You are on page 1of 39

Research Division

Federal Reserve Bank of St. Louis

Working Paper Series

Is Housing the Business Cycle? Evidence from U.S. Cities

Andra C. Ghent
Michael T. Owyang

Working Paper 2009-007A

March 2009


Research Division
P.O. Box 442
St. Louis, MO 63166

The views expressed are those of the individual authors and do not necessarily reflect official positions of
the Federal Reserve Bank of St. Louis, the Federal Reserve System, or the Board of Governors.
Federal Reserve Bank of St. Louis Working Papers are preliminary materials circulated to stimulate
discussion and critical comment. References in publications to Federal Reserve Bank of St. Louis Working
Papers (other than an acknowledgment that the writer has had access to unpublished material) should be
cleared with the author or authors.
Is Housing the Business Cycle? Evidence
from U.S. Cities

Andra C. Ghent and Michael T. Owyangy

keywords: local housing markets, residential investment, economic

‡uctuations, housing prices

We analyze the relationship between housing and the business cycle in a set of 36
US cities. Most surprisingly, we …nd that falls in house prices are often not followed by
declines in employment. We also …nd that the leading indicator property of residential
investment is not consistent across cities and that, at the national level, the leading
indicator property of residential investment is not robust to including …nancial factors
as control variables. [JEL: C32, E32, R11]

Kristie M. Engemann provided excellent research assistance. The views expressed herein are those of
the authors and do not re‡ect the o¢ cial positions of the Federal Reserve Bank of St. Louis or the Federal
Reserve System. We thank workshop participants at the 2008 Asian Real Estate Society Annual Meetings
and Baruch College for helpful comments.
Ghent (corresponding author): Dept. of Real Estate, Zicklin School of Business, Baruch College / CUNY;
phone 646-660-6929; email Owyang: Research Division, Federal Reserve Bank
of St. Louis; email
1 Introduction

In a recent paper, Leamer (2007) argues that real estate markets are grossly understudied
by macroeconomists interested in understanding business cycles. He asserts several stylized
facts about the behavior of the national housing market over the business cycle including
1) residential investment leads the business cycle and a fall in residential investment is a
reliable harbinger of a recession, and 2) volumes, rather than house prices, are what matter
for business cycles.
This paper furthers our understanding of the relationship between housing and the busi-
ness cycle by exploiting the cross-sectional variation in business cycles and housing markets
across Metropolitan Statistical Areas (MSAs). There exists a tremendous amount of vari-
ation across time in local real estate markets: for example, Glaeser and Gyourko (2007)
document that time dummies explain only about a quarter of the variation in city-level
house price changes, suggesting that most of the variation in house prices comes from city-
speci…c factors. Del Negro and Otrok (2007) similarly …nd that state and regional factors,
rather than national factors, drive the majority of the movement in house prices.
MSA-level variation enables us to identify any empirical regularities in the relationship
between the housing market and the business cycle more robustly than we can from national
data alone. While recent months mark the …rst time we have seen a large real decline in
house prices in the post-war period at the national level, as …gures 1 5 illustrate, several
cities in our sample experienced large and sustained declines in house prices over the last
25 years. Understanding how house prices a¤ect employment in these cities may help us
to understand whether the recession that began in December 2007 is due to factors that
simultaneously lowered house prices and employment or whether declines in house prices
themselves play a key driving role in economic activity.
We …rst seek to determine if the stylized facts Leamer identi…es at the national level
are also true in our sample of MSAs. Because residential investment is not available at
the MSA level, we proxy for residential investment using permit data. We ask whether

housing variables are robust leading indicators of employment after we control for national-
level factors. Smets (2007) points out that the appearance of residential investment as a
leading indicator may be due to cycles in interest rates. Since housing is known to be
sensitive to changes in interest rates (e.g., Hamilton, 2008), shocks to both short-term and
long-term interest rates may be common driving forces behind both residential investment
and employment. Our approach allows us to control for such interest rates.
We also assess the relationship between housing prices and employment over the business
cycle. Case, Quigley, and Shiller (2005), Slacalek (2006), Aron, Muellbauer, and Murphy
(2006), and Muellbauer (2007), among others, have found that a decline in housing wealth
lowers personal consumption at the macro-level. In some models, bearish economic senti-
ment on the part of consumers –perhaps caused by a decline in net worth due to a fall in
house prices –can lead to a decline in macroeconomic activity. However, Buiter (2008) ar-
gues that changes in housing prices redistribute wealth between those with a long position in
housing (young households) and households that e¤ectively have a short position in housing
(older households and households that plan to move to a di¤erent market) such that there
is no pure wealth e¤ect on aggregate consumption from changes in house prices.
It is thus a natural question what e¤ect changes in housing prices have on employment.
Earlier work has not reached a consensus on the relationship between the business cycle and
house prices: Iacoviello (2005) …nds that, at the national level, a decline in housing prices
leads to a decline in output. Davis and Heathcote (2005) …nd that the contemporaneous
correlation between national-level HP-…ltered house prices and output is 65 percent over the
1971-2001 period. In contrast, Kan, Kwong, and Leung (2004) examine annual city-level
data and …nd that the contemporaneous correlation between house prices and output growth
is less than 15 percent.
This paper contributes to a growing literature that uses regional data to analyze macro-
economic ‡uctuations. Carlino and DeFina (1998, 1999) and Fratantoni and Schuh (2003)
show, for example, that the e¤ects of monetary policy di¤er substantially across regions.

The majority of this literature attributes these di¤erences to industry composition and the
makeup of the regional banking sector. Housing, which may be thought of as a key com-
ponent in the propagation of monetary policy, has been largely neglected due to the highly
aggregated (i.e., states or larger) geographic unit of analysis. Owyang, Piger, and Wall
(2005) use state-level data to analyze what determines growth rates during recession and
expansion phases; Owyang, Piger, Wall, and Wheeler (2008) apply a similar framework to
U.S. cities. Owyang, Rapach, and Wall (2009) use a dynamic factor model to assess the
extent to which states comove in their business cycles and …nd substantial heterogeneity in
business cycles across states.
Consistent with Leamer’s …ndings, we …nd that house prices are poor leading indica-
tors. However, we …nd that increases in permits do not consistently raise employment in
individual cities in our VAR analysis. At the national level, we …nd that permit shocks do
not raise employment after we control for …nancial factors. We also explore the possibil-
ity that the housing market in‡uences employment over the business cycle in a nonlinear
fashion. In particular, we use a Markov-switching model with time-varying transition prob-
abilities (TVTPs) to ascertain whether housing variables are signi…cant in in‡uencing the
probabilities that cities move between expansion and recession regimes. Consistent with
our VAR results, the …ndings from our Markov-switching model do not support the notion
that housing variables drive the business cycle.
The remainder of the paper proceeds as follows: Section 2 describes our data and doc-
uments the correlations between housing indicators and employment at the national and
MSA level. Section 3 outlines our VAR model and uses it to identify whether, and which,
housing variables are good leading indicators for employment. Section 4 presents our TVTP
Markov-switching model and the results. Section 5 concludes.

2 Data

Data availability dictates our sample period and our choice of cities. We include all cities
for which all series are available beginning in 1982Q1, the earliest date for which permits
are available at the MSA level. Table 1 lists the cities in our sample and the de‡ator used
to convert nominal variables into real variables.1 Our nominal house price series are the
Freddie Mac Conventional Mortgage House Price Indices (CMHPIs).2 Real house prices are
constructed for the analysis.3
Finally, we include several national-level variables that may a¤ect both housing variables
and employment. To control for the stance of monetary policy, we include the federal funds
rate. We also include national-level in‡ation, the 30-year conventional mortgage rate, and
the spread between 3-month commercial paper and 90-day Treasury bills as a measure of
credit market turmoil.
Figures 1 through 5 graph our data. Both in units and in values, permits exhibit more
volatility than house prices or employment and thus are shown on a di¤erent scale. There
is also variation across cities in the volatility of permits; for instance, the scale of permits
in Ann Arbor is close to 20 times that of Bakers…eld. There is a substantial amount of
high-frequency variation in our permit series with all series exhibiting strong low-frequency
variation. This requires us to somehow …lter the data before establishing anything about
the relationships among the series.
Our sample excludes a few large and economically important cities because we lack either permit or
employment data for the full sample: permit data for San Francisco and New York begin in 1984 and 1988
while employment data for Boston and Washington, DC, start in 1990.
These data are constructed using the same method and data as the OFHEO price series such that the
di¤erences between the series are minute. The CMHPI methodology is based on repeat sales and implicitly
adjusts for changes in housing quality. Furthermore, the CMHPIs are available for a wider set of cities and
cover more years than the Case-Shiller indices. One limitation of these data, however, is that they include
only conforming loans, thus excluding both the top end of the market (homes that require a jumbo mortgage
to purchase) and homes purchased using sub-prime or alt-A mortgages, which might be lower-value homes.
To obtain real prices and real permits, we de‡ate by city-speci…c CPI de‡ators (excluding the shelter
component) when possible. For several cities in our sample, city-speci…c de‡ators are not available in which
case we de‡ate by the urban CPI de‡ator for that BLS region or the urban CPI de‡ator for large cities in
that BLS region as applicable. As CPIs are not available at the quarterly frequency for our full sample, we
interpolate quarterly values from annual data. This has the further advantage of deseasonalizing the data
in a consistent fashion as seasonally-adjusted CPIs are not available for most of our sample.

2.1 Correlations

2.1.1 Band-Pass Filtered

Tables 2 4 present the correlations between our three main housing variables and employ-
ment. As a …rst pass, we isolate the business cycle relationships by …ltering the data in
the frequency domain with the optimal band-pass …lter suggested by Corbae, Ouliaris, and
Phillips (2002) and Corbae and Ouliaris (2006).4 Our approach has the advantage of not
forcing us to take a stand on the relationship between the variables at low frequencies.5
Table 1 shows the contemporaneous correlation of each city’s ideal band-passed employ-
ment, ideal band-passed house prices, and ideal band-passed permits with their national
counterparts.6 The average correlation of city-level employment with national-level employ-
ment is 69 percent. City-level permits exhibit slightly less concordance with the national
level: The average correlation of city-level permits with national-level permits is 60 percent.
However, house prices at the city level exhibit much less concordance (29 percent on average)
with the national-level index.
At the national level, house prices exhibit mild procyclicality in leads and mild counter-
cyclicality in lags as table 2 shows. A slim majority of our MSAs also exhibit procyclical
prices. However, several of our MSAs exhibit counter-cyclical house price behavior.
The permit data exhibit a more consistent pattern as tables 3 and 4 show. Permits
measured in units (values) lead the cycle at the national level and in 31 (27) of our 36
Our de…nition of the business cycle frequency di¤ers somewhat from the more conventional de…nition
of frequencies associated with periods of 6-32 quarters. However, several postwar U.S. business cycles have
had periods greater than 32 quarters; for example, the expansion the NBER dates as beginning in 1991Q1
and ending in 2001Q1 lasted 40 quarters.
As Corbae and Ouliaris (2006) illustrate, the commonly used Baxter-King band-pass …lter has higher
leakage than the optimal-frequency …lter. That is, less of the variation from the undesired frequencies
appears in the …ltered series, and vice versa, with the optimal …lter than with the Baxter-King band-pass
…lter. Furthermore, the band-pass …lter will not consistently extract the cycle if the data is I(1) rather than
having a deterministic trend: when a series is I(1) there is leakage from the zero frequency into the other
frequency bands that does not disappear asymptotically.
The correlations for employment range from a low of 19 percent for Baton Rouge to a high of 98 percent
for Chicago and Cleveland. Not surprisingly, we see the least concordance with national-level employment
in cities heavily involved in energy production such as Bakers…eld (18 percent), Baton Rouge ( 19 percent),
Oklahoma City (25 percent), and Tulsa (32 percent).

MSAs. At the national level, the correlation between the value of permits and employment
is highest at a lead of 6 quarters. The timing of the relationship among our MSAs is
similar when permits are measured in units with 31 of our MSAs exhibiting the strongest
correlations at horizons of three to six quarters. The …nding that permits in units have a
more robust cyclical pattern than permits in values suggests that it is not simply a change
in the composition of housing over the business cycle that explains the leading indicator
property of residential investment.

2.1.2 Linearly Detrended

Tables 5 7 show these same correlations when we linearly detrend the data, leaving in
all the variation in the data at the high frequency and much if not all of the low-frequency
variation. Here, the results look starkly di¤erent. From these correlations, one would
conclude that permits and house prices behave countercyclically at the national level. These
are exactly the opposite conclusions of what we would conclude by looking at the band-pass-
…ltered correlations. Furthermore, several relationships amongst cities switch from being
countercyclical to procyclical and vice versa when the data is linearly detrended instead of
band-pass …ltered.
These results indicate that the relationship between housing and employment may be
frequency-dependent. In particular, the relationship between the trend in employment and
the trend in housing variables may be quite di¤erent from the relationship between the
variables at business cycle frequencies.

3 Linear Relationships

Correlations for the band-pass-…ltered data are suggestive of the relationship between em-
ployment and the housing market. Because there may be causal relationships between both
employment and the housing market, we use a VAR to look at the relationship between the
variables, which allows us to identify the e¤ect of housing market innovations on the business

Because we have a limited number of observations (104) with which to estimate the VAR
and many parameters to estimate for a VAR with four lags, we use principal components to
reduce the dimensionality of the national data.8 For each city i, we estimate the model

Xt;i = A0;i + Ai;i Xt i;i + C"t ; (1)

Xt;i = Ft1 ; Ft2 ; M SA0t;i ;

C is lower triangular, = CC 0 , Ft1 and Ft2 are two national-level …nancial factors, M SAt;i
is a vector of MSA-level variables, and p is set to 4.9 Because each of our cities is small
relative to the size of the U.S. economy, our identi…cation assumes that …nancial factors
do not respond contemporaneously to MSA-level shocks while all MSA-level variables can
respond to shocks to the …nancial factors. Since our interest is not in identifying shocks
to our national-level …nancial factors, the ordering of these factors does not matter. We
control for the type of national-level …nancial variables that Smets (2007) suspects drive the
leading indicator property of housing at the national level by constructing Ft1 and Ft2 from
N att;i = [ t ; F F Rt ; M Rt ; Spreadt ]0 : t is national-level in‡ation, F F Rt is the federal funds
rate, M Rt is the conventional mortgage rate, and Spreadt is the spread between the 3-month
commercial paper rate and the 90-day T-bill yield. The …rst two principal components, Ft1
Our VAR approach both exploits the cross-sectional variation in the data and allows us to include
national-level variables. By controlling for the national-level variables most likely to a¤ect housing and
employment, we are able to identify the e¤ect of shocks to our housing variables on employment. The
approach of using both macro- and city-level variables in a VAR is somewhat similar to the approaches
pursued by Barth and Ramey (2002), Davis and Haltiwanger (2001), and Fratantoni and Schuh (2003).
However, these papers are interested in identifying the e¤ect of aggregate variables on industry- and city-
level variables; our interest is the e¤ect of city-level shocks on city-level variables using aggregate variables
as controls.
Bernanke, Boivin, and Eliasz (2005) have previously used this data reduction technique to analyze the
e¤ects of monetary policy shocks without making restrictive assumptions about which data the Federal
Reserve can and cannot observe. Boivin, Giannoni, and Mihov (2007) have also used this technique to
study the e¤ect of price stickiness using industry-level data while controlling for a host of national factors.
We continue to use ideal band-passed data in our VAR analysis to ensure we accurately extract the
cyclical components of the data.

and Ft2 , combined capture approximately 92 percent of the variation in our national-level
variables, suggesting that we are accurately capturing the national factors that might drive
We estimate the model using the Gibbs sampler, a Bayesian technique in which the full
joint posterior is obtained from iterative draws from each parameter’s conditional distrib-
ution (see Gelfand and Smith, 1990; Casella and George, 1992; Carter and Kohn, 1994).
We assume that the set of parameters consisting of the VAR coe¢ cients and the variance-
covariance matrix are characterized by normal-inverse Wishart priors –that is,

A; N IW ( ; S; ; ) ;

where f ; S; ; g are hyperparameters that govern the shape of the prior. We set = 0,
S = IN , = 0, and = In , where n is the number of variables in the VAR and N is the
total number of VAR coe¢ cients (in this case, N = pn2 + n). Sampling from the posterior
is a straightforward implementation of Chib and Greenberg’s (1996) SUR sampler.10

3.1 Permits as a Leading Indicator

We …rst use a within-city bivariate VAR to identify how much information housing variables
contain about future employment beyond what is in the lags of employment. We then ask
how much of this information is due to housing speci…cally rather than common comovement
in housing variables and …nancial variables. Table 8 shows the e¤ect of shocks to permit
values at the national level and in our 36 cities in a simple bivariate VAR with four lags, i.e.,
Xt;i = [Empt;i ; P ermV alt;i ]0 , (2)

omitting the …nancial variables.11

For each draw of the model parameters, we can compute the impulse responses to a shock to any system
variable. To determine whether the impulse responses are signi…cantly di¤erent from zero, we examine the
10th and 90th percentiles of their posterior distributions.
The ordering of the city-level variables is motivated by our …ndings regarding the raw correlations
(employment tends to lag permit values) and based on the fact that we are using permit data rather than

We de…ne a city or the United States as reacting primarily positively if the cumulative
impulse response function after 20 quarters is positive. At the national level, employment
rises following a permit shock in the bivariate model and the impulse responses for the 80
percent coverage interval are positive for the …rst three quarters. However, in less than half
of our cities do we observe evidence that employment reacts positively to permit shocks.
Indeed, several cities exhibit negative responses to permit shocks. Moreover, we …nd no
clear pattern governing how city-level employment responds to permit shocks. We looked
at several reasons why changes in house prices may a¤ect employment di¤erently across
cities. We organized cities according to the growth rates of employment, real house prices,
permits, the ratio of house price growth to employment growth, the ratio of permit growth to
employment growth, and the ratio of house price growth to permit growth over the 25-year
period. None of these factors appear to in‡uence whether employment responds to permit
Next, we ask whether housing’s appearance as a leading indicator is due to its correlation
with …nancial factors, as Smets (2007) suggests. We estimate (1) with

M SAt;i = [Empt;i ; P ermV alt;i ]0 (3)

to ascertain how much of the strength of the relationship is due to national …nancial factors
a¤ecting permits. Table 9 contains these results. As the …rst row illustrates, using our
benchmark ordering, we now …nd that permit shocks do not increase employment even at
the national level. Rather, the impulse responses change signs. Furthermore, the strength
of permits as a leading indicator in several of the cities remains relatively weak: a majority
actual residential investment: As builders must obtain permits before they commence construction, it makes
sense to assume that even employment in the construction industry does not respond contemporaneously to
permit shocks.
For example, Oklahoma City and Tulsa had the lowest annual real house price growth ( 0:3 percent
and 0:6 percent), but both experienced increases in employment following permit shocks. House prices in
Baton Rouge and Memphis grew only slightly faster (at annual rates of 0:3 percent and 1:1 percent), but
permit shocks decrease employment signi…cantly in both. Similarly, house prices grew at nearly the same
annual rate in Jacksonville (2:52 percent) and Milwaukee (2:49 percent), and yet the two cities’employment
responses are opposites.

of cities continue to respond negatively to increases in permits.
Our identi…cation strategy at the city level assumes that national-level …nancial factors
cannot simultaneously respond to permits or employment. Since this is a much less plausible
assumption at the national level, we also consider two alternative orderings of the VAR to
test the robustness of the e¤ect of permit shocks on employment at the national level. Rows
2 and 3 of table 9 show the results when we order the variables in the VAR as

Xt;U S = [Empt;U S ; Ft;1 ; Ft;2 ; P ermV alt;U S ]0

Xt;U S = [Empt;U S ; P ermV alt;U S ; Ft;1 ; Ft;2 ]0 .

Consistent with our results for the benchmark model, permit shocks do not increase employ-
ment under these alternative sets of identifying assumptions.
We also consider the possibility that house prices drive permits. In light of the negative
correlations between house prices and employment, the poor performance of permits may
simply be due to omitted variable bias. We thus estimate (1) with

M SAt;i = [HPt;i ; Empt;i ; P ermV alt;i ]0 , (4)

where permits are ordered last.13

The results from estimating (4), shown in table 10, are substantively similar to those in
which house prices were excluded. Once again, we …nd no evidence that rises in permits
increase employment although the number of cities in which permit shocks signi…cantly
reduce employment falls once we control for the e¤ect of house prices: a mere nine cities
respond positively to permit shocks. It is worth noting that the oil-producing cities in our
Results from the previous section suggest that employment lags permit values. Moreover, house prices
have been shown to respond slowly to changes in macroeconomic conditions (see, for example, Mankiw and
Weil, 1989; and Poterba, 1991).

sample (Baton Rouge, Oklahoma City, and Tulsa) consistently respond positively to permit
shocks even after we control for house prices and …nancial factors. Beyond that, there
is no obvious commonality among the cities that respond positively (Ann Arbor, Atlanta,
Bakers…eld, Baltimore, San Diego, and Stockton).
The lack of a signi…cant response at the city-level in the bivariate results and at the
national level after we control for …nancial factors suggests that the pattern Leamer …nds
at the national level is due to …nancial factors a¤ecting both residential investment and
employment in all cities. After a large …nancial shock, permits and employment fall in
cities. Since national-level permits are simply an aggregation of individual cities’permits,
falls in national-level permits happen when there is a shock that a¤ects most cities. This
gives the misleading appearance that permits drive the business cycle when the true driver
is in fact …nancial shocks.

3.2 Real House Prices as a Leading Indicator

Our results so far suggest permits are not a good leading indicator for employment for
the majority of cities. We next ask whether decreases in house prices adversely a¤ect
Table 11 shows the response of employment to a house price shock in our model when
we control for both …nancial factors and permits. Far from house price increases being
consistently accompanied by rising employment, the responses of employment to house price
shocks di¤er dramatically across cities. A slim majority of cities respond positively to a
house price shock but several respond negatively.

4 A TVTP Markov-Switching Model

In the previous section, we found the the evidence supporting permits (linearly) drive em-
ployment is weak. Much research (e.g., Hamilton, 1989, 2005) suggests that the business
cycle is an inherently nonlinear phenomenon with asymmetric dynamics. As periods of eco-

nomic decline are rapid and sharp, with expansions being slow-paced and steadier, studying
only the linear relationship among variables may fail to capture important business cycle
relationships. In this section, we therefore ask if the state of the housing market in‡uences
whether the economy is in a recession or expansion phase. To address the issue, we adopt a
TVTP Markov-switching model along the lines suggested by Goldfeld and Quandt (1973).
Employment growth in city i, Empi;t , follows the process

Empi;t = si;t + "i;t ; "i;t N (0; i) : (5)

The mean of employment growth, si;t , depends on the regime according to

si;t = 0;i + 1;i si;t . (6)

The unobserved regime, si;t = si;t (zi;t ) 2 f0; 1g follows the homogeneous Markov-switching
process with transition probabilities

Pr (si;t = 0jsi;t 1 = 0; zi;t ) = pi (zi;t ) (7)

Pr (si;t = 1jsi;t 1 = 1; zi;t ) = qi (zi;t ) , (8)

where 0 is the expansion regime and 1 indicates the recession regime, such that 1i < 0. zi;t
is the housing variable that may drive the transitions between regimes.
We model the transition probabilities with latent variables

si;t = 0;i + z;i zi;t + s;i si;t 1 + ui;t , (9)

ui;t N (0; 1)

de…ned such that

Pr (si;t = 1) = Pr si;t > 0 .

Letting ( ) denote the standard normal cumulative distribution function, (9) implies that

pi (zi;t ) = 0i z;i zi;t (10)

qi (zi;t ) = 1 0i z;i zi;t s;i . (11)

Letting ( ) denote the standard normal probability density function,

@ Pr (si;t = 0jsi;t 1 = 0)
= z;i 0i z;i zi;t

@ Pr (si;t = 1jsi;t 1 = 1)
= z;i 0i z;i zi;t s;i ;

such that a negative value for z;i indicates that an increase in zi;t raises the probability of
staying in an expansion and lowers the probability of staying in a recession.14
We follow Filardo and Gordon (1998) in using Bayesian techniques to estimate the
model.15 This is accomplished through a straightforward application of the Gibbs sam-
pler. Brie‡y, the sampler is broken into four blocks to estimate the joint posterior for the
entire parameter vector

h i0
i = i; 0;i ; 1;i ; fsi;t gt=1 ; 0;i ; z;i ; s;i :

A single iteration of the Gibbs sampler begins with a draw from the posterior distribution
We allow the lagged six-quarter-moving-average of our …rst …nancial factor, house price growth, and
permit growth to in‡uence the transition probabilities. Using the six-quarter-moving-average has the ad-
vantage of both moderating the high-frequency variation in permits and allowing more distant lags to a¤ect
the probability of being in a recession in a parsimonious speci…cation.
For classical approaches to estimating TVTP Markov-switching models, see, for example, Diebold, Lee,
and Weinbach (1994), Durland and McCurdy (1994), Filardo (1994), Maheu and McCurdy (2000a,b), Perez-
Quiros and Timmermann (2000), and Masson and Ruge-Murcia (2005). In an earlier version of the paper,
we used classical techniques to estimate our model as well, but we found the estimates were very sensitive
to the starting values used to initialize the numerical maximization routine and thus felt Bayesian methods
suit our problem better.

(j 1) 2
of i conditional on values for i
. The prior for i is inverse-gamma, i.e.,

i Gamma (1; 1) :

The posterior is then

1+T 2T
j fsi;t gTt=1 ; 0;i ; 1;i Gamma ; ;
2 (1 + T ) (1 + y~i0 y~i )

(j 1) (j 1) (j 1)
where y~i is a T x1 vector with row i, y~i;t = yi;t 0;i 1;i si;t , and yi;t = Empi;t .
(j) (j) (j) (j 1) 0
We then draw 0;i , 1;i conditional on i and i
. If the prior for i =[ 0;i ; 1;i ] is

i 1 1i<0
N (^ i ; i I2 ) ;

where 1 is an indicator function that identi…es si;t = 1 as the recession state, ^ i =

T h i0
(j) (j) (j)
[1:5 yi ; 2 yi ], and yi = T1 yi;t , then the posterior from which we draw i = 0;i ; 1;i
i j fsi;t gt=1 ; i 1 1i<0
N i ; Ai i ;

+ Y•i0 Y•i I2 ^ i + Y•i yi , Y•i is a T x2 matrix with row t [1; si;t ],
1 0
where Ai = i I2 , i = Ai i

and y•i is a T x1 vector with row t as yt;i .

n oT
We next draw si;t recursively from a Bernoulli distribution using

P (si;t j•
yi ) / P (si;t jsi;t 1 ; zi;t ) P (si;t+1 jsi;t ; zi;t+1 ) f (yi;t jsi;t ) .

(j 1) (j)
Given and i = 0;i ; z;i ; s;i ; we apply Bayes’rule to obtain , the conditional
posterior probability that si;1 = 0, as

(j) p (zi;2 ) (j 1)
Pr jsi;2 = 0 = (j 1) + (1 (j 1) )
p (zi;2 ) q (zi;2 )) (1

(j) (1 p (zi;2 )) (j 1)
Pr jsi;2 = 1 = (j 1) + q (z ) (1 (j 1) )
(1 p (zi;2 )) i;2

n oT
We then use a data augmentation technique to draw a set of latent variables si;t
from truncated normal distributions [see Tanner and Wong (1987) and Albert and Chib
(j) (j)
(1993)]. If si;t = 0, si;t is drawn from the truncated normal distribution N 0;i z;i zi;t ; 1
(j) (j)
where the truncation is at 0 from the right since si;t is negative by de…nition if si;t = 0. If
(j) (j)
si;t = 1, si;t is drawn from the truncated normal distribution N 0;i z;i zi;t s;i ; 1

where the truncation is at 0 from the left.

n oT
We use si;t to generate draws for i. The prior for i is

0 2 31
B 61 0 07C
B 6 7C
i NB 6 7C
B0; 60 1000 07C .
@ 4 5A
0 0 1

The prior over z;i is relatively di¤use as we have little information regarding the likely scale
of the coe¢ cient; however, the results were qualitatively very similar for substantially more
di¤use and less di¤use priors over z;i . It can be shown (see, for example, Filardo and
Gordon, 1998) that the posterior distribution of i has the form

i N i ; A ;i ;

where A ;i = (I3 + Wi0 Wi ) 1 , i = A ;i Wi0 si , si = [si;0 ; si;1 ; :::; si;T ]0 , and Wi is a T x3

matrix with rows [1; zi;t , si;t ]. We repeat this process J = 10; 000 times for each city and
discard the …rst 5; 000 draws as burn-in iterations. See Filardo and Gordon (1998) for
additional details on the Gibbs sampling approach to estimating TVTP Markov-switching
Table 12 shows the 68 percent coverage intervals for each of our candidate leading indi-

cator variables on the transition probabilities in our TVTP Markov-switching models. The
…rst two columns show the signi…cance of the in‡uence of permits on p (zi;t ) and q (zi;t ) when
zi;t = [1; P ermM Ai;t 1 ]. The next two columns show the e¤ect of lagged house price
growth on employment by setting zi;t = [1; Pr iceM Ai;t 1 ]. The …nal columns show the
68 percent coverage intervals when zi;t = [1; F in1M Ai;t 1 ] ; where F in1 is the …rst principal
component constructed from the …nancial variables described in the previous section. Nei-
ther of our housing variables signi…cantly a¤ect the regime transition probabilities in any of
our cities or at the national level. The parameter estimates for the regime means and the
in‡uence of the current regime on transition probabilities are highly signi…cant and have the
expected signs, but are not shown in the interests of space. However, there is no evidence
that …nancial factors in‡uence the transition probabilities either.

5 Conclusions

We assessed the robustness of the relationship between housing and employment over the
business cycle in a set of 36 U.S. cities. We found permits to be a good leading indicator
of employment in less than half of the cities we analyze. We also found that permits are
not consistently a leading indicator for employment at the national level once we control for
…nancial factors. In short, we conclude that housing is not a major driver of the business
These results stand in some contrast to the previous literature linking housing to business
cycles. At a national level, housing appears to be an important driver of cyclical ‡uctuations.
If this causality truly held, we would expect the link to be preserved at the MSA level, as
housing spillovers into business cycle indicators (e.g., employment) should remain localized.
This may be especially important to verify at the local level in light of …ndings by Del Negro
and Otrok (2007) and others that housing shocks are primarily local phenomena. Our
results, however, appear at odds with the …nding that there exists a direct channel (whether
it be wealth e¤ects or housing capital e¤ects) from housing markets into employment.

Our results should not, however, be seen as evidence that there is no relationship between
housing and employment over the business cycle. Clearly, theoretical models such as Davis
and Heathcote (2005) illustrate that there is a relationship. However the lack of consistency
in the relationship at the city-level suggests that the relationship may be more complicated
than simple causal stories wherein a rise in house prices raises wealth, leads households to
consume more, and then leads to an economic expansion. Furthermore, the extent to which
we …nd frequency-dependence in the relationships suggests that increases in house prices
may have a di¤erent impact in the short term than in the long term.
While there is no question that housing played a pivotal role in the recession beginning
in December 2007, our …ndings suggest caution in assuming that future recessions will …t
the current pattern. The diversity we …nd across cities in the relationship between housing
and employment –in many cities, a rise in permits or house prices appears to be associated
with a subsequent decline in employment –is further evidence that, as Temin (1998) …nds,
business cycles are tremendously diverse.

Figure 1: Housing and Employment at the National Level (1982Q1=100)

Figure 2: Housing and Employment in US Cities
Employment = solid blue line; house prices = dashed red line; permits (values) = dotted green
line; permits (units) = dot-dashed pink line.

Figure 3: Housing and Employment in US Cities
Employment = solid blue line; house prices = dashed red line; permits (values) = dotted green
line; permits (units) = dot-dashed pink line.

Figure 4: Housing and Employment in US Cities
Employment = solid blue line; house prices = dashed red line; permits (values) = dotted green
line; permits (units) = dot-dashed pink line.

Figure 5: Housing and Employment in US Cities
Employment = solid blue line; house prices = dashed red line; permits (values) = dotted green
line; permits (units) = dot-dashed pink line.


Albert, James H. and Siddhartha Chib, 1993. “Bayesian Analysis of Binary and Polychoto-
mous Response Data”, Journal of the American Statistical Association, 88, 669-679.

Aron, Janine, John Muellbauer, and Anthony Murphy, 2006. “Housing Wealth and UK
Consumption”, Economic Outlook, 30, 11-20.

Barth, Marvin J. III and Valerie A. Ramey, 2002. “The Cost Channel of Monetary Trans-
mission” in Ben S. Bernanke and Kenneth S. Rogo¤, eds., NBER Macroeconomics
Annual 2001, 199-240.

Bernanke, Ben S., Jean Boivin, and Piotr Eliasz, 2005. “Measuring the E¤ects of Monetary
Policy: A Factor-Augmented Vector Autoregressive (FAVAR) Approach”, Quarterly
Journal of Economics, 120, 387-422.

Boivin, Jean, Marc P. Giannoni, and Ilian Mihov, 2007. “Sticky Prices and Monetary
Policy: Evidence from Disaggregated U.S. Data”, American Economic Review, forth-

Buiter, Willem H., 2008. “Housing Wealth Isn’t Wealth”, NBER Working Paper No. 14204.

Carlino, Gerald and Robert DeFina, 1998. “The Di¤erential Regional E¤ects of Monetary
Policy”, Review of Economics and Statistics, 80, 572-587.

___________, 1999. “The Di¤erential Regional E¤ects of Monetary Policy: Evidence

from the U.S. States”, Journal of Regional Science, 39, 339-358.

Carter, C.K. and R. Kohn, 1994. “On Gibbs Sampling for State Space Models”, Biometrika,
81, 541-553.

Case, Karl E., John M. Quigley, and Robert J. Shiller, 2005. “Comparing Wealth E¤ects:
The Stock Market versus the Housing Market”, Advances in Macroeconomics, 5, 1-32.

Casella, George and Edward I. George, 1992. “Explaining the Gibbs Sampler”, American
Statistician, 46, 167-174.

Chib, Siddhartha and Edward Greenberg, 1996. “Markov Chain Monte Carlo Simulation
Methods in Econometrics”, Econometric Theory, 12, 409-431.

Corbae, Dean and Sam Ouliaris, 2006. “Extracting Cycles from Nonstationary Data” in
Dean Corbae, Steven N. Durlauf, and Bruce E. Hansen, eds., Econometric Theory
and Practice: Frontiers of Analysis and Applied Research. Cambridge: Cambridge
University Press.

Corbae, Dean, Sam Ouliaris, and Peter C.B. Phillips, 2002. “Band Spectral Regression
with Trending Data”, Econometrica, 70, 1067-1109.

Davis, Morris A. and Jonathan Heathcote, 2005. “Housing and the Business Cycle”, Inter-
national Economic Review, 46, 751-784.

Davis, Steven J. and John Haltiwanger, 2001. “Sectoral Job Creation and Destruction
Responses to Oil Price Changes”, Journal of Monetary Economics, 48, 465-512.

Del Negro, Marco and Christopher Otrok, 2007. “99 Luftballons: Monetary Policy and the
House Price Boom Across U.S. States”, Journal of Monetary Economics, 54, 1962-1985.

Diebold, Francis X., Joon-Haeng Lee, and Gretchen C. Weinbach, 1994. “Regime Switching
with Time-Varying Transition Probabilities”in Colin P. Hargreaves, ed., Nonstationary
Time Series Analysis and Cointegration, 283-302. Oxford: Oxford University Press.

Durland, J. Michael and Thomas H. McCurdy, 1994. “Duration-Dependent Transitions in

a Markov Model of U.S. GNP Growth”, Journal of Business and Economic Statistics,
12, 279-288.

Filardo, Andrew J., 1994. “Business-Cycle Phases and Their Transitional Dynamics”,
Journal of Business and Economic Statistics, 12, 299-308.

Filardo, Andrew J. and Stephen F. Gordon, 1998. “Business Cycle Durations”, Journal of
Econometrics, 85, 99-123.

Fratantoni, Michael and Scott Schuh, 2003. “Monetary Policy, Housing, and Heterogeneous
Regional Markets”, Journal of Money, Credit, and Banking, 35, 557-589.

Gelfand, Alan E. and Adrian F.M. Smith, 1990. “Sampling-Based Approaches to Calculat-
ing Marginal Densities”, Journal of the American Statistical Association, 85, 398-409.

Glaeser, Edward L. and Joseph Gyourko, 2007. “Housing Dynamics”, Harvard Institute of
Economic Research Discussion Paper No. 2137.

Goldfeld, Stephen M. and Richard E. Quandt, 1973. “A Markov Model for Switching
Regressions”, Journal of Econometrics, 1, 3-15.

Hamilton, James D., 1989. “A New Approach to the Economic Analysis of Nonstationary
Time Series and the Business Cycle”, Econometrica, 57, 357-384.

_____________, 2005. “What’s Real About the Business Cycle?”, Federal Reserve
Bank of St. Louis Review, 87, 435-452.

_____________, 2008. “Daily Monetary Policy Shocks and New Home Sales”, Jour-
nal of Monetary Economics, 55, 1171-1190.

Iacoviello, Matteo, 2005. “House Prices, Borrowing Constraints, and Monetary Policy in
the Business Cycle”, American Economic Review, 95, 739-764.

Kan, Kamhon, Sunny Kai-Sun Kwong, and Charles Ka-Yui Leung, 2004. “The Dynamics
and Volatility of Commercial and Residential Property Prices: Theory and Evidence”,
Journal of Regional Science, 44, 95-123.

Leamer, Edward E., 2007. “Housing IS the Business Cycle”, NBER Working Paper No.

Maheu, John M. and Thomas H. McCurdy, 2000a. “Identifying Bull and Bear Markets in
Stock Returns”, Journal of Business and Economic Statistics, 18, 100-112.

_____________, 2000b. “Volatility Dynamics Under Duration-Dependent Mixing”,

Journal of Empirical Finance, 7, 345-372.

Mankiw, N. Gregory and David N. Weil, 1989. “The Baby Boom, the Baby Bust, and the
Housing Market”, Regional Science and Urban Economics, 19, 235-258.

Masson, Paul and Francisco J. Ruge-Murcia, 2005. “Explaining the Transition between
Exchange Rate Regimes”, Scandinavian Journal of Economics, 107, 261-278.

Muellbauer, John N., 2007. “Housing, Credit, and Consumer Expenditure”, Proceedings
of the Jackson Hole Symposium on Housing, Housing Finance, and Monetary Policy.

Owyang, Michael T., Jeremy Piger, and Howard J. Wall, 2005. “Business Cycle Phases in
U.S. States”, Review of Economics and Statistics, 87, 604-616.

Owyang, Michael T., Jeremy M. Piger, Howard J. Wall, and Christopher H. Wheeler, 2008.
“The Economic Performance of Cities: A Markov-Switching Approach”, Journal of
Urban Economics, 64, 538-550.

Owyang, Michael T., David E. Rapach, and Howard J. Wall, 2009. “States and the Business
Cycle”, Journal of Urban Economics, 65, 181-194.

Perez-Quiros, Gabriel and Allan Timmermann, 2000. “Firm Size and Cyclical Variations
in Stock Returns”, Journal of Finance, 55, 1229-1262.

Poterba, James M., 1991. “House Price Dynamics: The Role of Tax Policy and Demogra-
phy”, Brookings Papers on Economic Activity, 1991, 143-203.

Slacalek, Jirka, 2006. “What Drives Personal Consumption? The Role of Housing and
Financial Wealth”, manuscript, DIW Berlin.

Smets, Frank, 2007. “Commentary: Housing is the Business Cycle”, Proceedings of the
Jackson Hole Symposium on Housing, Housing Finance, and Monetary Policy.

Tanner, Martin A. and Wing Hung Wong, 1987. “The Calculation of Posterior Distributions
by Data Augmentation”, Journal of the American Statistical Association, 82, 528-540.

Temin, Peter, 1998. “The Causes of American Business Cycles: An Essay in Economic
Historiography”, NBER Working Paper No. 6692.