Journal of Real Estate Finance and Economics, 16: 1, 43±53 (1998)
# 1998 Kluwer Academic Publishers, Boston. Manufactured in The Netherlands.
Aggregation of Local Housing Markets
JOHN L. GOODMAN JR
National Multi Housing Council, 1850 M Street, NW, Suite 540, Washington, DC 20036
This article explores the implications of spatial aggregation for parameter estimates in models of the housing
sector. Using illustrative static and dynamic models and realistic assumptions about the ``true'' parameter values
at the level of the local market, the article characterizes and quanti®es the bias resulting from spatial aggregation
and then uses the results to shed light on some previous ®ndings from aggregate modeling of the housing sector.
Key Words: aggregation, estimation, housing markets, aggregation bias
``Housing markets are local'' is a common refrain of urban economists. Nonetheless, it is
standard econometric practice to estimate housing market models with geographically
aggregated data. In particular, models of housing construction and prices are commonly
estimated on national statistics.
Informal evidence abounds on the heterogeneity of local markets and potential prob-
lems with aggregate estimation. During the early 1990s, a recession gripped the econ-
omy, and single-family housing starts in 1991 fell to their nine-year lows for the nation.
Nonetheless, seven states had more starts that year than in any year during the 1980s (Joint
Center, 1994). Similarly, even as sales prices for existing homes nationally declined 2%
during 1990, prices rose in ®fty-three large markets and fell in only forty.
Franklin Fisher (1987) notes that addressing the aggregation problem is like asking,
``When is it possible to speak of `food' rather than of `applies, bananas, carrots, etc.'?''
This article examines when it is possible to speak of ``the housing market'' rather than of
Peoria, Jacksonville, Okmulgee, and so on. Economists analyze national housing data
because it is feasible and because, for macro purposes, we need to tell a national story. We
know intuitively that spatial aggregation is not right. This article attempts to determine
whether aggregation is wrong. That is, are the errors that result from aggregation large
enough to be economically meaningful? The article's objective is analogous to Geweke's
(1985) quanti®cation of the bias in macro modeling under a representative agent paradigm.
Previous work on aggregation bias in housing analysis has focused on the demand side
and differences in estimates from grouped and micro data, with an emphasis on
measurement problems. This work, done mostly in the 1970s (de Leeuw, 1971; Polinsky,
1977, 1979; Smith and Campbell, 1978), did not deal with spatial aggregation per se, and
there was not much attention paid to the implications of aggregation in dynamic models.
While that earlier work dealt with the grouping of individuals on estimates of demand, the
focus here is on grouping of local markets. Even though housing research has sometimes