Journal of Econometrics 44 (1990) 215-238.

North-Holland

SEASONAL INTEGRATION AND COINTEGRATION”

S. HYLLEBERG
Ur~ic~rsi&
of Aarhus, DK-00 Aurhus, Denmurk

R.F. ENGLE and C.W.J. GRANGER

Utrirre,siq of CuliJorniu San Diego, La Jollu. CA 92037, USA

B.S. YOO
Pe~~r~s$w~~iuState University, fJniuersi<v Park, PA, USA

This paper develops tests for roots in linear time series which have a modulus of one but which
correspond to seasonal frequencies. Critical values for the tests are generated by Monte Carlo
methods or are shown to be available from Dickey-Fuller or Dickey-Hasza-Fuller critical values.
Representations for multivariate processes with combinations of seasonal and zero-frequency unit
roots are developed leading to a variety of autoregressive and error-correction representations,
The techniques are used to examine cointegration at different frequencies between consumption
and income in the U.K.

1. Introduction
The rapidly developing time-series analysis of models with unit roots has
had a major impact on econometric practice and on our understanding of the
response of economic systems to shocks. Univariate tests for unit roots were
first proposed by Fuller (1976) and Dickey and Fuller (1979) and were applied
to a range of macroeconomic data by Nelson and Plosser (1982). Granger
(1981) proposed the concept of cointegration which recognized that even
though several series all had unit roots, a linear combination could exist which
would not. Engle and Granger (1987) present a theorem giving several repre-
sentations of cointegrated series and tests and estimation procedures. The
testing is a direct generalization of Dickey and Fuller to the hypothesized
linear combination.
All of this work assumes that the root of interest not only has a modulus of
one, but is precisely one. Such a root corresponds to a zero-frequency peak in

*The research was carried out while the first author was on sabbatical at UCSD and the last
author was completing his dissertation. The authors are indebted to the University of Aarhus,
NSF SESg7-05884, and SESW-04669 for financial sunnort. The data will be made available
through the Inter-university Consortium for Political add Social Research at the University of
Michigan.

0304-4076/90/$3.5OQ1990, Elsevier Science Publishers B.V. (North-Holland)

216 S. ff,~lleherg et (II., Seusonul integration und cointegrution

the spectrum. Furthermore, it assumes that there are no other unit roots in the
system. Because many economic time series exhibit substantial seasonality.
there is a definite possibility that there may be unit roots at other frequencies
such as the seasonals. In fact, Box and Jenkins (1970) and the many time-series
analysts influenced by their work implicitly assume that there are seasonal unit
roots by using the seasonal differencing filter.
This paper describes in section 2 various classes of seasonal processes and in
section 3 sets out to test for seasonal unit roots in time-series data both in the
presence of other unit roots and other seasonal processes. Section 4 defines
seasonal cointegration and derives several representations. Section 5 gives an
empirical example and section 6 concludes.

2. Seasonal time-series processes

Many economic time series contain important seasonal components and

there are a variety of possible models for seasonality which may differ across
series. A seasonal series can be described as one with a spectrum having
distinct peaks at the seasonal frequencies ws = 2rj/s, j = 1,. . . , s/2, where s
is the number of time periods in a year, assuming s to be an even number and
that a spectrum exists. In this paper, quarterly data will be emphasised so that
s = 4, but the results can be naturally extended in a straightforward fashion to
monthly data, for example.
Three classes of time-series models are commonly used to model seasonality.
These can be called:

(a) Purely deterministic seasonal processes,

(b) Stationary seasonal processes,
(c) Integrated seasonal processes,

Each is frequently used in empirical work often with an implicit assumption

that they are all equivalent. The first goal of this paper is to develop a testing
procedure which will determine what class of seasonal processes is responsible
for the seasonality in a univariate process. Subsequently this approach will
deliver multivariate results on cointegration at seasonal frequencies.
A purely deterministic seasonal process is a process generated by seasonal
dummy variables such as the following quarterly series:

x,=p, where p*= m. + Wit + mG%, + m3&. (2.1)

Notice that this process can be perfectly forecast and will never change its
shape.
 S. Ffvlieherg et al., Seasonal integration and cointegruiion 21-l

A stationary seasonal process can be generated by a potentially infinite

autoregression

‘P(B)x,= e,, E, i.i.d.,

with all of the roots of q(B) = 0 lying outside the unit circle but where some
are complex pairs with seasonal periodicities. More precisely, the spectrum of
such a process is given by

which is assumed to have peaks at some of the seasonal frequencies w,. An

example for quarterly data is

x, = p-4 + Et.

which has a peak at both the seasonal periodicities 7r/2 (one cycle per year)
and 7~ (two cycles per year) as well as at zero frequency (zero cycles per year).
A series x, is an integrated seasonal process if it has a seasonal unit root in
its autoregressive representation. More generally it is integrated of order d at
frequency 8 if the spectrum of x, takes the form

for w near B. This is conveniently denoted by

x, - Md).
The paper will concentrate on the case d = 1. An example of an integrated
quarterly process at two cycles per year is

x, = -x,-1 + E,, (2.2)

and at one cycle per year it is

x, = _X,_? + E,. (2.3)

The very familiar seasonal differencing operator, advocated by Box and

Jenkins (1970) and used as a seasonal process by Grether and Nerlove (1970)
and Bell and Hillmer (1985) for example, can be written as

(1 - B4)x, = E, = (1 - B)(l + B + B* + B3)x,

= (1 - B)(l + B)(l + B2)x,

= (1 - B)S( B)x,, (2.4)

218 ul., Semonal integralion and cointegrution

which therefore has four roots with modulus one: one is a zero frequency, one
at two cycles per year, and two complex pairs at one cycle per year.
The properties of seasonally integrated series are not immediately obvious
but are quite similar to the properties of ordinary integrated processes as
established for example by Fuller (1976). In particular they have ‘long
memory’ so that shocks last forever and may in fact change permanently the
seasonal patterns. They have variances which increase linearly since the start
of the series and are asymptotically uncorrelated with processes with other
frequency unit roots.
The generating mechanisms being considered, such as (2.2) or (2.4) are
stochastic difference equations. They generalize the ordinary 1(l), or Z,(l) in
the present notation, process. It is well known that an equation of the form

(1 -B)x,=q (2.5)
has two components to its solution: the homogeneous solution xlr where

(1 - B)x,, = 0

and the particular solution x2, given by

xzr = (l/(1 - WE,.

Thus xt = xlt + x2,, where xir = x,, (the starting value) and xZI = xiL$_j.
Clearly, if E[E~] = m f 0, then xZt will contain a linear trend mt.
The equation with S(B) = (1 + B)(l + B2),

S(B)x,=e,, (2.6)

also has a solution with two components. The homogeneous solution is

where ci, c2, c3 are determined from the starting conditions, plus the require-
ment that xlr is a real series, i.e., c2 and cg are complex conjugates. If
x-2 = x- t=xO=O so that th e starting values contain no seasonal, then
X 1* = 0.

The particular solution is

X& = [S(B)] -‘q,

 S. H~Mm-g et al., Seusonul integration ond cointegrution 219

and noting that

[S(B)] -’ = 4[1/(1+ B) + (1 - B)/(l+ B2)],

some algebra gives

r-1 int[( t - 1)/2]

$I=; c (-l)‘~,_,+ 1 j-l)JAsl_Z,,

,=a ;=o

where A = 1 - B and int[z] is the largest integer in z. The two parts of this
solution correspond to the two seasonal roots and to eqs. (2.2) and (2.3).
The homogeneous solutions to eqs. (2.9, (2.2) and (2.3) are given, respec-
tively, by

,--I

Slf =
, =o
c %/ for zero-frequency root,

r-l
szt= c (-l>G_j for the two-cycle-per-year root,
/=0

int[( f - I J/2]

S3r = c ( - W-2j for the one-cycle-per-year root.

,=o

The variances of these series are given by

V(s,,) = I+,,) = V(Sgr)= td,

so that all of the unit roots have the property that the variance tends to infinity
as the process evolves. When the series are excited by the same {E*} and t is
divisable by four, the covariances are all zero. At other values of t the
covariances are at most u2, so the series are asymptotically uncorrelated as
well as being uncorrelated in finite samples for complete years of data.
It should be noted that, if E[E~] = m # 0, all t, then the first term in x2t will
involve an oscillation of period 2. The complete solution to (2.6) contains both
cyclical deterministic terms, corresponding to ‘seasonal dummies’ plus long
nondeclining sums of past innovations or their changes. Thus, a series gener-
ated by (2.6) will have a component that is seasonally integrated and may also
have a deterministic seasonal component, largely depending on the starting
values. A series generated by (2.6) will be inclined to have a seasonal with peak
that varies slowly through time, but if the initial deterministic component is
large, it may not appear to drift very fast.
220 S. lf~~lleherg et al., Seasonul integrution und cointegrutim

If X, is generated by

(1 - B‘+, = E,. (2.7)

the equation will have solutions that are linear combinations of those for (2.5)
and (2.6).
A series with a clear seasonal may be seasonally integrated, have a determin-
istic seasonal, a stationary seasonal, or some combination. A general class of
linear time-series models which exhibit potentially complex forms of seasonal-
ity can be written as

4B)4B)(x,-P*)=~,, (2.8)
where all the roots of a(z) = 0 lie outside the unit circle, all the roots of
d(z) = 0 lie on the unit circle, and pLtis given as above. Stationary seasonality
and other stationary components of x are absorbed into a(B), while determin-
istic seasonality is in pr when there are no seasonal unit roots in d(B). Section
3 of this paper considers how to test for seasonal unit roots and zero-frequency
unit roots when other unit roots are possibly present and when deterministic
or stochastic seasonals may be present.
A pair of series each of which are integrated at frequency w are said to be
cointegrated at that frequency if a linear combination of the series is not
integrated at w. If the linear combination is labeled LX,then we use the
notation

x f - CZw with cointegrating vector (Y.

This will occur if, for example, each of the series contains the same factor
which is 1,(l). In particular, if

x,=LwJ,+x,. .Yt = vt + 7,)

where v, is Iw(l) and X, and Y, are not, then z, = X, - ayr is not 1,(l),
although it could be still integrated at other frequencies. If a group of series
are cointegrated, there are implications about their joint generating mecha-
nism. These are considered in section 4 of this paper.

3. Testing for seasonal unit roots

It is the goal of the testing procedure proposed in this paper to determine

whether or not there are any seasonal unit roots in a univariate series. The test
must take seriously the possibility that seasonality of other forms may be
present. At the same time, the tests for conventional unit roots will be
examined in seasonal settings.
 S. F<dleherg et nl., Seasonal integration und cointegration 221

In the literature there exist a few attempts to develop such tests. Dickey,
Hasza, and Fuller (1984), following the lead suggested by Dickey and Fuller
for the zero-frequency unit-root case, propose a test of the hypothesis u = 1
against the alternative a < 1 in the model x, = a~,_~ + a,. The asymptotic
distribution of the least-squares estimator is found and the small-sample
distribution obtained for several values of s by Monte Carlo methods. In
addition the test is extended to the case of higher-order stationary dynamics.
A major drawback of this test is that it doesn’t allow for unit roots at some but
not all of the seasonal frequencies and that the alternative has a very particular
form, namely that all the roots have the same modulus. Exactly the same
problems are encountered by the tests proposed by Bhargava (1987). In Ahtola
and Tiao (1987) tests are proposed for the case of complex roots in the
quarterly case but also their suggestion may at best be a part of a more
comprehensive test strategy. In this paper we propose a test and a general
framework for a test strategy that looks at unit roots at all the seasonal
frequencies as well as the zero frequency. The test follows the Dickey-Fuller
framework and in fact has a well-known distribution possibly on transformed
variables in some special cases.
For quarterly data, the polynomial (1 - B4) can be expressed as

(1 - B4) = (1 - B)(l + B)(l - iB)(l + iB)

= (1 - B)(l + B)(l + P), (3.1)

so that the unit roots are 1, - 1, i, and -i which correspond to zero

frequency, $ cycle per quarter or 2 cycles per year, and a cycle per quarter or
one cycle per year. The last root, -i, is indistinguishable from the one at i
with quarterly data (the aliasing phenomenon) and is therefore also interpreted
as the annual cycle.
To test the hypothesis that the roots of q(B) lie on the unit circle against
the alternative that they lie outside the unit circle, it is convenient to rewrite
the autoregressive polynomial according to the following proposition which is
originally due to Lagrange and is used in approximation theory.

Proposition. Any ( possibly infinite or rational) polynomial QI(B), which is

j&&e-valued at the distinct, nonzero, possibly complex points dl,. . , ep, can be
expressed in terms of elementary polynomials and a remainder as follows:

T(B) = i W(B)/&(B) +A(B)v**(B), (3.2)

X=1

where the X, are a set of constants, ‘p**(B) is a (possibly injinite or rational)

222 S. lf\Neherg et al., Seasonal integmtion and cointegrution

polynomial, and

S,(B) = 1 - ;B,
k

Proof. Let h, be defined to be

Ah= rp(dk)/ I-I q8kL

j#h

which always exists since all the roots of the 6’s are distinct and the
polynomial is bounded at each value by assumption. The polynomial

‘dB) - c &A(B)/&(B)
x=1
=dB) - 2 Qde,)n 6j(B)/S,(e,)
k=l j#h

will have zeroes at each point B = bJk.Thus it can be written as the product of
a polynomial, say cp**( B), and A(B). QED

An alternative and very useful form of this expression is obtained by adding

and subtracting A( B)CX, to (3.2) to get

cp(B)= i &A@)(1+ tdB))/‘%(B) +++?‘*(Bh (3.3)

h=l

where cp*( B) = cp**( B) + Ch,. In this representation ~(0) = q*(O) which is

normalized to unity.
It is clear that the polynomial q(B) will have a root at 8, if and only if
X, = 0. Thus testing for unit roots can be carried out equivalently by testing
for parameters X = 0 is an appropriate expansion.
To apply this proposition to testing for seasonal unit roots in quarterly data,
expand a polynomial (p(B) about the roots + 1, - 1, i, and -i as 8,,
k=l , . . . (4. Then, from (3.3)

cp(B) = h,B(l + B)(l + B2) + X2( -B)(l - B)(l + B2)

+ X3( -iB)(l - B)(l + B)(l - iB)

+ h,(iB)(l - B)(l + B)(l + iB)

+ cp*(B)(l - B4).
 S. EQNeherg et al., Seusonul integrution und mintegrution 223

Clearly, X, and X, must be complex conjugates since (p(B) is real. Simplify-

ing and substituting 7r, = --A,, rI = --A,, 2X, = -VT~ + in,, and 2X, = -v~ -
iTa, gives

q(B)= -7QB(l +B+B2+B3)-7T2(-B)(l -B+B2-B3)

- ( VT4 + 7r3B)(-B)(l - B2) + rp*(B)(l - B4). (3.4)

The testing strategy is now apparent. The data are assumed to be generated
by a general autoregression

T(B)X,=E,, (3.5)

and (3.4) is used to replace cp( B), giving

q*( B)y,, = ‘lil.h-1 + ?T2.h-l + T3.he2 + T44Y3t-1 + ‘17 (3.6)

where

yl, = (1 + B + B2 + B3)x, = S( B)x,,

y2, = - (1 - B + B2 - B3)x,,

y3, = - (1 - B2)x,, (3.7)

y,, = (1 - B4)x, = A4x,.

Eq. (3.6) can be estimated by ordinary least squares, possibly with additional
lags of y, to whiten the errors. To test the hypothesis that cp(B,) = 0, where 8,
is either 1, - 1, or + i, one needs simply to test that A, is zero. For the root 1
this is simply a test for 7r1= 0, and for - 1 it is r2 = 0. For the complex roots
X, will have absolute value of zero only if both V~ and r4 equal zero which
suggests a joint test. There will be no seasonal unit roots if r~* and either r3 or
rr4 are different from zero, which therefore requires the rejection of both a test
for 7r2 and a joint test for r3 and r4. To find that a series has no unit roots at
all and is therefore stationary, we must establish that each of the r’s is
different from zero (save possibly either n3 or r4). A joint test will not deliver
the required evidence.
The natural alternative for these tests is stationarity. For example, the
alternative to ~(1) = 0 should be ~(1) > 0 which means r1 < 0. Similarly, the
stationary alternative to cp( - 1) = 0 is ‘p( - 1) > 0 which corresponds to r2 < 0.
224 S. f~vlleherg et al., Seusonal integrution and cointegrutron

Finally, the alternative to lq( i)l = 0 is Iv(i)1 > 0. Since the null is two-dimen-
sional, it is simplest to compute an F-type of statistic for the joint null,
rrs = n, = 0, against the alternative that they are not both equal to zero. An
alternative strategy is to compute a two-sided test of “4 = 0, and if this is
accepted, continue with a one-sided test of r3 = 0 against the alternative
7s < 0. If we restrict our attention to alternatives where it is assumed that
r4 = 0, a one-sided test for 7~s would be appropriate with rejection for 7r3< 0.
Potentially this could lack power if the first-step assumption is not warranted.
In the more complex setting where the alternative includes the possibility of
deterministic components it is necessary to allow pr f 0. The testable model
becomes

which can again be estimated by OLS and the statistics on the V’S used for
inference.
The asymptotic distribution of the t-statistics from this regression were
analyzed by Chan and Wei (1988). The basic finding is that the asymptotic
distribution theory for these tests can be extracted from that of Dickey and
Fuller (1979) and Fuller (1976) for ri and r2, and from Dickey, Hasza, and
Fuller (1984) for r3 if r4 is assumed to be zero. The tests are asymptotically
similar or invariant with respect to nuisance parameters. Furthermore, the
finite-sample results are well approximated by the asymptotic theory and the
tests have reasonable power against each of the specific alternatives.
It is clear that several null hypotheses will be tested for each case of interest.
These can all be computed from the same least-squares regression (3.6) or (3.8)
unless the sequential testing of r3 and r4 is desired.
To show intuitively how these limiting distributions relate to the standard
unit-root tests consider (3.6) with q*(B) = 1. The test for rrl = 0 will have the
familiar Dickey-Fuller distribution if r2 = r3 = r4 = 0 since the model can be
written in the form

Yl, = (1 + “l)Yl,-1 f&I.

Similarly,

Y,, = - (1 + dh-1 + Et,

if the other V’S are zero. This is a test for a root of - 1 which was shown by
Dickey and Fuller to be the mirror of the Dickey-Fuller distribution. If yZr is
regressed on -y2,_i, the ordinary DF distribution will be appropriate. The
 S. f!rlleherg et (II., Smsonul integrution und cointegrutmn 225

third test can be written as

y,,= -(I +~3h-2+&

assuming r4 = 0 which is therefore the mirror of the Dickey-Hasza-Fuller

distribution for biannual seasonality. The inclusion of y,,_, in the regression
recognizes potential phase shifts in the annual component. Since the null is
that +rr3= r4 = 0, the assumption that r4 = 0 may merely reduce the power of
the test against some alternatives.
To show that the same distributions are obtained when it is not known
a priori that some of the 7~‘s are zero, two cases must be considered. First, if
the 7~‘s other than the one being tested are truly nonzero, then the process
does not have unit roots at these frequencies and the corresponding y’s are
stationary. The regression is therefore equivalent to a standard augmented
unit-root test.
If however some of the other r’s are zero, there are other unit roots in the
regression. However, it is exactly under this condition that it is shown in
section 2 that the corresponding y’s are asymptotically uncorrelated. The
distribution of the test statistic will not be affected by the inclusion of a
variable with a zero coefficient which is orthogonal to the included variables.
For example, when testing ri = 0, suppose 7r2= 0 but y, is still included in the
regression. Then y, and y, will be asymptotically uncorrelated since they have
unit roots at different frequencies and both will be asymptotically uncorrelated
with lags of y, which is stationary. The test for ri = 0 will have the same
limiting distribution regardless of whether y, is included in the regression.
Similar arguments follow for the other cases.
When deterministic components are present in the regression even if not in
the data, the distributions change. Again, the changes can be anticipated from
this general approach. The intercept and trend portions of the deterministic
mean influence only the distribution of ri because they have all their spectral
mass at zero frequency. Once the intercept is included, the remaining three
seasonal dummies do not affect the limiting distribution of ni. The seasonal
dummies, however, do affect the distribution of rZ, r3, and r4.
Table la gives the Monte Carlo critical values for the one-sided ‘t’ tests on
Ti3 r.2, and r3 in the most important cases. These are very close to the Monte
Carlo values from Dickey-Fuller and Dickey-Hasza-Fuller for the situations
in which they tabulated the statistics.
In table lb we present the critical values of the two-sided ‘t ’ test on r4 = 0
and the critical values for the ‘F’ test on n3 n ~7~= 0. Notice that the
distribution of the ‘t ’ statistic is very similar to a standard normal except when
the auxiliary regression contains seasonal dummies, in which case it becomes
fatter-tailed. The distribution for the ‘F’ statistic also looks like an F
 Table 1 a
Critical values from the small-sample distributions of test statistics for seasonal unit roots on 24000 Monte Carlo replications; data-generating - process
A,x, = P, - nid(O, 1).

Fractiles
‘t’:n, ‘t’ : n7 ‘t’:7 3
Auxiliary
regressions T 0.01 0.025 0.05 0.10 0.01 0.025 0.05 0.10 0.01 0.025 0.05 0.10
-
No intercept 48 - 2.12 ~ 2.29 - 1.95 ~ 1.59 ~ 2.61 - 2.21 - 1.95 ~ 1.60 - 2.66 ~ 2.23 - 1.93 - 1.52
No seas. dum. 100 - 2.60 ~ 2.26 ~ 1.97 ~ 1.61 ~ 2.61 ~ 2.22 - 1.92 - 1.57 - 2.55 -2.18 ~ 1.90 ~ 1.53
No trend 136 ~ 2.62 - 2.25 - 1.93 ~ 1.59 ~ 2.60 - 2.23 -1.94 - 1.61 ~ 2.5X ~ 2.21 - 1.92 ~ 1.56
200 - 2.62 - 2.23 -- 1.94 -1.62 - 2.60 - 2.24 - 1.95 ~ 1.61 ~ 2.5X ~ 2.24 - 1.92 - 1.55
Intercept 48 ~ 3.66 ~ 3.25 ~ 2.96 ~ 2.62 ~ 2.68 ~ 2.27 ~ 1.95 - 1.60 - 2.64 - 2.23 ~ 1.90 -1.52
No seas. dum. 100 ~ 3.41 - 3.14 - 2.88 - 2.58 - 2.61 - 2.24 ~ 1.95 ~ 1.60 ~ 2.61 ~ 2.23 -1.90 - 1.54
No trend 136 -3.51 -3.17 - 2.89 - 2.58 ~ 2.60 ~ 2.21 -1.91 ~ 1.58 - 2.53 - 2.18 - 1.88 -1.53
200 ~ 3.4X - 3.13 - 2.87 - 2.51 - 2.58 - 2.22 - 1.92 -1.59 -2.57 ~ 2.21 -1.90 -1.53
Intercept 4x ~ 3.71 - 3.39 - 3.08 - 2.12 - 3.75 - 3.37 ~ 3.04 ~ 2.69 -4.31 ~ 3.92 - 3.61 - 3.24
Seas. dum. 100 ~ 3.55 ~ 3.22 ~ 2.95 - 2.63 ~ 3.60 ~ 3.22 ~ 2.94 ~ 2.63 ~ 4.06 ~ 3.72 - 3.44 - 3.14
No trend 136 - 3.56 - 3.23 ~ 2.94 ~ 2.62 ~ 3.49 ~ 3.15 ~ 2.90 - 2.59 ~ 4.06 ~ 3.72 - 3.44 -3.11
200 - 3.51 ~ 3.18 - 2.91 - 2.59 - 3.50 - 3.16 - 2.89 - 2.60 - 4.00 - 3.67 ~ 3.38 ~ 3.07
Intercept 4X - 4.23 ~ 3.85 - 3.56 - 3.21 - 2.65 - 2.24 - 1.91 ~ 1.57 -- 2.68 - 2.21 -1.92 ~ 1.52
No seas. dum. 100 ~ 4.07 ~ 3.13 - 3.47 - 3.16 - 2.58 - 2.24 - 1.94 ~ 1.60 - 2.56 -2.19 ~ 1.89 -1.54
Trend 136 - 4.09 - 3.15 ~ 3.46 ~ 3.16 - 2.65 ~ 2.25 ~ 1.96 -1.63 ~ 2.56 ~ 2.20 ~ 1.90 -1.52
20 - 4.05 - 3.70 - 3.44 -3.15 ~ 2.59 - 2.25 ~ 1.95 -1.62 - 2.58 ~ 2.21 - 1.92 - 1.56
Intercept 4x - 4.46 ~ 4.04 - 3.71 -3.37 ~ 3.80 - 3.41 ~ 3.08 ~ 2.13 - 4.46 ~ 4.02 ~ 3.66 ~ 3.28
Seas. dum. 100 ~ 4.09 - 3.80 - 3.53 ~ 3.22 - 3.60 - 3.22 - 2.94 - 2.63 -4.12 - 3.76 - 3.48 - 3.14
Trend 136 - 4.15 ~ 3.X0 - 3.52 - 3.21 ~ 3.57 - 3.18 - 2.93 ~ 2.61 - 4.05 ~ 3.12 ~ 3.44 ~ 3.12
200 - 4.05 - 3.74 ~ 3.49 - 3.18 - 3.52 - 3.18 - 2.91 - 2.60 ~ 4.04 ~ 3.69 - 3.41 - 3.10
 Table 1b
Critical values from the small-sample distributions of test statistics for seasonal unit roots on 24000 Monte Carlo replications: data-generating process
Aqx, = E, - nid(0, 1).

Fractiles

Auxiliary 0.99 $
regressions T 0.01 0.025 0.05 0.10 0.90 0.95 0.975 0.99 0.90 0.95 0.975 P
d
No Intercept 48 ~ 2.51 -2.11 -1.76 - 1.35 1.33 1.72 2.05 2.49 2.45 3.26 4.04 5.02 2
No seas. dum. 100 - 2.43 - 2.01 -1.68 -1.32 1.31 1.67 2.00 2.40 2.39 3.12 3.89 4.X9 p
No trend 136 - 2.44 -1.99 - 1.68 - 1.31 1.30 1.66 1.99 2.38 2.41 3.14 3.86 4.81 @
200 ~ 2.43 - 1.98 ~ 1.65 - 1.30 1.29 1.67 1.97 2.36 2.42 3.16 3.92 4.81 k?
a
Intercept 4X -2.44 ~ 2.06 - 1.72 ~ 1.33 1.30 1.68 2.04 2.41 2.32 3.04 3.78 4.78 g
No seas. dum. 100 ~ 2.3X -1.99 ~ 1.68 - 1.30 1.28 1.65 1.97 2.32 2.35 3.08 3.81 4.77 ;
No trend 136 - 2.36 ~ 1.98 -1.6X -1.31 1.27 1.65 1.97 2.31 2.36 3.00 3.70 4.73
4.76 2
200 ~ 2.36 - 1.98 ~ 1.66 - 1.29 1.28 1.65 1.96 2.30 2.37 3.12 3.86 r’
2.
Intercept 4x ~ 2.X6 - 2.31 ~ 1.98 -1.53 1.54 1.96 2.35 2.81 5.50 6.60 7.68 9.22 s
Seas. dum. 100 - 2.78 ~ 2.32 - 1.96 ~ 1.53 1.52 1.93 2.29 2.73 5.56 6.57 7.72 x.74 2
No trend 136 - 2.12 - 2.31 ~ 1.96 - 1.52 1.51 1.92 2.28 2.71 5.56 6.63 7.66 8.92 4
200 ~ 2.74 - 2.33 -1.96 -1.54 1.53 1.95 2.32 2.78 5.56 6.61 7.53 8.93 $.
Intercept 48 - 2.41 - 2.05 - 1.70 - 1.33 1.26 1.64 1.96 2.37 2.23 2.95 3.70 4.64 2
No seas. dum. 100 ~2.38 - 1.97 - 1.65 -1.28 1.28 1.65 1.98 2.32 2.31 2.9X 3.71 4.70 a
4.57 ”
Trend 136 - 2.36 - 1.97 - 1.64 - 1.29 1.26 1.62 1.92 2.31 2.33 3.04 3.69 8
200 - 2.35 - 1.97 -1.66 ~ 1.29 1.26 1.64 1.96 2.30 2.34 3.07 3.76 4.66
Intercept 4X - 2.75 - 2.26 - 1.91 ~ 1.48 1.51 1.97 2.34 2.78 5.37 6.55 7.70 9.27
Seas. dum. 100 ~ 2.16 - 2.32 ~ 1.94 - 1.51 1.51 1.92 2.28 2.69 5.52 6.60 7.52 x.79
Trend 136 - 2.71 ~ 2.78 -1.94 -1.51 1.53 1.96 2.31 2.78 5.55 6.62 7.59 8.77
200 - 2.65 - 2.27 - 1.92 - 1.48 1.55 1.97 2.31 2.71 5.56 6.57 7.56 8.96
228 S. f~~Mwrg et of., Setlsonal integrution und comtegrution

distribution with degrees of freedom equal to two and T minus the number of
regressors in (3.6). However, when seasonal dummies are present, the tail
becomes fatter here as well.

4. Error-correction representation

In this section, an error-correction representation is derived which explicitly

takes the cointegrating restrictions at the zero and at the seasonal frequencies
into account. As the time series being considered has poles at different
locations on the unit circle, various cointegrating situations are possible. This
naturally makes the general treatment mathematically complex and notation-
ally involved. Although we treat the general case we will present the special
cases considered to be of most interest.
Let X, be an N X 1 vector of quarterly time series, each of which potentially
has unit roots at zero and all seasonal frequencies, so that each component of
(1 - B4).x, is a stationary process but may have a zero on the unit circle. The
Wold representation will thus be

(1 - B4)x, = C( B)E[, (4.1)

where F, is a vector white noise process with zero mean and covariance matrix
s2, a positive definite matrix.
There are a variety of possible types of cointegration for such a set of series.
To initially examine these, apply the decomposition of (3.2) to each element of
C(B). This gives

C(B) = i A,A(B)/G,(B) + C**(B)A(B),

h=l

where 6,(B) = 1 - (1/8,)B and A(B) is the product of all the S,(B). For
quarterly data the four relevant roots, 8,, are 1, - 1, i, and -i. which after
solving for the n’s becomes

C(B) = ‘P,[l + B + B2 + B3] + ‘P2[1 - B + B2 - B3]

+ ($ + q4B)[l - B2] + C**( B)(l - B4), (4.2)

where \k, = C( 1)/4, q2 = C( - 1)/4, ‘k3 = Re[ C( i)]/2, and \k4 = Im[ C( i)]/2.
Multiplying (4.1) by a vector CY’gives

(1 - B4)a’xr = a’C( B)E,.

Suppose for some OL= (Ye.aiC(l) = 0 = cw;\k,, then there is a factor of (1 - B)

 229

in all terms, which will cancel out giving

(1 +B+B2+B3)a;X,=a; { *2b + B2)l + (93 + ww + 4

+C**(B)[1+B+B2+B3]}Et,

so that (Y;x, will have unit roots at the seasonal frequencies but not at zero
frequency. Thus x is cointegrated at zero frequency with cointegrating vector
(pi, if o~~‘C(l) = 0. Denote these as

x, - CI, with cointegrating vector LX~.

Notice that the vector yi, = S(B) x, is 1,(l) since (1 - B)yl,= C(B)el, while
‘Y;Y~~is stationary whenever chic = 0 so that ylt is cointegrated in exactly
the sense described in Engle and Granger (1987). Since ylr is essentially
seasonally adjusted x, it follows that one strategy for estimation and testing
for cointegration at zero frequency in seasonal series is to first seasonally
adjust the series.
Similarly, letting _r2,= -(l - B)(l + B2)x,, (1 + B)y,, = - C( B)E, so that
y21 has a unit root at -1. If CX;C(-1) = 0, then a.$q2 = 0 and a;y,, will not
have a unit root at - 1. We say then that x, is cointegrated at frequency
w = t, which is denoted

x, - CII,, with cointegrating vector (Ye.

Finally denote y3, = -(l - B2)x,, which satisfies (1 + B2)y,, = - C( B)E,

and therefore includes unit roots at frequency i. If a;C(i) = 0 which implies
that a$q3 = a;‘k4 = 0, then a&, will not have a unit root at :, implying that

xt - CI,/4 with cointegrating vector 0~~.

Cointegration at frequency i can also occur under weaker conditions. Con-

sider the bivariate system:

(1 fB2)x,= [; !f+B2]%
in which both series are I 1,4(1) and there is no fixed cointegrating vector.
However, the polynomial cointegrating vector (PCIV), as introduced by Yoo
(1987) of (-B, 1) will generate a stationary series. It is not surprising with
seasonal unit roots, that the timing could make a difference. We now show
that the need for PCIV is a result purely of the fact that one vector is sought to
230 S. ~viieherg et ul., Seusonal integrution and cointegrution

eliminate two roots (+ i) and that one lag in the cointegrating polynomial is
sufficient.
Expanding the PCIV a(B) about the two roots (+ i) using (3.2) gives

a(B) = Re[ a( i)] + BIm[ a( i)] + a**( B)(l + B2)

= (ag + a,B) + a**( B)(l + B2),

so that the condition that a’(B) C(B) have a common factor of (1 + B2)
depends only on (Ye and Q. The general statement of cointegration at
frequency i then becomes

x, - C4,4 with polynomial cointegrating vector (Ye+ a,B,

if and only if (LX;+ aii)( +3 - q4k4i)= 0,

which is equivalent to a(i)‘C(i) = 0.

There is no guarantee that x, will have any type of cointegration or that
these cointegrating vectors will be the same. It is however possible that
(Yt= (Y2= (Yg, CY~ = 0, and therefore one cointegrating vector could reduce the
integration of the x series at all frequencies. Similarly if a2 = LYE,rx4 = 0, one
cointegrating vector will eliminate the seasonal unit roots. This might be
expected if the seasonality in two series is due to the same source.
A characterization of the cointegrating possibilities has now been given in
terms of the moving-average representation. More useful are the autoregressive
representations and in particular, the error-correction representation. There-
fore, if C(B) is a rational matrix in B, it can be written [using the Smith-
McMillan decomposition [Kailath (1980)], as adapted by Yoo (1987), and
named the Smith-McMillan-Yoo decomposition by Engle (1987)] as follows:

C(B) = U(B)-‘M(B)V(B)-‘, (4.3)

where M(B) is a diagonal matrix whose determinant has roots only on the
unit circle, and the roots of the determinants of U-‘(B) and V(B)-’ lie
outside the unit circle. This diagonal could contain various combinations of
the unit roots. However, assuming that the cointegrating rank at each fre-
quency is r, the matrix can be written without loss of generality as

(4.4)

where Zk is a k x k unit matrix. The following derivation of the error-correc-

tion representation is easily adapted for other forms of M(B).
 S. H,vlleherg et al., Seasonul integration and cointegrution 231

Substituting (4.3) into (4.1) and multiplying by U(B) gives

A4U(B)x,=M(B)V(B)-1~t. (4.5)
The first N - r equations have a A, on the left side only while the final r
equations have A, on both sides which therefore cancel. Thus (4.5) can be
written as

M(B)U(B)x,= V(B)_lq, (4.6)

with

M(B) = [A$- g. (4.7)

Finally, the autoregressive representation is obtained by multiplying by V(B)

to obtain

‘w)x,=Et, (4.8)
where

A(B) = V(B)M(B)U(B). (4.9)

Notice that at the seasonal and zero-frequency roots, det[A( e)] = 0 since A(B)

U,(B)’
has rank r at those frequencies. Now, partition U(B) and V(B) as

U(B) =
1 V(B)
[ a(B)’
=[b(B),
Y(B)],
’

where CX(B) and y(B) are N X r matrices and U,(B) and V,(B) are N x
(N - r) matrices. Expanding the autoregressive matrix using (3.3) gives

A(B) = II,B[l + B + B* + B3] - II,B[(l - B)(l + B’)]

+(n,- BI13)B[1 f B*] +A*(B)[l - B4],

with II, = - y(l)o1’(1)/4 3 - y&, II, = - y( - l)a( - 1)‘/4 = - y2~$, III, =

Re[y(i)a(i)‘]/2, and 117, = Im[y(j)Lu(i)‘]/2. Letting ~yi = ~~(1)/4, CY~=
a( - 1)/4, (Ye= Re[a(i)]/2, and 0~~= Im[a( i)] while yi = y(l), y2 = y( - l),
ys = Re[y( i)], and yq = Im[y( i)], the general error-correction model can be
232 S. I!!Vleherg et ~1.. Secrsonui integration und corntegrution

written

where A*(O) = C(0) = Z,v in the standard case. This expression is an error-
correction representation where both (Y, the cointegrating vector, and y, the
coefficients of the error-correction term, may be different at different frequen-
cies and, in one case, even at different lags. This can be written in a more
transparent form by allowing more than two lags in the error-correction term.
Add A,(y3a; + y4~; + y,aiB)x,_, to both sides and rearrange terms to get

- (~3 + Y,B)& + hB)~3t-z+ &tr (4.11)

where a*(B) is a slightly different autoregressive matrix from A*(B). The

error-correction term at the annual seasonal enters potentially with two lags
and is potentially a polynomial cointegrating vector. When (Ye= 0 or y4 = 0 or
both, the model simplifies so that, respectively, cointegration is contemporane-
ous, the error correction needs only one lag, or both.
Notice that all the terms in (4.11) are stationary. Estimation of the system is
easily accomplished if the (Y’Sare known apriori. If they must be estimated, it
appears that a generalization of the two-step estimation procedure proposed
by Engle and Granger (1987) is available. Namely, estimate the (Y’S using
prefiltered variables yi, y,, and y,, respectively, and then estimate the full
model using the estimates of the (Y’s. In the PCIV case this regression would
include a single lag. It is conjectured that the least-squares estimates of the
remaining parameters would have the same limiting distribution as the estima-
tor knowing the true LY’S just as in the Engle-Granger two-step estimator. The
analysis by Stock (1987) suggests that although the inference on the (Y’Scan be
tricky due to their nonstandard limiting distributions, inference on the esti-
mates of A*(B) and the y’s can be conducted in the standard way.
The following generalizations of the above analysis are discussed formally in
Yoo (1987). First, if r > 1 but all other assumptions remain as before, the
error-correction representation (4.11) remains the same but the (Y’S and y’s
now become N x r matrices. Second, if the cointegrating rank at the long-run
frequency is rO, which is different from the cointegrating rank at the seasonal
frequency, c,, (4.11) is again legitimate with the sizes of the matrices on the
right-hand side appropriately redefined. Thirdly, if the cointegrating vectors
 S. ff,hherg et ul., Seusonal integmtion and cointegmtion 233

al, a23 and (us coincide, equalling say, (Y, and (Ye= 0, a simpler error-
cointegrating model occurs:

A*(B)A,x, = y(B)&_, + E,, (4.12)

where the degree of y(B) is at most 3, as can be seen either from (4.10) or
from an expansion of y(B) using (3.2). For four roots there are potentially
four coefficients and three lags.
Finally, some of the cointegrating vectors may coincide but some do not. A
particularly interesting case is where a single linear combination eliminates all
seasonal unit roots. Thus suppose (Ye= (Ye= (Y, and CX~ = 0. Then (4.10)
becomes

A*( B)A,x, = yp;S(B)x,_, + y,( B)LY;Ax,_~ + E,, (4.13)

where y,(B) has potentially two lags. Thus zero-frequency cointegration

occurs between the elements of seasonally adjusted x. while seasonal cointe-
gration occurs between the elements of differenced x. This is the case exam-
ined by Engle, Granger, and Hallman (1989) for electricity demand. There
monthly electricity sales were modeled as cointegrated with economic vari-
ables such as customers and income at zero frequency and possibly at seasonal
frequencies with the weather. The first relation is used in long-run forecasting,
while the second is mixed with the short-run dynamics for short-run forecast-
ing.
Although an efficiency gain in the estimates of the cointegrating vectors is
naturally expected by checking and imposing the restrictions between the
cointegrating vectors, there should be no efficiency gain in the estimates of the
‘short-run parameters’, namely A*(B) and y’s, given the superconsistency of
the estimates of the cointegrating vectors. Hence, the representation (4.11) is
considered relatively general and the important step of model-building proce-
dure is then to identify the cointegratedness at the different frequencies. This
question is considered in the next section.

5. Testing for cointegration: An application

In this section it is assumed that there are two series of interest, xi1 and xZt,
both integrated at some of the zero and seasonal frequencies, and the question
to be studied is whether or not the series are cointegrated at some frequency.
Of course, if the two series do not have unit roots at corresponding frequen-
cies, the possibility of cointegration does not exist. The tests discussed in
section 3 can be used to detect which unit roots are present.
Suppose for the moment that both series contain unit roots at the zero
frequency and at least some of the seasonal frequencies. If one is interested in
234 S. Ffvlieherg et al., Seusonal integrution und cointegration

the possibility of cointegration at the zero frequency, a strategy could be to

form the static O.L.S. regression

xi, = A.xz, + residual,

and then test if the residual has a unit root at zero frequency, which is the
procedure in Engle and Granger (1987). However, the presence of seasonal
unit roots means that A may not be consistently estimated, in sharp contrast
to the case when there are no seasonal roots when 2 is estimated supereffi-
ciently. This lack of consistency is proved in Engle, Granger, and Hallman
(1989). If, in fact, xu and x2, are cointegrated at both the zero and the
seasonal frequencies, with cointegrating vectors CQand (Y, and with (pi f (Y$,it
is unclear what value of A would be chosen by the static regression. Presum-
ably, if (pi = LY,, then a will be an estimate of this common value. These
results suggest that the standard procedure for testing for cointegration is
inappropriate.
An alternative strategy would be to filter out unit-root components other
than the one of interest and to test for cointegration with the filtered series.
For example, to remove seasonal roots, one could form

%= W+l,, %,= W)X2t,

where S(B) = (1 - B”)/(l - B), and then perform a standard cointegration

test, such as those discussed in Engle and Granger (1987) on XIlr and ZZr. If
some seasonal unit roots were thought to be present in xir and xZt, this
procedure could be done without testing for which roots were present, but the
filtered series could have spectra with zeros at some seasonal frequencies, and
this may introduce problems with the tests. Alternatively, the tests of section 3
could be used, appropriate filters applied just to remove the seasonal roots
indicated by these tests, and then the standard cointegration tests applied. For
zero-frequency cointegration, this procedure is probably appropriate, although
the implications of the pretesting for seasonal roots has not yet been investi-
gated.
To test for seasonal cointegration the corresponding procedure would be to
difference the series to remove a zero-frequency unit root, if required, then run
a regression of the form
2
Axit = ‘2 a,Ax,,_, + residual,
, =0

and test if the residual has any seasonal unit roots. The tests developed in
section 3 could be applied, but will not have the same distribution as they
involve estimates of the (Y,. The correct test has yet to be developed.
A situation where the tests of section 3 can be applied directly is where
(pi = (Y,~and some theory suggests a value for this (Y, so that no estimation is
 S. I!~~lleherg et (II., Seasowl integrution und coinlegrurion 235

QE
E,
Ou1
oc
CO
- 0

I I * .
L ‘01 s ‘01
236 S. ~1fieherg et d., Seusonal inlegrution und cointegrution

required. One merely forms xlr - CXX~,and tests for unit roots at the zero and
seasonal frequencies.
An example comes from the permanent income hypothesis where the log of
income and the log of consumption may be thought to be cointegrated with
(Y= 1. Thus c - 4’ should have no unit roots using a simplistic form of this
theory, as discussed by Davidson, Hendry, Srba, and Yeo (1978), for instance.
To illustrate the tests, quarterly United Kingdom data for the period 1955.1
to 1984.4 were used with y = log of personal disposable income and c = log of
consumption expenditures on nondurables. The data are shown in fig. 1.
From the figure, it is seen that both series may have a random-walk
character implying that we would expect to find a unit root at the zero
frequency. However, the two series seem to drift apart whereby cointegration
at the zero frequency with cointegrating vector (1, - 1) is less likely. For the
seasonal pattern, it is clear that c contains a much stronger and less changing
seasonal pattern than y, although even the seasonal consumption pattern
changes over the sample period. Based on these preliminary findings, one may
or may not find seasonal unit roots in c and y or both, but cointegration at the
seasonal frequencies cannot be expected.
The tests are based on the auxillary regression (3.6) where G(B) is a
polynomial in B. The deterministic term is a zero, an intercept (I), an
intercept and seasonal dummies (I, SD), an intercept and a trend (I, Tr), or
an intercept. seasonal dummies, and a trend (I, SD, Tr).
In the augmented regressions nonsignificant lags were removed, and for c
and y this implied a lag polynomial of the form 1 - &B - &,B4 - &B’,
where C#Qwas around 0.85, $4 around -0.32, and I& around 0.25. For c - JJ
the lag polynomial was approximately 1 - 0.29B - 0.22B2 + 0.21B4. The ‘t’
statistics from these augmented regressions are shown in table 2.
The results indicate strongly a unit root at the zero frequency in both c, y,
and c -y implying that there is no cointegration between c and y at the
long-run frequency, at least not for the cointegrating vector [l, - 11.
Similarly, the hypothesis that c, y, and c -y are Z1,2(1) cannot be rejected
implying that c and y are not cointegrated at the biannual cycle either.
The results also indicate that the log of consumption expenditures on
nondurables are Z1,4(1) as neither the ‘F’ test nor the two ‘t’ tests can reject
the hypothesis that both r4 and V~ are zero. Such hypotheses are, however,
firmly rejected for the log of personal disposable income and conditional on
these results, c and y cannot possibly be cointegrated at this frequency or at
the frequency corresponding to the complex conjugate root, irrespective of the
forms of the cointegrating vectors. In fact, conditional on r4 being zero, the ‘t’
test on r3 cannot reject a unit root in c -y at the annual frequency in any of
the auxiliary regressions. The assumption that 7r4= 0 is not rejected when
seasonal dummies are absent and the joint ‘F ’ test cannot reject in these cases
either. When the auxiliary regression contains deterministic seasonals, both
 S. Hylieberg et ui., Seasonal integmtion und cointegrution 231

Table 2
Tests for seasonal unit roots in the log of UK consumption expenditure on nondurables c, in the
log of personal disposable income y, and in the difference c - .v; 1955.1-1984.4.
- -
Auxiliary ‘t’:n, ‘t’:7r2 ‘t’:n ‘t’:n, ‘F’:n,nlr,
VAR regressron
1 (zero frequency) (biannual) (annuil)

2.45 -0.31 0.22 -0.84 0.38

I - 1.62 -0.32 0.22 - 0.87 0.40
C I. SD - 1.64 - 2.22 -1.47 - 1.77 2.52
I, Tr - 2.43 -0.35 0.18 -0.85 0.38
I, SD. Tr ~ 2.33 - 2.16 - 1.53 - 1.65 2.43

2.61 -1.44 - 2.35h - 2.51h 5.6Xh

I -1.50 -1.46 - 2.3FSh - 2.51’ 5.15h
.V I, SD ~ 1.56 - 2.38 - 4.19h - 3.89h 14.74b
I. Tr - 2.13 -1.46 - 2.52h - 2.23b 5.46h
I. SD, Tr ~ 2.48 - 2.30 - 4.28’ - 3.46 13.74h

1.54 -1.10 - 0.98 - 1.17 1.19

I - 1.24 - 1.10 - 1.03 - 1.16 1.22
c--l, I,SD - 1.19 - 2.64 ~ 2.55 - 2.79’ 8.25h
I, Tr - 2.40 - 1.21 ~ 1.05 - 1.04 1.12
I,SD, TI ~ 2.48 - 2.84 - 2.72 - 2.4Rh 7.87h

“The auxiliary regressions were augmented by significant lagged values of the fourth difference
of the regressand.
‘Significant at the 5% level.

r4 = 0 and r3 n r4 = 0 are rejected leading to a theoretical conflict which can

of course happen with finite samples.

6. Conclusion

The theory of integration and cointegration of time series is extended to

cover series with unit roots at frequencies different from the long-run fre-
quency. In particular, seasonal series are studied with a focus upon the
quarterly periodicity. It is argued that the existence of unit roots at the
seasonal frequencies has similar implications for the persistence of shocks as a
unit root at the long-run frequency. However, a seasonal pattern generated by
a model characterized solely by unit roots seems unlikely as the seasonal
pattern becomes too volatile, allowing ‘summer to become winter’.
A proposition on the representation of rational polynomials allows reformu-
lation of an autoregression isolating the key unit-root parameters. Based on
least-squares fits of univariate autoregressions on transformed variables, simi-
lar to the well-known augmented Dickey-Fuller regression, tests for the
existence of seasonal as well as zero-frequency unit roots in quarterly data are
presented and tables of the critical values provided.
By extending the definition of cointegration to occur at separate frequen-
cies, the error-correction representation is developed by use of the Smith-
238 S. ~vlleherg et ul., Seusoncrl integruiion and cointegration

McMillan lemma and the proposition on rational lag polynomials. The error-
correction representation is shown to be a direct generalization of the well-
known form, but on properly transformed variables.
The theory is applied to the UK consumption function and it is shown that
the unit-elasticity error-correction model is not valid at any frequency as long
as we confine ourselves to only the consumption and income data.

References
Ahtola, J. and G.C. Tiao. 1987. Distributions of least squares estimators of autoregressive
parameters for a process with complex roots on the unit circle, Journal of Time Series Analysis
8, 1-14.
Barsky. R.B. and J.A. Miron, 1989. The seasonal cycle and the business cycle, Journal of Political
Economy 97. 503-534.
Bell, W.R. and S.C. Hillmer, 1984, Issues involved with the seasonal adjustment of economic time
series, Journal of Business and Economic Statistics 2, 291-320.
Bhargava, A., 1987, On the specification of regression models in seasonal differences. Mimeo.
(Department of Economics, University of Pennsylvania, Philadelphia, PA).
,Box, G.E.P. and G.M. Jenkins, 1970, Time series analysis, forecasting and control (Holden-Day.
San Francisco, CA).
Chan. N.H. and C.Z. Wei. 1988. Limiting distributions of least squares estimates of unstable
autoregressive processes, Annals of Statistics 16, 367-401.
Davidson. J.E.. D.F. Hendry, F. Srba, and S. Yeo, 1978, Econometric modelling of aggregate time
series relationships between consumer’s expenditure and income in the U.K., Economic
Journal 91, 704-715.
Dickey, D.A. and W.A. Fuller, 1979, Distribution of the estimators for autoregressive time series
with a unit root. Journal of the American Statistical Association 84, 4277431.
Dickey. D.A., H.P. Hasza. and W.A. Fuller 1984, Testing for unit roots in seasonal times series.
Journal of the American Statistical Association 79, 355-367.
Engle, R.F.. 1987, On the theory of cointegrated economic time series, U.C.S.D. discussion paper
no. X7-26, presented to the European meeting of the econometric Society. Copenhagen, 1987.
Engle. R.F. and C.W.J. Granger, 1987. Co-integration and error correction: Representation,
estimation and testing, Econometrica 55, 251-276.
Engle. R.F., C.W.J. Granger. and J. Hallman, 1989, Merging short- and long-run forecasts: An
application of seasonal co-integration to monthly electricity sales forecasting, Journal of
Econometrics 40. 45-62.
Fuller, W.A., 1976, Introduction of statistical time series (Wiley, New York, NY).
Grether, D.M. and M. Nerlove. 1970, Some properties of optimal seasonal adjustment, Economet-
rica 38, 682-703.
Hylleberg, S.. 1986. Seasonality in regression (Academic Press. New York, NY).
Kailath. T., 1980, Linear systems (Prentice-Hall, Englewood Cliffs, NJ).
Nelson. C.R. and C.I. Plosser, 1982, Trends and random walks in macroeconomic time series.
Journal of Monetary Economics 10, 129-162.
Nerlove. M., D.M. Grether, and J.L. Carvalho, 1979, Analysis of economic time series: A
synthesis (Academic Press, New York, NY).
Stock, J.H.. 1987, Asymptotic properties of least squares estimates of cointegrating vectors,
Econometrica 55, 1035-1056.
Yoo, S., 1987, Co-integrated time series: Structure. forecasting and testing, Ph.D. dissertation
(University of California. San Diego. CA).