Liquidity Dynamics across Public and Private
Markets
Shaun A. Bond1 and Qingqing Chang
Department of Finance
University of Cincinnati
November 2011 (this version March 2012)
Abstract
In this paper we investigate cross-asset liquidity between equity markets
and REITs and between REITs and private real estate markets. While
many studies have investigated REIT liquidity, and there is an emerging
interest in liquidity in the private real estate markets, there appears to be
little knowledge of the dynamics of cross-market liquidity. We find lower
levels of liquidity for REITs compared to a set of control firms matched
on size and book to market ratios. Commonality in liquidity is also lower
for REITs than the controls and the overall market. However, we do find
an important difference in share turnover for REITs, which appears to
have a higher level of commonality than found in other studies. We
suggest that this may be due to the financial crisis. Additionally we find
evidence of similar time-series variation in liquidity for public and private
real estate markets. We also find significant directional causality for most
liquidity proxies from the public to private real estate markets. Finally our
results show that there is strong contemporaneous correlation between
both public and private real estate market liquidity and the term spread and
real investment and consumption spending. REIT liquidity measures based
on intraday data also appear to contain important information not found in
measures constructed from daily returns.
Keywords: cross-market liquidity, REITs, commercial real estate.
JEL codes: R33, G11.
1
Corresponding author: Department of Finance and Real Estate, Lindner College of Business, University
of Cincinnati, OH, 45221-0195. Email: [email protected]. The authors wish to thank our discussant
Andy Naranjo, participants at the Tilburg Conference on Real Estate Securities, and an anonymous referee
for helpful comments.
I. Introduction
The financial crisis highlighted the important role played by liquidity in finance and real
estate markets. This has stimulated a range of research focusing on the dynamics and
cross-section commonality of liquidity, see for instance, Spiegel (2008), Korajczyk and
Sadka (2008), and Naes et al (2011). In this paper we examine liquidity measures and
transmission across public and private real estate markets. In particular, using data on
publicly traded real estate investment trusts (REITs) and trading data on commercial real
estate, we document the transmission of a liquidity shock from public to private markets.
Furthermore we examine the relationship between liquidity in real estate markets (both
public and private markets) to macroeconomic variances and show how the transmission
mechanism differs between public and private markets.
Using data on commercial real estate markets provides an appealing laboratory in which
to consider cross-asset commonality and transmission in liquidity. In this case the
underlying asset, commercial real estate, forms the main asset holding of publicly traded
REITs and is the source of constant performance monitoring by a number of professional
organization2. When considering all real assets markets, commercial real estate has the
most detailed and extensive, validated set of performance data of any alternative asset
class.
In terms of the dynamics of the liquidity process, we construct several recognized proxies
for liquidity, based on quote and trade prices, and daily trading data, for REITs, a sample
of control firms (matched for market capitalization and the book to market ratio), and the
2
Including the National Council of Real Estate Investment Fiduciaries (NCREIF), whose data is used in
this study.
2 overall market. Several papers have examined liquidity proxies using daily trading data,
see for instance, Marcato and Ward (2007), Brounen et al (2009), and Cannon and Cole
(2011). However, unlike these studies we include liquidity measures based on intraday
quote and trade prices for REITs.
To preview our results, we find that REITs generally have a lower level of liquidity than
either the controls or the overall market. After documenting the characteristics and
dynamics of these series, we then consider the transmission of a liquidity shock between
these series. Information about the transmission mechanism has implications for portfolio
diversification as many investors choose to hold REITs as a diversifying asset class.
Using Granger Causality tests, we find an intriguing result, even though the Amihud
liquidity proxy for the market tends to lead REIT liquidity (Amihud 2002), turnover in
REITs does appear to lead overall market turnover. We discuss possible reasons for this
in Section IV.
A further innovation in our study is the use of liquidity measures obtained from the direct
real estate market. To date very few studies have considered liquidity measures in the
direct real estate market and there is little understanding of the dynamic properties of this
series of interest (see Clayton et al 2008 for a related article). Using data from the MIT
Transactions-Based (TBI) Index (see Fisher, et al 2007), we examine liquidity measures
based on an Amihud-type of construct. Recently Bond and Slezak (2010) have used the
latter measure as a liquidity proxy in a portfolio optimization exercise and the results of
the present study would appear to support the use of this measure as a meaningful proxy
for liquidity. We document the dynamic relationship between these measures and show
some difference in the transmission mechanism between each measure and liquidity
3 measures in the public markets. Furthermore, we find evidence of a strong connection
between macroeconomic factors and liquidity in the private real estate market.
Our study extends the work of Marcato and Ward (2007) and Brounen et al (2009) in
three ways. Firstly, our sample covers the recent financial crisis and given the
significance of liquidity in the crisis, provides important information to researchers and
investors on the behavior of liquidity in crisis periods in these markets. Second, while our
paper examines only US markets, it incorporates measures of liquidity from the public
markets as well as private real estate markets. Most of the literature to date has primarily
focused on REIT liquidity. Finally, our study investigates the relationships between
macroeconomic factors and liquidity in real estate markets.
The outline of our paper is as follows. The next section reviews the relevant literature.
Following that we describe the data and liquidity measures used in our study in Section
III. Section IV outlines our approach to the factor decomposition of the liquidity
measures and Section V discusses our results. The paper concludes in Section VI.
II. Literature
A. Liquidity in public real estate markets
REIT liquidity has been the focus of a large stream of research. Bhasin, Cole and Kiely
(1997) and Clayton and MacKinnon (2000) have previously studied REIT liquidity with a
particular emphasis on within day trades. Both authors note the important changes in
REIT liquidity post 1993. However in both cases, the sample period studied was confined
4 to the 1990s. Ling et al (2011) finds an important connection between the availability of
mortgage financing and real estate market liquidity.
Recent papers by Marcato and Ward (2007) and Brounen et al (2009) extend the research
beyond US markets, and analyze REIT liquidity in an international setting. Even though
these studies have little emphasis on cross-market liquidity, they are helpful in
understanding the determinants of individual company liquidity. However, these studies
do not consider linkages between liquidity in the public and private markets.
B. Liquidity in private real estate markets
One approach to measuring liquidity risk in private commercial real estate markets is that
of Lin and Vandell (2007), Bond et al. (2007), and Cheng, Lin and Liu (2010). These
papers measure the volatility in asset returns over the (uncertain) time to sale (time on
market) and use this as a measure of liquidity risk. When the volatility of the time to sale
is taken into consideration, the ex-ante volatility of real estate returns can be much
greater than an ex-post measure of volatility calculated from historical real estate returns.
However, while this line of research clearly has significant merit, it is subject to the
limitation that it can be difficult to implement because of data availability. For instance, it
is very difficult to get reliable and timely data on time-on-market for commercial real
estate assets.
Another liquidity proxy that has been suggested for the commercial real estate market is a
measure based on the difference between an index of imputed seller reservation prices
and an estimate of the level of buyer’s reservation prices (See Fisher, Geltner and
Pollakowski 2007, and Fisher et al. 2003). A significant advantage of this measure is that
5 because it is a by-product of producing a transactions-based real estate index, it is
available at a quarterly frequency, and it is available over an extensive period of time
(from 1984 for the all property measure or from 1994 at the sector level). This measure
was used in Bond and Slezak (2010) to capture immediate liquidation costs in a multiasset portfolio optimization context. Their results show considerable promise for this
measure as a practical proxy of liquidity in commercial real estate markets.
To our knowledge the only other study that uses liquidity measures from the TBI index is
that of Buckles (2008). Rather than use the liquidity spread variable from the TBI,
Buckles estimates a liquidity index from a cointegrating regression between the constant
liquidity index and the supply index. He then presents a time series analysis of the
resulting series.
In summary, while some research has investigated the nature of daily REIT liquidity, to
our knowledge no other paper has considered commonality in REIT liquidity or
investigated the impact of the crisis of REIT liquidity. Furthermore we are not aware of
any paper that has considered the implications for private market real estate liquidity
using information on public market liquidity.
III. Data and liquidity measures
A. Data
The data used for the analysis come from four sources: the CRSP/Ziman REIT Database;
the CRSP daily/monthly stock returns; the TAQ database; and the transactions-based
index (TBI) by MIT center for Real Estate. Our analysis begins with all
NYSE/AMEX/NASDAQ firms in the Ziman REIT database with available daily data
6 since 1980. Between January 1980 and December 2010 there are over 500 REIT firms
with daily trading information.
Stocks are filtered in our sample based on the following criteria. For the daily and
intraday data: (1) We use only NYSE stocks to avoid any possibility of the results being
influenced by differences in trading protocols. (Chordia, Roll, and Subrahmanyam, 2001;
Korajczyk and Sadka, 2008). (2) If a firm changed listing from NASDAQ to NYSE
during the year, it was dropped from our sample for that year. There are no firms that
changed listing from NYSE to NASDAQ in our sample. (3) To avoid the influence of
overly high-priced stocks, stocks whose prices are below $1 and above $1000 are
excluded. (4) We excluded firms with less than 24 monthly observations.
For the intraday data: (1) Use only use best bid or offer (BBO)-eligible primary market
(NYSE) quotes. (2) Trades out of sequence, trades recorded before the opening or after
the closing time, and trades with special settlement conditions are eliminated. (3) We
discard negative bid-ask spreads and transaction prices. (4) To avoid after hours liquidity
effects, the first trade after the opening time is ignored. (5) In addition, only quotes that
satisfied the following filter conditions are retained: quotes in which the bid-ask spread is
positive and below $5; quotes in which the bid-ask spread divided by the midpoint of the
quoted bid and ask is less than 10% if the midpoint is greater than or equal to $50; and
quotes in which the quoted spread is less than 25% for midpoints less than $50.
Our major results based on a sample from September 1993 to December 2010, include
283 REITs firms with both available daily and intraday data. Our data begins in
September 1993 because the nature of the REIT market changed substantially after the
7 announcement of Revenue Reconciliation Act of 1993 in August 1993. We also consider
a long sample from January 1983 to December 2010, that contains 292 REITs firms.
To compare REITs shares with ordinary common shares (CRSP share codes 10 and 11),
we also construct a control group and a market sample. Fama and French (1992) argue
that common stock returns are related to firm size and book-to-market ratios. Barber and
Lyon (1997) document that constructing control firms by size and book-to-market
generates well-specified test statistics in most sample situations considered. As in Barber
and Lyon, we match a REITs sample firm with a control firm with the closest size and
book-to-market ratio 3 . Specifically, we chose a control firm with smallest absolute
differences in size and book-to-market ratio; if there are ties, we keep the closest match in
size. The market sample includes all ordinary common shares filtered by the criteria
mentioned earlier.
B. Liquidity measures
For each stock we define the following liquidity measures:
1. Amihud –– the daily average of absolute value of return divided by volume for asset i
in month t:
,
where
,
∑
,
,
,
is the return on asset i on day j of month t,
(1)
,
is the dollar volume (number
of shares multiplied by the transaction price) traded in asset i on day j of month t, and
is the number of trading days in month t. This measure is based the measure proposed in
3
Size and book-to-market are constructed as in Fama and French (1992, 1993).
8 ,
Amihud (2002). If
,
is observed less than 15 days for asset i in month t,
i is deleted from the sample for the month. We also scale
,
,
of asset
by the ratio of market
capitalization of the CRSP Ziman REIT Index at t-1 and at a reference date (first month
of the sample).
2. Turnover –– the ratio of monthly volume and shares outstanding:
∑
,
,
where
,
,
,
(2)
is shares outstanding of asset i at the end of month t.
3. Qspread – the quoted percentage spread is measured for each trade as the ratio of the
quoted bid-ask spread and the bid-ask midpoint
4.
,
Monthly estimates are a simple
average through the month:
,
where
,
,
,
⁄2 ,
,
,
and
∑
,
,
,
,
,
at the time of the jth trade of asset i in month t, and
,
(3)
are the ask and bid quotes prevailing
,
is the number of eligible trades of
asset i in month t.
4. Espread – the effective percentage half-spread is measured for each trade as the
absolute value of the difference between the transaction price and the quote midpoint:
4
The midpoint of the quotes as of five seconds prior to the trade is used as suggest by Lee and Ready (1991)
because of a delay in the time that bid and offer are quoted. Specifically, for the estimation, any quote
posted less than five seconds prior to a trade is ignored, and the first quote posted at least five seconds prior
to the trade is retained.
9 ,
where
,
,
∑
,
,
,
,
,
(4)
is the transaction price for the jth trade of asset i in month t.
The next four liquidity measures are price effects on trades estimated using intraday data,
distinguishing between the permanent and transitory effects. Permanent effects are
believed to be related to the private information revealed through the trading process, and
the transitory effects are the compensation to the market makers’ costs of making a
market, such as inventory and order processing (Ho and Macris, 1984; Glosten and
Harris, 1988; Madhavan and Smidt, 1991).
The estimation procedures of measuring the components of price impact are summarized
as follow. The method is an extension of the theory work in Glosten and Harris (1998),
developed in Sadka (2006) and applied in Korajczyk and Sadka (2008).
Let
denote the market maker’s expected value of the security, conditional on the
information set available at time t (t represents the event time of a trade)
|
where
is the order flow,
,
,
(5)
is a binary indicator variable that receives a value of (+1)
for a buyer-initiated trade and (-1) for a seller-initiated trade, and
is a public
information signal. Prices above the midpoint of the quoted bid and ask are considered
buyer-initiated; prices below the midpoint are considered seller-initiated. Trades whose
price equals the midpoint are discarded from the sample (Lee and Ready, 1991).
10 To estimate the permanent price effects, Sadka (2006) follow the formulation proposed
by Glosten and Harris (1988) and assume that
takes a linear form such that
Ψ
where Ψ and
(6)
are the fixed and variable permanent price impact costs, respectively. This
equation describes the innovation in the conditional expectation of the security value
,
through new information, both private (
and public (
. Note that information
induces a permanent impact on expected value.
To account for predictability in the order flow that is well documented in the literature,
Sadka (2006) adjust Eq. (6) to Eq. (7), assumed that market makers revise the conditional
expectation of the security value only according to the unanticipated order flow, not the
entire order flow at time t.
Ψ
(7)
Denote the unexpected sign of a trade as
signed volume of trade
,
, where
,
, where
,
, and unexpected
. Substituting the above
,
formulations in Eq. (7) and taking the first difference, we have:
∆
Ψ
,
,
(8)
The observed transaction price can be written as
Ψ
(9)
11 Ψ and
are temporary effects by the construction of Eq. (9), as they affect only
are not carried on to
and substituting ∆
. Taking the first differences of
and
from Eq.
(8) we have
∆
Ψ
,
Ψ∆
,
∆
(10)
Therefore, the components are obtained by the regression (11), estimated per firm per
month using OLS with corrections for serial correlation in the error term:
∆
Where
,
Ψ,
,,
,
Ψ, ∆
,,
,
,
∆
,
,
,
,
(11)
is a binary indicator variable of the jth trade of asset i in month t that receives
,
a value of (+1) for a buyer-initiated trade and (-1) for a seller-initiated trade,
order flow of the jth trade of asset i in month t,
,,
is the
,
is the unexpected signed volume of
trade measured as the fitted error term from a five-lag autocorrelation regression of the
order flow
,
,,
is the unexpected direction of trade calculated while imposing
normality of the error
, where
,,
Ε
,,
, and ∆ is the first difference
operator 5. Thus, the price component liquidity measures we research in this paper are:
5.
,
is the permanent variable component (PV) of price impact since it measures how
much the valuation of the asset changes given a shock to signed trading volume,
6.
,
.
is the transitory variable component (TV) of price impact since the effect of signed
volume for this trade,
of
,,
,
,
,
,
,
, has an effect of
on the price of trade
,
,
,
on the price of trade j, and effect
1, and no effect on subsequent prices.
5
see Sadka (2006) for more details
12 7. Ψ , is the permanent fixed component (PF) of price impact.
8. Ψ , is the transitory fixed component (TF) of price impact.
IV. Methodology
1. Factor decomposition of liquidity
Following Korajczyk and Sadka (2008), we examine a factor decomposition of each
liquidity measure and return in this section.
To avoid overweighting some liquidity measures over others due to their different scale,
we first standardize our liquidity measures. Define
1, 2 . Define ̂
observations on the ith liquidity measure (
and
matrix of
as the time-series mean,
as standard deviation of the cross-sectional average of liquidity measure i,
estimated from the sample up to time
,
as the
,
measure of the
̂
⁄
, where
matrix
,
1 . The standardized liquidity measure is
is the observations of the ith standardized liquidity
.
Assume that the data generating process for
,
is an approximate factor model:
,
where
is a
vector of factor sensitivities to the common liquidity shocks,
(3)
is a
matrix of shocks to liquidity measure i that are common across the set of n assets,
13 and
is an
matrix of asset-specific shocks to liquidity measure i. In other words,
are the systematic (undiversifiable) shocks that affect most of the assets, while
are
idiosyncratic (diversifiable) shocks that have weak link across the assets.
To estimate the factor model, Korajczyk and Sadka (2008) use Asymptotic Principal
Components (APC) analysis developed by Connor and Korajczyk (1986). Specifically, in
an approximate factor-model setting for a balanced panel (complete data), since
unobservable, we use a proxy for
is
consisting of the k largest eigenvectors of
Ω
.
(4)
Connor and Korajczyk (1986) show that this proxy gives asymptotically identical
∞) to those obtained if we were able to use the unobservable
estimates (as
Furthermore, because Ω is a
.
matrix, the computational burden of the principal
components (eigenvectors) of this matrix is independent of the cross-sectional sample
size, n. This implies that this method can be used for very large cross-sectional samples,
since for most panel data the time-series matrix has a much smaller dimension than the
corresponding cross-sectional crossproduct matrix used by standard factor analysis.
To accommodate missing data, we use the technique developed in Connor and Korajczyk
(1987). Specifically, we estimate each element of Ω by averaging over the observed data.
Let
be the data for liquidity measure i with missing data replaced by zeros. Define
as a
matrix for which
,
is equal to one if
,
is observed and is zero if
,
is
missing. Define
14 Ω ,,
,
,
(5)
,
where Ω ,, is the unbalanced panel equivalent of Ω in which the
,
element is defined
as the cross-sectional averages over the observed data only 6 . The estimates of latent
factors,
, are obtained by calculating the eigenvectors for the k largest eigenvectors of
Ω ,, .
We extract three principal components for each liquidity measures, and estimate the timeseries regression for each stock’s liquidity on the three extracted factors to demonstrate
the degree of commonality across assets for each liquidity measure. The regression model
is
,
where
is the
,·
̂, ,
(6)
1 vector of factor estimates for month t.
2. Granger causality and the testing method
To exam the relations among liquidity measures, and the relations between liquidity
measures and macroeconomic variables, we conduct the Granger Causality test under a
linear vector auto regression (VAR) framework7.
To describe the definition of Granger causality and the testing methodology (Granger,
1969), consider the case of two stationary time series. Denote
|
as the
While Ω in a balanced panel is guaranteed to be positive semi-definite, Ω ,, is not. However, similar to
Korajczyk and Sadka (2008), we have not found cases in which Ω ,, is not positive definite in large crosssections sample.
7
In this paper we do not consider the case of nonlinearity in the causality relationship as in the case of
Hiemstra and Jones (1994).
6
15 conditional probability distribution of
consisting of an
given the bivariate information set
-length lagged vector of
an
-length lagged vector of
and
, the time series
|
,
(
,
(
,…,
does not strictly Granger cause
|
,
,…,
) and
). Given lags
if:
1,2, …
The definition of causality used above is based entirely on the predictability of one time
contains information in past terms that helps to forecast
series. Specifically, if
this information is not contained in other series used in the predictor, then
Granger cause
and if
is said to
.
To test Granger causality, we employ the widely used vector autoregression (VAR)
framework:
∑
∑
∑
where the error term
,
and
,
∑
,
,
,
,
1,2, …
are uncorrelated white-noise.
The null and the alternative hypotheses are as below:
:∑
0 or
does not Granger cause
:∑
0 or
does Granger cause
16 The standard joint test (
lagged
-test) of exclusion restrictions is used to determine whether
can significantly forecast current . We reject the null hypothesis that
not Granger casue
if ∑
does
are jointly significantly different from zero. The above
method and testing framework can be generalized to the many variables situation as well.
V. Results
A. Common factors of liquidity measures
Table 1 reports the average
and the average adjusted-
of time-series regressions
using one, two, and three common factors. Results are reported for two sample periods:
1993:09 – 2010:12 as well as 1983:01 – 2010:12. Results for three different groups;
REITs firms, control firms, as well as the market are also reported.
The results obtained clearly indicate a commonality across assets for most liquidity
measures. However, the magnitude of the commonality was different among each of the
different groups. For example, the
of the Amihud measure for REITs was found to be
half of that of the control firms, and only one-third of the Market. For REITs, the
the Amihud measure was much smaller than the
of Turnover measure. The
values
of the Amihud measures for the REITs firms range from 6.64% to 22.8%, while the
the Turnover measure of REITs range from 26.74% to 37.81%. The
of
of
of Amihud
measure of market is consistent with the results of Korajczyk and Sadka (2008), while the
of the Turnover measure of market is nearly three times larger than that of the results
documented in Korajczyk and Sadka(2008).
17 The difference found could likely be the result of the different time periods we used in
our sample, which indicates that the Turnover measure’s commonality has been
increasing over the past ten years. The
of Amihud and Turnover measures for control
firms are similar for the period of 1993:09 – 2010:12. The most surprising result is that
the Turnover liquidity measure for REITs has the strongest commonality among all three
groups. It is also interesting that
of Turnover for REITs is significantly larger than that
of Amihud for REITs. This is inconsistent with the results of Korajczyk and Sadka
(2008), who show that the
of Turnover measure for the market is sizably smaller
among other measures. This difference could be due to the unique characteristics of
REITs Firms.
Figure 1 shows the time-series variations of the Amihud and turnover measure for REITs,
Control Firms and Market for periods 1993:09-2010:12. Both measures are calculated as
equally weighted averages across stocks. The first graph shows that the Amihud measure
of REITs is similar to that of control firms from 1996 to 2000, while it is consistently
higher over the remaining time periods. Note that the Amihud measure reflects illiquidity,
so a high value reflects a high price impact of trade, that is, low liquidity. This fact
indicates that, controlling for size and book-to-market ratio, the REIT firms are less
liquid than non-REIT firms. However, by controlling for size and book-to-market, we
rule out the possibility that the illiquidity of REITs firm is due to the fact that most of the
REITs firms are small value firms. The control group’s performance moves closely to
that of market after 2003. The turnover measures of three groups tell the same story with
the Amihud measure.
18 As for liquidity measures calculated from intraday data, Qspread, Espread, PV, PF, TV,
and TF, except Qspread, the
of the rest liquidity measures for REITs firms are
considerably smaller than those reported in Korajczyk and Sadka (2008) for the whole
market. For example, the
of TF for REITs firms are 13.51%, less than 1/3 of the
of
TF for the market (47.6%, see table 1 in Korajczyk and Sadka(2008)). This result
indicates that, to some extent, the liquidity of REITs firms share less commonality with
each other for a short period (intraday) compared to the liquidity of all the firms in the
market.
Figure 2 shows the time-series variations of Qspread, Espread, PV, PF, TV and TF for
REITs firms for periods 1993:01-2010:10. The first graph of bid-ask spread variables
shows that the liquidity of REITs firms is increasing over time, consistent with the results
for the Amihud and Turnover measures. However, an important difference compared to
the Amihud and Turnover measures, is the clear spike during 2008-2009, indicating that
REITs firms became very illiquidity during the financial crisis. The second and the third
graphs describe the price impact components. Permanent variables and transitory
variables vary in opposite directions for most of the time in our sample period, and they
all show evident spikes during financial crisis, suggesting that the liquidity measures
derived from intraday data are more sensitive to financial crisis or economic downturns
compared to the Amihud and Turnover measures derived from daily data.
B. Lead-lag relations between liquidity measures of REITs
To explore the relations between liquidity measures obtained both from inter- and
intraday data, we conduct Granger causality tests between pairs of liquidity measures of
19 REITs firms. For each pair we first test the null hypothesis that Column (1) does not
Granger cause Column (2), and then Column (3) does not Granger cause Column (4). We
perform the test by using a vector auto regression (VAR) framework.
The results of Granger causality tests in Panel A of Table 2 document strong evidence of
one way Granger causality of twelve pairs of liquidity measures of REITs firms at 1%
significant level. First of all, there is significant one-way Granger causality from
Amihud_MKT to Amihud, and Qspread to Amihud. Further investigation shows that
Amihud_MKT and Qspread for REITs firms have a bi-directional effect on each other
(Amihud_MKT lead Qspread at 1% level, while Qspread also lead Amihud_MKT at 5%
level).
Second, Turnover Granger causes TV, while both Espread and TF Granger cause
Turnover (the same with Amihud_MKT and Qspread, Espread and TF have significant
mutual effect on each other). However, there is no lead relation from Espread to TV, but
a one-way Granger causality from TV to Espread.
Last but not the least, as expected Amihud, Qspread also lead Espread, PF, and TF, while
there is no evidence suggesting reverse causality from the other seven liquidity measures
for REITs firms to Qspread, indicating that bid-ask spread contains information that have
influence on the following stock prices and especially the fixed components in prices.
C. REITs and the direct real estate market
Next we compare the liquidity measures of REITs firms to the liquidity measures of the
actual real estate market. We use a liquidity measure, Spread_TBI, derived from the
20 estimation of the transactions-based index (TBI) developed by the MIT center for real
estate. Construction of this measure is well discussed in Fisher, Geltner and Pollakowski
(2007), and Fisher et al (2003). Further discussion on the use of this series as a liquidity
measure is given in Bond and Slezak (2010). Using well-established econometric
techniques, two indices are created; the first represents the mid-point of the means of the
buyer and seller reservation distributions (which is referred to as the TBI index), and the
second is the mean of the buyers distribution (referred to as the constant-liquidity index).
The difference between these two indices can be thought of as being analogous to a bidask spread for real estate and is the measure used in this paper.
Figure 3 shows that Amihud measures of REITs, and Spread_TBI share similar patterns,
while the liquidity measure of REITs is more volatile. Turnover measures of REITs, on
the other hand, moves in the opposite direction with Amihud and Spread_TBI, since
Turnover measures ‘liquidity’ and Amihud and Spread_TBI measures ‘illiquidity’.
Because the results indicate similar patterns, we investigate the possibility of causality
between liquidity measures of REITs firms and the actual real estate market. We perform
the test by using a vector auto regression (VAR) framework. We performed the test on
the sample period from 1987:02 to 2010:04.
In panel B of Table 2, there is significant evidence of one-way Granger causality from
Turnover, Qspread, and Espread of REITs firms to Spread_TBI at 1% level. In terms of
transitory price impact variables, there is a two-way Granger causality relationship
between TV and Spread_TBI, and TF and Spread_TBI. This evidence suggests that the
21 fluctuation of REITs firms’ stock prices contain information on the direction of liquidity
for the actual real estate market.
Panel C repeats the analysis in Panel A for a period relating to the financial crisis. Using
only monthly data from the series derived from REIT and stock market data, there is
evidence to suggest to REIT liquidity measures led the liquidity measures for the overall
market during the crisis (measured as July 2007 to June 2010). This results holds for both
the Amihud and turnover measures, as well as for Qspread and Espread. This is an
intriguing result and may be due to the role that real estate played in the financial crisis.
D. Liquidity measures and macroeconomic variables
Table 3 shows contemporaneous correlations between liquidity measures, market
variables and macroeconomic variables of the U.S. We employ real GDP (GDPR), the
unemployment rate (UE), real consumption (CONSR), real investment (INV), money
supply (M2 and NONM18), and following a recent paper by Ling et al (2011) we also
include a measure of the tightening standards for commercial real estate loans
(TIGHTEN)9. We also use Excess market return (MKT), Term Spread (TERM), Credit
spread (Cred), volatility of S&P 500 (VIX), and spread in returns between value and
growth stocks (HML) to proxy for financial market variation.
8
NONM1 is non-M1 component of M2.
The GDPR is real gross domestic product, CONSR is real personal consumption expenditures, INV is real
private fix investments, and UE is unemployment rate for full time workers. All series are seasonally
adjusted. GDPR, CONSR, and INV are from the Federal Reserve Bank of St. Louis, and UE is from the
U.S. Bureau of Labor Statistics.
9
22 SPREAD_TBI are significantly positively correlated with two out of five financial
market variables, TERM, and CRED, indicating that actual market liquidity links tightly
to market interest rate structures. Note that the TURNOVER of REITs firms and
Spread_TBI are both negatively correlated to GDP growth. So when GDP growth
increases, the liquidity of REITs firms deteriorates but the liquidity in the actual real
estate market increases. In other words, the same causes that improve the liquidity in
actual real estate market also contribute to GDP growth. The correlation between the INV
growth and TURNOVER, ESPREAD, or SPREAD_TBI tells a similar story with those
liquidity measures’ correlations with GDP growth.
Panel B in the Table shows the correlations leading up to the financial crisis. During this
period, Credit Spreads are significantly inversely correlated with REIT liquidity measures.
However, during the financial crisis, this connection is not significant. However, the
credit tightening variable is found to positively correlated with turnover and inversely
related to the Amihud measure of illiquidity.
To investigate the lead-lag relations between liquidity measures and macroeconomic and
market variables, we conduct a similar Granger causality test and the results are reported
in Table 4. First of all, there is a significant one-way Granger causality from AMIHUD to
HML, indicating that the risk correlated to HML could be led by the liquidity risk
associated with the AMIHUD liquidity measure. Only one macro variables, CRED, out
of thirteen lead AMIHUD at 10% level.
23 Similarly, there is a significant one way Granger causality from TURNOVER to dUE,
and it is the only one liquidity measure out of nine has noticeable leading effect on the
change of unemployment rate. Moreover, two out of six market variables significantly
Granger cause TURNOVER at 1% level (TURNOVER_MKT, MKT, and TERM), and
five out of all thirteen variables significantly Granger cause TURNOVER at the 10%
level (VIX, HML, dINV, dUE, and CRED). Comparing the results of AMIHUD with
TURNOVER, it is clearly that the latter are heavily affected by macro and market
variables. Also note that from Fig. 3 we can see that the AMIHUD measure leads the
movement of the TURNOVER measures.
Second, the results of ESPREAD are very similar to the results of TURNOVER: more
than half of the thirteen variables significantly Granger cause both liquidity measures,
and both liquidity measure only have a few effect on the thirteen variables. dINV leads
ESPREAD at the 1% significance level, suggesting that the change of private investment
has significantly effect on the change of ESPREAD liquidity measure. HML also has a
significant Granger causal effect on ESPREAD at 1% level. dM2 and ESPREAD,
dCONSR and ESPREAD, and dTIGHTEN and ESPREAD have a mutual effect on each
other. Contrary to ESPREAD, QSPREAD does not Granger cause any of the thirteen
macro and market variables, and only two variables have significant effect on QSPREAD:
HML at 1% level, and dINV at 5% level.
Third, none of the four price impact components are Granger caused by any macro
variables, but all of them Granger cause several macro variables. For example, there is a
one-way Granger causality from PF to dTIGHTEN, but not the other way around. In
addition, two permanent price impact component, PV, and PF, have one way lead
24 relations with dM2, suggesting that the permanent effect in price could lead the changes
of money supply.
Last but not the least, only one of the macro variables leads SPREAD_TBI significantly
at 1% level, indicating that the movement of money market has significant effect on
private real estate market. There is a significant one way Granger causality from
SPREAD_TBI to dM2, VIX, AMIHUD_MKT, and TERM at the 10% level.
In results not reported, we also investigated the impact of the financial crisis on these
tests. Using a dummy variable for the time of the crisis, we recalculated the Granger
causality tests. Espread and Qspread showed a strong lag effect from most
macroeconomic variables. The lead effect for the Amihud variable on the HML variable
disappears. However, the turnover variable continues to be lead by many
macroeconomics series.
To summarize, for the private real estate market over the full sample period, the money
market variable (dNONM1) is the only variable among thirteen macro and market
variables that has significant effect on liquidity measure SPREAD_TBI. While, for public
real estate markets, the effects of macro and market variables on the liquidity measures
are mainly concentrated on the liquidity measures constructed from daily data but no
effect on the liquidity measures constructed from intraday data. This is not surprising as
the liquidity measures calculated from intraday data are designed to capture information
through the intraday trading process more quickly than the macro or market variables
calculated from daily or quarterly data. Moreover, contrary to the results in Ling et al
(2011), we do not find a significant bivariate lead-lag relation between the tightening
25 standards for commercial real estate loans (dTIGHTEN) and TURNOVER liquidity
measure. However, this might not be surprising given the simple bivariate relationships
investigated in the causality framework. A multivariate analysis may give rise to different
conclusions. Among the eight liquidity measures, dTIGHTEN and ESPREAD have
mutual effect on each other at 10% significant level, and the transitory fixed component
(TF) and the permanent fixed component (PF) significantly Granger cause dTIGHTEN.
The effect is strongest for the permanent fixed component of price impact, suggesting
that the fixed component of price impact has evident effect on the tightening standards
for commercial real estate loans. This result confirms the findings documented in Schnabl
(2011) that there exists a negative lead-lag relation between liquidity shocks and banks
lending.
Our results show important connections between these markets. These results are highly
relevant to real estate investors who might consider holding REITs in addition to private
market real estate to improve portfolio liquidity. The empirical results also have
implications for risk measurement for institutional investors and might suggest that
information from public markets or macroeconomic variables could be used to improve
risk exposure estimates for these investors. This paper has not investigated whether
trading strategies could be developed to take advantage of timing differences across
public and private markets. Such strategies may be unlikely to be profitable given the
large transaction costs of trading direct real estate.
VI. Conclusion
26 Our study has investigated a number of features of cross-market liquidity between public
and private commercial real estate markets. To our knowledge this has been the first
study to consider liquidity connections between these two markets. Furthermore it is one
of the first to investigate liquidity in commercial real estate markets following the
financial crisis.
Our initial set of results focuses on the REIT market and compares liquidity for REITs to
a set of controls matched for size and book to market, as well as overall equity market
liquidity. We note that controlling for size and book to market, REIT liquidity is lower
than for non-REIT firms. While our results show clear commonality in liquidity among
REITs, as well the control firms, using the Amihud liquidity measure shows that there is
less commonality in the REITs compared to the controls. Indeed, based on the R2
measure, the commonality in REITs is less than half that of the controls and only a third
of that of the overall market.
Like Brounen et al (2009), we do note important differences between the liquidity
measures. For instance, in terms of the measures of commonality reported for turnover,
for REITs we find this to be significantly higher than for the controls and the market.
These results also appear to be at odds with the findings of Korajczyk and Sadka (2008),
who generally find lower levels of commonality in liquidity based on turnover measures.
While this results requires further investigation it may be due in part to the financial crisis
and the way in which real estate related firms were particularly affected by the nature of
the crisis.
27 Furthermore, we found important information contained in liquidity measures constructed
using market microstructure variables that has not been previously identified. Firstly, the
extreme liquidity freeze that took place during the financial crisis is much more evident
when intraday data is used than daily data. Also, there is evidence that the liquidity
variables based on intraday data lead other liquidity measures for REITs.
During the financial crisis, we also found evidence that REIT liquidity measure led
overall market liquidity and turnover measures. This may be due to the real estate
intensive nature of the crisis.
In the second set of findings reported, we investigate cross-market liquidity between
public and private real estate markets. We find generally similar time-variation in the
liquidity measures for both real estate markets. Due to the similar time-series variation,
we test for any directional causality between the markets and find that generally the
causality runs from the public markets to the private markets. However, we do note the
finding of bi-directional causality between the TBI spread measure of liquidity and the
Amihud measure for REITs.
Our final set of results investigated a connection between macroeconomic factors and real
estate market liquidity. We find a strong association between real estate liquidity and the
term spread, and between real estate liquidity and changes in real investment and
consumption expenditure, as well as with the unemployment rate. Other strong
associations were found between liquidity and changes in GDP. An interesting difference
noted between the public and private markets, was that credit spreads did not appear to be
28 associated with liquidity in the private real estate market, but it was associated with
liquidity in the REIT market.
29 References
Amihud, Yakov. 2002. “Illiquidity and stock returns: cross-setion and time-series effects.”
Journal of Financial Markets 5:31-56.
Barber, Brad M. and John D. Lyon. 1997. “Detecting long-run abnormal stock returns: the
empirical power and speicification of test statistics.” Journal of Financial Economics 43:341-372.
Bhasin, V., R. A. Cole, and J. K. Kiely. 1997. “Changes in REIT Liquidity 1990-1994: Evidence
from Intraday Transactions.” Real Estate Economics 25:615-630.
Bond, Shaun A., Soosung Hwang, Zhenguo Lin, and Kerry D. Vandell. 2007. “Marketing Period
Risk in a Portfolio Context: Theory and Empirical Estimates from the UK Commercial Real
Estate Market.” Journal of Real Estate Finance and Economics 34:447-641.
Bond, Shuan A. and Steve L. Slezak. 2010. “The Optimal Portfolio Weight for Real Esate with
Liquidity Risk and Uncertainty Aversion.” Working Paper, University of Cincinnati.
Brounen, D., P. Eichholtz, and D. Ling. 2009. “The liquidity of property shares: an international
comparison.” Real Estate Economics 37:413-445.
Buckles, Brian W. 2008. “Liquidity Dynamics in Commercial Real Estate.” Journal of Real
Estate Portfolio Management 14:307-324.
Cannon, Susanne and Rebel Cole. 2011. “Changes in REIT liquidity 1988-2007: Evidence from
daliy data.” The Journal of Real Estate Finance and Economics 43:258-280.
Cheng, P., Z. Lin, and Y. Liu. 2010. “Illiquidity, Transaction Cost, and Optimal Holding Period
for Real Estate: Theory and Application.” Journal of Housing Economics 19:109-118.
Chordia, Tarun, Richard Roll, and Avanidhar Subrahmanyam. 2001. “Market liquidity and
trading activity.” The Journal of Finance 56:501-530.
Clayton, Jim and Greg MacKinnon. 2000. “Measuring and Explaining Changes in REIT
Liquidity: Moving Beyond the Bid-Ask Spread.” Real Estate Economics 28:89-115.
Connor, Gregory and Robert A. Korajczyk. 1986. “Performance measurement with the arbitrage
pricing theory: a new framdework for analysis.” Journal of Financial Economics 15:373-394.
Connor, Gregory and Robert A. Korajczyk. 1987. “Estimating pervasive economic factors with
missing observations.” Working paper no. 34. Department of Finance, Northwestern Universtiy.
Fama, Eugene F. and Kenneth R. French. 1992. “The cross-section of expected stock returns.”
Journal of Finance 47 427-465.
30 Fama, Eugene F. and Kenneth R. French. 1993. “Common risk factors in the returns on stocks
and bonds.” Journal of Financial economics 33:3-56.
Fisher, J., D. Gatzlaff, D. Geltner, and D. Haurin. 2003. “Controlling for the impact of variable
liquidity in commercial real estate price indices.” Real Estate Economic 31:269-303.
Fisher, J., D. Geltner, and H. Pollakowski. 2007. “A quarterly transactions-based index (TBI) of
institutional real estate investment performance and movements in supply and demand.” Journal
of Real Estate Finance and Economics 34:5-33.
Hiemstra, C. and J. Jones (1994) "Testing for Linear and Nonlinear Granger Causality in the
Stock Price-Volume Relation", Journal of Finance 49: 1639-1664.
Korajczyk, Robert A. and Ronnie Sadka. 2008. “Pricing the commonality across alternative
measures of liquidity.” Journal of Financial Economics 87:45-72.
Lin, Z. and K. Vandell. 2007. “Illiquidity and Pricing Biases in the Real Estate Market.” Real
Estate Economics 35:291-330.
Ling, D., A. Naranjo, and B. Scheick (2011) Credit Availability and Asset Pricing Spirals in
Illiquid Markets, Working paper.
Marcato, G and C. Ward. 2007. “Back from Beyond the Bid-Ask Spread: Estimating Liquidity in
International Markets.” Real Estate Economics 35:597-620.
Naes, R, JA Skjeltorp, and BA Qdegaard. 2011. “Stock market liquidity and the business cycle.”
Journal of Finance 66:139-176.
Sadka, Ronnie. 2006. “Momentum and post-earnings-announcement drift anomalies: The role of
liquidity risk.” Journal of Financial Economics 80:309-349.
Schnabl, Philipp. 2011. “The International Transmission of Bank Liquidity Shocks: Evidence
from an Emerging Market.” Working paper.
Spiegel, Matthew. 2008. “Patterns in cross market liquidity.” Finance Research Letters 5:2-10.
31 Fig. 1
Amihud and Turnover are the two liquidity measures. The Amihud (2002) measure is defined as
the monthly average of daily absolute value of return divided by dollar volume; Turnover is
defined as the ratio of monthly volume and shares outstanding. Both measures are calculated as
equally weighted averages across stocks. _REITs is liquidity measures of REITs firms, _control
is control group constructed by matching closest size and book-to-market with REITs firms, and
_market is liquidity measure of the whole market.
Amihud: 1993:09 ‐ 2010:12
25
20
15
10
5
199309
199404
199411
199506
199601
199608
199703
199710
199805
199812
199907
200002
200009
200104
200111
200206
200301
200308
200403
200410
200505
200512
200607
200702
200709
200804
200811
200906
201001
201008
0
Amihud_REITs
Amihud_control
Amihud_market
Turnover: 1993:09 ‐ 2010:12
12
10
8
6
4
2
199309
199404
199411
199506
199601
199608
199703
199710
199805
199812
199907
200002
200009
200104
200111
200206
200301
200308
200403
200410
200505
200512
200607
200702
200709
200804
200811
200906
201001
201008
0
Turnover_REITs
Turnover_control
Turnover_market
32 Fig. 2
Qspread, Espread, PV, PF, TV, and TF are the six liquidity measures calculated with intraday
data of REITs firms. Qspread is measured as the ratio of the quoted big-ask spread and the
bid-ask midpoint; Espread is measured as the absolute value of the difference between
the transaction price and the midpoint of quoted bid and ask divided by the bid-ask
midpoint; and four price impact components, PV, PF, TV, TF, are calculated as in Sadka
(2006).
Bid‐Ask Spread
0.03
0.025
0.02
0.015
0.01
0.005
199301
199308
199403
199410
199505
199512
199607
199702
199709
199804
199811
199906
200001
200008
200103
200110
200205
200212
200307
200402
200409
200504
200511
200606
200701
200708
200803
200810
200905
200912
201007
0
Espread
Qspread
33 ‐0.05
199301
199308
199403
199410
199505
199512
199607
199702
199709
199804
199811
199906
200001
200008
200103
200110
200205
200212
200307
200402
200409
200504
200511
200606
200701
200708
200803
200810
200905
200912
201007
‐0.00005
199301
199309
199405
199501
199509
199605
199701
199709
199805
199901
199909
200005
200101
200109
200205
200301
200309
200405
200501
200509
200605
200701
200709
200805
200901
200909
201005
Price Variable Impact
0.00015
0.0001
0.00005
0
‐0.0001
‐0.00015
Permanent Variable
Permanenant Fixed
Transitory Variable
Price Fixed Impact
0.2
0.15
0.1
0.05
0
‐0.1
Transitory Fixed
34 Fig. 3
Amihud, Turnover and Spread_TBI are the three liquidity measures of REITs firms. The Amihud
measure is defined as the monthly average of daily absolute value of return divided by dollar
volume; Turnover is defined as the ratio of monthly volume and shares outstanding. Both
measures are calculated as equally weighted averages across stocks. Spread_TBI are constructed
from the transactions-based index (TBI) developed by MIT center for real estate.
1.2
3.5
3
1
2.5
0.8
2
0.6
1.5
0.4
1
0.2
0.5
‐0.2
198702
198801
198804
198903
199002
199101
199104
199203
199302
199401
199404
199503
199602
199701
199704
199803
199902
200001
200004
200103
200202
200301
200304
200403
200502
200601
200604
200703
200802
200901
0
‐0.4
0
‐0.5
‐1
Amihud
Spread_TBI
Turnover
35 Table 1
and the average adjusted- of time-series regressions
This table reports the average
using one, two, and three common factors. Within-measure common factors are extracted
from both liquidity measures by APC method, and then for each variable and each stock,
we regress each of the liquidity measure on its common factors. The Amihud (2002)
measure is defined as the monthly average of daily absolute value of return divided by
dollar volume; Turnover is defined as the ratio of monthly volume and shares outstanding;
Qspread is measured as the ratio of the quoted big-ask spread and the bid-ask midpoint;
Espread is measured as the absolute value of the difference between the transaction price
and the midpoint of quoted bid and ask divided by the bid-ask midpoint; and four price
impact components, PV, PF, TV, TF, are calculated as in Sadka (2006). Before common
factor and regression analysis, for each stock, each liquidity measure is normalized every
month by its mean and standard deviation calculated up to the prior month.
REITs
Variable
Statistic
Amihud
Adj.
Turnover
Adj.
Qspread
Adj.
Espread
Adj.
PV(λ)
Adj.
PF(Ψ)
Adj.
TV(λ)
Adj.
TF(Ψ)
Adj.
Amihud
Adj.
Turnover
Adj.
Factor
1
Factor
2
Control Firms
Factor
2
Market
Factor
3
Factor
1
Factor
3
Factor
1
Factor
2
Factor
3
19.31
18.23
19.97
18.89
-
34.66
32.79
27.23
25.24
-
39.05
36.42
37.00
34.35
-
15.76
14.80
17.95
16.95
27.19
25.48
24.29
22.43
36.50
34.16
32.19
29.64
6.64
5.38
26.74
25.67
15.93
14.37
10.08
8.50
1.86
0.01
3.78
2.04
1.74
-0.11
4.84
3.11
13.57
11.19
34.95
33.01
35.96
33.38
14.84
11.86
3.44
-0.30
8.38
5.05
3.73
0.01
9.00
5.61
22.88
19.66
37.81
35.00
48.29
44.89
28.06
24.02
6.09
0.46
13.05
8.22
5.33
-0.39
13.51
8.57
Panel A. 1993:09 - 2010:12
11.29
18.42
21.76
10.10
16.23
18.56
13.50
20.32
29.57
12.32
18.10
26.58
-
8.60
7.43
23.58
22.55
13.88
11.65
31.94
30.07
18.57
15.38
36.24
33.55
Panel B. 1983:01 - 2010 :12
7.06
14.04
19.63
5.93
11.98
16.73
12.76
17.65
25.16
11.68
15.60
22.34
Table 2
The table shows Granger causality tests between liquidity measures derived from REITs
firms and liquidity measures derived from the transactions-based index (TBI). For each
measure, we first test the null hypothesis that the variables in Column (1) does not
Granger cause the variables in Column (2), and then variables in Column (3) does not
Granger cause variables in Column (4). The -statistics and p-value (in parentheses) are
reported for each test.
Panel A: Monthly liquidity measures from 1993:09 to 2010:12
pvalue
Null Hypothesis
(1)
(2)
pvalue
Null Hypothesis
(3)
(4)
Amihud
Turnover
27.930
0.000
5.579
0.349
Amihud_mkt
Turnover_mkt
1.754
0.781
13.302
0.021
Amihud_mkt
Turnover_mkt
Amihud
Qspread
2.417
0.491
Qspread
Amihud
26.211
0.000
Amihud
Espread
12.601
0.126
Espread
Amihud
7.865
0.447
Amihud
PV
2.397
0.302
PV
Amihud
0.625
0.732
Amihud
PF
3.613
0.823
PF
Amihud
5.475
0.602
Amihud
TV
2.675
0.263
TV
Amihud
3.147
0.207
Amihud
TF
4.535
0.716
TF
Amihud
2.702
0.911
Turnover
Qspread
5.884
0.318
Qspread
Turnover
14.786
0.011
Turnover
Espread
5.665
0.129
Espread
Turnover
11.722
0.008
Turnover
PV
2.553
0.466
PV
Turnover
2.544
0.467
Turnover
PF
8.746
0.033
PF
Turnover
0.788
0.852
Turnover
TV
53.898
0.000
TV
Turnover
16.374
0.498
Turnover
TF
3.299
0.348
TF
Turnover
18.958
0.000
Qspread
Espread
49.722
0.000
Qspread
Espread
4.406
0.819
Qspread
PV
1.017
0.313
PV
Qspread
0.734
0.392
Qspread
PF
19.783
0.000
PF
Qspread
1.326
0.250
Qspread
TV
3.721
0.054
TV
Qspread
0.148
0.701
Qspread
TF
26.741
0.000
TF
Qspread
4.413
0.731
Espread
PV
3.612
0.307
PV
Espread
0.758
0.859
Espread
PF
26.586
0.000
PF
Espread
11.389
0.010
Espread
TV
19.111
0.322
TV
Espread
51.709
0.000
Espread
TF
78.627
0.000
TF
Espread
91.971
0.000
PV
PF
6.539
0.478
PF
PV
6.608
0.471
PV
TV
0.399
0.528
TV
PV
2.683
0.101
PV
TF
18.805
0.009
TF
PV
3.991
0.781
PF
TV
18.006
0.207
TV
PF
75.542
0.000
Amihud
Turnover
Panel B: Quarterly liquidity measures from 1993:01 to 2009:03
Null Hypothesis
(1)
p-value
(2)
Null Hypothesis
(3)
p-value
(4)
Amihud
Spread_TBI
9.594
0.048
Spread_TBI
Amihud
7.482
0.113
Turnover
Spread_TBI
17.925
0.001
Spread_TBI
Turnover
1.723
0.787
Qspread
Spread_TBI
13.151
0.004
Spread_TBI
Qspread
5.251
0.154
Espread
Spread_TBI
21.876
0.009
Spread_TBI
Espread
11.334
0.254
PV
Spread_TBI
0.082
0.774
Spread_TBI
PV
2.431
0.119
PF
Spread_TBI
3.933
0.269
Spread_TBI
PF
4.952
0.175
TV
Spread_TBI
37.097
0.002
Spread_TBI
TV
30.580
0.015
TF
Spread_TBI
27.379
0.072
Spread_TBI
TF
144.416
0.000
38 Panel C: Monthly liquidity measures from 2007:07 to 2010:06
Null Hypothesis
p-value
Null Hypothesis
(1)
(2)
(3)
(4)
Amihud
Amihud
Amihud_mkt
Amihud_mkt
27.930
0.000
Turnover
Amihud
Amihud
Amihud
Amihud
Amihud
Amihud
Turnover
Turnover
Turnover
Turnover
Turnover
Turnover
Qspread
Qspread
Qspread
Qspread
Qspread
Espread
Espread
Espread
Espread
PV
PV
PV
PF
Turnover_mkt
Qspread
Espread
PV
PF
TV
TF
Qspread
Espread
PV
PF
TV
TF
Espread
PV
PF
TV
TF
PV
PF
TV
TF
PF
TV
TF
TV
7.817
13.739
11.824
6.875
0.896
5.578
5.803
1.595
2.054
3.032
1.536
2.472
2.393
0.099
0.008
0.019
0.143
0.925
0.233
0.214
0.810
0.726
0.553
0.820
0.650
0.664
21.687
6.433
3.140
4.806
0.000
0.169
0.535
0.308
11.016
6.283
0.026
0.179
11.034
2.830
2.911
4.252
0.495
1.023
0.026
0.587
0.573
0.373
0.974
0.906
3.302
0.509
Turnover_mkt
Qspread
Espread
PV
PF
TV
TF
Qspread
Espread
PV
PF
TV
TF
Espread
PV
PF
TV
TF
PV
PF
TV
TF
PF
TV
TF
TV
Turnover
Amihud
Amihud
Amihud
Amihud
Amihud
Amihud
Turnover
Turnover
Turnover
Turnover
Turnover
Turnover
Qspread
Qspread
Qspread
Qspread
Qspread
Espread
Espread
Espread
Espread
PV
PV
PV
PF
p-value
2.692
0.611
5.873
5.768
1.969
1.097
3.241
7.670
0.852
2.503
0.209
0.217
0.742
0.895
0.518
0.104
0.931
0.644
11.068
6.128
0.026
0.190
11.652
0.365
0.020
0.985
10.441
1.662
3.160
0.034
0.798
0.531
9.562
3.238
1.924
0.683
4.292
0.781
2.911
5.018
3.868
5.596
0.049
0.519
0.750
0.953
0.368
0.941
0.573
0.285
0.424
0.231
1.456
0.834
Table 3
This table shows the correlation coefficients between liquidity measures and market and macroeconomic variables of United States.
The p-values are reported parentheses. Amihud, Turnover, Qspread, Espread, PV, PF, TV, and TF are constructed with intra- or
interday data sets of REITs stocks. The Amihud measure is defined as the monthly average of daily absolute value of return divided
by dollar volume; Turnover is defined as the ratio of monthly volume and shares outstanding. Qspread is measured as the ratio of the
quoted big-ask spread and the bid-ask midpoint; Espread is measured as the absolute value of the difference between the transaction
price and the midpoint of quoted bid and ask divided by the bid-ask midpoint; and four price impact components, PV, PF, TV, TF, are
calculated as in Sadka (2006). Both measures are calculated as equally weighted averages across stocks. Spread_TBI are constructed
from the transactions-based index (TBI) developed by MIT center for real estate. TERM is term spread, CRED is credit spread, MKT
is excess market return, dGDPR is real GDP growth, dINV is growth in investment, dUE is growth in the unemployment rate, and
dCONSR is real consumption growth. dM2 is growth in the money supply M2 that includes M1 in addition to all time-related deposits,
savings deposits, and non-institutional money-market funds. dNONM1 is the growth in the non-M1 component of M2. VIX is the
volatility of S&P 500. HML is the spread in returns between value and growth stocks. dTIGHTEN is the net percentage of domestic
respondents tightening standards for commercial real estate loans10.
10
For further information of this variable, please refer to the Board of Governors of the Federal Reserve System's Senior Loan Officer Opinion Survey on Bank
Lending Practices release. http://www.federalreserve.gov/boarddocs/SnLoanSurvey/.
Panel A: 1987:Q1 – 2010:Q2
TERM
CRED
MKT
dGDPR
dINV
dCONSR
dUE
dM2
dNONM1
VIX
HML
dTIGHTEN
AMIHUD
TURNOVER
QSPREAD
ESPREAD
PV
PF
TV
TF
0.165
0.182
0.188
0.128
-0.080
0.522
0.034
0.784
0.060
0.630
0.164
0.184
-0.095
0.447
-0.208
0.091
-0.294
0.016
-0.061
0.626
0.166
0.181
0.021
0.866
0.025
0.843
-0.018
0.884
-0.147
0.235
-0.387
0.001
-0.359
0.003
-0.503
0.000
0.327
0.007
0.164
0.184
0.044
0.722
0.019
0.877
-0.030
0.807
0.020
0.870
0.024
0.849
0.107
0.388
0.008
0.950
0.102
0.410
0.082
0.507
0.226
0.066
-0.056
0.654
-0.155
0.210
-0.117
0.345
0.103
0.406
0.078
0.531
0.008
0.951
0.099
0.426
0.344
0.004
-0.090
0.471
-0.230
0.061
-0.316
0.009
-0.122
0.326
0.334
0.006
-0.079
0.524
-0.124
0.317
0.330
0.006
-0.023
0.857
0.001
0.991
0.310
0.011
0.279
0.022
0.097
0.433
-0.050
0.689
-0.106
0.393
-0.049
0.693
0.188
0.128
0.001
0.994
-0.128
0.301
0.127
0.306
0.075
0.547
-0.012
0.926
-0.145
0.243
-0.167
0.177
-0.117
0.348
0.022
0.860
0.026
0.834
0.079
0.525
-0.060
0.628
-0.048
0.698
-0.002
0.989
0.137
0.267
-0.259
0.034
0.134
0.279
-0.116
0.349
0.007
0.958
-0.218
0.076
0.044
0.722
0.130
0.295
0.108
0.386
-0.252
0.040
0.106
0.393
0.144
0.244
-0.153
0.217
-0.040
0.750
0.208
0.092
0.041
0.745
-0.102
0.413
0.085
0.494
-0.108
0.385
-0.163
0.189
-0.143
0.248
0.179
0.147
-0.261
0.033
-0.229
0.062
0.053
0.672
-0.095
0.443
-0.014
0.908
SPREAD
_TBI
0.431
0.000
0.446
0.000
-0.155
0.210
-0.556
0.000
-0.598
0.000
-0.591
0.000
0.629
0.000
-0.056
0.650
-0.313
0.010
0.247
0.044
0.002
0.990
-0.053
0.671
41 Panel B: 1987:Q1 – 2007:Q2
TERM
CRED
MKT
dGDPR
dINV
dCONSR
dUE
dM2
dNONM1
VIX
HML
dTIGHTEN
AMIHUD
TURNOVER
QSPREAD
ESPREAD
PV
PF
TV
TF
0.155
0.244
0.340
0.009
-0.187
0.160
-0.074
0.581
-0.133
0.320
0.087
0.517
0.110
0.411
-0.107
0.423
-0.231
0.081
0.095
0.480
0.082
0.543
0.105
0.435
-0.049
0.716
-0.331
0.011
0.005
0.970
-0.169
0.206
-0.096
0.473
-0.302
0.021
0.065
0.630
0.011
0.933
-0.021
0.877
-0.315
0.016
0.075
0.578
-0.107
0.426
-0.053
0.691
0.113
0.397
-0.016
0.907
0.156
0.241
0.143
0.283
0.277
0.036
-0.097
0.468
-0.161
0.226
-0.115
0.392
0.121
0.367
-0.052
0.698
0.092
0.491
-0.017
0.900
0.127
0.341
-0.003
0.980
0.172
0.196
0.140
0.293
0.292
0.026
-0.078
0.558
-0.137
0.305
-0.087
0.518
0.166
0.214
-0.071
0.599
0.094
0.483
0.265
0.044
0.244
0.065
0.035
0.795
0.096
0.474
0.072
0.591
0.141
0.290
0.028
0.836
0.038
0.777
-0.046
0.732
0.069
0.609
-0.039
0.773
0.059
0.658
-0.183
0.168
-0.353
0.007
0.102
0.445
0.245
0.064
0.250
0.059
0.240
0.070
-0.203
0.126
-0.190
0.153
-0.130
0.331
0.031
0.819
-0.191
0.152
0.112
0.402
0.003
0.985
0.140
0.296
-0.088
0.510
-0.176
0.187
-0.097
0.469
-0.089
0.508
0.075
0.577
0.042
0.754
-0.068
0.610
-0.061
0.651
0.146
0.273
-0.074
0.583
-0.099
0.461
-0.420
0.001
0.152
0.255
0.246
0.063
0.310
0.018
0.218
0.101
-0.296
0.024
-0.363
0.005
-0.187
0.160
-0.215
0.105
-0.185
0.164
0.047
0.727
SPREAD
_TBI
0.346
0.008
0.086
0.521
-0.081
0.544
-0.127
0.343
-0.093
0.488
-0.151
0.259
0.150
0.260
-0.245
0.064
-0.417
0.001
-0.253
0.056
0.035
0.792
-0.020
0.882
42 Panel C: 2007:Q3 – 2010:Q2
AMIHUD
TERM
CRED
MKT
dGDPR
dINV
dCONSR
dUE
dM2
dNONM1
VIX
HML
dTIGHTEN
0.956
0.000
0.430
0.288
0.344
0.404
-0.384
0.348
-0.424
0.295
-0.163
0.699
0.693
0.057
-0.278
0.506
-0.451
0.262
0.332
0.422
0.272
0.515
-0.925
0.001
TURNOVER
-0.490
0.218
0.334
0.419
-0.736
0.038
-0.485
0.223
-0.173
0.682
-0.670
0.069
-0.027
0.949
0.859
0.006
0.794
0.019
0.406
0.319
-0.501
0.206
0.748
0.033
QSPREAD
0.486
0.222
0.861
0.006
-0.120
0.777
-0.738
0.037
-0.918
0.001
-0.428
0.290
0.948
0.000
0.125
0.767
0.121
0.774
0.800
0.017
-0.243
0.562
-0.427
0.292
ESPREAD
0.731
0.039
0.860
0.006
-0.139
0.742
-0.780
0.023
-0.819
0.013
-0.433
0.284
0.875
0.005
0.156
0.712
0.075
0.861
0.741
0.035
-0.170
0.688
-0.580
0.132
PV
PF
TV
TF
0.389
0.341
0.102
0.810
0.558
0.150
-0.076
0.857
-0.239
0.569
-0.233
0.579
0.491
0.217
-0.176
0.677
-0.452
0.261
0.106
0.803
0.470
0.240
-0.408
0.315
-0.234
0.578
0.347
0.400
-0.679
0.064
-0.406
0.319
-0.263
0.529
-0.116
0.784
0.047
0.912
0.419
0.301
0.648
0.082
0.564
0.146
-0.632
0.093
0.196
0.642
-0.483
0.225
-0.043
0.919
-0.471
0.239
0.162
0.702
0.253
0.545
0.281
0.500
-0.573
0.138
0.232
0.580
0.551
0.157
-0.275
0.509
-0.268
0.521
0.666
0.072
0.272
0.515
0.303
0.466
0.210
0.617
-0.331
0.423
-0.579
0.133
-0.290
0.487
0.718
0.045
-0.107
0.802
-0.275
0.510
0.510
0.197
-0.076
0.857
-0.463
0.247
SPREAD
_TBI
0.680
0.064
0.715
0.046
0.039
0.927
-0.691
0.058
-0.696
0.055
-0.575
0.136
0.912
0.002
0.068
0.873
-0.089
0.835
0.649
0.081
-0.007
0.988
-0.573
0.138
43 Table 4
The table shows Granger causality tests between liquidity measures macro variables. For each measure, we first test the null hypothesis that if each
of liquidity measures does not Granger cause macro variables, and then whether each of the macro variables does not Granger cause any liquidity
measures. The -statistics and p-value (in parentheses) are reported for each test.
Null Hypothesis
Null Hypothesis
dM2
dNONM1
VIX
HML
AMIHUD_MKT
dGDPR
dINV
dCONSR
dUE
MKT
TERM
CRED
dTIGHTEN
AMIHUD
4.805
7.630
2.718
5.101
0.308
0.106
0.606
0.277
11.997
1.602
7.581
0.302
5.837
0.693
1.575
0.017
0.808
0.108
0.990
0.212
0.952
0.813
8.283
1.942
0.082
0.747
AMIHUD
dM2
dNONM1
VIX
HML
AMIHUD_MKT
dGDPR
dINV
dCONSR
dUE
MKT
TERM
CRED
dTIGHTEN
0.604
1.767
4.646
0.963
0.778
0.326
17.069
2.094
5.122
5.437
6.542
2.058
1.907
1.730
5.978
3.795
0.002
0.718
0.275
0.245
0.162
0.725
0.753
0.785
0.201
0.435
Null Hypothesis
Null Hypothesis
dM2
dNONM1
VIX
HML
6.553
3.518
0.162
0.475
0.041
0.015
dM2
dNONM1
VIX
HML
7.514
4.853
1.318
6.102
0.111
0.303
0.858
0.192
9.991
12.368
TURNOVER_MKT
21.409
0.000
TURNOVER_MKT
5.904
0.207
2.744
0.602
8.980
5.194
0.062
0.268
dGDPR
dINV
dCONSR
0.588
0.500
0.823
dUE
10.363
0.035
dUE
MKT
TERM
CRED
dTIGHTEN
21.469
21.113
8.092
7.318
0.000
0.000
0.088
0.120
MKT
TERM
CRED
dTIGHTEN
2.823
3.354
1.521
15.27
6
4.717
3.040
2.505
4.694
dM2
dNONM1
VIX
HML
dGDPR
dINV
dCONSR
dUE
MKT
TERM
CRED
dTIGHTEN
20.614
13.054
3.812
0.000
0.011
0.432
31.777
4.526
0.000
0.340
13.992
17.768
12.979
9.730
6.569
3.225
0.007
0.001
0.011
0.045
0.161
0.521
8.051
4.352
1.865
1.849
2.297
0.542
0.090
0.360
0.761
0.764
0.681
0.969
9.153
2.099
3.530
1.893
6.344
0.057
0.718
0.473
0.756
0.175
8.846
0.065
dM2
dNONM1
VIX
HML
dGDPR
dINV
dCONSR
dUE
MKT
TERM
CRED
dTIGHTEN
7.817
0.099
dGDPR
dINV
dCONSR
TURNOVE
R
ESPREAD
TURNOVER
ESPREAD
0.004
0.318
0.551
0.644
0.320
Null Hypothesis
Null Hypothesis
dM2
dNONM1
VIX
HML
dGDPR
dINV
dCONSR
dUE
MKT
TERM
CRED
dTIGHTEN
dM2
dNONM1
VIX
HML
dGDPR
dINV
dCONSR
dUE
MKT
TERM
CRED
dTIGHTEN
QSPREAD
PV
6.636
5.215
7.425
0.156
0.266
0.115
20.034
5.003
0.001
0.287
12.387
5.292
3.679
6.012
6.834
1.996
2.129
0.015
0.259
0.451
0.198
0.145
0.737
0.712
1.945
0.042
3.209
3.392
2.578
3.368
1.446
2.448
2.296
0.511
1.693
0.592
0.746
1.000
0.523
0.495
0.631
0.498
0.836
0.654
0.682
0.972
0.792
0.964
QSPREAD
DM2
DNONM1
VIX
HML
DGDPR
DINV
DCONSR
DUE
MKT
TERM
CRED
DTIGHTEN
5.009
2.493
0.940
1.774
0.876
1.171
4.931
2.177
0.639
1.735
4.146
7.485
0.286
0.646
0.919
0.777
0.928
0.883
0.295
0.703
0.959
0.784
0.387
0.112
PV
dM2
dNONM1
VIX
HML
dGDPR
dINV
dCONSR
dUE
MKT
TERM
CRED
dTIGHTEN
8.382
18.605
3.073
2.198
3.562
7.410
0.079
0.001
0.546
0.700
0.469
0.116
8.308
1.074
2.521
0.874
4.826
1.092
0.081
0.898
0.641
0.928
0.306
0.896
46 Null Hypothesis
Null Hypothesis
dM2
dNONM1
VIX
HML
dGDPR
dINV
dCONSR
dUE
MKT
TERM
CRED
dTIGHTEN
dM2
dNONM1
VIX
HML
dGDPR
dINV
dCONSR
dUE
MKT
TERM
CRED
dTIGHTEN
PF
TV
3.290
2.727
1.151
0.994
1.714
0.436
0.936
1.142
0.433
0.763
0.458
0.826
0.511
0.605
0.886
0.911
0.788
0.979
0.919
0.888
0.980
0.943
0.978
0.935
2.132
1.381
2.236
1.020
1.997
2.992
2.004
0.546
1.012
0.793
1.891
1.725
0.712
0.848
0.693
0.907
0.736
0.559
0.735
0.969
0.908
0.939
0.756
0.786
PF
TV
dM2
dNONM1
VIX
HML
dGDPR
dINV
dCONSR
dUE
MKT
TERM
CRED
dTIGHTEN
10.009
6.874
6.834
1.105
1.379
0.202
4.626
1.647
0.040
0.143
0.145
0.894
0.848
0.995
0.328
0.800
8.268
4.739
0.082
0.315
13.362
13.875
0.010
0.008
dM2
dNONM1
VIX
HML
dGDPR
dINV
dCONSR
dUE
MKT
TERM
CRED
dTIGHTEN
8.504
6.211
3.644
1.259
3.808
5.990
0.075
0.184
0.456
0.868
0.433
0.200
9.296
6.867
5.392
0.434
6.652
4.038
0.054
0.143
0.249
0.980
0.156
0.401
47 Null Hypothesis
Null Hypothesis
dM2
dNONM1
VIX
HML
dGDPR
dINV
dCONSR
dUE
MKT
TERM
CRED
dTIGHTEN
dM2
dNONM1
VIX
HML
AMIHUD_MKT
dGDPR
dINV
dCONSR
dUE
MKT
TERM
CRED
dTIGHTEN
TF
SPREAD
_TBI
6.067
6.168
6.511
11.392
0.474
3.063
3.835
2.054
0.898
1.410
4.215
6.542
0.194
0.187
0.164
0.023
0.976
0.547
0.429
0.726
0.925
0.842
0.378
0.162
4.319
0.365
13.610
1.540
6.546
6.380
6.687
4.138
0.555
3.765
3.471
5.080
5.290
4.072
0.009
0.820
0.162
0.173
0.153
0.388
0.968
0.439
0.482
0.279
0.259
0.396
TF
SPREAD
_TBI
dM2
dNONM1
VIX
HML
dGDPR
dINV
dCONSR
dUE
MKT
TERM
CRED
dTIGHTEN
3.931
2.823
0.560
2.092
0.816
1.471
6.820
6.891
1.693
1.666
2.493
0.415
0.588
0.967
0.719
0.936
0.832
0.146
0.142
0.792
0.797
0.646
8.868
0.065
dM2
dNONM1
VIX
HML
AMIHUD_MKT
dGDPR
dINV
dCONSR
dUE
MKT
TERM
CRED
dTIGHTEN
8.230
1.003
0.084
0.909
8.451
0.705
0.076
0.951
12.274
4.075
0.616
4.137
1.192
6.209
0.015
0.396
0.961
0.388
0.879
0.184
8.161
5.208
5.937
0.086
0.267
0.204
48