Asymmetric Information and the Predictability of
Real Estate Returns*
Michael Cooper
Krannert School of Management, Purdue University
David H. Downs
Terry College of Business, University of Georgia
Gary A. Patterson
School of Management, SUNY-New Paltz
Abstract
This paper examines the relation between systematic price changes and the heterogeneity of
investors’ information sets in real estate asset markets. The empirical implications rely on a
theoretical economy in which information asymmetry alters the dynamic relation between returns
and trading volume. We employ a filter-rule methodology to determine predictability in returns
and augment the return-based conditioning set with trading volume. The additional conditioning
information is necessary since the model is underspecified when predictability is based on
returns alone. Our results provide new insight into the co-existence of informational and noninformational exchange in the speculative markets for real estate assets. Specifically, we find
that the predictability of real estate returns is generally more indicative of portfolio rebalancing
effects than an adverse selection problem. Importantly, these results are unique in addressing
the time-variation in information asymmetry.
Forthcoming Journal of Real Estate Finance and Economics
Final version: October 1998
JEL Classification: G10, G14, R33
Key Words: Information, Predictability, Real Estate
*Address Correspondence to: David H. Downs, Terry College of Business, University of Georgia, Athens
GA 30602-6255 or email to [email protected]
We wish to thank Crocker Liu and the participants of the AREUEA and FMA meetings for their helpful
suggestions. We are especially grateful to Chinmoy Ghosh.
Asymmetric Information and the Predictability of Real Estate Returns
1. Introduction
The predictability of asset returns has been the focus of a large body of academic research with studies
attributing this apparent phenomenon to informational inefficiency, investor irrationality, time variation in
risk premia, and other market-specific effects. Recently, Mei and Gao (1995) examine whether the shortterm predictability of real estate assets is exploitable in an economically meaningful sense.1 They find that
the real estate security market is efficient with respect to trading profits, and thus, the real estate security
markets are not accessible to competent arbitrageurs. Their study portrays a general condition of efficiency
in the real estate markets without considering specific market conditions that may alter the behavior of
asset prices. In contrast, other studies suggest that market conditions may influence informational
efficiency and, consequently, asset prices. Damodaran and Liu (1993) conduct a study that focuses upon
events producing periods of asymmetric information that affect price movements
in real estate assets. They examine a sample of REITs that choose to reappraise themselves, an action that
endows the REIT insiders with private information. By identifying an event that heightens information
asymmetry, their study demonstrates that the trading activity of informed investors could influence the price
formation process. However, the extent to which informed trading might influence the predictability of real
estate assets is largely an empirical question.
In this paper we study the predictability of real estate returns for evidence of information-based
trading in a speculative market for real estate assets. We conduct our research of speculative markets as
presented by Wang (1994) and assume the existence of heterogeneous traders and asymmetric information.
In this economy, investors trade for informational and non-informational purposes. The informed investors
possess heterogeneous endowments of (1) private information about the future cash flows of the underlying
asset and (2) private investment opportunities. In this market, the uninformed investor rationally extracts
information from prices and other public signals to estimate expected returns. Consequently, the two
1
classes of investors trade competitively based on (a) the non-informational motives of the uninformed, (b)
the informational motives of the informed (i.e., trading based on private information), and (c) the noninformational motives of the informed (i.e., portfolio rebalancing to accommodate private investment
opportunities). Thus, our paper extends the prior research of Mei and Gao (1995) and Damodaran and Liu
(1993) by examining predictability in short-term returns in a market containing information-based trading
in real estate securities and where investors are heterogeneous in their information and private investment
opportunities (Wang, 1994).2
The heterogeneities in Wang’s model serve to characterize the popular idea of REIT insiders as
private-market participants capitalizing on financing opportunities in the publicly traded markets. Wang
shows that heterogeneity among investors gives rise to different dynamic relations between trading volume
and returns.3 In essence, a high return accompanied by high volume implies low future returns (price
reversals) if informed investors are trading for changes in their private investment opportunities and not
because of private information. Yet when informed investors condition their trades upon private
information, then high future returns (price continuations) are expected when high returns are accompanied
by high trading volume. The model demonstrates that the underlying motivation behind investor behavior
produces different volume-return interactions that affect the pattern of return behavior. Wang’s model
allows us to characterize the nature of investor heterogeneity by examining the pattern of expected returns
that emerges from the interaction between returns and trading volume.
A testable implication of the dynamic relation between returns and trading volume is that the return
reversals documented by Mei and Gao (1995) will be correlated with volume. To test this possibility we
form contrarian portfolios with the aid of a filter rule methodology (Cooper, 1998). This approach avoids
the criticism levied against previous short-horizon contrarian papers that base their portfolio weights on the
cross-sectional distribution of lagged returns.4 More importantly, the filter method offers flexibility for the
detection of nonlinearities in the predictability of price changes. We test the effect of volume on the
2
autocorrelation of weekly returns with the realized portfolio returns acting as proxies for the expected
returns in the Wang model.
We find strong evidence of nonlinearities in the predictability of real estate returns when we
introduce volume into the trading rule. Specifically, the price-volume dynamic differs between high and low
volume periods, where the high volume periods reflect the exchange of real estate assets motivated by
private information. We also observe predictable patterns of return reversals when we form portfolios using
a filter rule based only upon lagged prices, which is consistent with earlier papers. Importantly, the study
highlights the adverse selection problem faced by investors who trade in public real estate markets where
representative insiders may have both private information and private investment opportunities.
The remainder of this paper is organized as follows. Section 2 describes the application of a filter
rule to determine return predictability. In Section 3 we describe the data and then present our empirical
results in Section 4; the analysis focuses on the price-volume dynamic for speculative trading of real estate
assets. We conclude in Section 5.
2. Empirical methods
2.1 Improving signal quality in short-term predictability
To address the testable implication concerning predictability (i.e., that return reversals are
correlated with volume), we employ two significant modifications to the overreaction portfolio formation
methodology used by Mei and Gao (1995). These modifications are designed to boost the “signal-to-noise”
ratio of the security selection process used to form contrarian portfolios. Specifically, our modifications to
the signal extraction process include (1) the use of filters, and (2) the use of a conditioning variable other
than price changes, namely volume.
The filter-rule method allows us to screen on the magnitude of lagged returns and percentage
changes in volume when forming loser and winner portfolios. In contrast, prior short-term contrarian
papers’ portfolio formation methodology (Lehmann, 1990, and Mei and Gao, 1995) typically emphasizes
3
forming portfolios by investing in all securities in their sample, giving greater weight to securities with
larger relative lagged cross-sectional returns. Including stocks regardless of lagged return magnitudes
results in inclusion of securities into the overreaction portfolios that may not be subject to investor
overreaction.5 In contrast, the filter portfolio formation method includes an asset in a loser (winner)
portfolio only if its lagged weekly return moved down (up) by a threshold amount. Hence, our method will
provide a more sensitive measure of predictability for analysis of the price-volume dynamic. Other papers
that use variations of the filter-rule method to boost the sensitivity of their analysis include Alexander
(1961), Fama and Blume (1966), Sweeney (1986, 1988), Brown and Harlow (1988), Lakonishok and
Vermaelen (1990), Bremer and Sweeney (1991), Corrado and Lee (1992), Cox and Peterson (1994), and
Fabozzi, Ma, Chittenden, and Pace (1995).
Our second method to improve the signal-to-noise ratio in a weekly overreaction portfolio strategy
is to utilize variables not directly derived from a security’s price. Because of the scarcity of
macroeconomic and microeconomic time series variables at shorter time intervals, a natural choice would
be to examine the time series properties of volume as it relates to subsequent weeks’ return behavior.
Theoretical papers that have taken this approach include Blume, Easley and O’Hara (1994), Campbell,
Grossman and Wang (1993), and Wang (1994) who present models suggesting there is valuable information in the time series of lagged volume for predicting a security’s price movement. Conrad, Hameed,
and Niden (1994) examine the interaction between lagged percentage changes in transactions, lagged
returns, and subsequent weekly returns to individual NASDAQ securities. They employ an overreaction
portfolio weighting scheme that produces returns from negative autocorrelation. Motivated by these results,
we incorporate a lagged, individual security volume measure into the overreaction portfolio formation rules.
Additionally, the joint use of volume and return filters allows this paper to examine the heterogeneity of
investor behavior.
4
2.2. Filter-rule methodology
The methodology we use is a first-order filter rule where lagged information from one week (i.e.,
returns, or returns and volume) is used to predict future returns. In all, six strategies are examined. The
first two strategies are price-only strategies. For example, if last week’s return is negative, it falls into the
strategy of “loser-price” filter. Hence, the two price-only strategies are “loser-price” and “winner-price,”
and they form portfolios that provide a baseline for interpreting the price-volume results. The remaining
four strategies incorporate both price and volume information. For example, if last week’s return and
percentage change in volume for a security are each negative, the security is assigned to a “loser-price |
low-volume” filter strategy. Likewise, the four price and volume strategies are “loser-price | low-volume”,
“loser-price | high-volume”, “winner-price | low-volume”, and “winner-price | high-volume.”
Past week’s returns are classified as winners or losers using the following criteria:

For k = 0, 1,..., 4 :

Return states = 

For k = 5 :

loserk * A if − k*A > Ri, t −1 ≥ −(k + 1) * A

winner k * A if k*A ≤ Ri, t −1 < (k + 1) * A
loserk * A if Ri, t −1 < − k*A

winner k * A if Ri, t −1 ≥ k * A
(1)
where:
Ri , t is the non-market adjusted return for security i in week t
k is the filter counter that ranges from 0, 1, …5.
A is a parameter equal to 2 percent.
The low and high states for percentage change in individual security weekly volume (termed "volume
returns”) are defined to be:

For k = 0, 1,...,4 :

Volume return states = 

For k = 5 :

low k * B if − k * B > VR i,t −1 ≥ −(k + 1) * B

high k *C if k * C ≤ VRi, t −1 < (k + 1) * C
low k * B if VR i, t −1 < − k * B

high k *C if VRi,t −1 ≥ k * C
(2)
where:
VRi ,t is the volume return for security i in week t
k is the filter counter that ranges from 0, 1, …5.
B is a parameter equal to 15 percent.
5
C is a parameter equal to 50 percent.
The percentage change in individual security weekly volume, termed “volume returns,” adjusted for the
number of outstanding shares of a security, are defined as:
Vi ,t Vi ,t −1 
VRi ,t = 
−

 Si ,t Si ,t −1 
Vi ,t −1 


 Si ,t −1 
(3)
where:
Si ,t is the number of outstanding shares for security i in week t
Vi ,t is the weekly volume for security i in week t .
Thus, k*A, k*B, and k*C are the grid increments for the price filters, low volume filters, and high volume
filters, respectively. For each of the strategies, the applicable price and volume filters are varied over their
respective domains, resulting in thirty-six sets of price and volume filter combinations for the price and
volume strategies.
The specific filter breakpoints are determined by examining the overall sample distributions of the
weekly price return and volume return from our sample of REITS and then choosing appropriate
filter values to span the distributions.6 Specifically, the return filters start at zero percent and increment in
steps of two percent, to a maximum (minimum) of positive (negative) ten percent for winner (loser) filters.
The low-volume return filters begin at zero percent and increment in steps of 15 percent to a minimum of
negative 75 percent. The high-volume return filters start out at zero percent and rise to a maximum of 250
percent in increments of fifty percent. The skewness in the volume return distribution produces the
asymmetry in the filters for volume.
For each combination of filter values, the securities whose lagged weekly returns (or returns and
volume) meet the filter constraints are formed into equally-weighted portfolios during week t. All portfolios
are held for a period of one week and then liquidated. The resulting portfolio’s mean return is calculated for
weeks in which non-zero positions are held. If mean returns of the portfolios are significantly different from
6
zero, this is taken as evidence in favor of return predictability. Thus, the null hypothesis of no predictability
is that the mean return of a portfolio equals zero.7
We use moment conditions to calculate test statistics for the mean returns since there may be some
dependence in the time series of portfolio returns, both contemporaneously and across periods. Specifically,
moment conditions are estimated with a generalized method of moments estimator (Hansen, 1982), and
Newey and West (1987) weights are employed on the variance/covariance matrix to compute the mean and
standard errors of the time series of trades for each portfolio and to perform comparisons between the
means of different strategies. Comparing the mean returns in a GMM framework has the advantage of
controlling for contemporaneous and time series correlations in the portfolio returns.
3. Data
To examine the interactions between lagged returns and volume, we construct a data set of Wednesdayclose to Wednesday-close weekly returns and weekly volume for 301 Real Estate Investment Trusts
(REITs) in the CRSP file between 1973 and 1995. Securities are included in the sample for week t if they
have daily volume in each of the previous ten trading days. Since the weights placed on individual securities
to form portfolios are based on non-market adjusted returns, the portfolio returns associated with our filterbased strategy should not emanate from index autocorrelation.8
Table 1 reports sample statistics for the data set. The mean market equity over the entire sample
period is 119 million dollars and the average share price is approximately 15.4 dollars. REITs with a share
price less than 5 dollars are screened out of the sample as a precaution against bid-ask bounce effects. The
cross-sectional average of individual security weekly autocorrelation coefficients is –7.07 percent at the
first lag and –2.77 percent at the second lag. The negative autocorrelation is consistent with overreaction
for individual stocks or it may indicate the existence of a bid-ask spread effect. For this reason we also
report the four day return’s (e.g., the skip-day returns) first-order autocorrelation of -3.24 percent. The
7
magnitude of this statistic strongly suggests that negative autocorrelation induced by the bid-ask spread is
not driving the negative autocorrelations exhibited in the full weekly returns.
In addition, Table 1 presents descriptive statistics for the volume return (percentage change in
volume) measure used with the filter strategies. The measure of volume we use, VRit as defined in
equation 3, is the average percentage change in weekly volume, and over the 1294 week sample period,
VRi,t averages 67.4 percent.
4. Strategies that condition on price and volume
The empirical analysis in this paper relies on the use of information from trading volume to augment a
simple, lagged price filter rule. In turn, we attempt to determine whether predictability in real estate returns
is related to volume and interpret our findings in the context of the return-volume dynamics of Wang
(1994). Theoretical and empirical works suggest there is valuable information in the time series of lagged
volume for predicting a security’s price movements (Blume, Easley and O’Hara (1994), Campbell,
Grossman and Wang (1993) and Conrad, Hameed, and Niden (1994)). The inclusion of volume, which
represents the trading activity of investors, as well as an analysis of the behavior of portfolio returns, may
also provide important information about the level of heterogeneity of investors in the real estate market.
By concurrently examining return behavior and the investors’ information sets, Wang (1994) argues that
one may more accurately identify those periods where the predictability of returns is attributable to the
information heterogeneity in asset markets.
Figure 1 illustrates the general pattern in weekly portfolio returns when portfolio construction is
conditioned on different values of lagged returns and lagged volume. We find that conditioning on negative
percentage changes in volume results in increased negative return-autocorrelations for the more extreme
winner and loser filters. In contrast, conditioning on positive changes in volume results in decreased
negative return-autocorrelations for the more extreme winners and losers. Thus, we observe an inverse
8
relation between volume and the autocorrelation of returns, which supports critical tenets of the Wang
model. Namely, that prices alone are not sufficient to resolve the identification problem of investor
heterogeneity.
Table 2 presents detailed results for the graphical relations shown in figure 1 and allows us to
focus on a major element of our results. Specifically, we find the behavior of portfolio returns in week t
reveals two distinct interactions between price and volume based upon the level of lagged volume. Low
volume portfolios (i.e., stocks with a negative percentage change in volume from week t-2 to week t-1)
experience considerably greater reversals (i.e., stocks with positive (negative) returns in week t-1
experience larger magnitude negative (positive) returns in week t) than the high volume portfolios (i.e.,
stocks with a positive percentage change in volume from week t-2 to week t-1). Additionally, portfolios at
the extreme price | low volume filters consistently have greater reversals than do the portfolios with
comparable price-only filters. Panel A of Table 2 shows that portfolios from the loser-price | low-volume
strategy yield larger weekly portfolio returns than those produced by the price-only strategies shown in the
“No volume filter” row. For example, the average weekly returns approach 3 to 4 percent when we jointly
condition on extreme losers and low volume. When interpreted in the context of Wang’s model, an
environment with a greater proportion of trades motivated by private investment opportunities will produce
the observed reversals in portfolio returns.
The returns from winner portfolios also experience greater reversals when volume is added to the
portfolio formation process. Panel C shows the results of the winner-price | low-volume strategy where a 0.376 percent (t-statistic = -1.676) weekly return from the “greater than 10%” filter for the price-only
strategy decreases to –3.330 percent (t-statistic = -3.21) when low-volume (<-75%) REITs are considered.
The return pattern associated with low volume is particularly evident at higher absolute magnitude price
filters in both winner (>10%) and loser (<-10%) portfolios. Overall, we observe large increases in the level
of reversals from incorporating volume information into the more extreme price filter rules. One
interpretation of this result is that transitory shifts in noninformational demand are more pronounced and
9
persistent when volume is low and returns are either very high or very low. This finding is consistent with
the argument that less active stocks are problematic not because there are too many informed traders, but
because there are too few uninformed ones (Easley, Kiefer, O’Hara, and Paperman, 1996). Downs and
Güner (1998) document a similar phenomenon among publicly traded real estate firms. Their paper shows
that the higher levels of information asymmetry contribute to a less-liquid, less-active REIT market,
perhaps due to the adverse selection problem faced by uninformed investors.
In contrast to the preceding discussion, conditioning on high volume lowers the magnitude of return
reversals, though the autocorrelation pattern remains strongly negative. The loser-price | high-volume results in Panel B of Table 2 reveal a trend of decreasing return reversals across the price filters for increasing levels of volume. For example, loser portfolios formed by the “-10 percent or less” price-only filter
generated weekly returns of 2.20 percent (t-statistic = 6.34). At the same price filter, but also conditioning
on a weekly volume filter of greater than 250 percent, weekly returns diminish to 1.47 percent (t-statistic =
1.87). The same pattern of decreased reversals in subsequent weekly portfolio returns is seen in Panel D of
Table 2 when a winner-price | high-volume strategy is used to form portfolios. For both losers and winners,
the increased return reversals found in portfolios that condition on low volume are more evident at higher
price filters.9 These results suggest that the information content during periods of high volume reflects a
market that is responding to private information trades as well as trades motivated by changing investment
opportunities.
The empirical results show that the magnitude of return predictability varies considerably between
high and low volume periods. These results, with different price-volume dynamics, are consistent with an
economic model in which the relative proportion of trades motivated by private information and by
heterogeneous investment opportunities will affect the behavior of expected returns. By interpreting these
test results in the context of Wang’s model, the periods with high volume contain a greater proportion of
private information which leads to less predictable reversals in portfolio returns; the reverse is observed in
periods of low trading volume.
10
4.1. Robustness
As a summary measure of the price-volume dynamic relation, we report the correlation of volume and
subsequent returns. This statistic allows us to formally test the relation between volume and future returns
and, as proposed by Wang (1994), to identify a dominant trading behavior in the speculative exchange of
real estate assets. Though the correlation analysis can not separate the data into high and low volume
periods, it measures the general price-volume relation that may help identify whether private information or
investment opportunities is, on average, the primary motivator of trading activity. For the entire sample
period, the correlation between the absolute value of weekly portfolio returns and the lagged volume filters
is negative and significant (-0.17 with a p-value of 0.038). This finding, interpreted in the context of
Wang’s model, supports the dominance of non-informationally motivated trading over informationally
motivated trading, which our earlier tests more clearly identify to be strongest during periods of low
volume.
We also consider the robustness of our results by drawing attention to the volume measure, VRit , used
in the filter-based method. Our findings show that the incorporation of volume improves the predictability
of returns, but in a manner not entirely consistent with a model in which investors trade because of their
differences. In other words, Wang (1994) emphasizes price changes accompanied by high volume when
identifying the type of information that produces price reversals or continuations. We find that large
absolute magnitude price changes (in week t-1) accompanied by high volume (e.g., the percentage change
in volume from week t-2 to week t-1 is positive) will reverse (in week t) but this pattern is stronger during
low volume periods. To ensure that our volume measure is not biasing the test results, we construct volume
measures using other time horizons. Specifically, we examine returns to strategies that condition on longer
horizon volume measures, so we construct two volume measures that employ an average of the last 4 and
20 weeks of volume to form trading shocks relative to longer term volume expectations. The substitute
volume measures are defined as:
11
m
V − (1 / m) ∑ V
it
i, t − j
j =1
VR
=
it , m
m
(1 / m) ∑ V
i, t − j
j =1
(4)
where m is equal to 4 or 20, the number of weeks used to form the volume average for security i in week t.
Similar to the calculations for one-week volume returns, weekly portfolio returns are calculated
using the four “price | volume” strategies (e.g., loser-price | low-volume). Subsequently, we construct the
summary measure of the price-volume dynamic relation as above. Recall this is the correlation of the absolute value of the weekly portfolio returns and these alternate volume measures.
The correlation coefficients between weekly portfolio returns and the lagged 4-week and 20-week
volume measures are both negative and significant at -0.16 (p=0.05) and -0.20 (p=0.02), respectively.
These results, with their negative relationships between volume and expected returns, support our earlier
test results which suggest that speculative trading in real estate assets is dominated by portfolio rebalancing
and not private information. However, this strict test precludes the dynamic nature of information
asymmetry and the existence of a weak dominance in informational trading over other motives. For this
reason, we turn to our final set of results.
4.2 Additional analysis conditioning on price and volume
To examine the association between expected returns and the return-volume dynamic, we run a crosssectional regression of the average of all week t portfolio returns (RET) on week t-1 returns (RET_LAG):
RET = 0.48 + -0.11 ⋅ RET_LAG
(8.79) (-13.64)
Adj. R2 = .56, N = 144.
t-statistics in parentheses
(5)
To the extent that RET measures the expectation of future returns conditioning on current returns
for each price-volume filter, the significant negative parameter estimate is consistent with earlier studies
that find return reversals for real estate securities (Mei and Gao, 1995). Additionally, we use information
12
within trading volume to resolve the identification problem associated with investor heterogeneity. We
conduct an alternative test where we regress the average of all week t portfolio returns (RET) on week t-1
returns with an emphasis on high volume periods by the use of an interactive dummy variable. The dummy
variable HI_VOL has a value of 1 for all filter levels where volume in week t-1 is greater than or equal to
50 percent over the prior week’s volume. HI_VOL assumes a value of 0 otherwise.
RET = 0.48 + -0.12 ⋅ RET_LAG + 0.04 ⋅ RET_LAG ⋅ HI_VOL
(8.79) (-12.31)
(2.61)
Adj. R2 = .58, N = 144.
t-statistics in parentheses
(6)
The coefficients on the RET_LAG and the interactive term are both significant at the 99 percent
level, but the point estimates have opposite signs.10 The test shows that an environment exists where low
(high) returns typically imply high (low) expected returns for real estate securities. Yet the positive
coefficient on the interactive term suggests that information in high trading volume periods dampens the
reversal effect that is found in periods with lower volume. This mitigating effect on return reversals is
consistent with the trading behavior of heterogeneously informed investors.
Finally, we examine the robustness of the price volume relationship across high and low vacancy
periods.11 This analysis allows us to assess the influence of the economic condition associated with the
underlying property market on the price-volume dynamic. As such, we construct an aggregate vacancy rate
measure for income-producing properties in the U.S. since 1980. The data are obtained from the 1997 U.S.
Bureau of the Census, Abstracts of the United States. A simple method to gauge the effect of occupancy
rates on the lagged return, lagged volume, subsequent reversal relationship is to form equally-weighted
portfolios of stocks in the bottom or top half of the price filters (i.e., week t-1 returns less than –6 percent
(loser-price) or greater than 6 percent (winner-price)) and the bottom half of the volume return filters (i.e.,
week t-1 volume returns less than 0 percent (low volume)). We name these portfolios loser-low and
winner-low, respectively, as they are formed by averaging the returns in the three most right columns of
Table 2, Panel A (loser-low) and the three most right columns of Table 2, Panel C (winner-low).
13
The pattern that emerges is one of greater reversals in high vacancy rate years relative to low
vacancy years for both the loser-low and winner-low portfolios. For example, the loser-low's average
weekly return is 3.85 percent (1.83 percent) in high (low) vacancy years (paired t-statisitic = 2.42). The
winner-low portfolio has average weekly returns of -1.44 percent in high vacancy periods versus returns of
-0.818 percent in the low vacancy periods (paired t-statistic = 0.95). Thus, the dampening of return
reversals during periods of low vacancy suggests that informed investors are trading on their private
information to extract what profits are available from public-market real estate. Just as our previous tests
found a relative increase in asymmetric information during high volume periods, the greater activity
occurring during strong real estate markets may provide opportunities for insiders to exploit their private
information more easily.
The greater return reversals observed during high vacancy years suggest there is a relative decrease
in asymmetric information during weak real estate markets. In the context of Wang’s model, the heightened
return reversals evident in high vacancy periods suggest that the trading of insiders in the real estate
securities market is driven by their private investment opportunities. In other words, the information
advantage of REIT insiders is less prominent in their trades than the need to rebalance portfolios in pursuit
of the private investment (e.g., vulture) opportunities often associated with a depressed real estate market.
5. Conclusion
Our study concludes that non-informational trading activity strictly dominates trading that is motivated by
the private information of informed investors. We arrive at this assessment by examining the predictability
of real estate returns in the context of the model in Wang (1994) where investor heterogeneity, in terms of
investment opportunities and information, leads to alternative specifications of a price-volume dynamic.
Our observance of strict dominance is based on a price-volume relation that exists across all periods and, in
this sense, is consistent with the Mei and Gao (1995) approach to documenting real estate market
overreaction. However, Damodaran and Liu (1993) provide compelling evidence that information
14
asymmetry, a principal determinant of the price-volume dynamic, changes across periods. Consequently,
our study attempts to reconcile some of the apparent contradictions in the real estate literature.
Our analysis of the predictability of real estate returns, conditioning on volume, demonstrates that
reversals are less pronounced during periods of high volume. This result is consistent with a weak-form
dominance of informationally motivated trading, which may explain why Mei and Gao do not find
economically significant price reversals. Most importantly, our findings contribute to the understanding of
the time-variation in the adverse selection problem of real estate investors.
Intuition suggests that a sophisticated investor with private information about publicly-traded real
estate assets might also have competing private-investment opportunities. Our research suggests that the
predictability in real estate returns is more a function of the rebalancing effects associated with the latter
endowment opportunities than the market corrections generated by the former asymmetric information (i.e.,
private information) opportunities. Importantly, our research also suggests that the risk of trading with a
more-informed investor is higher during periods of active trading as well as during periods when occupancy
rates are high.
15
Acknowledgements
This paper is an extended version of the second half of an earlier working paper with the same title. The
first half of the earlier working paper focuses specifically on trading strategies (Cooper, Downs and
Patterson, 1998). We wish to thank Crocker Liu, participants of the AREUEA and FMA meetings and two
anonymous reviewers for their helpful suggestions. We are especially grateful to Chinmoy Ghosh.
16
References
Alexander, S. (1961). “Price Movements in Speculative Markets: Trends or Random Walks,” Industrial
Management Review 2, 7-26.
Ball. R., S. Kothari, and C. Wasley. (1995). “Can We Implement Research on Stock Trading Rules?”
Journal of Portfolio Management 21 (Winter), 54-63
Blume, L., D. Easley, and M.O’Hara. (1994). “Market-Statistics and Technical Analysis: The Role of
Volume,” Journal of Finance 49, 153-181.
Bremer, M., and R. Sweeney. (1991). “The Reversal of Large Stock-Price Decreases,” Journal of Finance
46, 747-754.
Brown, K., and W. Harlow. (1988). “Market Overreaction: Magnitude and Intensity,” The Journal of
Portfolio Management 14 (Winter), 6-13.
Campbell, J.Y., S.J. Grossman and J. Wang. (1993). “Trading Volume and Serial Correlations in Stock
Returns,” The Quarterly Journal of Economics 108, 905-939.
Conrad, J., M. N. Gultekin, and G. Kaul. (1997). “Profitability and Riskiness of Contrarian
Portfolio Strategies,” Journal of Business and Economic Statistics 15, 379-386.
Conrad, J., A. Hameed, and C. M. Niden. (1994). “Volume and Autocovariances in Short-Horizon
Individual Security Returns,” Journal of Finance 49, 1305-1329.
Cooper, M. (1998). “The Use of Filter Rules and Volume in Uncovering Short-Term Overreaction,”
Review of Financial Studies, forthcoming.
Cooper, M., D.H. Downs, and G.A. Patterson. (1998). “Real Estate Securities and a Filter-based, Shortterm Trading Strategy,” Journal of Real Estate Research, forthcoming.
Corrado, C., and S. Lee. (1992). “Filter Rule Tests of the Economic Significance of Serial Dependencies
in Daily Stock Returns,” The Journal of Financial Research 15, 369-387.
Cox, D., and D. Peterson. (1994). “Stock Returns following Large One-Day Declines: Evidence on ShortTerm Reversals and Longer-Term Performance,” Journal of Finance 49, 255-267.
Damodaran, A., and C. Liu. (1993). “Insider Trading as a Signal of Private Information,” The Review of
Financial Studies 6, 79-119.
Downs, D.H., and Z.N. Güner. (1998). “Is the Information Deficiency in Real Estate Evident in Public
Market Trading?” Real Estate Economics, forthcoming.
Easley, D., N. Kiefer, M. O’Hara, and J. Paperman. (1996). “Liquidity, Information, and Infrequently
Traded Stocks,” Journal of Finance 51, 1405-1436.
Fabozzi F., C. Ma, W. Chittenden, and R. Pace. (1995). “Predicting Intraday Price Reversals,” The
Journal of Portfolio Management 21 (Winter), 42-53.
17
Fama, E.F., and M.E. Blume. (1966). “Filter Rules and Stock-Market Trading,” Journal of Business 39,
226-241.
Hansen, L. P. (1982). “Large Sample Properties of Generalized Method Of Moments Estimators,”
Econometrica 50, 1029-1054.
Keim, D. B., and A. Madhavan. (1997). “Transaction Costs and Investment Style: An Inter-exchange
Analysis of Institutional Equity Trades,” Journal of Financial Economics 46, 265-292.
Lakonishok, J., and T. Vermaelen. (1990). “Anomalous Price Behavior around Repurchase Tender
Offers,” Journal of Finance 45, 455-477.
Lehmann, B. (1990). “Fads, Martingales, and Market Efficiency,” Quarterly Journal of Economics 105,
1-28.
Ling, D., A. Naranjo, and M. Ryngaert. (1998). “The Predictability of Equity REIT Returns: Time
Variation and Economic Significance,” Journal of Real Estate Finance and Economics, forthcoming.
Ling, D., and M. Ryngaert. (1997). “Valuation Uncertainty, Institutional Involvement, and the
Underpricing of IPOs: The Case of REITs,” Journal of Financial Economics 43, 433-456.
Liu, C., and J. Mei. (1992). “Predictability of Returns on Equity REITs and Their Comovement with
Other Assets,” Journal of Real Estate Finance and Economics 5, 401-418.
Lo, A.W., and A.C. MacKinlay. (1990). “When are Contrarian Profits Due to Stock Market
Overreaction?,” Review of Financial Studies 3, 175-205.
Mei, J., and C. Liu. (1994). “The Predictability of Real Estate Returns and Market Timing,” Journal of
Real Estate Finance and Economics 8, 115-135.
Mei, J., and B. Gao. (1995). “Price Reversal, Transaction Costs, and Arbitrage Profits in the Real Estate
Securities Market,” Journal of Real Estate Finance and Economics 11, 153-165.
Newey, W.K., and K.D. West. (1987). “A Simple, Positive Semi-definite, Heteroskedasticity and
Autocorrelation Consistent Covariance Matrix”, Econometrica 55, 703-707.
Sims, C.A. (1984). “Martingale-Like Behavior of Prices and Interest Rates,” Working
paper, University of Minnesota.
Sweeney, R.J. (1986). “Beating the Foreign Exchange Market,” Journal of Finance 41, 163-182.
Sweeney, R.J. (1988). “Some New Filter Rule Tests: Methods and Results,” Journal of Financial and
Quantitative Analysis 23, 285-300.
Wang, J. (1994). “A Model of Competitive Stock Trading Volume,” Journal of Political Economy 102,
127-168.
Wang, K., S. Chan and G. Gau. (1992). “Initial Public Offerings of Equity Offerings: Anomalous
Evidence using REITs,” Journal of Financial Economics 31, 381-410.
18
Table 1. Summary statistics for the REIT sample (N=301), period from 1973 through 1995.
Mean
Median
Std. Dev.
Min.
Max.
ρ1 (s.d.)
5 day return (%)
0.267
0.0
3.914
-46.078
112.00
-7.073
(13.647)
4 day return (%)
0.205
0.0
3.532
-40.298
96.296
-3.236
(11.402)
VRi ,t (%)
67.363
-2.152
643.521
-100.00
1049.00
-17.810
(12.226)
Capitalization
($, millions)
119
57.2
168
0.018
1760
Price
($ per share)
15.430
13.75
8.947
5.00
132
Five day return is a Wednesday-to-Wednesday close weekly holding period return. Four day return is a “skipday” Wednesday-to-Tuesday close four day holding period return. REITs with prices less than $5 per share
are excluded from the sample. The mean, median, and standard deviation of capitalization and price are
calculated across time and across securities. The statistic ρ1 is the average first-order autocorrelation
coefficient of weekly returns of individual stocks. The population standard deviation is given in parentheses.
Since the autocorrelation coefficients are not cross-sectionally independent, the reported standard deviations
cannot be used to draw the usual inferences; they are presented as a measure of cross-sectional variation in
the autocorrelation coefficients.
19
Weekly Portfolio Returns
5%
4%
Percent return per week
3%
2%
1%
0%
<-10%
<-8 and >=-10
<-6 and >=-8
<-4 and >=-6
<-2 and >=-4
<0 and >=-2
>=0 and <2
Lagged weekly price
>=2 and <4
return (%)
>=4 and <6
>=6 and <8
>=8 and <10
-1%
-2%
-3%
<-75
<-60 and >=-75
<-45 and >=-60
<-15 and >=-30
<0 and >=-15
>= 10
<-30 and >=-45
Lagged weekly volume return (%)
None
>=0 and <50
>=50 and <100
>=100 and <150
>=200 and <250
>=150 and <200
>=250
-4%
Figure 1. Weekly portfolio returns conditioning on percentage changes in price and volume.
This figure presents the average weekly portfolio returns based upon a contrarian trading strategy using
price-volume filters. The greatest return reversals occur with the loser-price | low-volume portfolios in the
upper right quadrant. The upper left quadrant presents the performance of the loser-price | high-volume
portfolios. The lower left and right quadrants respectively reflect the average weekly returns of the winnerprice | high-volume and winner-price | low-volume portfolios.
20
Table 2. Weekly portfolio returns to price and volume strategies
_______________________________________________________________________________
Panel A: Loser-price | Low-volume
Lagged weekly return filter (%)
No
volume
filter
<0 and
≥ -15
<-15 and
≥ -30
Volume
Filter
(%)
<-30 and
≥ -45
<-45 and
≥ -60
<-60 and
≥ -75
<-75
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
<0 and
≥ -2
0.202
1.944
1279
2.677
0.157
2.645
833
1.567
0.201
2.958
878
1.874
0.225
3.003
879
2.004
0.265
2.900
834
2.687
0.158
2.608
713
1.690
0.014
2.790
682
0.127
<-2 and
≥ -4
0.561
2.519
1260
6.624
0.650
4.068
602
4.008
0.503
3.660
641
3.455
0.415
3.852
632
2.757
0.361
3.400
579
2.586
0.415
4.181
440
2.274
0.190
3.533
370
1.080
<-4 and
≥ -6
0.835
3.342
1097
7.843
0.911
4.369
285
3.892
0.866
3.874
273
3.065
0.449
4.892
332
1.796
0.860
4.304
279
3.425
0.858
5.393
203
2.188
1.156
4.514
164
2.937
<-6 and
≥ -8
1.212
4.236
786
7.579
1.251
5.426
111
2.727
1.426
4.684
121
3.234
0.955
4.726
109
2.402
1.342
5.275
121
3.453
1.106
5.303
86
2.011
2.118
6.590
73
2.800
<-8 and
≥ -10
1.354
5.499
496
5.271
0.889
5.394
50
0.890
2.188
7.734
49
1.696
2.271
7.166
48
2.494
-0.738
7.668
42
-0.531
1.200
4.582
33
1.206
1.451
6.923
29
1.411
<-10
2.200
7.008
477
6.338
3.476
5.764
41
3.953
1.240
7.727
44
1.840
4.655
10.640
38
2.570
3.039
9.187
42
2.433
3.292
7.596
19
1.558
3.784
10.699
18
3.085
21
Table 2. (continued)
_________________________________________________________________________________
Panel B: Loser-price | High-volume
Lagged weekly return filter (%)
No
volume
filter
≥ 0 and
<50
≥ 50 and
<100
Volume
Filter
(%)
≥ 100 and
<150
≥ 150 and
<200
≥ 200 and
<250
≥ 250
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
<0 and
≥ -2
0.202
1.944
1279
2.677
0.118
2.724
1095
1.262
0.398
2.911
834
3.827
0.382
3.012
604
2.684
0.436
2.701
401
3.777
0.169
3.218
316
1.012
0.461
3.309
621
3.355
<-2 and
≥ -4
0.561
2.519
1260
6.624
0.664
3.741
909
5.394
0.458
4.174
613
2.448
0.376
3.430
410
2.187
0.366
3.666
243
1.173
0.507
3.512
179
2.005
0.351
3.733
460
1.976
<-4 and
≥ -6
0.835
3.342
1097
7.843
0.823
4.709
523
3.672
0.615
3.926
329
2.651
0.943
5.413
206
2.564
1.024
6.504
142
1.692
1.359
4.446
94
3.363
0.811
4.108
274
3.548
<-6 and
≥ -8
1.212
4.236
786
7.579
1.100
4.724
243
3.182
1.413
4.515
152
3.835
1.744
4.155
106
4.668
1.499
6.423
65
2.496
1.225
4.881
44
0.910
1.182
4.993
169
2.746
<-8 and
≥ -10
1.354
5.499
496
5.271
0.706
5.499
113
1.309
2.353
8.555
87
2.238
0.420
5.198
56
1.063
0.933
7.696
45
0.804
2.102
6.567
23
1.555
1.475
5.452
99
2.902
<-10
2.200
7.008
477
6.338
1.892
7.887
96
2.871
1.659
6.493
89
2.582
1.345
7.841
54
1.355
0.289
6.726
50
0.057
1.505
5.611
32
2.144
1.470
7.941
148
1.869
22
Table 2. (continued)
_________________________________________________________________________________
Panel C: Winner-price | Low-volume
Lagged weekly return filter (%)
No
volume
filter
<0 and
≥ -15
<-15 and
≥ -30
Volume
Filter
(%)
<-30 and
≥ -45
<-45 and
≥ -60
<-60 and
≥ -75
<-75
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
≥ 0 and
<2
0.192
1.499
1278
3.597
0.221
2.537
865
2.659
0.334
2.960
894
3.123
0.266
2.616
879
2.922
0.179
2.706
876
1.964
0.114
2.575
777
1.057
0.205
2.954
780
1.887
≥ 2 and
<4
0.121
2.165
1256
1.665
0.216
3.626
606
1.274
0.302
3.688
649
2.057
0.087
3.203
606
0.660
0.305
3.483
555
1.981
0.037
3.756
442
0.068
-0.026
3.588
385
-0.235
≥ 4 and
<6
0.063
3.177
1151
0.643
0.207
3.826
336
0.994
0.149
4.077
313
0.679
-0.086
3.813
338
-0.335
-0.465
3.760
261
-1.682
-0.417
3.883
208
-1.383
-0.532
3.879
149
-1.611
≥ 6 and
<8
-0.086
4.003
925
-0.643
-0.301
4.635
154
-0.941
-0.841
4.371
171
-2.648
-0.029
4.107
165
0.024
0.726
6.019
119
1.209
-0.706
4.312
95
-1.420
-1.035
4.907
101
-1.807
≥ 8 and
<10
-0.021
4.612
662
-0.064
0.841
5.548
71
1.470
-1.238
4.647
59
-2.536
-0.169
4.840
69
-0.294
-0.675
4.357
73
-1.176
-1.876
5.956
34
-1.084
-0.978
5.326
44
-1.318
≥ 10
-0.376
5.687
748
-1.676
0.025
5.796
78
-0.130
0.529
7.124
89
0.542
-0.524
6.245
72
-0.764
0.457
8.986
67
0.374
-0.359
6.257
40
-0.890
-3.330
6.681
34
-3.208
23
Table 2. (continued)
______________________________________________________________________________
Panel D: Winner-price | High-volume
Lagged weekly return filter (%)
No
volume
filter
≥ 0 and
<50
≥ 50 and
<100
Volume
Filter
(%)
≥ 100 and
<150
≥ 150 and
<200
≥ 200 and
<250
≥ 250
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
Mean (%)
Stand. dev.
N
t-Statistic
≥ 0 and
<2
0.192
1.499
1278
3.597
0.183
2.529
1098
2.265
0.406
2.636
845
4.150
0.273
2.991
629
2.136
0.345
3.113
421
2.240
0.250
3.623
321
1.324
0.331
3.130
655
2.778
≥ 2 and
<4
0.121
2.165
1256
1.665
0.040
2.774
919
0.410
0.129
3.170
644
1.084
0.104
3.426
445
0.607
0.020
3.422
260
0.110
0.353
4.582
174
0.907
0.376
3.540
471
1.999
≥ 4 and
<6
0.063
3.177
1151
0.643
0.154
6.095
648
0.635
0.213
4.388
403
0.792
0.311
4.366
259
1.183
-0.496
3.453
164
-2.012
-0.431
4.027
98
-0.789
-0.156
3.933
321
-0.653
≥ 6 and
<8
-0.086
4.003
925
-0.643
-0.177
4.426
359
-0.807
0.273
5.045
238
0.820
-0.041
4.675
143
0.139
-0.439
3.795
80
-0.706
0.294
4.439
53
0.553
-0.148
4.445
182
-0.377
≥ 8 and
<10
-0.021
4.612
662
-0.064
0.386
6.198
180
1.215
-0.337
4.961
131
-0.732
0.075
4.264
87
0.406
0.303
6.059
47
0.562
-0.005
6.408
37
-0.025
-0.096
4.912
97
-0.267
≥ 10
-0.376
5.687
748
-1.676
-0.438
7.158
196
-1.154
-0.752
5.313
145
-1.889
-0.786
6.872
119
-0.966
-0.575
6.256
77
-0.728
0.375
5.879
52
0.105
-0.639
5.669
229
-1.666
Panels A, B, C, and D give the corresponding portfolio’s means, standard deviations, and t-statistics for a
mean equal to zero null hypothesis for the four joint price and volume strategies for weeks in which equity
positions are held. Securities are included in a given portfolio if the lagged weekly return and lagged volume
return (percentage changes in volume) meet the filter conditions for both lagged return and lagged volume.
Four price-volume strategies are examined: loser-price | low-volume, loser-price | high-volume, winner-price
| low-volume, and winner-price | high-volume in panels A, B, C, and D, respectively. A "No volume filter"
corresponds to a price-only strategy and is included for comparison purposes with the volume strategies.
The sample consists of REITs for the period from January 1973 to December 31, 1995. N is the number of
portfolio weeks the strategy traded at the respective price and volume filter levels out of a possible 1294
weeks. The t-statistics are robust to heteroskedasticity and autocorrelation.
24
Notes
1
In contrast to Mei and Gao (1995), Cooper, Downs and Patterson (1998) use a filter-based trading strategy and
find relatively strong evidence of short-term predictability for real estate securities. See also Liu and Mei (1992),
Mei and Liu (1994), and Ling, Naranjo and Ryngaert (1998) for evidence on the predictability of real estate returns
based on macro economic forecasting factors such as yield spreads, dividend yields, and capitalization rates on
equity REITS.
2
Whether a particular case of return predictability is attributable to market inefficiencies or time-varying risk
premia is often a contentious point, especially in longer horizon predictability. Lehmann (1990) and others suggest
that this disagreement may be resolved by examining the predictability of short-term (weekly) stock returns based
on the assumption that expected returns are not likely to change over a week. Specifically, Lehmann cites Sims
(1984) who hypothesizes that as time intervals shorten, prices should follow a random walk because there should
be few systematic changes in valuation over daily and weekly periods if information arrival is unpredictable. Thus,
we examine weekly return horizons. Furthermore, our study differs from previous research as we (1) employ a
filter-based portfolio construction methodology, (2) include volume as an additional forecast variable, and (3)
interpret our finding based on a theoretical framework in which the heterogeneity across investors gives rise to
different price-volume dynamics.
3
Wang (1995) derives a relation between current period returns (Rt) and volume (Vt), and expected returns (Rt+1),
E[ Rt +1 | Rt , Vt ] ≅ (φ 0 − φ1Vt 2 ) Rt . Here, φ0 and φ1 are constants and the sign of φ1 depends on the information
asymmetry between the two types of investors. Specifically, if φ1 < 0 then trading is dominated by informational
motives and φ1 > 0 indicates that trading is dominated by non-informational motives. An inherent feature of the
Wang model is the emphasis on trading motives of the informed (i.e., competitive trading to benefit from private
information or competitive trading to re-balance portfolios due to a shift in private investment technology.) For
this reason, the implications of the model tend to highlight high volume. One might expect the dynamic between
current and expected returns to be similar for low volume scenarios, although the magnitude of the reversal or
25
continuation may differ from the case where volume is high. See Wang (1995) for a complete discussion of the
implications.
4
Lehmann (1990) was the first to examine short-horizon reversals using a relative cross-sectional weighting
method. Mei and Gao apply a similar method to examine reversals in the real estate securities market. Several
papers subsequent to Lehmann provide alternative explanations for the profits found by employing cross-sectional
weighting methods. Lo and MacKinlay (1990), for example, show that up to 50% of Lehmann’s contrarian profits
are due to lagged forecastability across large and small securities. Other important citations include Ball, Kothari,
and Wasley (1995) and Conrad, Gultekin, and Kaul (1997).
5
The filter method may more closely correspond with the academic evidence on the psychology of overreaction.
Related studies (see DeBondt (1989) for a review) show that individuals tend to overreact to a greater degree when
confronted with a large information shock relative to their prior base-rate expectations. This realization leads
DeBondt and Thaler (1985) to postulate an overreaction hypothesis that states; “(1) Extreme movements in stock
prices will be followed by extreme movements in the opposite direction; (2) The more extreme the initial
movement, the greater will be the subsequent adjustment.” This hypothesized predictable behavior, manifested in
extreme price movements, forms the basis of the filter rules. In these rules, a security is included in a portfolio only
if its lagged return is within the filter level. Thus, by employing filters on lagged returns, we are able to screen
stocks for “large” past price movements which may likely be investor overreaction, and subsequently eliminate
securities that experienced smaller lagged returns (or those that may be noise to a contrarian strategy). Cooper
(1998) examines large capitalized NYSE and AMEX stocks and finds that weekly contrarian strategies based on
filter rules generally earn greater weekly profits than do portfolios formed from relative cross-sectional weighting
rules.
6
The price and volume filter breakpoints are determined by using each variable’s overall sample distribution
percentiles of approximately 1, 2.5, 5, 10, 25, 50, 75, 90, 95, 97.5, and 99 percent. As with all filter-based
methods, the primary goal in setting the breakpoints is to generate maximum dispersion in the return and volume
26
distributions. As such, the filter breakpoints are chosen to span the distribution of return and volume conditioning
variables and, therefore, are independent of the results.
7
We follow the practice of other short-horizon contrarian papers and report mean equal to zero t-statistics. We
also calculate t-statistics (not reported in the paper) by subtracting the unconditional weekly mean return of the
sample from the return of each filter portfolio and find that this measure of excess returns produces little variation
in the reported t-statistics.
8
Our method employs weights conditioning on raw returns. As such, the profits from the filter strategies will be
based upon individual security autocovariances and individual security unconditional mean weekly returns. This is
an important point as Conrad, Gultekin and Kaul (1997) and Lo and MacKinlay (1990), using a profit
decomposition originally derived in Lehmann (1990), show that contrarian strategies that base their weights on a
security’s deviation from an equally-weighted index of those securities result in a large percentage of profits
attributable to positive autocovariances of the returns of an equally-weighted portfolio of the component assets.
As we will report, the average weekly unconditional return of the REIT sample is 0.267 percent. This
mean return is relatively small compared to the magnitude of the profits from many of this paper’s filters
strategies, suggesting that the primary source of predictability is individual security autocovariance.
9
Determining whether there are “arbitrage” opportunities in the publicly-traded real estate markets is an
interesting topic which takes us away from our main objective of studying the characteristics of predictability in the
context of informational and non-informational motives for trading real estate. However, a casual observation of
the more extreme filters suggests the possibility for profitable trades. Keim and Madhavan (1997) report roundtrip total execution costs of 0.96 percent (price impact, bid-ask spreads, and commission costs) calculated from
actual trades placed by 21 institutional investors on the smallest quintile of NYSE securities over the 1991 to 1993
period for medium sized trades. The issue of trading costs is more fully explored in Mei and Gao (1995) and
Cooper, Downs and Patterson (1998).
27
10
We investigate alternative specifications of the relation between expected returns and returns, which include
omitting the 36 extreme portfolios (i.e., price filters < -10%, >=10%, and volume filters < -75%, >= 250%). The
results of this regression test do not change.
11
Liu and Mei (1992) show that REITs are a hybrid security in that returns are influenced not only by stock market
conditions but by conditions in the property markets, as well. Wang, Chan, and Gau (1992), and Ling and
Ryngaert (1997) find evidence supporting the influence of the changing real estate market environment on real
estate returns. We are indebted to Crocker Liu for suggesting this test.
28