Aggregate and Regional Disaggregate Fluctuations
by
Danny T. Quah
LSE Economics Department
CENTRE FOR ECONOMIC PERFORMANCE
DISCUSSION PAPER NO. 275
Decemb er 1995
This paper is pro duced as part of the Centre's Programme on National Economic
Performance.
I thank I IES in Sto ckholm for its hospitality. I am grateful also to seminar
participants at IIES, the LSE Macro Breakfast Group, and the University of Edin-
burgh, as well as Fischer Black, David Canning, Fabio Canova, and Costas Meghir
for helpful suggestions. Iowe sp ecial thanks to an anonymous referee who carefully
read and helpfully criticized an earlier version of the pap er. All calculations were
p erformed using the author's econometrics shell t
s
r
f
.
Aggregate and Regional Disaggregate Fluctuations
by
Danny T. Quah
LSE Economics Department
Decemb er 1995
ABSTRACT
This paper mo dels uctuations in regional disaggregates as a nonsta-
tionary, dynamically evolving distribution. Doing so enables study of
the dynamics of aggregate uctuations jointly with those of the rich
cross-section of regional disaggregates. For the US, the leading state|
regardless of which it happ ens to b e|contains strong predictive power
for aggregate uctuations. This eect is dicult to understand if only ag-
gregate disturbances aect aggregate business cycles through aggregate
propagation mechanisms. Instead, a b etter picture might be one of a
\wave" of regional dynamics, rippling across the national economy.
Suggested pagehead:
Disaggregate Fluctuations
Keywords:
aggregate disturbance, business cycle, distribution dynamics, regional
uctuation, sto chastic kernel
JEL Classication:
C32, C33, E32
Communications to:
D. T. Quah, LSE, Houghton Street, London WC2A 2AE.
[Tel: +44-171-955-7535, Fax: +44-171-831-1840, Email:
dquah@lse.ac.uk
]
1. Introduction
Macro economics, by denition, concerns aggregate economic variables. And, tra-
ditionally, macro empirics hews to this same discipline. In whichever mainstream
version|real business cycle, aggregate demand and aggregate supply, or new
Keynesian|theoretical and empirical macro economics studies the dynamic re-
sp onse of aggregate variables to hypothesized aggregate disturbances.
Departures from this fo cus exist, but are for the most part minor. In one
instance, disaggregates are analyzed only to provide an aggregation theory, i.e.,
only to understand the macro implications of modelling the underlying micro units.
The disaggregates themselves bear but auxiliary interest. In a second instance,
the researcher might study empirically the b ehavior of consumers and rms, say
in cross-section or panel data mo delling, to understand their resp onses to changes
in their environment. Often, the parameters of those disaggregates are then just
presented as if immediately having implications for macro economic b ehavior. Such
work views disaggregates as providing only
more
data (b eyond aggregate time
series), not
dierent
data. The latter, by contrast, is the view that this pap er
adopts.
There are, of course, counter-examples to the crude characterization just
given. Interactions between individual income distributions and macroeconomic
dynamics (e.g., Galor and Zeira [12] and Persson and Tab ellini [19]), b etween rel-
ative prices and aggregate ination (e.g., Lach and Tsiddon [15]), and between
sectoral imbalance and aggregate unemployment (e.g., Evans [11] and Lilien [16])
are instances where disaggregate analysis has contributed insights for understand-
ing macroeconomic uctuations. In the same vein are the ideas that cross-sectional
spillovers can cumulate for aggregate growth and uctuations (e.g., Durlauf [8] and
Long and Plosser [18]) and that gross labor ows|rather than just net ones|are
informative for macro economic business cycles (e.g., Davis and Haltiwanger [7]).
All these counter-examples share an imp ortant distinctive feature. This is that
there is signicant
two-way
interaction b etween aggregate and disaggregate b ehav-
ior: aggregates aect disaggregates, and disaggregates in turn aect aggregates.
{2{
Because the interaction is two-way, it contradicts the standard assumption, for
instance, in panel data work where aggregate variables might aect disaggregates,
but not vice versa. Moreover, as the income distribution and relative price exam-
ples make clear, the op erative economic mechanism sometimes involves a relation
between dierent parts of the disaggregates distribution: interaction b etween rich
and p o or, or tradeos between high- and low-priced commo dities. Then, summary
statistics of the distribution|say a conditional mean or cross-sectional variance|
will be inappropriate for understanding the relation between disaggregates and
aggregates.
1
What is needed, instead, is a way to analyze exibly the dynamics of
an entire distribution (or rich cross-section) of disaggregates.
Few econometric to ols extant are appropriate for this. This pap er seeks to add
to those to ols. It explores theoretical and empirical mo delling of the joint dynamics
of aggregate and regional disaggregate output. The regional disaggregates studied
b elow|the states in the US|are large enough compared to aggregate US output
that one cannot casually dismiss the p otential eects of disaggregate dynamics on
the aggregate. At the same time, there are many enough regional disaggregates to
make apparent the mo delling diculties: standard vector time-series metho ds, for
instance, will not do for mo delling the dynamics of a 50 by 1 random vector.
2
If one
were to turn then to the joint dynamics of European Union regional disaggregates|
1
The easiest way to see this is through an example. Supp ose that it is income
inequality that matters for aggregate uctuations and growth, as, e.g., in Galor
and Zeira [12, 19] and Persson and Tab ellini [12, 19]. Which income inequality
measure should one use in empirical analysis? Theory do esn't always provide an
answer since the simplied distributions that app ear in a theoretical mo del are
only suggestive of more general economic forces at work. Atkinson's classic pap er
[1] shows how alternative inequality measures imply substantively dierent|and
p otentially contradictory|views on the inequality actually extant.
2
Post-War quarterly time-series now contain 200 observations. But a VAR
mo del for a 50 by1vector already has 2500 free parameters in the rst-order lag
matrix co ecient; the variance-covariance matrix for the innovation contributes
another 1275. Quah and Sargent [29] attempt to control this parameter prolif-
