consumptionand
Disposable Income: Evidence from
Twelve OECD Countries*
FUCHUN JIN
Colgate University
Hamilton,New York

I. Introduction

One of the implications of the rationalexpectation-permanentincome hypothesis is that con-

sumptionand disposable income are cointegrated.Campbell [3], using tests that are developed
in Engle and Granger[7], concludes that this implication holds on the quarterlyaggregate data
of the U.S. He finds that although a Phillips-Perrontest could not reject the nonstationarityof
the disposable income and consumptionprocesses, an augmentedDickey-Fullertest rejects the
existence of a unit root in the residuals obtained from the least squareregressionof disposable
income on consumption.Campbelland Clarida [4] confirmthose conclusions with the quarterly
aggregatetime series data of Canada.
These tests, however, may yield differentconclusions if they are applied to data from other
countries. This is because the unit root tests in the literaturehave ratherweak power when they
are appliedto individualtime series of moderatelength, particularlywhen the alternativeis close
to the null. Indeed, Attfield, Demery and Duck [1],using quarterlyaggregatetime series data of
the U.K.,' find that Phillips-Perrontest strongly rejects (at the 1% level) the nonstationarityof
the disposableincome process. They also find no evidence (at even the 10%level) againsta unit
root in the residuals of the regression between disposable income and consumption with aug-
mented Dickey-Fullertests of 1 or 4 lags. Campbelland Clarida [4] indicate that, for the United
Kingdom, the unit root hypothesis for the disposableincome is rejectedat the 5% level, and the
hypothesis of no cointegrationbetween consumptionand disposableincome can only be rejected
at the 10% level.2
The conflicting results documentedin this literaturereveal that empirical findings of non-
stationarityof both consumptionand disposableincome, as well as cointegrationbetween them,
may simply be due to the weak power of the unit root tests that they are based upon. In this

*Theauthorwould like to thankPaulEvans,TomMichl, Bob Turnerandan anonymousrefereefor helpfulcomments.

The usual disclaimer applies.
1. Their paper, which is a critique of MacDonaldand Speight [10], correctedan error of data constructionin the
latter.
2. The dataused in Campbelland Clarida[4] are differentfrom those in Attfield,Demery, and Duck [1].The former
run from 55:1 to 84:4 while the latterrun from 66:1 to 86:3. Campbelland Clarida [4] also made some adjustmentsto
their data &la Blinder and Deaton [2].

77

This content downloaded from 142.51.1.212 on Tue, 15 Dec 2015 17:54:59 UTC
All use subject to JSTOR Terms and Conditions
78 FuchunJin

paper I use the national accounts data from twelve OECD countries to test the nonstationarity
of consumptionand disposable income as well as the cointegrationof them that is implied by
the RE-PIH. The econometric method used in this paper is based on the theory of testing unit
roots with panel datadeveloped by Levin and Lin [9]. The advantageof using panel datais that a
substantialgain in the power of the test can be obtained,even with a moderatenumberof cross
sections and time series observations.
The findingsof this paperconfirmthatconsumptionand disposableincome are cointegrated.
They also providecomparisonsbetween the conclusionsobtainedusing only individualtime series
data and those obtained using pooled data of time series and cross sections. Specifically, I find
that residualbased cointegrationtests cannot reject the hypothesis that consumptionand income
are not cointegratedwhen performedon individualtime series dataof each country.Nevertheless,
when the data are pooled as a panel with time series and cross sections, the evidence strongly
rejects the hypothesis of no cointegrationbetween consumptionand disposable income.
The rest of the paper is organizedas follows. In section II, I briefly review the cointegra-
tion of consumptionand disposable income as an implication of the RE-PIH. I also show that
this cointegrationrelationshipshould hold regardlessof whether some consumers are liquidity
constrained.In section III, I briefly review the econometric methods of testing unit roots and
cointegrationwith panel data.Dataused in the paperare describedin section IV. Empiricalresults
are presented in section V. Section VI contains the results of Monte Carlo simulationsfor the
finite sample distributionsof the test statistics used in statistical inferences. The conclusion of
the paper follows in section VII.

II. RE-PIH: Its Cointegration Implications

The starting point of this paper is the consumption function under the rational expectation-
permanentincome hypothesis;i.e.,

Ct= /3{(1 + r)At-1+ Z[1/(1 + r)]iEtWt+i} (1)

i=O

where Ct, Wt and At are respectively, consumption,labor income and asset holding in period
t. The real interest rate r is assumed to be a constant and equal to the consumer's subjective
time discount rate. Et is the mathematicalexpectation conditional on all informationavailable
to the consumer in period t. / is the marginalpropensityto consume for a unitary increase in
the consumer'spermanentincome.
Following Campbell [3], if the tth period consumptionCt can be describedby equation (1),
it should be cointegratedwith the tth period disposableincome Yt,which is equal to Wt+ rAt-1.
This can be seen from the following,

- - + +
3[(1 + r)/r]Yt Ct = /3[(1 + r)/r]Yt /3{(1 + r)At-1 Z[1/(1
i=O0
r)]iEtWt+i}

=
/3[(1 + r)/r](Wt + - /3(1 + - !3[1/(1 + r)]iEtWt+i
rAt-l) r)Atil i=O0
00 00

= -'[1/(1 + r)l]iWt- !Z[1/(1 + r)]i

EtWt+i
i=0 i=0

This content downloaded from 142.51.1.212 on Tue, 15 Dec 2015 17:54:59 UTC
All use subject to JSTOR Terms and Conditions
 COINTEGRATION OF CONSUMPTION AND INCOME 79
00

= -3 Z[1/(1 + r)]i(EtWt+i- Wt)

i=O

00oi
= -3Et{Z[1/(1 + r)]i•
AWt+j}
i=1 j=1

= -3 Etj{ [1/(1 + r)VAWt+j} [1/(1 + r)]'

j=1 i=O
00

= -3[(1+ r)/r] [1/(1 + r)]iEAWt+i. (2)

i=1

If AWt+i for i = 1, 2... are covariance stationary,then the right hand side of equation
(2) should also be stationarysince the coefficients [1/(1 + r)]' sum to 1/r. This implies that
disposableincome and consumptionare cointegratedwith cointegratingvector (1, -/3[(1 +r)/r]).
In the special case when 0 = r/(1 + r), equation (2) has a straightforwardeconomic inter-
pretation.Underthe RE-PIH,consumptionequalsthe annuityvalue of permanentincome. Saving
forecasts declines in future labor income, because only when currentincome is high relative to
permanentincome will a rationalconsumer save.
A similarresult can be obtainedif 3 = r/(1 + r), and a fraction A of total disposablelabor
income in the economy accrues to liquidity constrainedconsumers (type 1) who consume their
currentdisposable labor income, while the remainder1 - A accrues to individualswho are per-
manent income consumers (type 2), setting their consumptionlevel according to equation (1).
Under these assumptions,

= Wit = AW~
Clt

C2t = [r/(1 + r)]{(1 + r)At-1 + Z[1/(1 + r)]i'EtW2,t+i

i=O
00

= [r/(1 + r)]{(1 + r)At-I + (1 - A) Z[1/(1 + r)]iEtWt+i}

i=0

where Wit and W2t representthe parts of Wtthat accrue to the two types of consumersin period
t, and Clt and C2tare the consumptionsin period t of the two types of consumers.
Total consumptionis the sum of consumptionby both types of consumers
00

C,= Ct + C2t= AWt+ [r/(1 + r)]{(1 + r)At-I + (1 - A) Z[1/(1 + r)]iEtWt+i}.

i=0

Similar manipulationsas in equation (2) give rise to the following condition:

00

Y,- C, = -(1 - A) [1/(1 + r)]i (3)

i=1 EtAWr+i.
As a result,consumptionshouldbe cointegratedwith total disposableincomeregardlessof whether
and how many consumersare liquidity constrained.Therefore,one need not be concerned with
whether households are liquidity constrained.

This content downloaded from 142.51.1.212 on Tue, 15 Dec 2015 17:54:59 UTC
All use subject to JSTOR Terms and Conditions
80 FuchunJin

III. Review of the Econometric Methodology

It is well known in the literaturethat traditionaltests for unit roots in univariatetime series,
such as the Dickey-Fuller[5; 6; 8] and Phillips-Perron[14;16] tests, have very low power against
local stationaryalternative.Levin and Lin [9] propose a method to test for unit roots in panel
data. They show that their test statistics are asymptoticallynormallydistributedand that there is
a dramaticimprovementin power over the standardunivariatetests. Theirprocedureis described
as follows.
Considera group of N countries observed annuallyover T years. We assume that for each
country i, Zit follows the simplest case of an AR(1) in t, and has a country specific effect r7i,i.e.,

, + + eit i = 1, 2,..., N; t = 1, 2,..., T.

Zit = pZi,t-1 (4)
7i
Assume first that Eitare i.i.d. with E (eit) = 0 and var(eit) = a2 for all 1 < i < N, 1 < t < T,
and that E eit 2+A < +oo for some A > 0 for all i, t. Estimation of equation (4) by the fixed
effect model yields
N T N T
P= {- -(Zit - Zi)(Zi,t-1- i)}/{ (Zi,t-1- i)2
i=1 t=1 i=1 t=1

where
T
= Zit.
Zi (1/T)1-
t=1

Levin andLin [9]proved thatunderthe null

hypothesis p 1, i'(1- 1)- (Vi•/7T/P8rT) of = T
is asymptotically distributed as normal with mean 0 and variance 10.2, and -
1i.25[t, (iVP7T/
I-t/8T)] is asymptotically distributed as standard normal when N - oc and T 00,3 where
N T

tp
= [ (Zi,,t1z- )2]1/2( 1)/
i=1 t=1

and
N T
& = {(1/NT) E - - - Z)]21/2
-[(Zit Zi) p(Zit1
i=1 t=1

Furthermore,if ViN/T -* 0, then T(•W 3 - 1) + 3vli is asymptoticallydistributedas normal

with mean 0 and variance10.2, while 1/.25tp + v/1.875v is asymptoticallydistributedas stan-
dardnormal.In other words, the presence of individualspecific fixed effects causes the unit root

3. Thedefinitions
for/L7T and1L8Tare
T

L7T
=
E{[1/(To2)] - = -1/2 + o(T)
-(Zi,-t Zi)Ei,
t=1
T
= - = 1/6 + o(T).
LrT E{[1/(T2O2)] Z(Zi,t-1 i)2}
t=1

See LevinandLin [9]for moredetail.

This content downloaded from 142.51.1.212 on Tue, 15 Dec 2015 17:54:59 UTC
All use subject to JSTOR Terms and Conditions
 COINTEGRATIONOF CONSUMPTIONAND INCOME 81

test statisticsto have a non-centralasymptoticnormaldistributionwith a variancesmallerthan 1.

The downwardshift of the mean of this distributionis proportionalto v/N.
Levin and Lin [9] have also proved that those asymptotic properties are independentof
whethera constantintercept,a time trend,or time specific fixed effects are includedin the model,
and of what the values of the individualspecific effects are. By Monte Carlo simulation,Levin
and Lin [9] tabulatedcriticalvalues for the purposeof testing unit root in finite samplesof various
N and T.
In the more general case when 6it are serially correlated,Levin and Lin [9] suggest either
to include AZi,t-k(k = 1, 2,..., K) (a la the Augmented Dickey-Fuller test), or to correct for
the serial correlationdirectly by a nonparametricapproach,using certain window sizes and lag
length (a la the Phillips-Perrontest). It is noted in their paper that the asymptoticpropertiesof
p under the null will be independentof the serial correlationsin Eitafter these corrections.
In this paper, the following general model is utilized to test unit roots and cointegration:
R

60 + 61it+i +Vt + + +Eit

AZit= pZi,t_ Z
j=1
OjAZi,t-j

where i = 1, 2,...N, t = 1, 2,..., T. 61t is a linear time trend and vt is the time specific fixed
effect. p is estimatedby a panel transformationof the data,which subtractsthe individualmeanand
time specific mean from all variablesinvolved in the above equation,followed by an estimationof
R

+
AZit = pZt-1 + Z OAZt
j
j=1 -it
where variableswith a tilde are those after the panel transformation.
The asymptotic distributionof the t-statistic for 3 follows that of the non-centralnormal
describedabove since it is not affected by inclusion of 60o,61t, vt and AZi,t-j (j = 1,..., R). The
critical values for the finite sample size of the data set used in this paper are obtainedby Monte
Carlo simulation.

IV. The Data

Data for the following twelve OECDcountriesareused in my empiricalanalysis:Australia(AUS),

Austria(AUT), Belgium (BEL), Canada(CAN), Finland(FIN), Germany(GER),Greece (GRE),
Japan(JPN), Norway (NOR), Switzerland(SWI), the United Kingdom (UKD), and the United
States (USA). The data necessary for this study are not generally availablefor the other OECD
countries.
Data on the following variables are obtained from the computer tapes of OECD [12; 13].
They are annualand from 1960 to 1988 for all twelve countries (The names in the parentheses
are the ones used in the data tapes from which they are extracted):
(a) Private final consumption expenditure,valued at current prices (POPC); (b) National
disposable income, valued at currentprices (NDI); (c) Generalgovernmentconsumptionexpen-
diture, valued at currentprices (POGC); (d) Mid-yearestimates of population (POP); (e) Price
indices of privatefinal consumptionexpenditure(IDXPRIPC,1985 = 100); (f) Price indices of
gross domestic product (IDXPRIGDPE,1985 = 100); (g) Price indices of general government

This content downloaded from 142.51.1.212 on Tue, 15 Dec 2015 17:54:59 UTC
All use subject to JSTOR Terms and Conditions
82 FuchunJin

consumptionexpenditure(IDXPRIGC,1985 = 100); (h) Generalgovernmentnet saving, valued

at currentprices (SAVGG).
Real per capitaprivatefinalconsumptionexpenditure(C,) is constructedby dividingnominal
private final consumptionexpenditure(POPC)by both the price index of privatefinal consump-
tion expenditure(IDXPRIPC)and mid-yearestimatesof population(POP). Real per capitaprivate
disposable income (Y,) is constructedby dividing real private disposable income by the mid-
year estimates of population (POP), where the real private disposable income is real national
disposable income less real general governmentdisposable income.4
A potentialproblemwith the dataused in this paperexists because of the differencebetween
consumptionand consumptionexpenditure.This problem,however, is reducedto certain extent
by two featuresof the data.Annualdataareused insteadof datafrom more frequentobservations;
it is reasonableto assume that consumptionis close to consumptionexpenditurein a year's time,
even thoughsome goods will last longer thanone year. Aggregationacrosshouseholdsalso helps
to reduce the extent of the difference between consumptionand consumptionexpenditure;the
unusuallyhigh consumptionspending of some households in a particularyear, when they make
durableconsumptionpurchases,are likely to be offset in the aggregatedata by the low spending
of others that have made durableconsumptionpurchasesearlier.In the rest of this paper, I will
take privateconsumptionexpenditureas privateconsumptionwithout furtherexplanation.

V. Empirical Results

For the sake of comparison,I first list in TablesI and II the results of Dickey-Fuller(DF) tests,
AugmentedDickey-Fuller(ADF) tests andPhillips-Perrontests for unitroots in per capitaprivate
disposableincome and per capitaprivateconsumptionfor twelve OECDcountries.The ADF tests
include one to threelags of A Ytin TableI, and one to threelags of ACt in TableII. In both tables,
one to fourth order Newey-West [11]corrections are used in the Phillips-Perrontest statistics.
Furthermore,all tests are performedfor equationsboth with and without a linear time trend.
The results in both tables are overwhelminglyin favor of the unit root hypothesis in dispos-
able income and consumption.With only one exception,the values of all test statisticsin TableI
are insignificantat the 10%level. In TableII, only one test statisticis significantat the 5% level
and only two are significantat the 10%level. Appreciablymore of these test statisticsshouldhave
been significantmerely by chance: three at the 5% level and six at the 10%level in each table.
TableIII lists the resultsof the cointegratingregressionsfor all twelve countries.The residu-
als of these regressionsare subsequentlyused in testing cointegrationbetween consumptionand
disposable income. The results of Engle-Granger(EG) test, AugmentedEngle-Granger(AEG)
test and Phillips-Ouliaris[15]test are listed in TableIV. One to three lags of At are included in
the AugmentedEngle-Grangertest, while one to fourth orderNewey-Westcorrectionsare used
in the Phillips-Ouliaristest.
The results in Table IV indicate that only in three of the twelve countries is there strong
evidence againstno cointegrationbetween consumptionand disposableincome. The EG test and

4. Real nationaldisposable income and real general governmentdisposable income are obtained, respectively,by
dividing nominalnationaldisposableincome (NDI) by price indices for gross domestic product(IDXPRIGDPE),and di-
viding nominalgeneralgovernmentdisposableincome by price indices for generalgovernmentconsumptionexpenditure
(IDXPRIGC),where the nominal general governmentdisposableincome is the sum of general governmentconsumption
expenditure(POGC) and general governmentsaving (SAVGG).

This content downloaded from 142.51.1.212 on Tue, 15 Dec 2015 17:54:59 UTC
All use subject to JSTOR Terms and Conditions
 COINTEGRATION
OF CONSUMPTION
AND INCOME 83

TableI. Dickey-Fuller,
AugmentedDickey-FullerandPhillips-Perron
Test:PerCapitaPrivate
DisposableIncome(Y,),TwelveCountries,1960-1988,Annual
DF and ADF Tests Phillips-PerronTests
Lags Lags
Trend 0 1 2 3 1 2 3 4
AUS No -1.50 -1.91 -1.54 -1.25 -1.61 -1.58 -1.60 -1.62
Yes -1.19 -1.23 -1.50 -1.21 -1.22 -1.30 -1.30 -1.29
AUT No 0.21 -0.19 -0.32 -0.51 0.20 0.24 0.19 0.17
Yes -1.95 -1.83 -1.37 -2.95 -2.14 -2.14 -2.26 -2.32
BEL No -0.06 -0.24 -0.37 -0.55 -0.02 -0.01 -0.00 -0.01
Yes -2.12 -1.70 -1.81 -1.68 -2.23 -2.30 -2.33 -2.39
CAN No 1.17 0.57 0.96 0.62 1.11 1.20 1.29 1.38
Yes -2.28 -2.31 -2.35 -2.49 -2.50 -2.48 -2.44 -2.40
FIN No -0.03 0.15 0.04 0.35 -0.04 0.04 0.13 0.22
Yes -2.59 -3.10 -2.56 -2.29 -2.88 -2.83 -2.73 -2.64
GER No 0.31 -0.14 -0.30 -0.66 0.27 0.28 0.29 0.30
Yes -2.00 -2.31 -2.25 -2.33 -2.28 -2.34 -2.37 -2.37
GRE No -1.11 -1.16 -1.88 -1.66 -1.18 -1.26 -1.28 -1.29
Yes -1.48 -1.33 -0.98 -1.14 -1.57 -1.50 -1.52 -1.55
JPN No -0.86 -1.08 -1.54 -1.75 -0.90 -0.94 -0.92 -0.92
Yes -1.17 -1.17 -0.78 -1.44 -1.30 -1.25 -1.37 -1.39
NOR No 0.88 0.55 0.95 0.90 0.86 0.95 1.00 1.14
Yes -1.62 -2.33 -2.03 -2.72 -1.87 -1.87 -1.87 -1.78
SWI No -1.07 -0.87 -1.26 -1.04 -1.11 -1.15 -1.17 -1.19
Yes -2.25 -2.32 -1.96 -2.32 -2.46 -2.38 -2.39 -2.36
UKD No 0.55 0.05 0.37 0.90 0.41 0.54 0.77 1.04
Yes -2.17 -3.56* -2.05 -2.46 -2.56 -2.50 -2.33 -2.16
USA No 0.06 -0.09 0.09 -0.05 0.01 0.04 0.10 0.18
Yes -2.07 -2.90 -2.42 -2.48 -2.41 -2.40 -2.29 -2.13
at the 10%level.
*Significant

AEG test with 1 lag are significantat the 5% level for Greece and U.S., while the EG test for
Austria,AEG test with 2 lags for Greece and AEG test with 4 lags for US are significantat the
10% level. As for the Phillips-Ouliaristest, Austria, Greece and U.S. are all significant at the
5% level, regardlessof the numberof lags included in the Newey-Westcorrection.These results
conform well with the conflicting ones documentedby Campbell [3], Campbelland Clarida [4]
and Attfield, Demery and Duck [1],thus providingstrong motivationfor investigatingthe issues
using panel data.
Formalevidence of unit roots in consumptionanddisposableincome, andtheircointegration,
are provided in Tables V, VI and VII. In TableV,

R
+
Yit = PY, t-1 f + Zit
j=1 Yit-J

is estimated with R = 1, 2, 3 and 4. Similar panel estimation is applied to consumption and

This content downloaded from 142.51.1.212 on Tue, 15 Dec 2015 17:54:59 UTC
All use subject to JSTOR Terms and Conditions
84 FuchunJin

TableII. Dickey-Fuller,
AugmentedDickey-Fuller Test:PerCapitaPrivate
andPhillips-Perron
Consumption (C,), TwelveCountries,1960-1988,Annual
DF and ADF Tests Phillips-PerronTests
Lags Lags
Trend 0 1 2 3 1 2 3 4
AUS No -1.10 -1.70 -1.52 -1.63 -1.09 -1.13 -1.18 -1.21
Yes -1.03 -1.67 -0.78 -0.89 -1.28 -1.22 -1.16 -1.13
AUT No -0.05 -0.05 -0.28 -0.32 -0.02 -0.00 -0.01 -0.03
Yes -2.03 -1.66 -1.53 -1.78 -2.09 -2.13 -2.20 -2.29
BEL No -0.42 -0.74 -0.72 -0.76 -0.44 -0.44 -0.45 -0.45
Yes -1.12 -1.48 -1.57 -2.46 -1.38 -1.48 -1.60 -1.65
CAN No 1.02 0.13 0.13 0.10 0.84 0.73 0.69 0.68
Yes -1.78 -2.28 -2.79 -2.75 -2.10 -2.24 -2.30 -2.31
FIN No 0.74 0.68 0.70 0.64 0.66 0.69 0.65 0.65
Yes -1.34 -2.04 -1.62 -2.74 -1.74 -1.74 -1.84 -1.84
GER No -0.38 -0.44 -0.58 -0.97 -0.40 -0.40 -0.40 -0.40
Yes -1.42 -2.91 -1.82 -2.01 -1.84 -1.94 -1.94 -1.88
GRE No -1.79 -1.68 -2.11 -2.50 -1.80 -1.81 -1.70 -1.66
Yes -0.27 -0.50 -0.33 -1.34 -0.36 -0.34 -0.47 -0.53
JPN No -0.72 -0.65 -1.08 -1.03 -0.72 -0.74 -0.74 -0.75
Yes -1.45 -1.75 -1.55 -1.91 -1.64 -1.63 -1.64 -1.64
NOR No -0.22 -0.41 0.21 0.75 -0.27 -0.24 -0.19 -0.14
Yes -2.66 -3.99** -3.27* -2.65 -2.98 -2.91 -2.75 -2.62
SWI No -1.89 -1.19 -1.54 -1.49 -1.73 -1.72 -1.79 -1.90
Yes -1.35 -2.09 -1.40 -1.59 -1.61 -1.63 -1.59 -1.52
UKD No 2.88 1.69 1.91 2.07 2.51 2.48 2.57 2.76
Yes 1.15 -0.17 0.62 0.50 0.62 0.69 0.96 1.38
USA No 0.53 -0.11 0.29 0.12 0.41 0.44 0.53 0.64
Yes -1.94 -3.32* -2.43 -2.47 -2.39 -2.43 -2.33 -2.20
at the 10%level.
*Significant
at the 5%level.
**Significant

of Yt= a + ,Ct + et, OLS,TwelveCountries,1960-1988,Annual

TableIII. Estimation

& t(&) /3 t(/3)

AUS -217.19 -0.77 1.12 29.59
AUT -18557.14 -14.50 1.34 84.24
BEL -26149.88 -4.69 1.23 58.45
CAN -1434.62 -6.56 1.30 50.79
FIN 1080.20 1.19 1.03 33.39
GER 753.53 4.22 1.09 83.72
GRE -22427.25 -4.83 1.34 67.90
JPN -248050.89 -15.66 1.37 99.72
NOR -1769.58 -1.33 1.17 39.62
SWI 481.16 0.87 1.20 40.49
UKD 112.54 0.99 1.07 29.90
USA 350.15 2.52 1.08 69.28

This content downloaded from 142.51.1.212 on Tue, 15 Dec 2015 17:54:59 UTC
All use subject to JSTOR Terms and Conditions
 COINTEGRATIONOF CONSUMPTIONAND INCOME 85

Table IV. Engle-Granger,AugmentedEngle-Grangerand Phillips-OuliarisTest for Cointegration(At),

Twelve Countries, 1960-1988, Annual

EG andAEGTests Tests
Phillips-Ouliaris
Lags Lags
0 1 2 3 1 2 3 4
AUS -2.39 -2.20 -2.20 -1.64 -2.50 -2.55 -2.51 -2.46
AUT -3.40* -2.84 -3.19 -2.93 -3.57** -3.61** -3.57** -3.51**
BEL -2.55 -1.88 -1.88 -1.89 -2.57 -2.64 -2.69 -2.68
CAN -1.86 -1.95 -1.88 -1.84 -2.00 -2.00 -1.99 -1.94
FIN -2.00 -2.04 -2.22 -2.05 -2.13 -2.21 -2.21 -2.20
GER -2.44 -2.43 -3.15 -2.54 -2.68 -2.80 -2.76 -2.66
GRE -3.69** -3.71** -3.49* -3.03 -3.88** -3.80** -3.70** -3.63**
JPN -2.32 -2.52 -2.70 -2.10 -2.56 -2.60 -2.47 -2.37
NOR -2.59 -2.37 -1.50 -1.34 -2.79 -2.67 -2.61 -2.59
SWI -2.95 -2.79 -1.82 -2.13 -3.12 -3.03 -3.06 -3.06
UKD -1.44 -2.34 -1.69 -1.44 -1.87 -1.89 -1.77 -1.59
USA -3.73** -4.22** -3.08 -3.48* -3.90** -3.85** -3.83** -3.83**
*Significantat the 10%level.
**Significantat the 5% level.

Table V. Testing Unit Root in Disposable Income with Panel Data

lags of A?it included

1 2 3 4

Yit-1 -0.0029 (-0.90) -0.0035 (-1.09) -0.0048 (-1.71) -0.0045 (-1.50)

A 0.7434 (18.98) 0.4805 (8.69) 0.2716 (5.39) 0.2287 (3.74)
A iyt-l 0.3589 (6.50) 0.0854 (1.63) 0.0798 (1.48)
it-2
A 0.5712 (11.33) 0.5494 (10.13)
'i,t-3
A Yit-4 0.0689 (1.11)

Table VI. Testing Unit Root in Consumptionwith Panel Data

lags of ACit included

1 2 3 4

Ci,t-1 -0.0003 (-0.11) -0.0003 (-0.11) 0.0000 (0.02) 0.0008 (0.31)

0.8389 (24.79) 0.6206 (10.94) 0.4923 (9.05) 0.4366 (7.05)
ACi,t-1
ACi,t-2 0.2781 (4.84) 0.0122 (0.20) 0.0150 (0.24)
ACi,t-3
0.4537 (8.23) 0.3954 (6.33)
ACi,t-4
0.1074(1.71)

the residual of the cointegratingregression, with their results presented in Table VI and VII
respectively.
The estimates of the p's in all three regressions are listed in the first rows of each table,
with the tp statistics for them listed in the parenthesisimmediately following. These tp's have
been appropriatelycorrected for loss of the degree of freedom due to panel transformationof
the original data. It is obvious that t,'s for disposable income are all negative, two of the t,'s for

This content downloaded from 142.51.1.212 on Tue, 15 Dec 2015 17:54:59 UTC
All use subject to JSTOR Terms and Conditions
86 FuchunJin

TableVII. TestingCointegration
betweenConsumption
andDisposableIncomewithPanelData

lags of /Mit
included

1 2 3 4
(-9.11) -0.6561 (-10.00) -0.6189 (-8.01) -0.6100 (-7.00)
i,t-1 --0.5217
0.2488 (4.25) 0.3453 (5.46) 0.3243 (4.59) 0.3289 (3.95)
Atl
aUi,t-2 0.2536 (4.17) 0.2300 (3.30) 0.2166 (2.80)
Ui,t-3 -0.1021 (-1.57) -0.1164 (-1.57)
Eai,t-4 0.0561 (0.84)

consumptionare negative while the other two are positive. More importantly,all of them have
small absolute values.
The tP's for the residuals of the cointegratingregressions are, however, all negative with
large absolutevalues. Accordingto the Monte Carlo simulationof Levin and Lin [9], the critical
values of tP for T = 25 and N = 10 are -5.94 for the 1% significance level, -5.42 for 5%,
and -5.14 for 10%. By these critical values, we cannot reject the unit root hypothesis in either
consumptionor disposableincome at any significancelevel. We can, nevertheless,stronglyreject
the hypothesis of no cointegrationbetween the two.
The finite sample critical values of Levin and Lin [9] used above are obtained in a model
with no lagged differenceterms. Furthermore,because the residualsare generatedby the cointe-
grating regression, unit root hypothesis will be rejected too often if the critical values for raw
data are used to test for cointegrationbased on the generatedresiduals.For these reasons, I will
use Monte Carlo experimentsto get finite sample critical values in the next section.

VI. Monte Carlo Simulations

Three sets of Monte Carlo simulationsare performedin this paper to get the finite sample dis-
tributionof the test statistics.The critical values, which are obtainedin models calibratedto the
data employed in this paper, are then used to make more appropriateinferences in the unit root
and cointegrationtests with panel data.
In the first simulation,I obtain the critical values for testing unit roots in disposableincome
with panel data. This is achieved by first estimatingthe model under the null hypothesis that
p = 0, i.e.,
R
A =i +
~t +0Ai,t-j it.
j=1

The null model above, with its coefficients 0*'s calibratedto the data set of this paper, are used
along with 10,000 draws of samples to obtain simulateddata. Each draw of the iiit'scome from
a standardnormalrandomnumbergenerator.The simulateddata, after appropriatepanel trans-
formation,are then used to estimate the alternativemodel
R
t = + Z
+Ri, t-j +it*,.
Pit-1
j=1

The resultsof the tp'sprovidean approximatedistributionof the statisticfor testingunitroots with

This content downloaded from 142.51.1.212 on Tue, 15 Dec 2015 17:54:59 UTC
All use subject to JSTOR Terms and Conditions
 OFCONSUMPTION
COINTEGRATION ANDINCOME 87

TableVIII.Results
of MonteCarloSimulations
forFiniteSample
Critical
Values
DisposableIncome Consumption Residual
Minimum -6.83 -7.34 -8.99
1% -5.65 -5.59 -6.55
5% -5.07 -4.99 -5.92
10% -4.76 -4.66 -5.58
25% -4.23 -4.14 -5.03
Median -3.66 -3.57 -4.44
75% -3.11 -3.01 -3.82
90% -2.61 -2.52 -3.19
95% -2.32 -2.24 -2.82
99% -1.80 -1.69 -2.06
Maximum -0.84 -0.27 0.75

panel data. Column 1 in Table VIII summarizesthe results of this simulationfor the particular
case of R = 1. In the simulation,startingvalues, Yi,i and Yi,2,are taken to be zero.
In the second simulation,I obtain the critical values for testing unit root in consumption
with panel data, with the same method used in the first simulation.The resultsfor the case when
R = 1 are summarizedin column 2 of Table VIII.
In the third simulation,I obtain the critical values for testing cointegrationbetween con-
sumptionand disposable income with panel data. This is achieved by first estimatingthe model
for consumptionunder the nonstationarynull hypothesis that p = 0, i.e.,
R

,it= j=1 +,t-jit,

and the model for cointegratingresidualunderthe nonstationarynull hypothesis that p = 0, i.e.,
S

j=1

The null models above, with theircoefficients and (f's calibratedto the data set of this paper,
are used along with 10,000 drawsof samples to ii'sobtain simulateddata. Each draw of the
iiit'sand
,it's, assumed to be independent of each other, come from a standard normal random number
generator.Note thatthe simulateddata for Yitare producedthroughthe cointegratingregressions
listed in Table III.
The simulateddata for Yitand Citare first used to obtain the cointegratingresiduals.These
residuals are then appropriatelysubtractedby their individual as well as time specific means.
Finally they are used to estimate the alternativemodel
S
= +
1jit.
ait P6i,t-i Z
j=1
?J'zi,t- +

A total of 10,000 sets of data provide an approximatedistributionof the tP statistic for testing
cointegrationwith panel data. The results are listed in column 3 of Table VIII.
The criticalvalues listed in TableVIII arenot very far from those of Levin and Lin [9], espe-
cially those for testing unit roots in the raw data.The critical values for testing cointegrationare,

This content downloaded from 142.51.1.212 on Tue, 15 Dec 2015 17:54:59 UTC
All use subject to JSTOR Terms and Conditions
88 FuchunJin

as expected, generallylargerin absolutevalue. However, we can still comfortablyreject the null

of no cointegrationbetween consumptionand disposableincome at all conventionalsignificance
levels.

VII. Concluding Remarks

This paper employs nationalaccounts data from twelve OECD countries to test the nonstation-
arity of consumptionand disposable income, as well as their cointegration.Using the data as
a panel, tests are unable to reject the unit root hypothesis for either consumptionor disposable
income. When residual-basedcointegrationtests are performed on individual time series, one
cannot reject the null hypothesis that consumptionand disposable income are not cointegrated;
in comparison,when the data are deployed as a panel, the evidence stronglyrejects the null hy-
pothesis of no cointegration.I thereforeconclude that the results from individualcountries are
due to small sample size, and that consumptionand disposable income are indeed cointegrated
as implied by the rationalexpectation-permanentincome hypothesis.

References
1. Attfield,C. L. F., D. Demery and N. W. Duck, "Savingand RationalExpectations:Evidence for the U. K." The
EconomicJournal,December 1990, 1269-76.
2. Blinder, Alan S. and Angus S. Deaton, "The Time Series ConsumptionFunctionRevisited."BrookingsPapers
on EconomicActivities,1985:2, 465-511.
3. Campbell,John Y., "Does Saving Anticipate Declining LaborIncome? An AlternativeTest of the Permanent
Income Hypothesis."Econometrica,November 1987, 1249-73.
4. - and RichardH. Clarida. "Savingand PermanentIncome in Canadaand the United Kingdom,"in Eco-
nomic Effects of the GovernmentBudget, edited by ElhananHelpman et al. Cambridge,Mass.: MIT Press, 1988, pp.
122-41.
5. Dickey, David A. and Wayne A. Fuller, "Distributionof the Estimatorsfor AutoregressiveTime Series with
a Unit Root." Journalof the AmericanStatisticalAssociation,June 1979, 427-31.
6. - , "LikelihoodRatio Statistics for AutoregressiveTime Series with a Unit Root." Econometrica,June
1981, 1057-72.
7. Engle, Robert F and Clive W. J. Granger,"Co-integrationand ErrorCorrection:Representation,Estimation,
and Testing."Econometrica,March 1987, 251-76.
8. Fuller, Wayne A. Introductionto StatisticalTimeSeries. New York:John Wiley, 1976.
9. Levin, Andrewand Chien-FuLin. "UnitRoot Tests in Panel Data: Asymptotic and Finite-SampleProperties."
University of Californiaat San Diego, Departmentof Economics WorkingPaper 92-23, May 1992.
10. MacDonald,Ronaldand Alan E. H. Speight, "Consumption,Saving and RationalExpectations:Some Further
Evidence for the U. K." The EconomicJournal,March 1989, 83-91.
11. Newey, Whitney K. and KennethD. West, "ASimple, Positive Definite, Heteroskedasticityand Autocorrelation
ConsistentMatrix."Econometrica,May 1987, 703-8.
12. Organizationfor Economic Co-operationand Development,Departmentfor Economics and Statistics.National
Accounts VolumeI: Main Aggregates.Paris:OECD, 1990.
13. - . NationalAccounts VolumeII: Detailed Tables.Paris:OECD, 1990.
14. Phillips, Peter. C. B., "Time Series Regression with Unit Roots." Econometrica,March 1987, 277-301.
15. - and Sam Ouliaris, "AsymptoticPropertiesof ResidualBased Tests for Cointegration."Econometrica,
January1990, 165-93.
16. - and Pierre Perron, "Testingfor a Unit Root in Time Series Regression."Biometrica,1988, 335-46.

This content downloaded from 142.51.1.212 on Tue, 15 Dec 2015 17:54:59 UTC
All use subject to JSTOR Terms and Conditions