momentum in housing - uw · pdf filebeen shown to contribute to stock market momentum are ......
TRANSCRIPT
Momentum in Residential Real Estate
Eli Beracha and Hilla Skiba
Abstract: This paper examines whether there is return momentum in residential real
estate in the U.S. Case and Shiller (1989) document evidence of positive return
correlation in four U.S. cities. Similar to Jegadeesh and Titman’s (1993) stock market
momentum paper, we construct long-short zero cost investment portfolios from more
than 380 metropolitan areas based on their lagged returns. Our results show that
momentum of returns in the U.S. residential housing is statistically significant and
economically meaningful during our 1983 to 2008 sample period. On average, zero cost
investment portfolios that buy past winning housing markets and short sell past losing
markets earn up to 8.92% annually. Our results are robust to different sub-periods and
more pronounced in the Northeast and West regions. While zero cost portfolios of
residential real estate indices is not a tradable strategy, the implications of our results can
be useful for builders, potential home owners, mortgage originators and traders of real
estate options.
Keywords: Momentum, Residential Real Estate, Predictable Returns, Zero Cost Portfolios
Eli Beracha: Department of Finance, College of Business, Bate 3129, East Carolina University, Greenville, NC
27858; Email: [email protected]; Tel: 252-328-5824.
Hilla Skiba (corresponding author): Department of Economics and Finance 3985, 1000 E. University Ave.,
University of Wyoming, Laramie, WY 82071; Email: [email protected]; Tel: 307-766-4199.
1
1. Introduction
Since the early days of the finance literature on market efficiency, there has been
evidence suggesting that buying past winners may generate positive abnormal returns (Levy
(1967)). Following these findings, many finance practitioners have taken advantage of buying and
selling stocks based on their momentum or their relative lagged strengths. As a response to a vast
academic literature on contrarian investment strategies from the 1980s, Jegadeesh and Titman
(1993) show that zero cost portfolio strategies that buy stocks that performed well and short sell
stocks that performed poorly in prior periods generate significant positive returns over 3- to 12-
month holding periods, independent of the systematic risk of the portfolios.
Momentum in stock returns is related to many factors. Jegadeesh and Titman (2001) find
that illiquid stocks in the U.S. market generate higher momentum profits. Zhang (2006)
documents a stronger momentum in stocks for which information asymmetry is higher and
returns are more volatile. A significant relationship between turnover and momentum is
documented by Lee and Swaminathan (2000). Ali and Trombley (2006) show that momentum is
positively related to short sale constraints. In a behavioral based study, Daniel, Hirschleifer, and
Subrahmanyam (1998) show that investors’ overconfidence and self-attribution biases are
positively related to momentum in stocks. Similarly, Chui, Titman and Wei (2008) document a
positive relationship between cross-country momentum and country specific overconfidence
proxied by cultural individualism. In their cross-country study, Chui, Titman, and Wei also find
that in addition to overconfidence, factors that are significantly and positively related to
momentum include uncertainty and transaction cost.
The literature on the efficiency of the housing market dates back less than 25 years to
studies by Hamilton and Schwab (1985) and Linneman (1986). However, many factors that have
been shown to contribute to stock market momentum are characteristics of the housing markets as
well. Several real estate papers that study the degree of real estate market efficiency point out
2
imperfections that are present in the housing markets. Specifically, these characteristics include:
Information asymmetry due to high information cost, high transaction cost, absence of short
selling, and infrequent trading (see for example Gau (1984), (1987), Atteberry and Rutherford
(1993), Fu and Ng (2001)). Case and Shiller (1989) point out that the dominance of individuals
with consumption rather than investment view in regard to housing and the lack of professional
traders in the market contribute to its inefficiency. In the absence of large sophisticated investors
and in the presence of large transactions costs, it is possible for housing prices to deviate from
fundamentals throughout time. Case and Shiller (1989) are also the first ones to document
momentum and predictability in housing returns. Using a sample of four U.S. metropolitan areas
from 1973-1986, they confirm that last period’s returns predict future price movements in
housing prices.
In this paper we use more than 380 metropolitan statistical areas (MSA) in the U.S. to
examine whether momentum in residential real estate exists during the 1983 to 2008 sample
period. Our results show long and statistically significant momentum effect in the U.S. housing
markets. Using an autoregressive (AR) model we find that area-specific real estate return at time t
is related to returns earned in previous quarters. Particularly, the models suggest that quarter t-1
has mostly a slight mean reverting effect on the return real estate market experiences during
quarter t, while the return during the four quarters spanning t-5 to t-2 positively correlates with
current return. This means that MSAs experiencing returns above or below the U.S. housing
average during quarters t-5 through t-2 are likely to earn returns above or below the U.S. average
during quarter t, respectively. The positive effect of quarters t-5 to t-2 is especially strong in the
early and late part of our sample. As a robustness check, we also employ a dynamic panel
generalized method of moments estimation to the sample and its sub-periods, and confirm the
initial finding.
In order to gain more insight to the positive correlation between current and lagged
returns and to the economic significance of momentum in real estate, we employ long-short
3
portfolio strategies. Similar to Jegadeesh and Titman’s (1993) analysis on momentum of stocks,
we examine the extent to which long-short portfolio strategies on housing MSA indices produce
positive abnormal returns. Specifically, we construct zero cost portfolios that buy housing indices
of MSAs that performed better than average in the past quarter(s) and short sell housing indices
of MSAs that performed worse than average in the past quarter(s). Our results show that zero
cost portfolios that are based on one to four quarters of MSA housing performance and held for
one to four quarters before rebalancing, earn up to 8.92% on an annual basis during the 25-year
sample period. The magnitude of the returns on the zero cost portfolios is especially impressive
given that a traditional buy-and-hold strategy of the comprehensive U.S. housing index returns
only 4.69% on an annual basis during the same time period.
As robustness checks, we test the same long-short portfolio strategy on five separate sub-
periods and on four broad geographic regions. Overall, we find that the momentum effect is
robust to different time and region specifications. However, the momentum effect is especially
pronounced in the West region and during the 2004 to 2008 period.
Obviously, constructing long-short portfolios of houses is not necessarily a tradable
strategy in itself. Nevertheless, our results provide timing insight to potential home buyers,
builders1, and mortgage lenders
2, as well as to traders of housing derivative contracts that are now
available on the Chicago Mercantile Exchange (CME). Specifically, the results presented in this
paper shed light on the magnitude of the momentum effect in the U.S. housing markets. Our
results show that greater magnitude of momentum is associated with more extreme previous
periods’ winners and losers and when momentum is based on a longer period of past
performance.
1 For example, a builder may alter the decision of when and where to build based on area-specific
momentum information and the projected delivery time of the structure. Similarly, potential home buyers
may choose to delay their purchase if they have sufficient information that negative momentum exists in
their area. 2 Mortgage lenders can use momentum information on housing to better estimate the future value of their
collateral.
4
The rest of the paper is organized in the following way. Section 2 develops hypotheses
while reviewing literature on housing and stock market momentum. Section 3 shows the data and
methodology. Section 4 reviews the main results and Section 5 concludes.
2. Momentum and Testable Hypotheses
2.1 Momentum in Stock Prices
Momentum in stock prices is a well-established phenomenon since seminal work by Levy
(1967), who documents profitability of buying stocks with relative strength. In a more recent
paper, Jegadeesh and Titman (1993) show that a portfolio strategy that buys stocks that have
performed well and shorts stocks that have performed poorly in the past period(s) generates
significant positive returns over the 3- to 12-month holding periods, independent of the
systematic risk of the portfolios. Part of the abnormal return generated during the first year,
however, dissipates in the two years that follow the first year. Jegadeesh and Titman (2001)
confirm their general momentum finding with out of sample data. Also, Chan, Jegadeesh, and
Lakonishok (1996) and Rouwenhorst (1997) show that buying of past winners in the U.S. and
foreign stock markets is a profitable strategy. The opposing stream of finance literature focuses
on contrarian strategies. Among others, De Bondt and Thaler (1985) and Lakonishok, Shleifer,
and Vishny (1994) document positive profits from buying past losers. Overall, there is evidence
of profitability in both momentum and contrarian strategies in the stock market, so that in the
short run traders profit from momentum and in the long run they profit from return reversals.
In a recent stream of finance literature momentum in stocks has been linked to volatility
and idiosyncratic risk. Ang, Hodrick, Xing, and Zhang (2006) find that portfolios of stocks
formed based on their idiosyncratic volatility have lower average returns. Lee and Swaminathan
5
(2000) provide a link between trading volume and price momentum. Firms with high past
turnover earn lower future returns and the past trading volume predicts the magnitude and the
persistence of price momentum.
In a comprehensive cross-country study on country momentum, Titman, Chui, and Wei
(2008) show that behavioral overconfidence, also shown by Daniel, Hirschleifer, and
Subrahmanyam (1998), partially explains country specific momentum along with uncertainty,
volatility, and transaction costs.
Momentum in stocks has also been linked to momentum in mutual funds. Hendricks,
Patel, and Zeckhauser (1993) and Goetzmann and Ibbotson (1994), among others, document
persistent performance in mutual funds from one period to another. Although Carhart (1997)
shows that a strategy that buys the top decile of mutual funds based on their last year’s
performance and shorts the bottom decile of mutual funds yields an 8% return, momentum in
stocks as documented by Jegadeesh and Titman (1993, 2001) help explain part of this abnormal
return along with other common risk factors in stock returns.
2.2 Momentum in Housing
Case and Shiller (1989) show that housing prices in the U.S. do not appear efficient. By
studying resale housing data from 1970 to 1986 from Atlanta, Chicago, Dallas, and San
Francisco, Case and Shiller conclude that there is persistence through time in the change of
housing prices, and that quarterly abnormal returns in housing prices are predictable based on last
period’s changes. Their finding implies that there is a profitable trading rule in the housing
markets for those buyers who can time the purchase of their homes.
Case and Shiller (1990) document forecastability of prices and excess returns in Atlanta,
Chicago, Dallas, and San Francisco, adding support to their earlier argument about inefficiency in
the market of single-family homes. Case and Shiller show that owner-occupied home prices in
6
these four markets have a tendency to change, for more than one year, in the same direction as
they did in the previous quarter. The authors also provide evidence suggesting that the ratio of
construction to price, changes in adult population, and increases in income per capita affect the
excess returns and price changes over the subsequent year. Additionally, Case and Shiller
document weak evidence of opposite negative relationship between returns and appreciation
lagged by more than one quarter.
In a more recent paper on forecastability of housing prices, Gupta and Miller (2008)
show that home prices are predictable for nearby metropolitan areas. The authors show that
appreciation of Los Angeles housing prices causes appreciation in housing prices in Las Vegas
directly and appreciation in Las Vegas housing prices causes appreciation in Phoenix housing
prices directly (appreciation of housing in Los Angeles causes Phoenix home-price appreciation
indirectly). Los Angeles housing prices are shown to be exogenous in Gupta and Miller’s study.
In a related work to Case and Shiller (1989, 1990) by Abraham and Hendershott (1993),
the authors illustrate that lagged returns explain twice the return in housing markets in volatile
coastal cities relative to the inland cities. Abraham and Hendershott (1996) show that bubbles
tend to form in housing markets in areas where lagged returns have higher explanatory power,
and that the downward swings are also greater in those areas after housing bubbles burst. This is
true especially in the Northeast and California. In the Midwest, however, where the lag return is
positive but smaller in magnitude compared to the coastal areas, the period following the bubble
shows only moderate price drops. According to their model, determinants of real housing price
appreciation can be divided into two groups. The first group includes variables that affect changes
in equilibrium prices such as growth in real income, real construction costs, and changes in real
after-tax interest rate and the second group is the adjustment dynamics that include the lagged
real appreciation and deviation from equilibrium prices.
The momentum effect is also shown to be significant in Real Estate Investment Trusts
(REITs). Chui, Titman, and Wei (2003) find significant momentum in the U.S. REITs from 1983
7
to 1999. Similarly, Brounen (2008) finds strong support for performance persistence in REITs.
Hung and Glascock (2008) document momentum in REITs and show that momentum is higher
during the up markets and for those REITs with higher dividend/price ratios. Hung and Glascock
(2008) also show that momentum is positively related to volatility. The authors find that REITs
with the lowest past returns have higher idiosyncratic risks than REITs with the highest past
returns. They conclude that idiosyncratic risks can partially explain momentum.
While many aspects of real estate momentum have been explored, the existence and the
economic significance of a broad based real estate momentum trend in the U.S. is, to our
knowledge, yet to be investigated. Our paper aims to fulfill this gap in the literature.
2.3 Hypothesis Development
The market for real estate is characterized by many of the same factors that are
significantly related to momentum in stock prices. These factors, or market imperfections, that
have been discussed mainly in real estate market efficiency studies include: Low liquidity, high
transaction cost, limited information that leads to more uncertainty, lack of professional traders,
short sale constraints, and uniqueness of properties that makes valuation more difficult (Gau
(1984), (1987), Atteberry and Rutherford (1993), Fu and Ng (2001), and Case and Shiller (1989),
Shiller (2007) among others). Of these market imperfections, transaction costs, liquidity,
uncertainty, and short sale constraints are all positively related to momentum in stocks (Chui,
Titman, and Wei (2008), Jegadeesh and Titman (2001), Zhang (2006), Ali and Trombley (2006)).
In addition to the factors mentioned above, behavioral characteristics of investors can
cause momentum in stocks. Daniel, Hirschleifer, and Subrahmanyam’s (1998) model relates
investors’ overconfidence and self-attribution biases positively to momentum in stocks. Chui,
Titman, and Wei (2008) also find that cross-country overconfidence is positively related to
momentum.
8
It is reasonable to assume that overconfidence is also present in the housing market.
Buyers and sellers in the market for residential housing consist largely of individuals who have
limited and infrequent experience with housing transactions. Shiller (2007, 2008) asserts that a
significant factor in the recent housing boom propelled by the notion that a house is a great
investment. According to Shiller, a psychological feedback mechanism helped spread that notion
and caused home prices to reach inflated levels. Shiller argues that fundamentals could not
explain housing prices during the boom and offers the explanation of “social epidemic of
optimism”. This optimism, speculative psychology, and investors’ overconfidence fuel positive
momentum and housing prices which in turn lead to sharper declines.
Additionally, sophisticated large investors are practically absent in housing markets.
Gervais and Odean (2001) show that experience is an important determinant of overconfidence,
so that those investors who have been trading for the shortest periods have the greatest levels of
overconfidence, and with more experience investors are better able to understand their own
abilities. Investors with limited experience base the view of their own ability mainly on prior
performance. As a result, positive returns in the past fuel overconfidence the most among less
experienced investors. Similarly, the authors show that individual stock market investors are more
overconfident in their abilities compared to institutional investors. In a related experimental work
by Bloomfield, Libby, and Nelson (1999) the authors find that less informed investors are more
overconfident compared to informed investors.
Based on the studies mentioned above, it is likely that the average buyers in the
residential housing market are, on average, overconfident in their abilities. This characteristic
should contribute to price momentum in housing because when buyers and sellers with limited
experience happened to be engaged in a housing transaction that proved to be lucrative, they are
likely to be willing to pay premiums on their consecutive homes.
Piazzesi and Schneider (2009) provide additional support to our overconfidence argument
and its relation to momentum. The authors find from survey evidence that during the latest boom
9
in the housing markets, there was always a group of buyers who thought that prices would further
increase. The size of this “momentum cluster” increased as the prices of homes increased. In
addition, Piazzesi and Schneider find that even a small number of optimistic investors can have a
large positive effect on housing prices.
To conclude, factors that cause momentum in stocks are similar in nature to
characteristics of housing markets. For this reason, we expect to find momentum in the housing
markets as well. Case and Shiller (1989, 1991) show that there is a positive and significant
relationship between lagged and future returns in four major markets in the U.S. Our first goal is
to examine the existence and persistence of housing momentum in the U.S. in large scale3. We
extend Case and Shiller’s (1989, 1991) sample of 4 cities to more than 380 MSAs and cover a 25-
year period (from 1983 to 2008). Our first hypothesis then becomes:
H1: There is positive price appreciation momentum in the housing markets in the U.S.
If momentum is present in any financial market, future returns become predictable. This
means that above average positive past performance will be followed by above average positive
future performance for some time after the measuring period. Similarly, lower than average past
returns will be followed by lower than average returns in the future periods. This may allow for a
hypothetic tradable strategy of buying past winners and selling past losers.
Our goal is to test if there are opportunities for positive abnormal returns with zero
investment portfolios. Jegadeesh and Titman (1993) show that zero investment portfolios of U.S.
stocks that short sell past losers and buy past winners earn significant positive abnormal returns.
These returns are economically significant and persistent. Like Jegadeesh and Titman, we
hypothesize that portfolios of U.S. housing indices formed based on their past performance earn
abnormal positive returns. More formally:
3 See data section for more detail about the FHFA housing indices.
10
H2: Zero cost buy-sell portfolios formed based on housing markets’ past performance earn
economically significant positive returns in the following holding periods.
3. Data, Methodology, and Robustness Checks
3.1 Data
We obtain Housing Price Indices (HPI) for more than 380 MSAs from the Federal
Housing Finance Agency (FHFA), which was formed in 2008 partially from the Office of Federal
Housing Enterprise Oversight (OFHEO)4. The indices include quarterly observations from which
we derive quarterly housing price changes. Our sample covers the period between the first quarter
of 1983 and the third quarter of 2008, a total of 103 quarters. The HPIs are based on modification
of Case and Shiller’s (1989) weighted repeated sales methodology. The FHFA estimates each
HPI using only repeated sales or refinancings of single-family residential properties financed
through a conforming loan. FHFA defines a repeated sale when the same physical address
originates at least two mortgages and those mortgages are purchased by either Freddie Mac or
Fannie Mae. The use of repeated sales of the same physical address controls for properties’
characteristics and reduces the effect of changes in construction quality over time on housing
prices.
The FHFA data are broad in coverage. In 1983 the indices are available for 181 of the
U.S.’s MSAs and gradually increase to cover 381 MSAs in 2008 (See Table 1). One limitation of
the indices is that only conforming conventional loans and sales of single-family detached
properties are included. However, inclusion of only the conforming conventional loans makes
4 The OFHEO index data are available at http://www.fhfa.gov.
11
HPIs less sensitive to outliers induced by subprime mortgages and other risky mortgage
financings.5
3.2. Methodology
To test the momentum in returns of the MSAs from 1983 to 2008, we first run a basic
autoregression, where the dependent variable is the quarterly return on each MSA’s HPI net of
the return on the broad U.S. housing. The independent variables are the lagged returns on each
MSA housing market net of the return on the broad U.S. housing market for periods t-1 through t-
n. More formally:
itUSAtiMSAtUSAtiMSAtUSAtiMSA RRRRRR ...)()( 2,2,,21,1,,1,,, (1)
Where RMSAi,t is the quarterly return on an MSA i in time period t and RUSA,t is the quarterly return
on the US comprehensive housing index in time period t. The lags in our study are quarterly lags.
A positive and statistically significant coefficient on any of the independent variables
would provide evidence of momentum. Specifically, a positive 2 coefficient, for example,
would suggest that housing markets in MSAs that outperform the broad U.S. housing market in
period t-2 will, on average, outperform the broad U.S. housing market during period t. The same
positive coefficient would also imply that housing markets in MSAs that underperform the broad
US housing market in period t-2 will, on average, underperform the broad U.S. housing market
during period t. Therefore, positive coefficients in equation (1) would support our first
hypothesis.
5 For more detail about the index construction see Calhoun (1996) and OFHEO’s website at
http://www.fhfa.gov
12
We begin our autoregressive analysis with one independent factor, for time t-1. In the
case that we find the coefficient for this factor to be positive and statistically significant we run
the autoregressive model with two factors, for times t-1 and t-2. We continue to add factors until
the coefficient on the added factor is no longer positive. By gradually adding factors to the
autoregressive model we are able to determine not only if momentum in returns on housing
exists, but also how long, on average, the momentum lasts. We also use the Akaike Information
Criterion (AIC) to determine which specification of equation (1) is the most appropriate.
In order to test the validity of our second hypothesis and to measure the economic
significance of return momentum in residential real estate, we construct zero cost buy-sell
portfolios as employed on stocks by Jegadeesh and Titman (1993). We form relative strength,
zero investment portfolios of the MSAs’ HPIs based on their J- quarter lagged returns and hold
them for K quarters, after which the portfolios are rebalanced. In the beginning of every holding
period, the FHFA’s MSA indices are ranked based on their J-quarter lagged returns. The lowest
past return portfolio is the sell portfolio and the highest past return portfolio is the buy portfolio.
The portfolios’ returns are the annualized equally weighted returns over the K-quarter holding
periods of the buy and sell portfolios. The buy portfolio is then:
,
1n K
J
N
W MSA
n p
R RN
(2)
Where RW is the annualized return on a long past winners portfolio that buys the top pth percentile
of MSA indices based on their J period lagged return and holds the winner portfolio for K-
quarters. Under different scenarios we let p take the value of 15th, 30
th, and 50
th percentile.
Similarly the sell portfolio is:
13
,
1n K
J
N
L MSA
n p
R RN
(3)
Where RL is the annualized return on a short past losers portfolio that sells the bottom pth
percentile of MSA indices based on their J-period lagged return and holds the losers for K
quarters. Here again, p takes a values of 15th, 30
th, and 50
th percentile under different scenarios.
Combining equations (2) and (3), we form the buy-sell zero investment portfolio:
Z W LR R R (4)
Where RZ is the abnormal return on a zero cost portfolio that buys the top pth percentile of MSA
indices based on their J-period lagged return and sells the bottom pth percentile of MSA indices
based on their J-period lagged return and holds the portfolios for K quarters. If there is no
momentum in returns on MSA indices, the abnormal return RZ from equation 4 will not be
positive and not significantly different from zero.
3.3. Robustness Checks
Because of the dynamic panel structure of our data, standard panel estimation techniques
may not be appropriate to estimate the lagged coefficients of equation (1) and cause them to be
biased. For this reason, as a robustness check, we employ Arellano-Bond’s (1991) dynamic panel
estimator to confirm our AR analysis. In order to do so, we calculate the first difference of
equation (1) which removes the independent and identically distributed random effects from the
panel. Then, the coefficients of the lags are estimated via Arellano-Bond generalized method of
moments estimator using lagged levels of the dependent variable as instruments in the estimation.
14
Additionally, we repeat the dynamic GMM analysis for five consecutive 5-year periods of our
data to detect if our results are triggered by a time-specific event.
It is possible that the results we observe using the long-short portfolio strategy on the full
sample are driven by a particular geographic region or a specific time period. In order to confirm
that our results are robust to different time specifications we employ the long-short portfolio
strategy on five consecutive 5-year periods. Similarly, we divide our sample to four geographic
regions to confirm that our results are not driven by a particular area. Our geographic four-region
classification is based on the Census Regions and Division of the United States, which divides the
U.S. into Northeast, Midwest, South, and West6. If we observe momentum during all sub-periods
and in all geographic areas, we can conclude that our results are broad and do not exist only
during a unique time period or in a specific geographical region.
4. Results
4.1. Momentum in Housing Prices
Table 1 reports selected summary statistics for the MSA indices in our sample. The first
two rows show the number of MSAs in the beginning and end of the sample period, while the
remainder of the table provides information on average and extreme returns of the data along with
standard deviations. The average annual appreciation of the U.S. housing prices during the time
period is 4.69%. As a comparison, the top 15th, 30
th, and 50
th percentile portfolios appreciated at
6.39%, 6.01%, and 5.46% respectively on an annual basis, and the bottom 15th, 30
th, and 50
th
percentile portfolios appreciated at 2.67%, 3.11%, and 3.46% respectively on an annual basis
during the same time period.
6 www.census.gov/geo/www/us_regdiv.pdf provides detailed region classification at the state level.
15
Next, we test for the return persistence in the MSA indices using the autoregressive
analysis from equation (1). The AR results are reported in table 2 for up to seven quarterly lags.
The explanatory power of the lagged returns is positive and significant for all lags between two
and five quarters. Surprisingly, the first period’s lag gradually turns negative and significant when
we increase the number of lags in the regression. The coefficients of the lags between two and
five periods are large in magnitude, statistically significant, and much larger than the first
periods’ negative coefficients or coefficients of the lags greater than five quarters. The second,
third and fourth lag seem to have the largest explanatory power. The sixth lag is also positive, but
small in magnitude and not statistically significant7.
Overall, the results reported in table 2 provide support to hypothesis one. The results
suggest that areas with above (below) U.S. average return during quarters t-2 through t-5 are
likely to outperform (underperform) the average return on residential real estate in the U.S. during
quarter t. The results also hint of a slight mean-reverting effect that is taking place during quarter
t-1. To reaffirm the momentum evidence presented in tables 2, and to provide economic meaning
to the momentum effect, we present the long-short portfolio strategy results in table 3.
4.2. Momentum Using Long-Short Strategy
Table 3’s panels A, B, and C show the annualized returns on long portfolios using
equation (2), short portfolios using equation (3), and zero cost long-short portfolios derived from
equation (4). The zero cost portfolios are formed based on lagged returns from one to four
quarters and are held for one to four quarters before rebalancing. The sample consists of quarterly
returns on the MSAs’ indices from the first quarter of 1983 to the third quarter of 2008.
7 The specification suggested by the AIC is an AR model with five lags.
16
In panel A, the long portfolio buys housing indices that are lagged J-quarters’ winners, or
in the top 15% performers of the U.S. MSAs available to us at that time. Similarly, the short
portfolio sells housing indices that were lagged J-quarters’ losers, or in the bottom 15%
performers of the U.S. MSAs of the sample. The long portfolios’ average annualized return varies
from 6.15% to 9.90%. The highest long return (9.90%) is generated by the portfolio that buys
winners based on four quarters returns and holds the index for two quarters. The least profitable
long portfolio (6.15%) buys winners based on one quarter’s returns and holds the indices for one
quarter.
Short portfolios generate positive annualized returns in all base-holding period
combinations. However, these returns are always lower compared to their long portfolio
counterparts. The short portfolios’ returns range from 0.82% to 4.68%. The lowest short return
(0.82%) is generated by the portfolio that sells losers based on past four quarters’ returns and
holds the indices for three quarters. The highest short portfolio return (4.68%) is in the portfolio
that shorts losers based on their one quarter return and holds them for one quarter.
Returns on zero cost long-short portfolios that are constructed based on momentum
strategy range from 1.34% to 8.92% on an annual basis. From the combinations tested, the
strategy that buys and shorts housing indices based on one quarter’s return and holds them for one
quarter performs the worst, while the combination that buys and shorts housing indices based on
four quarters’ return and holds them for three quarters performs the best. Overall, panel A’s
results suggest that when holding the base period constant, the abnormal returns increase with the
holding period from one to two to three quarters, but are slightly lower for four quarter holding
periods compare to three quarter holding periods. Also, while holding the holding period
constant, the abnormal return increases with a longer base period. The positive, consistent, and
economically significant abnormal returns generated by the momentum based zero cost portfolios
supports our hypothesis 2.
17
In panel B the analysis is repeated from panel A, except that instead of the 15% cutoffs
for winners and losers, we use the top 30% and bottom 30% of the MSA indices to construct the
buy and sell portfolios. The long portfolios’ average annualized returns range from 5.64% to
8.12%. The highest long return (8.12%) is generated by the portfolio that buys winners based on
four quarters returns and holds the index for four quarters. The least profitable long portfolio
(5.64%) buys winners based on one quarter’s returns and holds the indices for one quarter. The
results are smaller in magnitude compared to panel A, but behave similarly across different base-
holding period combinations, so that the highest long returns are generated by portfolios that have
longer base and holding periods.
Short portfolios generate positive annualized returns in all base-holding period
combinations similarly to panel A. The short portfolios’ returns range from 1.67% to 4.15%. The
lowest short return (1.67%) is generated by the portfolio that sells losers based on past four
quarters’ returns and holds the indices for three quarters. The highest short portfolio return
(4.15%) is in the portfolio that shorts losers based on their one quarter return and holds them for
one quarter. Again, these are the same base-holding periods that generate the highest and lowest
returns in panel A for the long and short portfolios.
Returns on zero cost long-short portfolios range from 1.42% to 6.31% on an annual basis.
From all combinations, the strategy that buys and shorts housing indices based on one quarter’s
return and holds them for one quarter performs the worst, while the combination that buys and
shorts housing indices based on four quarters’ return and holds them for three quarters performs
the best. These are again the same portfolios that produced the highest and lowest long-short
returns in panel A. Also, the results suggest that when holding the base period constant, the
abnormal returns increase with the holding period from one to two to three quarters, but are
slightly lower for four quarter holding periods compared to three quarter holding periods. Finally,
while holding the holding period constant, the abnormal return increases with a longer base
period. The findings provide more support for our hypothesis 2.
18
Finally, in panel C we repeat the analysis from panels A and B with 50% cutoffs for
determining winners and losers of the MSA indices to construct the buy and sell portfolios. The
long portfolios’ average annualized returns vary from 5.04% to 6.83% and the short portfolios’
returns range from 2.40% to 3.97%. The highest and lowest returns for both long and short
portfolios are generated by the same portfolios from panel B.
Returns on zero cost long-short portfolios range from 1.03% to 4.21% on an annual basis.
Again, similar to panels A and B, the strategy that buys and shorts housing indices based on one
quarter’s return and holds them for one quarter performs the worst, while the combination that
buys and shorts housing indices based on four quarters’ return and holds them for three quarters
performs the best. The results also suggest that the abnormal returns increase with longer holding
period and longer base period.
Figures 1 to 3 illustrate performance of the best performing long-short portfolios that buy
winners and sell losers based on their four quarters lagged return and hold the winning and losing
indices for three quarters after which the portfolios are rebalanced. Figure 1 shows the 15th
percentile cutoff results, figure 2 shows the 30th percentile, and figure 3 the 50
th percentile cutoffs
of winner and loser MSA indices.
In figure 1, since 1983 the long-short zero investment portfolio accumulated almost an
800% return compared to the US housing index that generated just over 220% return over the
same time period. Note however, that the roughly 220% generated by the U.S. housing index is a
simple buy and hold long strategy. As a comparison, the long momentum based portfolio returned
roughly 1000% over the same time period.
In figure 2, since 1983 the long-short zero investment portfolio accumulated almost
400% return and the long momentum based portfolio returned over 600% over the same time
period. Smaller accumulated return for the momentum portfolios is shown in figure 3. The long-
short zero investment portfolio accumulated nearly 200% return and the long momentum based
portfolio returned roughly 400% over the same time period. Once again, both figure 2 and figure
19
3 show that momentum based strategies outperform traditional buy and hold investment
strategies.
Overall, the findings support hypotheses 1 and 2, that there exists positive momentum in
the housing markets, and that zero investment long-short portfolios are able to produce positive
and economically significant abnormal returns. The zero investment long-short portfolios
generate up to 8.92% on an annual basis. The magnitude of this return is especially impressive
when compared to the buy and hold return on the U.S. housing index, which during the same
period is 4.69% per annum. The abnormal returns are increasing with the holding period and base
period, and the highest abnormal returns are produced by a strategy that uses 15th percentile
cutoffs in determining the winning and losing MSAs based on their lagged returns. The abnormal
returns are lower for strategies that use 30th percentile cutoff and the lowest for strategies that use
50th percentile as the cutoff points in forming long-short portfolios.
4.3. Robustness Checks
Table 4 reports results from the dynamic GMM estimation for five sub-periods (1983-
1988, 1989-1993, 1994-1998, 1999-2003, and 2004-2008)8. The results are similar in nature to
the results presented in table 2. Lags t-2 to t-4 are generally positive, statistically significant, and
large in magnitude during all sub-periods. However, their magnitude and significance level is
especially large during the 2004-2008 time period, and also during this time period the first
quarter’s negative return is much smaller in magnitude or even positive depending on the
specification. These results are consistent with our initial AR analysis that momentum is present
in all time periods, while showing that momentum is stronger during the later time period9.
8 Results of the full sample Arellano-Bond estimation are similar to our autoregression results. We do not
include the results for brevity, and they are available upon request. 9 The results from the full sample Arellano-Bond dynamic GMM estimation are very similar to the results
in table 2 in both significance and magnitude.
20
Table 5 presents the results of a long-short strategy for the five consecutive 5-year sub-
periods spanning the full 25 year sample. Building on the results from table 3, we construct our
long-short portfolios based on previous 4 quarters returns and a holding period of 3 quarters. We
again use top and bottom 15%, 30%, and 50% cutoff to identify winners and losers. Overall, the
results show that momentum in real estate is not unique to particular time period. Also, similar to
what is shown in table 3, a stricter definition of previous winners and losers (top and bottom
15%) yields higher return compared with a more relaxed definition (top and bottom 50%).
Even though the results show positive housing price momentum in all sub-periods, the
magnitude of the momentum is not constant across time. The momentum profits are higher in the
early and the late part of the sample and lower in the middle. The long-short strategy yields an
annual geometric return of as high as 14.88% during the 2004-2008 period and 10.79% during the
1983-1988 period. The returns during the 1989-1993 and 1999-2003 periods are still high, but
less than 8%, and the return during the 1994-1998 period is lower than 4.50% annually.
Table 6 displays the yield earned on the long-short portfolios in four different regions of
the U.S.. The results show the robustness of the momentum effect in housing prices subject to
geographical segmentation, but once again, with different magnitudes. The return on a
momentum based zero cost portfolio is highest in the West region (up to 10.02% annually) and
lowest in the Midwest region (up to 2.35% annually). The annual return on the momentum based
portfolio in the Northeast and the South is as high as 7.70% and 4.50% respectively. It appears
that areas in which home prices are higher and more volatile on average (West and Northeast)
have a stronger momentum effect compared with less expensive and volatile regions such as the
South and the Midwest.
5. Conclusion
21
Since Case and Shiller (1989), there have been several papers that document a positive
relationship between lagged returns and current returns in housing markets. Housing returns are
forecastable and there is strong evidence that housing markets, because of the transactions costs
and the nature of buyers and sellers participating in them, are probably less efficient than more
liquid markets with large institutional investors.
Our first contribution is documenting the momentum effect in a broad sample of housing
price indices in the U.S. over a long time period. Our dataset includes more than 380
Metropolitan Statistical Areas (MSAs). On average, quarterly MSA-specific lagged housing
returns over the U.S. lagged housing return are positively related to current MSA housing returns
over the U.S. housing return returns. This positive relation begins, as early as five quarters prior
to the current quarter with some signs of mean-reversion during the most recent quarter.
Similar to Jegadeesh and Titman’s (1993) stock market momentum paper, we form long-
short portfolios of the housing indices. We construct the buy portfolio from winning housing
markets and the short portfolio from losing housing markets based on their lagged returns. The
returns from the long-short portfolios are statistically significant and economically meaningful.
The annualized returns from these zero cost long-short portfolios are as high as 8.92% on an
annual basis. Such rate of return is about twice as large in magnitude compared with the return on
a traditional buy and hold strategy that is based on the broad US housing market index during the
same time period.
Our finding also confirms that the momentum effect in home prices exists regardless of
the time period or geographical region. However, the effect appears to be more pronounced in the
West and Northeast regions and during the 2004-2008 time period.
In reality, buying and selling of all of the housing indices used in this paper is not
possible as a tradable strategy. However, the Chicago Mercantile Exchange has tradable futures
on the major U.S. housing indices based on larger geographical areas, so there may be
opportunities for profitable trading strategies. Additionally, information about return momentum
22
and relative strength in housing can be useful for builders, potential home owners, and mortgage
originators.
23
6. References
Abraham, Jesse M. and Patric H. Hendershott (1993). Patterns and Determinants of Metropolitan
House Prices, 1977-1991, Real Estate and Credit Crunch, 18-42.
Abraham, Jesse M. and Patric H. Hendershott (1996). Bubbles in Metropolitan Housing Markets,
Journal of Housing Research 7(2), 191-207.
Ali, Ashiq and Mark Trombley (2006). Short Sales Constraints and Momentum in Stock Returns,
Journal of Business Finance & Accounting 33(3-4), 587-615.
Ang, Andrew, Robert J. Hodrick, Yuhang Xing, and Xiaoyan Zhang (2006). The Cross-Section
of Volatility and Expected Returns, Journal of Finance 61(1), 259-299.
Arellano, Manuel and Stephen Bond (1991). Some Tests of Specification for Panel Data: Monte
Carlo Evidence and an Application to Employment Equations, The Review of Economic
Studies 58, 277-297.
Atteberry, William L., Ronald C. Rutherford, and Mark E. Eakin (1993). Industrial Real Estate
Prices and Market Efficiency, Journal of Real Estate Research 8(3), 377-386.
Bloomfield, Robert J., Robert Libby, and Mark W. Nelson (1999). Confidence and the Welfare of
Less-Informed Investors, Accounting, Organizations and Society 24(8), 623-647(25).
Brounen, Dirk (2008). The Boom and Gloom of Real Estate Markets, Working Paper.
Calhoun, Charles A. (1996). OFHEO House Price Indices: HPI Technical Description, Office of
Federal Housing Enterprise Oversight.
Carhart, Mark M. (1997). On Persistence in Mutual Fund Performance, Journal of Finance 52(1),
57-82.
Case, Karl E. and Robert J. Shiller (1989). The Efficiency of the Market for Single-Family
Homes, American Economic Review 79(1), 125-137.
Case, Karl E. and Robert J. Shiller (1990). Forecasting Prices and Excess Returns in the Housing
Market, Real Estate Economics 18(3), 253-273.
Chan, Louis K., Narasimhan Jegadeesh, and Josef Lakonishok (1996). Momentum Strategies,
Journal of Finance 51(5), 1681-1711.
Chui, A. C. W., S. Titman, and K.C.J. Wei (2003). Intra-industry Momentum: The Case of
REITs, Journal of Financial Markets 6, 363-387.
Chui, A. C. W., S. Titman, and K.C.J. Wei (2008). Individualism and Momentum around the
World, Working paper.
Clayton, Jim (2003). Rational Expectations, Market Fundamentals and Housing Price Volatility,
Real Estate Economics 24(4), 441-470.
Daniel, Kent, David Hirschleifer, and Avanidhar Subrahmanyam (1998). Investor Psychology
and Security Market Under- and Overreactions, Journal of Finance 53, 1839-1886.
De Bondt Werner F. M. and Richard Thaler (1985). Does the Stock Market Overreact?, Journal
of Finance, 40(3), 793-805.
Fu Ng (2001). Market Efficiency and Return Statistics: Evidence from Real Estate and Stock
Markets Using a Present Value Approach, Real Estate Economics 29(2), 227-250.
Gau, George W. (1984). Weak Form Tests of the Efficiency of Real Estate Investment Markets,
Financial Review 19(4), 301-320.
24
Gau, George W. (1987). Efficient Real Estate Markets: Paradox or Paradigm, Real Estate
Economics 15(2), 1-12.
Gervais, Simon and Terrance Odean (2001). Learning To Be Overconfident, Review of Financial
Studies 14, 1-27.
Goetzmann, William N. and Roger R. Ibbotson (1994). Do Winners repeat? Patterns in Mutual
Fund Performance, Journal of Portfolio Management 20, 9-18.
Green, William H. (2003). Econometric Analysis (5th edition), Upper Saddle River, NJ: Prentice
Hall.
Gupta, Rangan and Stephen M. Miller (2008). “Ripple Effects” and Forecasting Home Prices in
Los Angeles, Las Vegas, and Phoenix, Working Paper.
Hamilton, B and R. Schwab (1995). Expected Appreciation in Urban Housing Markets, Journal
of Urban Economics 18, 103-118
Hendricks, Darryll, Jayendu Patel, and Richard Zeckhauser (1993). Hot Hands in Mutual Funds:
Short-Run Persistence of Performance, 1974-1988, Journal of Finance 48, 93-130.
Hung, Kathy and John L. Glascock (2009). Volatilities and Momentum Returns in Real Estate
Investment Trusts, Working Paper.
Hung, Kathy and John L. Glascock (2008). Momentum Profitability and Market Trend: Evidence
from REITs, Journal of Real Estate Finance and Economics 37, 51-69.
Jegadeesh, Narasimhan and Sheridan Titman (1993). Returns to Buying Winners and Selling
Losers: Implications for Stock market Efficiency, Journal of Finance 48(1), 65-91.
Jegadeesh, Narasimhan and Sheridan Titman (2001). Profitability of Momentum Strategies: An
Evaluation of Alternative Explanations, Journal of Finance 56, 699-720.
Lakonishok Josef, Andrei Shleifer, and Robert W. Vishny (1994). Contrarian Investment,
Extrapolation, and Risk, Journal of Finance 49(5), 1541-1578.
Lee, Charles M.C. and Bhaskaran Swaminathan (2000). Price Momentum and Trading Volume,
Journal of Finance 55(5), 2017-2069.
Levy, Robert (1967). Relative Strength as a Criterion for Investment selection, Journal of
Finance 22, 595-610.
Linneman, P. (1986). An Empirical Test of the Efficiency of the Housing Market, Journal of
Urban Economics 20, 140-154.
Piazzesi, Monika and Martin Schneider (2009). Momentum Traders in the Housing Market:
Survey Evidence and a Search Model, NBER Working Paper 14669.
Rouwenhorst, Geert K. (1998). International Momentum Strategies, Journal of Finance 53(1),
267-284.
Shiller, Robert J. (2007). Understanding Recent Trends in House Prices and Home Ownership.
NBER Working Paper.
Shiller, Robert J. (2008). Historic Turning Points in Real Estate, Eastern Economic Journal 34, 1-
13.
Zhang, X. Frank (2006). Information Uncertainty and Stock Returns, Journal of Finance 61(1),
105-137.
25
Table 1. Descriptive Statistics of MSAs’ Historical Housing Returns
Table 1 shows selected descriptive statistics of Metropolitan Statistical Areas’ (MSAs) housing
returns during the sample period from the first quarter of 1983 to third quarter in 2008. The
average annual appreciation as well as the standard deviation are for the broad U.S. housing index
during the sample period. The average annual appreciations for the top and bottom 15%, 30%,
and 50% are based on hindsight selection of MSAs.
Number of MSAs 1983 181
Number of MSAs 2008 381
Avg. annual appreciation 83-08 – U.S. 4.69%
Stdv. annual appreciation 83-08 – U.S. 3.41%
Avg. annual appreciation 83-08 – Top 15% (Hindsight) 6.39%
Avg. annual appreciation 83-08 – Bottom 15% (Hindsight) 2.67%
Avg. annual appreciation 83-08 – Top 30% (Hindsight) 6.01%
Avg. annual appreciation 83-08 – Bottom 30% (Hindsight) 3.11%
Avg. annual appreciation 83-08 – Top 50% (Hindsight) 5.46%
Avg. annual appreciation 83-08 – Bottom 50% (Hindsight) 3.46%
26
Table 2. Autoregressive analysis
Table 2 shows the results from the autoregression defined in equation (1)*. Equation (1) includes
one through seven quarterly lags. The dependent variable is the quarterly return on each MSA
housing market net of the average return on the broad U.S. housing market, and the independent
variables are the lagged returns on each MSA housing market net of the return on the broad U.S.
housing market for the lagged periods. All MSA observations are included in each quarter for
which the lagged index exists. T-statistics are reported in parenthesis below the coefficients.
*
itUSAtiMSAtUSAtiMSAtUSAtiMSA RRRRRR ...)()( 2,2,,21,1,,1,,,
Constant t-1 t-2 t-3 t-4 t-5 t-6 t-7
-0.0009
(-6.60)
0.0733
(3.94)
-0.0008
(-7.33)
0.0335
(2.15)
0.1738
(17.27)
-0.0006
(-6.60)
-0.041
(-0.27)
0.1760
(18.59)
0.2067
(28.02)
-0.0005
(-5.30)
-0.0317
(-2.07)
0.1586
(18.26)
0.2248
(29.55)
0.1236
(17.54)
-0.0003
(-3.40)
-0.0396
(-2.49)
0.1470
(15.95)
0.2239
(30.58)
0.1531
(20.25)
0.0811
(10.72)
-0.0001
(-1.75)
-0.0417
(-2.64)
0.1441
(14.52)
0.2362
(31.38)
0.1615
(20.81)
0.0976
(12.29)
0.0044
(0.59)
-0.0001
(-0.68)
-0.0337
(-2.12)
0.1473
(15.03)
0.2393
(31.03)
0.1644
(20.89)
0.0971
(12.14)
0.0039
(0.52)
-0.0026
(-0.42)
27
Table 3. Returns on Momentum Based Portfolios
Table 3’s panels A, B, and C show the annualized geometric returns on long momentum
portfolios defined in equation (2)*, short momentum portfolios defined in equation (3)**, and
zero cost long-short momentum portfolios defined in equation (4)***. The long and short
momentum portfolios are formed based on lagged returns on the top and bottom MSA performers
during the base J-period timeframe and are held for K periods. The sample consists of quarterly
returns to the MSAs from the first quarter of 1983 to the third quarter of 2008. In panel A,
momentum strategies are based on the top and bottom 15% of performers during the base period,
while panel B and panel C use the top and bottom 30% and 50% of performers during the base
period, respectively.
*,
1n K
J
N
W MSA
n p
R RN
**
,
1n K
J
N
L MSA
n p
R RN
***
Z W LR R R
Panel A: Returns on Momentum Strategies Based on Past Top and Bottom 15% Performers
Base Period Holding period Long Return Short Return Long – Short
Return
4 quarters 1 quarter 9.37% 1.82% 7.37%
4 quarters 2 quarters 9.90% 0.97% 8.78%
4 quarters 3 quarters 9.86% 0.82% 8.92%
4 quarters 4 quarters 9.83% 1.47% 8.32%
3 quarters 1 quarter 8.99% 2.50% 6.30%
3 quarters 2 quarters 9.48% 1.49% 7.83%
3 quarters 3 quarters 9.74% 0.96% 8.65%
3 quarters 4 quarters 9.67% 1.71% 7.92%
2 quarters 1 quarter 8.04% 3.23% 4.63%
2 quarters 2 quarters 8.84% 2.02% 6.65%
2 quarters 3 quarters 9.14% 1.55% 7.44%
2 quarters 4 quarters 9.16% 2.23% 6.88%
1 quarter 1 quarter 6.15% 4.68% 1.34%
1 quarter 2 quarters 7.28% 3.01% 4.12%
1 quarter 3 quarters 7.54% 2.47% 4.95%
1 quarter 4 quarter 7.53% 3.08% 4.42%
28
Panel B: Returns on Momentum Strategies Based on Past Top and Bottom 30% Performers
Base Period Holding period Long Return Short Return Long – Short
Return
4 quarters 1 quarter 7.40% 2.48% 4.80%
4 quarters 2 quarters 7.91% 1.83% 5.98%
4 quarters 3 quarters 8.05% 1.67% 6.31%
4 quarters 4 quarters 8.12% 2.22% 5.88%
3 quarters 1 quarter 7.16% 2.75% 4.29%
3 quarters 2 quarters 7.67% 2.04% 5.52%
3 quarters 3 quarters 7.97% 1.72% 6.18%
3 quarters 4 quarters 8.01% 2.25% 5.74%
2 quarters 1 quarter 6.62% 3.22% 3.30%
2 quarters 2 quarters 7.33% 2.46% 4.77%
2 quarters 3 quarters 7.64% 2.18% 5.38%
2 quarters 4 quarters 7.68% 2.64% 5.02%
1 quarter 1 quarter 5.64% 4.15% 1.42%
1 quarter 2 quarters 6.43% 3.05% 3.29%
1 quarter 3 quarters 6.81% 2.82% 3.93%
1 quarter 4 quarters 6.77% 3.22% 3.54%
Panel C: Returns on Momentum Strategies Based on Past Top and Bottom 50% Performers
Base Period Holding period Long Return Short Return Long – Short
Return
4 quarters 1 quarter 6.15% 2.87% 3.20%
4 quarters 2 quarters 6.53% 2.48% 3.98%
4 quarters 3 quarters 6.65% 2.40% 4.21%
4 quarters 4 quarters 6.83% 2.77% 4.06%
3 quarters 1 quarter 5.93% 3.08% 2.78%
3 quarters 2 quarters 6.40% 2.61% 3.73%
3 quarters 3 quarters 6.57% 2.47% 4.06%
3 quarters 4 quarters 6.72% 2.87% 3.84%
2 quarters 1 quarter 5.64% 3.36% 2.21%
2 quarters 2 quarters 6.16% 2.84% 3.26%
2 quarters 3 quarters 6.34% 2.69% 3.61%
2 quarters 4 quarters 6.50% 3.09% 3.40%
1 quarter 1 quarter 5.04% 3.97% 1.03%
1 quarter 2 quarters 5.66% 3.34% 2.26%
1 quarter 3 quarters 5.91% 3.13% 2.75%
1 quarter 4 quarters 6.04% 3.55% 2.48%
29
Table 4. Autoregressive Model using Dynamic Panel Estimation
Table 4 shows the five sub-period autoregression results from Arellano-Bond dynamic GMM estimation. First, equation (1) is first differenced to remove the
independent and identically distributed random effects from the panel. Then, the coefficients of the lags are estimated via Arellano-Bond generalized method of
moments estimator using lagged levels of the dependent variable as instruments in the estimation. Absolute values of T-statistics are reported below the
coefficients.
1983-1988 1989-1993 1994-1998
Dependent Variable: RMSA,t-RUS,t Dependent Variable: RMSA,t-RUS,t Dependent Variable: RMSA,t-RUS,t
t-1 -0.333 -0.366 -0.349 -0.270 -0.209 -0.196 -0.179 -0.223 -0.167 -0.170 -0.212 -0.228 -0.254 -0.337 -0.247 -0.281 -0.327 -0.379 -0.410 -0.418 -0.394
11.24 11.81 7.93 6.14 4.21 3.56 3.11 5.74 3.79 4.55 5.15 5.63 6.62 11.39 10.65 11.20 14.31 16.43 18.28 16.92 14.64
t-2 -0.045 -0.051 0.068 0.118 0.127 0.148 0.108 0.148 0.095 0.072 0.042 0.002 -0.031 -0.046 -0.103 -0.158 -0.176 -0.151
1.50 1.32 1.59 2.87 3.20 3.51 3.18 4.95 2.90 1.84 1.30 0.06 1.36 1.82 4.06 5.80 5.82 4.44
t-3 0.009 0.143 0.195 0.197 0.213 0.117 0.129 0.096 0.079 0.091 0.071 0.060 0.011 -0.017 0.001
0.19 3.95 5.00 5.28 5.66 3.97 2.91 1.86 1.77 2.63 3.54 2.55 0.45 0.60 0.04
t-4 0.133 0.178 0.192 0.201 0.086 0.074 0.094 0.129 0.056 0.069 0.056 0.083
4.09 6.08 6.42 6.98 2.33 1.42 2.45 3.80 2.68 3.07 2.23 2.79
t-5 0.084 0.095 0.083 0.011 0.007 0.055 0.078 0.081 0.102
3.23 2.83 2.28 0.23 0.15 2.07 4.60 4.49 4.87
t-6 0.047 0.044 0.012 0.009 0.022 0.052
1.82 1.59 0.70 0.69 1.18 2.82
t-7 -0.019 -0.037 0.049
1.00 2.41 2.72
1999-2003 2004-2008
Dependent Variable: RMSA,t-RUS,t Dependent Variable: RMSA,t-RUS,t
t-1 -0.192 -0.166 -0.193 -0.193 -0.228 -0.258 -0.341 0.335 0.255 0.096 0.044 0.035 -0.014 -0.078
6.10 4.41 4.87 4.62 5.16 5.55 7.68 6.81 8.10 3.36 1.41 0.97 0.37 2.10
t-2 0.025 0.038 0.032 -0.019 -0.063 -0.139 0.425 0.394 0.326 0.292 0.273 0.239
0.83 1.10 0.92 0.54 1.64 3.42 21.13 18.72 16.42 11.78 10.86 9.24
t-3 0.076 0.102 0.083 0.034 -0.005 0.330 0.331 0.311 0.288 0.298
3.00 4.34 3.46 1.23 0.16 13.85 13.08 11.83 9.59 10.06
t-4 0.075 0.083 0.065 0.021 0.210 0.218 0.232 0.245
4.23 4.27 3.31 0.96 7.04 6.49 7.50 6.86
t-5 0.046 0.052 0.037 0.063 0.102 0.140
1.93 2.08 1.47 2.56 4.22 5.32
t-6 -0.001 0.026 -0.034 -0.002
0.02 0.97 1.40 0.08
t-7 0.018 -0.085
0.84 3.22
30
Table 5. Sub-Period Returns on Momentum Based Portfolios
Table 4 shows the sub-period annualized geometric returns on long momentum portfolios defined
in equation (2)*, short momentum portfolios defined in equation (3)**, and zero cost long-short
momentum portfolios defined in equation (4)***. The long and short momentum portfolios are
formed based on lagged returns on the top and bottom MSA performers during the base J=4
periods timeframe and are held for K=3 periods. The 25 years (1983 to 2008) covered by the data
are divided into five consecutive 5-year periods.
*,
1n K
J
N
W MSA
n p
R RN
**
,
1n K
J
N
L MSA
n p
R RN
***
Z W LR R R
Sub-period Cutoff for
Winners/Losers
Long Return Short Return Long – Short
Return
1983-1988 15% 12.04% 1.15% 10.79%
1983-1988 30% 9.11% 1.73% 6.31%
1983-1988 50% 7.19% 2.47% 4.69%
1989-1993 15% 8.02% 0.64% 7.38%
1989-1993 30% 6.46% 1.45% 7.33%
1989-1993 50% 5.47% 1.98% 3.46%
1994-1998 15% 5.92% 1.37% 4.44%
1994-1998 30% 5.29% 1.88% 4.99%
1994-1998 50% 4.89% 2.58% 2.28%
1999-2003 15% 11.49% 3.47% 7.95%
1999-2003 30% 9.12% 3.66% 3.34%
1999-2003 50% 7.35% 3.78% 3.54%
2004-2008 15% 12.96% -2.05% 14.88%
2004-2008 30% 11.15% -0.03% 11.14%
2004-2008 50% 8.94% 1.45% 7.46%
31
Table 6. Regional Analysis of Returns on Momentum Based Portfolios
Table 5 shows the annualized geometric returns momentum portfolios for four different
geographical regions. The long momentum portfolios defined in equation (2)*, short momentum
portfolios defined in equation (3)**, and zero cost long-short momentum portfolios defined in
equation (4)***. The long and short momentum portfolios are formed based on lagged returns on
the top and bottom MSA performers during the base J=4 periods timeframe and are held for K=3
periods. The sample consists of quarterly returns to the MSAs from the first quarter of 1983 to the
third quarter of 2008.
*,
1n K
J
N
W MSA
n p
R RN
**
,
1n K
J
N
L MSA
n p
R RN
***
Z W LR R R
Region Cutoff for
Winners/Losers
Long Return Short Return Long – Short
Return
Northeast 15% 9.92% 2.20% 7.70%
Northeast 30% 9.08% 2.72% 6.34%
Northeast 50% 8.12% 3.65% 4.46%
West 15% 10.38% 0.09% 10.02%
West 30% 9.64% 1.06% 8.40%
West 50% 8.22% 2.22% 5.87%
South 15% 6.31% 1.67% 4.50%
South 30% 5.69% 2.12% 3.52%
South 50% 4.97% 2.73% 2.20%
Midwest 15% 5.19% 2.79% 2.35%
Midwest 30% 4.96% 3.08% 1.86%
Midwest 50% 4.59% 3.26% 1.31%
32
Figures 1, 2, and 3. Cumulative Return on Momentum Strategies
Figures 1, 2, and 3 illustrate the performance of the long, short, and long-short momentum
strategies that are based on four quarters lagged return and three quarters holding period before
they are rebalanced. All MSAs that have lag return information available are included in each
quarter from the first quarter of 1983 to the third quarter of 2008. As a comparison, the figures
also demonstrate the performance of the buy and hold strategy on the broad U.S. housing index.
In figure 1 momentum strategies are based on the top and bottom 15% of performers during the
base period, while figure 2 and figure 3 use the top and bottom 30% and 50% of performers
during the base period, respectively.
Figure 1
Returns on Neutral, Long and Short Strategies
(15% cutoff)
0
200
400
600
800
1000
1200
19
83
19
85
19
87
19
90
19
92
19
94
19
96
19
99
20
01
20
03
20
05
20
08
Time
Ind
ex
(1
98
3:Q
2=
10
0)
Long Momentum
Short Momentum
Long-Short Momentum
USA Long
33
Figure 2
Returns on Neutral, Long, and Short Strategies
(30% cutoff)
0.00
100.00
200.00
300.00
400.00
500.00
600.00
700.00
800.001
98
3
19
85
19
87
19
90
19
92
19
94
19
96
19
99
20
01
20
03
20
05
20
08
Time
Ind
ex
(1
98
3:Q
2 =
10
0)
Long Momentum
Short Momentum
Long-Short Momentum
USA Long