Journal of Real Estate Finance and Economics, 18:1, 89±105 (1999)
# 1999 Kluwer Academic Publishers, Boston. Manufactured in The Netherlands.
Housing Inventory and Completion
N. EDWARD COULSON
Department of Economics, Penn State University, University Park, PA 16802,
The model of Hendry (1986) is extended to a vector autoregressive system, in order to examine the sources of
¯uctuations in housing completions and inventory. The time-series properties of the completion rate are examined
as well as those of the accumulated inventory. Shocks to income, interest rates, materials price, and housing price
are examined. The completion rate appears to be more or less constant, unaffected by these shocks, so that
inventory is evidently the control variable for builders.
Housing starts are usually considered the single most informative indicator of the state
of the construction and housing markets and so have been the subject of a fair amount of
research in real estate economics, starting at least with Maisel (1963) and including later
treatments by Rosen (1984), Thom (1985), McGarvey and Meador (1991), and others. By
contrast, research on housing completions is sparse, which is unfortunate since housing
completions more accurately represent the state of new supply brought to market.
There is one obvious, though not compelling, reason for this lack of interest in
completions. The empirical modeling of completions may seem to be redundant once
starts have been adequately modeled insofar as completions are simply the foregone
consequence of starts. One could in that light imagine a model of the form C
where C and S are completions and starts at time t, so that y
allegedly gives the percentage
of starts completed in i periods. While this simple distributed lag model has some appeal
(and has been used in the literature), it seems to model completions in a purely
technological way, merely asking the question: What is the average number of months
from start to completion? Or more accurately: What is the distribution of building times?
There are a couple of reasons that this approach cannot be quite right. For instance, the
seasonal patterns of starts and completions are quite different. Starts are far more seasonal
than completions (Coulson and Richard, 1996). This may be due to the fact that seasonal
cost differentials are greater for the foundation and framing work that occurs at the time of
a start than the (often indoor) ®nishing work, which occurs during completion. Thus the y
might vary seasonally.
Second, it leaves no room for changing economic conditions to affect the pattern of
completions. The model above might be altered in two ways. One could add variables to