You are on page 1of 32

2008 V36 2: pp.

371–402

REAL ESTATE
ECONOMICS

Time-Series Characteristics and Long-Run


Equilibrium for Major Australian
Office Markets
Anthony J. De Francesco∗

While much attention has focused on the modeling of office property markets,
little emphasis has been put on distinguishing between permanent and tempo-
rary effects. This article attempts to address this issue in the context of the rental
adjustment mechanism and the demand–employment relationship for major
Australian central business district office markets. It is shown that, by allowing
the natural vacancy rate and the work–space ratio to be endogenously deter-
mined, it offers richer model specifications that permit a partitioning between
long-run and short-run influences. This is achieved by employing econometric
techniques that examine the stochastic behavior of time series data. It is found
that, while equilibrium relationships exist (between the vacancy rate and rent,
and demand and employment), other macroeconomic variables are found to
be relevant cyclical determinants.

Introduction
Over the last two decades, much attention has focused on the modeling of
commercial office property markets. This has mainly been a response to the
pronounced nature of office rental cycles, characterized by extensive periods
of booms and busts. Accordingly, this has necessitated numerous studies that
have identified key influences on office markets to better understand the in-
teractions between: demand and supply dynamics, demand and employment,
supply and rents and rents and vacancies. These relationships are often consid-
ered to be the building blocks of empirical office property market econometric
models. Some studies exploring these relationships include Wheaton (1990),
Hendershott and Kane (1995), DiPasquale and Wheaton (1996), Wheaton, Torto
and Evans (1997), Hendershott, Lizieri and Matysiak (HLM 1999), Wheaton
(1999) and Sivitanidou and Sivitanides (1999). With reference to Australian
office markets, relevant studies include the work by Hendershott (1995, 1996,
2000), Hendershott, MacGregor and Tse (HMT 2002) and Tonelli, Cowley and
Boyd (2004).

School of Economics, University of New South Wales, Sydney, 2052, Australia and
Colonial First State, Sydney, New South Wales, 2000, Australia or adefrancesco@
colonialfirststate.com.au.

C 2008 American Real Estate and Urban Economics Association
372 De Francesco

This article extends the research in this field by modeling the major Australian
Central Business District (CBD) office markets of Sydney and Melbourne. The
size of the Australian office sector is approximately 20 million square meters1
(sqm) of which the CBD office market represents about 14 million (or 70% of
total stock). Major CBD office markets include Sydney, Melbourne, Brisbane,
Canberra, Perth, Adelaide and Hobart. Sydney is the largest market, with an
estimated stock of 4.7 million sqm, followed by the Melbourne market, which,
although significant, is considerably smaller in size with 3.4 million sqm. The
combined market share of the Sydney and Melbourne CBD office markets is
58% of total CBD office stock.

The article also revisits the modeling strategy of key relationships for the office
market, specifically, the rental adjustment mechanism and the demand equation.
The article uses a time-series modeling approach and employs cointegrating
analysis. This modeling strategy is motivated by key space market variables,
such as rents, vacancies and demand, displaying approximate nonstationary
characteristics. Importantly, it is shown that conventional models for the rental
mechanism and demand are enriched by allowing the natural vacancy rate
and the work–space ratio to be endogenous, that is, dependent on a series of
intertemporal variable determinants.

Furthermore, this article attempts to identify short-run and long-run deter-


minants in an effort to better understand the interplay between market cy-
cles and trends for the Sydney and Melbourne CBD office markets. While
past studies have contributed toward identification and estimation of key de-
terminants of office property markets, little attention has been given to dis-
cerning whether such determinants have a permanent or transitory influence.
Arguably, this issue is of greater relevance for many real estate market data
series, which are typically of small samples or time spans with much variabil-
ity. It is found that various macroeconomic variables significantly influence
the equilibrium and cyclical nature of office markets; however, these influ-
ences vary across the two markets, reflecting their differences in space market
fundamentals.

This article is organized as follows. The next section provides a brief discussion
of the data employed in this study. The third section presents an overview of the
cyclical behavior of the Sydney and Melbourne CBD office markets. The fourth
section embarks on a discussion on modeling office property markets. First, the
error-correction model (ECM) is outlined as an adequate modeling framework
that distinguishes between short-run and long-run effects. Second, discussion

1
Figures as of January 2006 and sourced from the Property Council of Australia (2006).
Time-Series Characteristics and Long-Run Equilibrium 373

centers on the basic models for the rental adjustment mechanism and office de-
mand and how these models can be extended to incorporate short-run dynamics.
The fifth section presents empirical results. Specifically, it reports test results
for nonstationarity for a variety of pertinent space market and macroeconomic
variables. Cointegration test results and long-run parameter estimates are also
reported. The sixth section presents concluding remarks.

Data
This study uses annual time-series data spanning the period 1970 to 2003.
The data set comprises various space market variables and macroeconomic
indicators. A brief description of each data type is outlined below.

Space Market Variables


The primary real estate data used in this research are sourced from The Real
Estate Intelligence Service (REIS) Report for Australia produced by Jones
Lang La Salle (JLL 2004). The JLL data set provides a consistent reporting
framework for space markets, split by geographical location and sector across
Australia and the Asian-Pacific region. The report provides lengthy historical
time series for both real space market measures and pricing variables. Pricing
information includes various rental types, investment yields and capital value
indices. While the Property Council of Australia (PCA) also produces an office
market report, it does not report rental data and its historical time series does
not extend back to 1970.

Total supply of office space (S t ) represents total available stock of office build-
ings at a given point in time. The change in supply (S t ) is often called net
additions. This measure is obtained by monitoring the level of new comple-
tions, refurbishments and withdrawals from stock. Total demand (D t ) repre-
sents occupied office space, which includes space that has been leased up but
not necessarily physically occupied. The change in demand (D t ) is called
net absorption, reflecting additional space leased less space relinquished by
the tenant. Vacancies (At ) represent the excess supply of space available in
the market, computed as the difference between available stock and occupied
space. Vacancy data are obtained by tracking actual vacancy levels by building
across grades. The vacancy rate (V t ) is computed by dividing vacancies by
available stock. With the exception of the vacancy rate, all of the above series
are expressed in sqm.

Rental data used in this study refer to market rents, which is generally based on
new leasing transactions. The nominal rent (Rnt ) represents the average prime
nominal gross effective rent, which is a simple average of market rents captured
374 De Francesco

on premium and A-grade asset buildings. The classification of building quality


by grade generally follows that adopted by the PCA. The gross effective rent
includes outgoings (i.e., regular occupancy costs which typically include local
and state taxes/rates, insurance, property management charges, security and
cleaning) but excludes incentives. Effective rents are computed by deducting
the present value of incentives, amortized over the term of the lease, which is
assumed to be 10 years, from the face rent (i.e., the market rentals as represented
on leases). All rents are expressed as $/sqm. The real rent (Rrt ) is computed by
deflating the nominal gross effective rent by the inflation rate. The inflation rate
is measured using the Consumer Price Index (CPI). The CPI series is sourced
from the Reserve Bank of Australia’s (RBA) statistical tables, accessible at
www.rba.gov.au.

Macroeconomic Indicators
Key macroeconomic variables used in this study include economic growth,
interest rates, employment and unemployment.

Economic output (Y t ) is represented by the seasonally adjusted real gross do-


mestic product (GDP) measure. The unemployment rate (u t ) represents a spliced
data series. It combines data sourced from Table G07 between 1978 and 2003
and data sourced from Table 4.15 Occasional Paper #8 between 1969 and 1997.
Total employment (ETt ) and industry employment (EIt ) also represent spliced
data series. Data for these series are sourced from Table G07 between 1978 and
2003 and data sourced from Table 4.3 Occasional Paper #8 between 1969 and
1997. The nominal interest rate (i nt ) is represented by the 90-day bill rate or the
10-year Treasury bond rate series, sourced from Tables F1 and F2. The real in-
terest rates (i rt ) is constructed by deducting the annualized inflation rate (based
on the CPI series) from the nominal interest rate series. All data are sourced
from the RBA statistical tables.

Two CBD white-collar employment (WCE) data series are used for each CBD
office market: one based on total industries (EwceTt ) and the other based on
specific industries (EwceIt ). For the Sydney CBD market, specific industries
include: finance and insurance, property and business services and information
and technology. For the Melbourne CBD market, industries include: finance
and insurance, property and business services, information and technology and
the public sector. WCE data are derived by splicing data series sourced from
Access Economics (2004) (for the period 1980 to 2003) with stand-alone office
workforce data series sourced from BIS-Shrapnel (2004) (for the period prior
to 1980).
Time-Series Characteristics and Long-Run Equilibrium 375

Movements in Demand, Supply and Real Rents


The synchronicity between supply and demand for the Sydney and Melbourne
CBD office markets is illustrated by linear detrending the series, which are
depicted in Figures 1A and 1B. These charts highlight the phases of the real
estate supply and demand cycles as noted in Born and Pyhrr (1994) and Pyhrr,

Figure 1 

A Space Market Dynamics


Sydney CBD Office Market
400 1000

300 900

800
200
space (000 sqm)

700

RGER ($/sqm)
100
600

0 500

400
-100
300
-200
detrended demand 200
-300 detrended supply
100
R.G.E.R.
-400 0
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003

B Space Market Dynamics


Melbourne CBD Office Market
600 700
detrended demand
450 detrended supply 600
R.G.E.R.
300 500
space (000 sqm)

RGER ($/sqm)

150 400

0 300

-150 200

-300 100

-450 0
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003
376 De Francesco

Roulac and Born (1999). In an attempt to trace movements in excess supply


and demand fundamentals of the physical space market to prices, the real gross
effective rent is also shown. While the number of observations is small, the
span of the data is deemed reasonable as it covers the two major economic
downturns of 1982–1983 and 1990–1991 as well as the property downturns of
the mid 1970s and early 1990s.

With reference to the Sydney office market, up to the early 1980s supply and
demand essentially moved in line. The large divergence experienced in the
late 1980s and early 1990s highlights the severity of the recession due to over
supply and dwindling demand. Kummerow (1999) notes that the oversupply
in Australia is partly explained by financial deregulation in the 1980s, which
led to a swift rise in the availability of capital funding. The fall in demand
was the result of a government-induced tightening of monetary policy in an
attempt to stabilize an overheating economy and restore a deteriorating external
balance position. The co-movement of supply and demand from 1995 to early
2000 signals the Sydney market reverting back to equilibrium. Since 2000
the market has experienced significant erosion in demand while supply has
remained relatively moderate. This was mainly attributable to a combination of
slowing domestic economic activity, softening employment growth, especially
in full-time employment, and growing business uncertainty due to a variety
of exogenous global events, such as the terrorist bombing of the World Trade
Towers in the United States on September 11, 2001, and the start of the Iraq
War in 2003.

Furthermore, Figures 1A and 2A illustrate that periods of excess supply or


rising vacancies have been associated with a fall in the real rent and vice versa.
From the early 1970s to 1978, real rents in Sydney had been falling, then
experienced a gradual rise (over the next 10 years) peaking in 1988 and then
falling more sharply (over the next 5 years) reaching a trough in 1993, then
again experiencing a gradual recovery up until 2000 when the market again
experienced a softening in rentals. Movements in the vacancy rate have also
been dramatic, rising sharply to 13% in 1976, bottoming out at 1.2% in 1987 and
followed by another steep rise to 23% in 1992. This feature is regarded by some
as a reflection of different phases of the market cycle. For instance, Gordon,
Mosbaugh and Canter (1996) argue that movements in vacancy rates (or office
market volatility) are likely to be affected by different factors at different stages
of the property cycle. Others regard these events as evidence of speculative
bubbles, that is, where variations cannot be explained by market fundamentals.
Hendershott (2000), in exploring the existence of property asset bubbles in the
Sydney CBD office market, argues that the excessive construction and vacancies
in the late 1980s is a reflection of excess price volatility or an asset price bubble.
Time-Series Characteristics and Long-Run Equilibrium 377

It is interesting to note that the market exhibits prolonged periods of high and
then low for vacancy rates and rents. This time-series feature is often referred
to as persistence or a slow mean-reverting process,2 a characterization com-
monly found in macroeconomic indicators such as unemployment rates and
interest rates. Grenadier (1995) argues that market persistence is underpinned
by two phenomena. First, the reluctance of owners to adjust occupancy levels
despite large shifts in renter demand and second, the occurrence of periods of
sustained overbuilding, both of which are caused by a combination of demand
uncertainty, adjustment costs and construction lags. Hendershott (2000) com-
ments that property markets have to be mean reverting given the incentives of
developers to build when values rise substantially above replacement cost and
vice versa. In this context, one can consider supply and demand, on average,
to be synchronized and vacancies to represent stochastic deviations from equi-
librium. On a practical note, even in equilibrium one would not expect zero
vacancies due to a costly matching process between tenants and buildings as
well as uncertainty over future space requirements. However, one may expect
the amplitude of vacancy cycles to be reduced through rising market efficiency
and greater alignment of management interests with capital.

With reference to the Melbourne CBD market, a comparison of Figure 1A


with 1B shows that, while the rental cycles for the two markets are generally
in line, the underlying behavior of supply and demand fundamentals are quite
different. Since the 1990s, while demand has shown mild perturbations away
from trend, supply-side fluctuations have been more pronounced and hence
primarily contributed toward the movements in rents. For instance, the rapid
decline in the real rent in the early 1990s was the result of a huge oversupply
against a fairly steady demand. In contrast, the slow recovery in the real rent
in the late 1990s was underpinned by large withdrawals from stock—mainly
conversions to residential—against relatively sluggish demand. More recently,
while the rise in the vacancy rate for the Sydney market was mainly due to a
sluggish demand, the rise in the vacancy rate for the Melbourne market was
mainly due to a mini supply-side cycle.

Modeling Strategy and Office Market Relationships


This section first outlines a modeling strategy that allows for a partitioning of
variables that have a temporary influence from those which have a permanent
influence. Second, it outlines some of the basic models for the rental adjustment
mechanism and illustrates how the model can be recast in an ECM framework

2
See Nelson and Plosser (1982) and Campbell and Mankiw (1987, 1989).
378 De Francesco

when making the natural vacancy rate endogenous. Third, discussion focuses on
office demand where it is shown that the basic model of demand and employment
can be expanded to incorporate either short-run dynamics or additional variables
that constitute part of the long-run relationship. This is facilitated by allowing
the work–space ratio to be endogenously determined.

Modeling Strategy
Like many economic and financial time-series data, real estate series also exhibit
nonstationary behavior, as noted in Myer, Chaudhry and Webb (1997), Wilson,
Okunev and Webb (1998) and Shilton (2000). This is primarily due to the
nature of the data, characterized by low frequencies (commonly annual) with
relatively short time spans.3 Under this case, meaningful estimation is obtained
when the order of integration for the dependent variable is of the same order of
integration as the regressors or any linear combination of the regressors. Such
equations are referred to as balanced equations. When the order of integration
is not the same, then applying operation of least squares (OLS) estimation will
result in inconsistent estimated parameters. Hence, the application of standard
statistical procedures may produce misleading inferences.4

When modeling with nonstationary variables, a long-run equilibrium relation-


ship between two or more variables can be formally examined using a cointe-
grating model of the form

y1t = θt + b y2t + vt , (1)

where y 1t represents a scalar dimension I(1) time-series variable so that first


differencing of the variable induces stationarity or I(0), y2t is a g × 1 vector of
g I(1) variables and [v t , y2t ] = (L)ε t where ε t = [ε 1t , ε 2t ] is an i.i.d. process
with mean zero, finite variance, E(εt ε t ) =  ε and fourth-order moments. In this
model, if it can be shown that the variables y 1t and y2t are I(1) and v t is I(0), then
y 1t and y2t are said to be cointegrated and a long-run equilibrium relationship
is found to exist. Under this scenario, b represents the cointegrating vector.

While Model (1) has merit in its simplicity, it ignores the role of short-run
dynamics. The relevance of such short-term influences is noted by Scott and

3
Hakkio and Rush (1991) argue that data with long time spans are required to discern
whether the time series are cointegrated. However, Maddala and Kim (1998) argue that it
is not the long time spans of the time series that govern the detection of cointegration but
rather the speed of adjustment toward equilibrium, as this defines the long-run period.
4
For a further discussion on this topic, refer to Banerjee et al. (1993) and Marmol
(1996).
Time-Series Characteristics and Long-Run Equilibrium 379

Judge (2000) who time-series analyze the commercial property cycle in Britain
by decomposing the cyclical component of property values from the long-term
component. As such, an alternative modeling strategy that encompasses both
the long-run equilibrium relationship and the short-run dynamics is the use of an
ECM. The main characteristic of an ECM is the notion of an equilibrium long-
run relationship and the inclusion of past disequilibrium as explanatory variables
in the dynamic behavior of current variables. Following Boswijk (1994), a
single-equation ECM of a time series y 1t conditional upon y2t can be expressed
as follows
y1t = θt + a y2t + δ(y1t−1 − b y2t−1 )

p−1 (2)
+ (γ j y1t− j + β j y2t− j ) + vt .
j=1

In comparison to Model (1), it can be clearly noted that Model (2) captures the
long-run equilibrium component as set out in (1).

Over the last decade, the literature on cointegration has broadened to consider
exogenous stationary variables or covariates5 to capture additional multivari-
ate information on short-run dynamics as part of the modeling strategy for
integrated time series. The relevance of properly accommodating for these
short-run dynamics in time-series methods is emphasized in studies by Hansen
(1995), Seo (1998) and De Francesco (2001). These studies illustrate that the
inclusion of these covariates in tests for unit roots and cointegration have the
potential to deliver substantial power6 gains, if they indeed are part of the data-
generating
q process. Model (2), augmented with correlated stationary covariates
terms, i=0 xt−i , can be expressed as

y1t = θt∗ + a y2t + δ(y1t−1 − b y2t−1 + θ + )


p−1 
q
+ (γ j y1t− j + β j y2t− j ) + i xt−i + vt∗ . (3)
j=1 i=0

An important part of the modeling strategy involves testing for long-run rela-
tionships. While a vast number of tests have been proposed in the literature, the
more commonly applied tests to real estate markets include Engle-Granger’s

5
Refer to Hansen (1995) or Seo (1998) for a technical definition of a stationary covariate.
6
The power refers to the probability that the null hypothesis will be rejected given that
an alternative hypothesis is true. Low power suggests that the test is not very informative
to discriminate between the null and alternative hypothesis. However, when power is
high, the test can be considered to be very informative.
380 De Francesco

(EG 1987) Augmented Dickey-Fuller (ADF) test and Johansen (JH 1988, 1991)
maximum likelihood (ML) ratio trace test, based on a vector-autoregressive
(VAR) framework. An application of these methods to property markets can
be found in Chaudhry, Christie-David and Sackley (1999) and Chaudhry, Myer
and Webb (1999). While tests for identifying the presence of long-run equilib-
rium in Model (1) usually entails an examination of the estimated residuals,
tests conducted on Model (2) or (3) are motivated by the Granger representa-
tion theorem, which says that if a set of I(1) variables are cointegrated they can
be regarded as being generated by an ECM. One test of this type, based on a
single-equation framework, is the Wald test (W) proposed by Boswijk (1994).

Models of Rental Adjustments


The modeling of office rents is largely underpinned by the inverse relationship
between the vacancy rate and the real rent as depicted in Figures 2A and 2B.
These model specifications are generally based on two approaches. The first
involves an explicit modeling of the key determinants of supply and demand that
make up vacancies, usually within a system framework. The second approach,
which is the focus in this section, models the rental adjustment mechanism in
terms of a reduced-form single equation.

Consider the traditional rental adjustment formulation,

 ln Rr t = rr t = α(V̄t − Vt−q ) (4.1)

V̄t = β  Zt−q . (4.2)

Equation (4.1) specifies that percentage change in the real rental value (ln Rrt )
is caused by deviations of nominal vacancy rates (V t−q ) from the structural or
natural vacancy rate (V̄t ). Note that α (> 0) denotes the speed of rental adjust-
ment to equilibrium and q is the lag order intertemporally affecting V̄t . Other
variants of the rental adjustment model, as proposed by Clapp, Pollakowski
and Lynford (1992), involve replacing the dependent variable ln Rrt with the
change in the real rent, Rrt . Equation (4.2) specifies the function for the nat-
ural vacancy rate where Zt−q represents the matrix of determinants for the
natural vacancy rate and β  represents the corresponding parameter matrix that
requires estimation. Simply, the model implies that a rise (fall) in the vacancy
rate above (below) the natural vacancy rate will see rents falling (rising). Rosen
and Smith (1983) define the natural vacancy rate in a manner analogous to the
natural unemployment rate as the vacant stock required to facilitate the search
needs of tenants and landlords looking for office space and tenants, respectively.
Alternatively, Shilling, Sirmans and Corgel (SSC 1987) argue that the natural
vacancy rate is analogous to the optimal or desired inventory of vacant units
Time-Series Characteristics and Long-Run Equilibrium 381

Figure 2 

A Vacancies versus Rents


Sydney CBD Office Market
25% 1000
vacancy rate (lhs)
900
R.G.E.R. (rhs)
20% 800
700
vacancy rate

15% 600

real rent
500

10% 400

300
5% 200

100

0% 0
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003

Vacancies versus Rents


B
Melbourne CBD Office Market
25% 700
vacancy rate (lhs)
R.G.E.R (rhs) 600
20%
500
vacancy rate

15%
real rent

400

300
10%

200
5%
100

0% 0
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003

that maximizes landlords’ expected profits, hence depends on their expectations


with respect to office space demand and the marginal cost of holding vacant
units.

This model formulation provides a general specification, which encompasses


numerous models outlined in the literature. Model variations primarily hinge
on the specification of the natural vacancy rate or, more precisely, on the
382 De Francesco

composition of elements of Zt−q . In early research (Rosen 1984, SSC 1987)


the natural vacancy rate was assumed to be constant over time. In terms of
Model (4), this implies that

Zt−q = 1 (5.1)

and

V̄t = β0 . (5.2)

However, in later studies, such as Sivitanides (1997), the assumption of a con-


stant vacancy rate was relaxed, allowing it to be endogenously determined by
space market variables, such as net absorption, completions and changes in
vacancies as well as office employment growth. Mueller (1999), in examining
rental growth in office and industrial markets in the United States, finds that
demand and supply or vacancy rate is the major determinant of rental growth
rates. Although Sivitanides does not comment on the order of integration of
these variables, they can be considered to be stationary as their corresponding
level series can be characterized as being nonstationary. These stationary dy-
namic terms, denoted X t−q , are captured within the vector Zt−q . In terms of
Model (4), the modified specification can be represented by

Zt−q = [1,Xt−q ] (6.0)

and

V̄t = β0 + β  Xt−q . (6.1)

The natural vacancy rate may also be dependent on other cyclical (anticipated
and unanticipated) macroeconomic variables. For instance, Heckman (1985),
in modeling rental levels, considers not only vacancy rates but also economic
GNP, total employment and local unemployment rates. D’arcy, McGough and
Tsolacos (DMT 1997) find that the change in real GDP and the level of the
real short-term interest rate explain movements in European real office rents.
Also in emphasizing market dynamics, Slade (2000) comments that market
participants value the determinants of office rents differently during periods of
market decline, trough and recovery. Apart from deviations of the vacancy rate
from the natural vacancy rate, Hendershott (1995, 2002) argues deviations in
effective rents from the equilibrium rent as another key driver.

To accommodate for a variety of potential determinants (like the ones mentioned


above) which have permanent and short-run influences on the natural vacancy
rate, Model (6) can be generalized further by partitioning the vector Zt−q into
I(1) weakly exogenous variables (in the sense of Engle and Granger 1987)
denoted by Xt−q and I(0) exogenous stationary covariates denoted by Xt−q .
Under this formulation,
Time-Series Characteristics and Long-Run Equilibrium 383

Zt−q = [1, Xt−q , Xt−q ] (7.0)

and the natural vacancy rate Equation (4.2) now becomes


 
1
  
V̄t = β  Z t−q = β0 β1 β2    
 Xt−q  = β0 + β1 Xt−q + β2 Xt−q . (7.1)
Xt−q

Furthermore, substituting (7.1) into (4.1), the general estimation model7 can be
rewritten as

rrt = b0 + b1 Xt−q + b2 Xt−q − b3 Vt−q + vt (7.2)

and the implied natural vacancy rate is estimated as

V̄t = (b0 /b3 ) + (b1 X̄t−q /b3 ) + (b2 X̄t−q /b3 ),

where X̄t−q and X̄t−q represent mean values.

Model (7) provides a general structure that encompasses the traditional rent
adjustment equation, which may seem appropriate over long time spans, but it
also accommodates for the vacancy rate and rental level variables that may be
characterized as nonstationary I(1) processes over short time spans. As noted
earlier, a modeling strategy that attempts to synthesize between short-run and
long-run effects and the nonstationarity characteristics of pertinent variables is
provided by the ECM framework. Indeed, Model (7) can also give rise to an
ECM, depending on the specification of Zt−q . In illustrating this, consider that
the structural elements influencing the natural vacancy rate is given by
 
Zt−1 = 1, rrt−1 , Vt−1 , rrt−1 , X t−1 ∗
. (8.1)

Equation (8.1) implies that the natural vacancy rate is a function of permanent
influences, being the lagged real rental level, and a series of short-term influ-
ences, which include lagged changes in the vacancy rate, the real rent and an
exogenous factor. Substituting (8.1) into (4.1) with q = 1 and rearranging terms
gives

rrt = θ + δ[rrt−1 − bV t−1 ] + b21 Vt−1 + b22 rt−1 + b23 X t−1 + vt . (8.2)

7
Specification (7.2) can be generalized such that V t− p denotes an error-correction term
between supply and demand. That is, r rt = b0 + b 1 Xt−q + b 2 Xt−q − b3 [S t− p − ϕ D t− p ]
+ v t , where [S t− p − ϕ D t− p ] denotes an error-correction term with lag term p, and that
both S t and D t are I(1). Note that, when the market is in disequilibrium, ϕ is interpreted
as the short-run parameter, determining the vacancy rate level in the market at a given
point in time. However, when the market is in equilibrium,Vt− p = V̄ , and both S t− p and
D t− p will be cointegrated. In this case, ϕ represents the cointegrating parameter.
384 De Francesco

Clearly, Equation (8.2) is a specific case8 of the ECM in Equation (3). The error-
correction term consists of the real rent and vacancy rate. This specification
allows for the identification and estimation of the natural vacancy rate. It also
facilitates estimation of the corresponding long-run equilibrium rental level; a
feature that Mourouzi-Sivitanidou (2002) comments is absent from traditional
rental adjustment models.

Models of Office Demand


Attempts to forecast office demand usually involves modeling of office employ-
ment. Rosen (1984) estimated office demand as a function of total employment
in finance, insurance and real estate as well as the real rent. Wheaton (1987),
in his neoclassical office market model, estimates net absorption as a function
of employment and employment growth as well as the level of the real rent
and lagged demand. The following discussion attempts to synthesize many of
the empirical models of office demand under a common modeling strategy that
encompasses both long-run equilibrium and short-run dynamics.

Consider the simple relationship between demand for CBD office space (D t )
and office employment (E t ), which is typically expressed as
β
D t = ωt E t , (9.1)
where ωt (> 0) denotes the efficiency ratio. When β = 1, ωt is commonly
interpreted as the work–space ratio; the average amount of office space (in
sqm) per employed office worker. The above specification implies that a firm
will not increase its employment of labor without increasing the use of office
space by ωt .

Taking natural logarithm of (9.1) gives


dt − βet = wt (9.2a)
ωt = ω0 , (9.2b)
where d t = ln D t , et = ln E t and w t = ln ωt .

From (9.2), demand and employment will exhibit a long-run relationship, if


both d t and et are I(1) and w t is stationary I(0).

One case in which the component w t will be stationary is if ωt is an exoge-


nous constant, ω0 . This assumption is often made when modeling property
markets (see Lentz and Tse 1999) and seems reasonable as in the long-run the

8
The first two components of Equation (8.2) correspond to Equation (2 ) in HMT (2002).
Time-Series Characteristics and Long-Run Equilibrium 385

work–space ratio is expected to exhibit upper and lower bounds. For instance,
it is constrained from below, as it is impractical to completely devote total floor
space to employees. Conversely, it is bounded from above by an organization’s
profitability constraint. That is, it is uneconomical to devote an entire floor space
to a single employee (see Hakfoort and Lie 1996).9 In Model (9.2), if long-run
equilibrium does exist then it implies that demand for office space is principally
determined by office employment or WCE.

Although ωt is assumed constant or fixed in the long run, over shorter time
periods it can display deviations around its long-run mean which might be
associated with peaks and troughs of the business cycle. For instance, when
market conditions are sluggish and labor is shed, there tends to be an under-
utilization of space, also referred to as hidden vacancies. In this situation, the
work–space ratio rises. This was experienced by the Melbourne CBD office
market with the recession of the early 1990s. BIS-Shrapnel (2004b) report that
the work–space ratio increased from 23 in 1990 to 28 in 1992. Conversely,
when market conditions tighten and employment is strong, there tends to be an
overutilization of space, also referred to as pent-up demand. In this scenario,
the work–space ratio falls. These scenarios generally arise with unanticipated
changes in market conditions.

Hakfoort and Lie (1996), even though they do not distinguish between short-
and long-term effects, argue that the work–space ratio is variable and attempt
to analyze its determinants. They argue that demand for office space per worker
is conditional on the profit-maximization objective of the firm as it can be
considered to be another input into the production process of the office-using
firm. They comment that the office space per worker is a function of rent,10
expected growth of the firm and its associated uncertainty of this growth, the
lease period, substitution opportunities on the market and search and adjustment
costs.11

If Model (9.2) is misspecified, it might be due to the restrictive exogeneity


assumption for ωt . By relaxing this assumption and making ωt endogenous,
the model can be extended where the work–space ratio is a function of broader
macroeconomic and financial variables, denoted by the variable Z t−q . If Z t−q
represents I(1) variables (X t ), then Equation (9.2b) is reformulated as
9
Note that, while ωt is constant in a particular market, it may not be uniform across
markets due to variations in relative marginal costs of labor and office space.
10
This is based on the neoclassical model where office space per worker is chosen such
that its marginal productivity equates to marginal cost, that is, the real rent.
11
Higher search costs for new space and adjustment costs faced by the firm when taking
up office space (e.g., investment outlays in new equipment and layout or interiors) lower
the flexibility of firms to substitute existing space for new space.
386 De Francesco

λ
ωt = ω0 X t−q (9.3a)

and substituting (9.3a) into (9.1) gives


λ β
Dt = ω0 X t−q Et

or

dt − βet − λ ln(X t−q ) = ln(ω0 ). (9.4)

If Z t denotes only stationary variables (X t ), then it will impact on the short-
run dynamics of the model. Under this specification, Equation (9.2b) can be
reformulated as
 λ
ωt = ω0 e(X t−q ) , (9.3b)

with the corresponding demand equation given by

dt − βet − λX t−q = ln(ω0 ). (9.4/ )

As with the rental adjustment model, the demand model is enriched, allowing
for a greater variety of influences to determine the work–space ratio. As part of
the modeling strategy, the task is to find suitable candidates for Z t . A guide to
this can come from theory or prior research. HMT (2002) and Wheaton (1999)
postulate that demand is a joint function dependent on employment and the
real effective rent. Letting Z t denote the space real rental rate, Rrt ($ per sqm),
Equation (9.4) becomes
β
Dt = ω0 Rrt−λ E t , (9.5)

where λ and β denote the constant rental (or price) elasticity and the income
elasticity of the demand for space, respectively. This is the demand model
specified in HMT (2002). By restricting β = 1 one obtains the demand function
postulated by Wheaton (1999).

As mentioned earlier, HLM (1999), in modeling the London office market, ar-
gue that the real interest rate (i rt ) affects the space market through its effects
on the equilibrium rent. Under this scenario, one could also explore the pos-
sibility of the interest rate to be a long-term determinant of demand. This can
be encompassed in Model (9.3a) by substituting eir t in place of X t . Clearly,
the specification of Equation (9.4) provides a general model structure with in
which a variety of empirical models are nested.

Empirical Results
Empirical results are based on annual data spanning between the years 1970
and 2003. Because the data samples under empirical investigation may be
Time-Series Characteristics and Long-Run Equilibrium 387

considered relatively small, a variety of tests and estimation methods are ap-
plied for confirmatory data analysis. Before presenting modeling results for
particular relationships, the stochastic properties of individual time series are
examined.

Testing for Nonstationarity


This section presents results on nonstationarity tests for selected space market
variables as well as macroeconomic indicators. In all, space market level series
variables are considered to be nonstationary. While one may expect the vacancy
rate and rent series to be stationary over long time spans, they are considered to
display approximate nonstationary behavior over the given sample period. The
following provides a more detailed discussion of the results.

In determining the nonstationary properties of the data series, the ADF test for
the null of a unit root is conducted on a truncated sample between 1974 and
2003 to allow for possible lags in estimation. Test results for various space
market variables for the Sydney and Melbourne office markets are reported in
panels A and B of Table 1A, respectively. Test results for space market variables
are reported for level and first difference transformations. Due to the trending
nature of supply (S t ) and demand (D t ) variables, their tests were conducted with
linear time trends. Test results for supply, demand, vacancy, vacancy rate and
logged rentals suggest these series exhibit nonstationarity with their computed
statistics being smaller in absolute value than the 5% critical value (c.v.).12
However, test results for supply, demand and vacancies for the Sydney market
and the vacancy rate for the Melbourne market may be considered marginal as
the null is rejected at the 10% level.

Given the relatively short time span of the data exhibiting long swings, there is
concern that the above tests may have low power against relevant alternatives.13
With respect to tests for unit roots, Shiller and Perron (1985) and Perron (1989)
using Monte Carlo experiments find that power depends more on the time span
of the data set rather than the frequency of observations. Indeed, test results on
selected variables displayed significant variability when estimated over different
sample periods.

12
Decision rules are based on the 5% level of significance. This follows from the work
of Neyman and Pearson (1928) who assume that a type I error (i.e., reject the null when
it is in fact true) is likely to be more serious in practice than a type II error (i.e., not reject
the null when it is in fact false).
13
One could also consider applying the KPSS (Kwiatkowski et al. 1992), which tests for
the null of stationarity. The test requires the estimation of the long-run variance, which
can be computed using the Bartlett window and the truncation parameter, , based on
the procedures outlined in Andrews (1991).
Table 1A  Tests for the null of nonstationarity—space market variables.
Estimation period: 1974–2003 (30 observations)

Panel A: Sydney CBD Panel B: Melbourne CBD


Office Market Office Market
388 De Francesco

ADF1 Tests ADF2 Tests


Deterministic Standard3 Bootstrapped4 Standard3 Bootstrapped4
Series Symbol Terms c.v. c.v. c.v. c.v.
Levels
Office stock St Constant, trend −3.29∗ −3.29∗ −2.23 −2.23
Occupied space Dt Constant, trend −3.45∗ −3.45∗ −1.39 −1.39
Vacancy At Constant −2.78∗ −2.78∗ −2.60 −2.60∗
Vacancy rate Vt Constant −2.78∗ −2.78∗ −2.91∗ −2.91∗
Logged real rent rrt Constant −2.59 −2.59 −2.34 −2.34
First difference
Office stock St Constant/none −3.44∗∗ n.a. −2.52∗∗ n.a.
Occupied space Dt Constant/none −3.42∗∗ n.a. −4.55∗∗∗ n.a.
Vacancy rate V t None −3.44∗∗∗ n.a. −3.38∗∗∗ n.a.
Logged real rent rrt None −3.40∗∗∗ n.a. −3.12∗∗∗ n.a.

Note: ∗ , ∗∗ , ∗∗∗ denote significance at the 10%, 5% and 1% level, respectively. ADF = Augmented Dickey-Fuller; c.v. = critical value.
1
Autoregressive lag lengths for Sydney variables (listed down the column): 1, 1, 1, 1, 1, 2, 1, 0 and 0.
2
Autoregressive lag lengths for Melbourne variables: 1, 0, 1, 1, 2, 1, 0, 0 and 1.
3
Stanard critical values for ADF test taken from MacKinnon (1996).
4
Critical values for ADF test computed from bootstrapping exercise based on 10,000 replications.
Time-Series Characteristics and Long-Run Equilibrium 389

Table 1B  Tests for the null of nonstationarity—macroeconomic variables.


Estimation period: 1974–2003 (30 observations)

Deterministic Lag ADF1


Series Symbol Coverage Terms Length Test
Real GDP Yt National Constant, trend 2 −0.36
Employment - total ETt National Constant, trend 2 −3.00
Employment - industry EIt National Constant, trend 0 −2.25
Employment - full time EFTt National Constant, trend 1 −3.84∗∗
WCE - total EwceTt Sydney Constant, trend 1 −2.86
Melbourne Constant, trend 1 −2.57
WCE - industry EwceIt Sydney Constant, trend 1 −3.17
Melbourne Constant, trend 1 −2.64
Unemployment rate ut National Constant 1 −3.19∗∗
NSW Constant 1 −3.08∗∗
VIC Constant 1 −2.78∗
Treasury bond rate inT 10Yt Nominal Constant 0 −0.93
irT 10Yt Real Constant 1 −2.04
Bank bill rate inB90Dt Nominal Constant 2 −0.74
irB90Dt Real Constant 2 −1.67

Note: ∗ , ∗∗ , ∗∗∗ denote significance at the 10%, 5% and 1% level, respectively. ADF
= Augmented Dickey-Fuller; GDP = gross domestic product; WCE = white-collar
employment.
1
Critical values for ADF test taken from MacKinnon (1996).

In an attempt to address the robustness of the ADF test results, a bootstrapped


exercise was conducted to obtain in-sample critical values based on 10,000
replications. Bootstrapped critical values were found to be close in value to
standard critical values reported in MacKinnon (1996), resulting in inferences
being generally unchanged. The only difference emerged with the vacancy level
series for the Melbourne CBD office market where the bootstrapped critical
values suggest rejection of the null at the 10% level.

To further gauge the order of integration on these series, tests were also applied
on their first differences. Tests results for net additions, net absorption, the
change in the vacancy rate and the change in logged rentals indicate rejection
of the null of nonstationarity as computed test statistics are all larger in absolute
value than the 5% critical value. This suggests that these differenced series are
stationary or integrated of order zero. These results adds support that corre-
sponding level transformations of these series are deemed to be characterized
as I(1) nonstationary series.

The ADF test was also applied to a variety of macroeconomic indicators with re-
sults reported in Table 1B. Test results indicate that most of the macroeconomic
390 De Francesco

series are considered to be nonstationary with the null not being rejected at the
10% level. In particular, WCE series (based on total industry and selected in-
dustries) were found to be nonstationary. The only exceptions were the national
full-time employment and the unemployment rate series; both are considered
to be stationary with rejection of the null at the 5% level.

Rental Adjustments Equations


Because the real rent and vacancy rate series were found to exhibit nonstationary
behavior, these series can be modeled using cointegrating analysis. This is
achieved using the ECM framework. Five variations of the rental adjustment
model (Models (a) to Model (e)) were estimated for the Sydney and Melbourne
CBD markets with results reported in panels A and B of Table 2, respectively.

Model (a) captures the long-run relationship between the logged real rent and
the vacancy rate. Estimated coefficients for the vacancy rate and logged rent
have the expected sign and are statistically significant. The dynamic vacancy
rate term is also significant. The models are considered to provide a moderate
fit of the data with adjusted-R 2 ( R̄ 2 ) values of 0.62 and 0.42 for the Sydney
and Melbourne markets, respectively. With reference to the logged real rent,
the long-run vacancy rate coefficient is interpreted as the relative change in the
real rent for a given absolute change in the vacancy rate. This implied estimated
long-run coefficients (denoted as V ce ) for the Sydney and Melbourne markets
are –6.1 and –5.3, respectively. Furthermore, the implied natural vacancy rate
for the Sydney and Melbourne markets are 7.9% and 10.8%, respectively. The
natural vacancy rate value for the Sydney market is within the range of estimates
provided in HMT (2002), albeit estimates are based on different sample periods.

Models (b) and (c) provide a richer specification by incorporating variables


that capture short-run dynamics. Candidate variables included the change in
the national or state unemployment rate (u t ), the change in various employ-
ment series (E t ), the change in the real bond rate (i rT10Y t ) and the change
in real GDP (Y t ). The lag order of these variables was determined by gaug-
ing the t-ratio and Akaike information criterion (AIC) statistics. For the Syd-
ney market, significant explanators included national total employment and
industry-specific employment. For the Melbourne market, such terms included
the national unemployment rate and industry-specific WCE. While the specific
short-run determinants for each market are considered plausible, their signif-
icance suggests that short-run movements in rentals are prone to a variety of
cyclical macroeconomic factors. As such, diagnostics suggest that these richer
models have improved explanatory power; R̄ 2 values are higher, the AIC and
Bayesian information criterion (BIC) statistics are lower and the Durbin-Watson
(DW) statistic tends toward two.
Table 2  Rental adjustment equations: CBD office market. Estimation period: 1974–2003 (30 observations)

Panel A: Sydney CBD Office Market Panel B: Melbourne CBD Office Market
Model Model Model Model Model Model Model Model Model Model
Variable Statistic (a) (b) (c) (d) (e) (a) (b) (c) (d) (e)
Constant coeff. 3.103 3.436 3.114 3.746 3.463 1.309 2.359 2.047 2.084 2.716
t ratio (4.229) (4.930) (5.161) (4.781) (5.435) (2.508) (4.171) (3.631) (3.438) (4.372)
rrt −1 coeff. −0.461 −0.521 −0.475 −0.565 −0.533 −0.202 −0.406 −0.360 −0.332 −0.426
t ratio (4.102) (4.839) (5.138) (4.665) (5.426) (2.405) (4.133) (3.714) (3.358) (4.275)
V t −1 coeff. −2.828 −2.824 −2.581 −3.197 −2.781 −1.079 −2.066 −1.914 −1.526 −1.781
t ratio (5.394) (5.800) (5.913) (5.926) (6.183) (3.361) (4.880) (4.656) (4.203) (5.063)
irT10Yt−1 coeff. 1.078 0.736 1.196 1.855
t ratio (2.050) (1.693) (2.292) (3.323)
V t coeff. −1.866 −1.077 −1.002 −1.618 −0.928 −1.567 −1.406 −1.452 −1.706
t ratio (2.700) (1.475) (1.627) (2.360) (1.590) (3.355) (3.444) (3.264) (4.026)
ut coeff. 3.686 3.706
t ratio (3.094) (3.266)
ETt coeff. 3.27E-04
t ratio (2.266)
EIt coeff. 0.002 0.002
t ratio (3.664) (3.718)
EWCEIt−1 coeff. 0.009
t ratio (1.885)
Y t coeff. −4.51E-06
t ratio (2.312)
irT10Yt coeff. 0.887 1.210 1.228 1.395
t ratio (1.252) (2.092) (1.968) (2.410)
V ce −6.139 −5.424 −5.433 −5.659 −5.221 −5.345 −5.083 −5.321 −4.595 −4.178
V̄ t 7.9% 7.9% 7.9% 8.0% 8.0% 10.8% 11.0% 10.9% 11.1% 11.3%
R̄ 2 0.622 0.674 0.744 0.654 0.774 0.422 0.565 0.605 0.504 0.580
DW 1.562 1.649 1.876 1.521 1.814 1.511 1.680 1.757 1.502 1.418
AIC −1.681 −1.801 −2.044 −1.715 −2.119 −1.905 −2.162 −2.234 −2.006 −2.148
BIC −1.494 −1.567 −1.810 −1.435 −1.792 −1.718 −1.929 −1.954 −1.725 −1.821
Time-Series Characteristics and Long-Run Equilibrium 391
392 De Francesco

Model (d) expands the ECM term to include the real interest rate. In other
words, it is assumed that the long-run equilibrium relationship, represented
by the vector Xt−q in Equation (7.0), incorporates not only the vacancy rate
and real rent, but also the real interest rate. This is motivated by HMT (2002)
who note that the equilibrium rent varies with the real risk-free interest rate
while other relevant components are taken to be constant. Finally, Model (e)
extends this model to also include short-term dynamics. Using the real 10-year
Treasury bond rate as a proxy for the risk-free rate, estimated coefficients have
the expected sign and are significant. The positive coefficients for the risk-free
rate are consistent to the equilibrium rent coefficient reported in Table 1 of
HMT (2002).

Overall, the estimated natural vacancy rate is around 8% for the Sydney market
and 11% for the Melbourne market. This divergence in the natural vacancy
rate is not surprising as it reflects different supply and demand responses to
(economic) shocks. With regard to supply, the Sydney market is relatively con-
strained, with limited development locations due to geographical boundaries
and the higher economic rent required by developers for construction. In con-
trast, the Melbourne market has a relatively large number of easily accessible
development sites available, a feature that has been reinforced with the inclu-
sion of the Dockland precinct as part of the CBD market over recent years. With
regard to demand, the tenancy profile for the Sydney market is characterized
by large tenants heavily skewed toward the industries of finance, insurance and
legal services whereas the tenancy profile for the Melbourne market is charac-
terized by slightly smaller sized tenants with proportionally higher exposure to
the government and information technology sectors. This suggests that demand
for office space is relatively more elastic in the Melbourne market. As such, a
positive employment shock in the Melbourne market will tend to increase de-
mand for new space, triggering a mini-supply cycle due to the fast response by
developers and hence keeping the vacancy rate higher than in a more constrained
market.

The statistical significance of the error-correction term in Models (a) to (e)


tentatively provides support for the existence of a long-run relationship: either
between the real rent and vacancy rate or between the real rent, vacancy rate
and real interest rate. In exploring this further, formal tests of cointegration
are conducted. EG and JH test results for the Sydney and Melbourne mar-
kets are reported in panels A and B of Table 3, respectively. In contrast to the
EG test results, which indicate no presence of cointegrating relationships, the
JH test results indicate the presence of cointegrating relationships. This may be
explained by the richer specification offered by the JH tests. For the Sydney mar-
ket, the JH trace statistic indicates the existence of a cointegrating relationship
between the real rent, vacancy rate and real interest rate with the computed
Time-Series Characteristics and Long-Run Equilibrium 393

Table 3  Cointegration test results: rental–vacancy rate relationship.


Estimation period: 1974–2003 (30 observations)

EG Test1 JH Test2
AR VAR Raw
Lag Test Lag Test Adjusted
Model Variable Coefficient Length Statistic Coefficient Length Statistic Statistic
Panel A: Sydney CBD Office Market
[r, V ] V t −1 −4.099 1 −2.21 32.535 2 16.10∗ 13.42
[r, V, ir ] V t −1 −4.141 1 −2.49 −5.045 1 46.31∗∗∗ 40.14∗∗∗
irt −1 1.651 2.028
Panel B: Melbourne CBD Office Market
[r, V] V t −1 −2.850 2 −1.97 −8.302 2 15.90∗ 13.25
[r, V, ir ] V t −1 −3.168 1 −2.07 −4.176 1 29.51∗ 25.57
irt −1 2.743 4.353
[r, V] V t −1 n.a. n.a. n.a. −5.702 2 24.47∗∗∗ 20.39∗∗
with
covariate

Note: ∗ , ∗∗ , ∗∗∗ denote significance at the 10%, 5% and 1% level, respectively. Figures
in brackets denote lag orders.
1
EG tests conducted on cointegrating residuals based on linear regression with sample
period 1970 to 2003. Significance is based on conventional EG critical values taken
from Phillips and Ouliaris (1990).
2
Statistics refer to Johansen raw trace statistics with size-adjusted statistics reported.
Significance is based on conventional JH critical values taken from Table B10 Case 2,
Hamilton (1994).

test statistic being larger than the critical value at the 5% level. This is also
supported when the computed test statistics are adjusted for size distortion.14
Notably, the result supports the expanded rental adjustment process proposed
by HMT (2002). For the Melbourne market, initial tests indicate weak support
for a cointegrating relationship. However, the JH test indicates the presence of
a cointegrating relationship between the real rent and vacancy rate15 when the
national unemployment rate is included as a stationary covariate term. The fail-
ure to find a cointegration with the real interest rate possibly suggests that other
components of the equilibrium rent, such as replacement cost and depreciation
rate, are not generally constant. As such, the presence of cointegration with the
unemployment rate may simply reflect this variability.

14
Using Monte Carlo, Ho and Sørensen (1996) find evidence of severe size distortions in
the Johansen LR tests when the ratio of data points to the number of parameters is small.
To correct for potential size distortion, the small sample correction factor proposed by
Reinsel and Ahn (1992) is adopted. The correction factor used is [(T − k)/T ], where
T is the number of observations and k is the number of parameters in the unrestricted
VAR.
15
Test results were also conducted on the [r, v, i] model with the stationary national
unemployment rate. However, results were disregarded as the JH ML procedure produced
meaningless estimates.
394 De Francesco

In all, results highlight that rental determination for the Sydney and Melbourne
markets differ not only in terms of short-run influences, but also in terms of long-
run influences. This difference is not surprising when one considers that both
markets vary significantly in size and have variations in rental pricing schemes
depending on the state of the market. On this latter point, market pricing of
rents in Sydney is a mixture of face and effective leasing deals, whereas in
Melbourne deals tend to be effective when the market is tight and face when
market conditions are sluggish.

Demand, Employment and Cointegrating Analysis


Figures 3A and 3B show trends for occupied space, total WCE (WCET) and
industry-specific WCE (WCEI) for the Sydney and Melbourne markets, re-
spectively. Notably, the series were normalized to facilitate meaningful com-
parisons. A visual inspection of demand (occupied space) and WCE for the
Sydney market in Figure 3A shows that the two series generally trend together.
The divergence of the two series in the early 1970s reflects a sharp rise in the
work–space ratio associated with a slowing economy, coupled with rising un-
employment and the rise in real average earnings. Figure 3B shows a weaker
co-movement between demand and employment for the Melbourne market.
Disparities are evident in the early 1990s, highlighted by the fall-away in WCE
due to the severe impact of the recession, and the 2000s.

In considering the simple long-run relationship between demand and employ-


ment as represented by Model 9.2, use is made of cointegration analysis since
the two series are found to be nonstationary. The EG test, the W test and the
JH test statistic were applied with results reported in panels A and panel B of
Table 4 for the Sydney and Melbourne markets, respectively. Results for the
Sydney market generally detect a long-run relationship between demand and
WCE, both for the total and industry-specific series. These results support the
strong co-movement behavior present in Figure 3A. In contrast, test results for
the Melbourne market indicate no presence of cointegration between the de-
mand series with either of the two WCE series. These results suggest that the
WCE is not a sole determinant of the office demand, which is indicated by the
disparities between the two series highlighted in Figure 3B.

Corresponding cointegrating estimates, computed using three estimation tech-


niques being, OLS, nonlinear least squares (NLS), and Johansen’s ML method,
are reported in panel A of Table 5. With reference to the Sydney market,
cointegrating parameter values have the correct positive sign but vary across
estimation technique. For instance, with reference to model [d,eWCET ], estimates
based on OLS, NLS and ML were 0.973, 0.707 and 0.811, respectively. Given
that these parameter values are less than one, it implies that a unit percentage
Time-Series Characteristics and Long-Run Equilibrium 395

Figure 3 

A Trends in Demand and Employment


Sydney CBD Office Market
2.5
2.0
1.5
normalised scale

1.0
0.5
0.0
`
-0.5
-1.0
occupied space
-1.5
WCE-total
-2.0
WCE-industry
-2.5
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003

B Trends in Demand and Employment


Melbourne CBD Office Market
2.5

2.0

1.5
normalised scale

1.0

0.5

0.0
`
-0.5

-1.0 occupied space


-1.5 WCE-total
WCE-industry
-2.0
1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003

change in WCE corresponds to less than a unit percentage change in demand.


Interestingly, much lower cointegrating parameter values were found when the
industry specific WCE series was used.

The potential variation in parameter estimates and lack of a cointegrating re-


lationship for the Melbourne market may be due to the omission of additional
(short-run or long-run) determinants of the work–space ratio, implying that the
396 De Francesco

Table 4  Cointegration test results. Estimation period: 1974–2003 (30 observations)

JH Test4
Model Covariate1 EG Test2 AIC W Test3 AIC Raw Adjusted AIC
Panel A: Sydney CBD Office Market
Bivariate models
[d, ewceT ] None −3.82∗∗ −5.26 13.49∗∗ −4.84 19.32∗∗ 16.10∗∗ −8.86
[d, ewceI ] None −4.27∗∗ −5.24 16.23∗∗∗ −4.86 30.85∗∗∗ 27.77∗∗∗ −8.05
[d, ewceT ] unt −3.90∗∗ −5.29 - - 21.44∗∗∗ 15.72∗∗ −8.93
ut −4.00∗∗ −5.34 14.85∗∗ −4.83 - - -
irT10Yt −4.91∗∗∗ −5.46 - - - - -
[d, ewceI ] ut −4.52∗∗∗ −5.35 18.35∗∗∗ −4.88 31.73∗∗∗ 27.50∗∗∗ −8.17
irT10Yt −5.27∗∗∗ −5.40 18.52∗∗∗ −4.87 - - -
Higher order models
[d, ewceI , rr ] None −3.94∗∗ −5.15 51.11∗∗∗ −5.05 46.20∗∗∗ 40.04∗∗∗ −9.16
[d, ewceI , ir ] None −3.41 −4.21 32.64∗∗∗ −4.77 46.09∗∗∗ 39.94∗∗∗ −3.06
[d, ewceI , rr ] ut −4.29∗∗ −5.20 52.19∗∗∗ −5.15 47.43∗∗∗ 39.52∗∗∗ −9.27
[d, ewceI , ir ] ut −3.77∗∗ −4.26 33.58∗∗∗ −4.83 45.05∗∗∗ 37.54∗∗∗ −3.12
Panel B: Melbourne CBD Office Market
Bivariate models
[d, ewceT ] None −2.38 −4.16 4.07 −4.19 5.07 4.56 −8.09
[d, ewceI ] None −2.33 −4.13 4.38 −4.27 4.97 4.48 −7.84
[d, ewceT ] ut −1.88 −4.62 3.73 −4.58 10.23 8.87 −8.35
[d, ewceI ] ut −1.84 −4.56 2.85 −4.58 7.92 6.86 −8.10
Higher order models
[d, ewceI , rr ] None −2.32 −4.13 3.79 −4.24 14.48 12.55 −9.39
[d, ewceI , ir ] None −3.27 −3.41 17.38∗∗ −4.45 24.30 21.06 −3.30
[d, ewceI , rr ] ut −2.11 −4.28 2.75 −4.54 12.75 10.62 −9.70
yt - - 3.93 −4.37 15.78 13.15 −9.59
[d, ewceI , ir ] yt - - 23.92∗∗∗ −4.61 27.45 22.87 −3.56

Note: ∗ , ∗∗ , ∗∗∗ denote significance at the 10%, 5% and 1% level, respectively.


1
The optimal lag length for each variable was based on the minimum AIC statistic from
a sequence of lag lengths.
2
ADF tests conducted on cointegrating residuals with lag order selected on the basis of
minimum AIC statistic. Significance is based on conventional EG critical values taken
from Phillips and Ouliaris (1990).
3
Wald tests conducetd on ECM with lag length selected on the basis of minimum AIC
statistic. Significance is based on conventional W critical values taken from Table B2,
Boswijk (1994).
4
Statistics refer to Johansen raw trace statistics with size-adjusted statistics reported.
Significance is based on conventional JH critical values taken from Table B10, Hamilton
(1994).

work–space ratio is endogenously determined. In exploring the relevance of


short-run dynamics (or covariates), modified cointegration tests are conducted,
which include the change in the unemployment rate (u t ), the change in logged
real GDP (y t ), and the change in the real interest rate (i rt ). These were
Time-Series Characteristics and Long-Run Equilibrium 397

Table 5  Cointegrating estimates for office CBD demand. Estimation period:


1974–2003 (30 observations)

NLS2 JH3
OLS1
No No No
Model Variable Covariate Covariate Covariate Covariate Covariate
Panel A: Sydney CBD Office Market
Bivariate models
[d, ewceT ] ewceT 0.973 0.707 0.696 0.811 0.747
[d, ewceI ] ewceI 0.529 0.423 0.450 0.418 0.395
Higher order models
[d, ewceI , rr ] ewceI 0.529 0.402 0.429 0.422 0.392
rr −0.001 −0.148 −0.136 −0.124 −0.143
[d, ewceI , ir ] ewceI 0.495 0.370 0.413 0.389 0.370
ir 0.008 0.009 0.007 −0.013 −0.011
Panel B: Melbourne CBD Office Market
Bivariate models
[d, ewceT ] ewceT 1.009 0.598 0.980 0.525 0.148
[d, ewceI ] ewceI 0.782 0.382 0.674 0.366 0.143
Higher order models
[d, ewceI , rr ] ewceI 0.803 0.396 0.719 0.775 0.820
rr 0.086 0.232 0.218 0.285 0.250
[d, ewceI , ir ] ewceI 0.692 0.593 0.661 0.590 0.613
ir 0.016 0.022 0.020 0.024 0.025
1
OLS regression includes drift term but no trend component.
2
Phillips and Loretan’s (1991) nonlinear least squares (NLS).
3
Johansen maximum likelihood estimation method.

chosen to reflect broad cyclical macroeconomic influences. While the inclu-


sion of covariates showed an improvement over the basic models, indicated by
the lower AIC statistics, the modified test results still indicate no presence of
cointegration for the Melbourne market.

Another possibility for the absence of cointegration for the Melbourne market
may be that relevant determinants of the work–space ratio are not stationary
variables but rather I(1) variables that form part of the long-run equilibrium
relationship. A closer scrutiny of Figure 3B adds some support to this view
with a change in trend for the WCE series around the early 1990s. Potential
candidates include the logged real rent (r rt ) and the real bond rate (i rT10Y t ).
Test results generally indicate the presence of cointegration when either of
these two variables is included in the Sydney model. However, test results are
only slightly encouraging for the Melbourne model with the detection of a
398 De Francesco

cointegrating relationship based only on the W test when the real interest rate
is added to the model.16

Parameter estimates for these higher order models are also reported in Table 5.
With reference to model [d,ewceI ,rr ], estimates for the Sydney market seem
generally plausible. Estimates for the income and price elasticities are appro-
priately signed in accordance with the model postulated in Equation (9.5).
Estimated long-run coefficient values for the WCEI are positive, ranging be-
tween 0.392 and 0.529. These are considered low, as one would expect these
values to lie within the neighborhood of one with the presence of cointegration.
(An auxiliary study was undertaken to gauge the sensitivity of these estimates
over different sample periods. Parameter estimates were found to be close to
one when the sample period ended in 2000. This highlights the discontinuity
between demand and employment since 2000 as noted in Figure 3A, possibly
due to recent exogenous influences as mentioned in the third section.) NLS and
ML estimated17 values for the real rent are negative, ranging between –0.124
and –0.148. While these values are low when compared to Wheaton’s (1999)
–0.4 value used in his simulation study, they are closer to the –0.24 estimated
value reported in HMT (2002).

With reference to model [d,ewceI ,ir ], estimates for the Sydney market indicate
a positive relationship between demand and WCEI, however, the sign for the
real interest rate varies across estimation method. The coefficient value is posi-
tive when estimated using OLS or NLS, however it is negative when estimated
using JH’s ML method. The positive coefficient values are considered dubious.
Based on HLM (1999), a rise in the real interest rate should translate into a rise
in the real rental level, inducing a reduction in net absorption. Estimates for
the Melbourne market also indicate a positive co-movement between demand
and WCEI and demand and the real interest rate. These results are consid-
ered tenuous in light of the test statistics generally indicating no presence of
cointegration.

Conclusion
In this article I modeled the CBD office markets of Sydney and Melbourne
in the context of two key relationships: the rental adjustment mechanism
and the demand–employment relationship. Time series cointegrating analysis

16
This result is tenuous as the estimated parameter values fro the real rent has the
incorrect sign.
17
The relatively low OLS estimated long-run value for the real rent variable may once
again reflect lack of lag-dynamics as noted for the bivariate models.
Time-Series Characteristics and Long-Run Equilibrium 399

is employed, motivated by space market variables displaying approximate


nonstationary behavior. Furthermore, employing an ECM framework, the anal-
ysis attempts to distinguish between long-run equilibrium and short-run deter-
minants. This is motivated by real estate markets being influenced not only by
permanent effects but also cyclical macroeconomic influences. All these fac-
tors are encompassed within the formal modeling framework by allowing the
natural vacancy rate and work–space ratio to be endogenously determined.

The article presented a general, richer, specification for the rental adjustment
model by allowing the natural vacancy rate to be endogenously determined. The
traditional rental adjustment model is a particular case nested within the general
model where the natural vacancy rate is assumed to be exogenous. For the
Sydney market, empirical results suggest that long-term determinants of rentals
include not only the vacancy rate, but also the real interest rate. This is consistent
with the work of Hendershott (1995). However, for the Melbourne market,
only the vacancy rate was found to have a permanent influence on rentals. In
accordance with these markets exhibiting departures from equilibrium, a variety
of macroeconomic indicators were found to have significant short-term impact
on rents.

The article also examined the demand–employment relationship. While formal


tests indicate that WCE is a long-term determinant of demand for the Sydney
office market, tests failed to find a cointegrating relationship for the Melbourne
market. The basic model was expanded, by allowing the work–space ratio to
be endogenously determined, in an attempt to consider additional short-run
and long-run determinants of office demand. While a variety of macroeco-
nomic variables were found to have a transitory influence on office demand
for both the Sydney and Melbourne markets, the real rent and real interest rate
were found to be pertinent long-term drivers of demand for the Sydney market
only.

Future research on this topic can proceed in numerous ways. First, this study
can be extended to other smaller CBD office markets in Australia to compar-
atively assess their key determinants with the larger markets of Sydney and
Melbourne. Second, this study could also consider the existence of common
factors across CBD office markets in an attempt to examine the transmission of
(external) shocks across markets. Third, the study can be expanded to explore
other pertinent relationships such as the supply and rent.

Presented at the 11th European Real Estate Society (ERES) Conference, Milan, Italy
(June 2–5, 2004). This article is a modification of Chapter 8 of my Ph.D. Thesis, submitted
in 2001. I am grateful to the referees for useful comments and suggestions on an earlier
version of this article. All errors are my own.
400 De Francesco

References
Access Economics. 2004, March. Detailed Employment Forecasts.
Andrews, D.W.K. 1991. Heteroskedasticity and Autocorrelation Consistent Covariance
Matrix Estimation. Econometrica 59: 817–858.
Banerjee, A., J.J. Dolado, J.W. Galbraith and D.F. Hendry. 1993. Co-Integration, Error
Correction, and the Econometric Analysis of Non-Stationary Data. Oxford, UK: Oxford
University Press.
BIS-Shrapnel. 2004a. Commercial Property Market Forecasts and Strategies: Sydney
2003–2017.
——. 2004b. Commercial Property Market Forecasts and Strategies: Melbourne 2003–
2017.
Born, W.L. and S.A. Pyhrr. 1994. Real Estate Valuation: The Effect of Market and
Property Cycles. Journal of Real Estate Research 9: 455–484.
Boswijk, H.P. 1994. Testing for an Unstable Root in Conditional and Structural Error
Correction Models. Journal of Econometrics 63: 37–60.
Campbell, J.Y. and N.G. Mankiw. 1987. Permanent and Transitory Components in
Macroeconomic Fluctuations. American Economic Review Papers and Proceedings 77:
111–117.
——. 1989. International Evidence on the Persistence of Economic Fluctuations. Journal
of Monetary Economics 23: 319–333.
Chaudhry, M.K., R.A. Christie-David and W.H. Sackley. 1999. Long-Term Structural
Price Relationships in Real Estate Markets. Journal of Real Estate Research 18: 335–
354.
Chaudhry, M.K., F.C.N. Myer and J.R. Webb. 1999. Stationary and Cointegration in
Systems with Real Estate and Financial Assets. The Journal of Real Estate Finance and
Economics 18: 339–349.
Clapp, J., H.O. Pollakowski and L. Lynford. 1992. Intrametropolitan Location and Office
Market Dynamics. AREUEA 20: 229–258.
D’Arcy, E., T. McGough and S. Tsolacos. 1997. National Economic Trends, Market Size
and City Growth Effects on European Office Markets. Journal of Property Research 14:
297–308.
De Francesco, A. 2001. The Use of Exogenous Short-run Dynamics in Analysing Long-
Run Equilibrium Relationships in Economics. Dissertation. School of Economics, Uni-
versity of New South Wales.
DiPasquale, D. and W.C. Wheaton. 1996. Urban Economics and Real Estate Markets.
Englewood Cliffs, NJ: Prentice-Hall.
Engle, R.F. and C.W.J. Granger. 1987. Co-Integration and Error Correction: Represen-
tation, Estimation, and Testing. Econometrica 55: 251–276.
Gordon, J., P. Mosbaugh and T. Canter. 1996. Integrating Regional Economic Indicators
with the Real Estate Cycle. Journal of Real Estate Research 12: 469–501.
Grenadier, S. 1995. The Persistence of Real Estate Cycles. The Journal of Real Estate
Finance and Economics 10: 95–121.
Hakfoort, J. and R. Lie. 1996. Office Space per Worker: Evidence from Four European
Markets. Journal of Real Estate Research 11: 183–195.
Hakkio, C.S. and M. Rush. 1991. Cointegration: How Short is the Long Run. Journal
of International Money and Finance 9: 75–88.
Hamilton, J.D. 1994. Time Series Analysis. Princeton, NJ: Princeton University Press.
Hansen, B.E. 1995. Rethinking the Univariate Approach to Unit Root Testing: Using
Covariates to Increase Power. Econometric Theory 11: 1148–1172.
Time-Series Characteristics and Long-Run Equilibrium 401

Heckman, J. 1985. Rental Price Adjustment and Investment in the Office Market. Real
Estate Economics 13: 33–47.
Hendershott, P. 1995. Real Effective Rent Determination: Evidence from the Sydney
Office Market. Journal of Property Research 12: 127–135.
——. 1996. Rental Adjustment and Valuation of Real Estate in Overbuilt Markets:
Evidence from the Sydney Office Market. Journal of Urban Economics 39: 51–67.
——. 2000. Property Asset Bubbles: Evidence from the Sydney Office Market. The
Journal of Real Estate Finance and Economics 20: 67–81.
Hendershott, P.H. and E. Kane. 1995. Office Market Values During the Last Decade,
How Distorted Have Appraisals Been? Real Estate Economics 23: 101–117.
Hendershott, P., C.M. Lizieri and G.A. Matysiak. 1999. The Workings of the London
Office Market. Real Estate Economics 27: 365–386.
Hendershott, P., B.D. MacGregor and R.Y.C. Tse. 2002. Estimation of the Rental Ad-
justment Process. Real Estate Economics 30: 165–183.
Ho, M.S. and B.E. Sørensen. 1996. Finding Cointegration Rank in High Dimensional
Systems Using the Johansen Test - An Illustration Using Data Based Monte Carlo
Simulations. Review of Economics and Statistics 78: 31–43.
Johansen, S. 1988. Statistical Analysis of Cointegrating Vectors. Journal of Economic
Dynamics and Control 12: 231–254.
——. 1991. Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian
Vector Autoregressive Models. Econometrica 59: 1551–1580.
Jones Lang LaSalle. 2004. Real Estate Intelligence Service – Australia. Quarter 1.
Kummerow, M. 1999. A System Dynamics Model of Cyclical Office Oversupply. Jour-
nal of Real Estate Research 18: 233–255.
Kwiatkowski, D., P.C.B. Phillips, P. Schmidt and Y. Shin. 1992. Testing the Null Hy-
pothesis of Stationarity Against the Alternative of a Unit Root. Journal of Econometrics
54: 159–178.
Lentz, G.H. and K.S.M. Tse. 1999. Supply Adjustments to Demand Shocks in the
Commercial Real Estate Market. Real Estate Economics 27: 231–262.
MacKinnon, J.G. 1996. Numerical Distribution Functions for Unit Root and Cointegra-
tion Tests. Journal of Applied Econometrics 11: 601–618.
Maddala, G.S. and I. Kim. 1998. Unit Roots, Cointegration and Structural Change.
Cambridge, UK: Cambridge University Press.
Marmol, F. 1996. Nonsense Regressions Between Integrated Processes of Different
Orders. Oxford Bulletin of Economics and Statistics 58: 525–536.
Mourouzi-Sivitanidou, R. 2002. Office Rental Processes: The Case of U.S. Metropolitan
Markets. Real Estate Economics 30: 317–344.
Mueller, G.R. 1999. Real Estate Rental Growth Rates at Different Points in the Physical
Market Cycle. Journal of Real Estate Research 18: 131–150.
Myer, N.F.C., M.K. Chaudhry and J.R. Webb. 1997. Stationarity and Co-Integration in
Systems with Three National Real Estate Indices. Journal of Real Estate Research 13:
369–381.
Nelson, C.R. and C.I. Plosser. 1982. Trends and Random Walks in Macroeconomic Time
Series: Some Evidence and Implications. Journal of Monetary Economics 10: 139–162.
Perron, P. 1989. Testing for a Random Walk: A Simulation Experiment of Power When
the Sampling Interval Is Varied. B. Raj, editor. Advances in Econometrics and Modeling.
Dordrecht, The Netherlands: Kluwer Academic Publishers.
Phillips, P.C.B. and M. Loretan. 1991. Estimating Long-Run Economic Equilibria. Re-
view of Economic Studies 59: 407–436.
402 De Francesco

Phillips, P.C.B. and S. Ouliaris. 1990. Asymptotic Properties of Residual Based Tests
for Cointegration. Econometrica 58: 165–193.
Property Council of Australia. 2006, January. Australian Office Market Report.
Pyhrr, S.A., S.E. Roulac and W.L. Born. 1999. Real Estate Cycles and their Strategic
Implications for Investors and Portfolio Managers in the Global Economy. Journal of
Real Estate Research 18: 7–68.
Reinsel, G.C. and S.K. Ahn. 1992. Vector Autoregressive Models with Unit Roots and
Reduced Rank Structure: Estimation, Likelihood Ratio Tests, and Forecasting. Journal
of Time Series Analysis 13: 353–375.
Rosen, K. 1984. Toward a Model of the Office Building Sector. Real Estate Economics
12: 261–269.
Rosen, K. and L. Smith. 1983. The Price Adjustment Process for Rental Housing and
the Natural Vacancy Rate. American Economic Review 73: 779–786.
Scott, P. and G. Judge. 2000. Cycles and Steps in British Commercial Property Values.
Applied Economics 32: 1287–1297.
Seo, B. 1998. Statistical Inference on Cointegration Rank in Error Correction Models
with Stationary Covariates. Journal of Econometrics 85: 339–385.
Shiller, R. and P. Perron. 1985. Testing the Random Walk Hypothesis: Power Versus
Frequency of Observation. Economics Letters 18: 381–386.
Shilling, J.D., C.F. Sirmans and J.B. Corgel. 1987. Price Adjustment Process for Rental
Office Space. Journal of Urban Economics 20: 90–100.
Shilton, L. 2000. Random Walks and Cointegration of the ACLI and NCRIEF. Real
Estate Economics 28: 435–465.
Sivitanides, P.S. 1997. The Rent Adjustment Process and the Structural Vacancy Rate
in the Commercial Real Estate Market. Journal of Real Estate Research 13: 195–209.
Sivitanidou, R. and P. Sivitanides. 1999. Does the Theory of Irreversible Investments
Help Explain Movements in Office-Commercial Construction. Real Estate Economics
28: 623–661.
Slade, B.A. 2000. Office Rent Determinants During Market Decline and Recovery.
Journal of Real Estate Research 20: 357–380.
Tonelli, M., M. Cowley and T. Boyd. 2004. Forecasting Office Building Rental Growth–
Using a Dynamic Approach. Paper presented at the PRRES Tenth Annual Conference
in Bangkok, Thailand.
Wheaton, W.C. 1990. Vacancy, Search and Prices in a Housing Market Matching Model.
Journal of Political Economy 98: 1270–1293.
——. 1999. Real Estate “Cycles”: Some Fundamentals. Real Estate Economics 27:
209–230.
Wheaton, W.C., R.G. Torto and P. Evans. 1997. The Cyclic Behaviour of the London
Office Market. The Journal of Real Estate Finance and Economics 15: 77–92.
Wilson, P., J. Okunev and J. Webb. 1998. Step Interventions and Markets Integration:
Tests in the U.S., U.K., and Australian Property Markets. The Journal of Real Estate
Finance and Economics 16: 91–123.

You might also like