Master Thesis
Earnings and Price Momentum:
Insights from REIT’s on the
Performance of and Relation
Between the Two Anomalies
Author:
Jochem J. Bron
Supervisor:
Dr. Milena T. Petrova
A thesis submitted in partial fulfillment of the requirements
for the degree of Master of Science in Business Economics with a double
specialisation in Finance and Real Estate Finance
in the
Amsterdam Business School
University of Amsterdam
This document is written by Student Jochem J. Bron who declares to take full responsi-bility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of com-pletion of the work, not for the contents.
Abstract
Faculty of Economics and Business University of Amsterdam
Master of Science in Business Economics with a double specialisation in Finance and Real Estate Finance
Earnings and Price Momentum: Insights from REIT’s on the Performance of and Relation Between the Two Anomalies
by Jochem J. Bron
Earnings and price momentum are two prevailing anomalies of the efficient market hy-pothesis. Although one would expect a relation between the two, because stock prices are driven by their underlying fundamentals, like earnings, the source of this relation is yet to be discovered. Some attribute the relation to macro economic variables, while others find a relation between the two forms of momentum when controlling for these systematic effects. Here the performance and relation of earnings and price momentum is tested for European and US real estate investment trusts (REIT’s) by first making use of a portfolio analysis based on ten percentile portfolios sorted on the previous 3- to 12-month returns, standardized unexpected earnings or a combination of both. Then, with the use of a four-factor asset pricing model the relation is tested on a risk adjusted basis and finally
Fama & Macbeth (1973) cross-sectional regressions are used to test if earnings and/or price momentum are priced. The main conclusion based on the results is that a earnings and/or price momentum strategy works, although there are differences between the per-formance of European and US REIT’s. Furthermore the results suggests that there is a negative relation between earnings and price momentum at the inter-industry level.
First and foremost I would like to express my deep gratitude to Dr. Milena Petrova, my research supervisor, for her patience and advice while writing my thesis. Further I would like to thank Dr. Jeroen Ligterink for guiding me in the right direction while finding the right research topic during the thesis seminar. I would also like to thank Dr. Marc Bron and Dr. Jiyin He, who were always willing to give advice for writing my thesis and introduced me to LateX. Further my grateful thanks are also extended to Maarten Bron and Nora Drentje, who I could always disturb for i.e. a coffee break and my complaints about writing a thesis, and to Jerry Vermeire and Tedros Medhin for being the best of friends. Finally, I wish to thank my parents for their support and encouragement throughout my study.
Statement of Originality i Abstract ii Acknowledgements iii List of Tables v 1 Introduction 1 2 Literature Review 5 3 Methodology 13 3.1 Portfolio Analysis . . . 13
3.2 Asset Pricing Models . . . 17
4 Data 20
5 Results 24
5.1 Portfolio Analysis . . . 24
5.2 Asset Pricing Models . . . 38
6 Robustness Checks 45
6.1 Portfolio Analysis . . . 45
6.2 Asset Pricing Models . . . 57
7 Conclusion 64
A Appendix: Figures 68
Bibliography 73
4.1 Distribution of REIT’s in Europe . . . 21
4.2 Descriptive Statistics . . . 22
5.1 Monthly Price Momentum Portfolio Returns - US Sample . . . 25
5.2 Monthly Price Momentum Portfolio Returns - EU Sample . . 27
5.3 Monthly Earnings Momentum Portfolio Returns . . . 29
5.4 Monthly Two-Way Momentum Portfolio Returns - US Sample 32 5.5 Monthly Two-Way Momentum Portfolio Returns - EU Sample 36 5.6 Time-Series Regressions - US Sample . . . 39
5.7 Time-Series Regressions - EU Sample . . . 41
5.8 Fama & Macbeth (1973) Cross-Sectional Regressions . . . 43
6.1 Crisis Period - Price Momentum - US Sample . . . 46
6.2 Crisis Period - Price Momentum - EU Sample . . . 48
6.3 Crisis Period - Earnings Momentum . . . 50
6.4 Crisis Period - Two-Way Momentum - US Sample . . . 52
6.5 Crisis Period - Two-Way Momentum - EU Sample . . . 54
6.6 Monthly Momentum Portfolio Returns - One Month Skipped 56 6.7 Normal Period - Time-Series Regressions - US Sample . . . 59
6.8 Crisis Period - Time-Series Regressions - US Sample . . . 60
6.9 Normal Period - Time-Series Regressions - EU Sample . . . 62
6.10 Crisis - Time-Series Regressions - EU Sample . . . 63
Introduction
In finance there are two classical asset pricing anomalies, one, first reported by Levy
(1967), called the price momentum effect and the other anomaly called post-earnings-announcement-drift, or earnings momentum. The former anomaly indicates that stocks that have been performing well over the past 3- to 12-months will continue to perform well over the next 3- to 12-months. Although this effect was first reported in 1967, it took until 1993 to become accepted within the field of finance afterJegadeesh & Titman(1993) published their results. Since than institutions slowly started to actively trade on price momentum and introduced funds based on the price momentum effect (Sias,2007; John-son, 2011). For now these funds focus mainly on global equitys and with the increasing supply of momentum strategy funds the price momentum profits could be reduced and perhaps eventually vanish within this asset class. However for now the price momentum funds are still profitable, for example, on 1 September 2014 the Robeco Momentum Equi-ties I EUR fund had returned 16.58 percent since its introduction in 2012 (Robeco,2014). Further the latter anomaly, earnings momentum, which indicates that good-news firms, stocks with positive unexpected earnings announcements, have a tendency to continue to drift up, while bad-news firms, stocks with negative unexpected earnings announcements, have a tendency to continue to drift down (Feng, Price & Sirmans,2013). Like with price momentum, institutions actively trade on earnings momentum and although some stud-ies actually find that the size of the effect is gradually declining over time, the anomaly is still economically significant (Ke & Ramalingegowda, 2005; Campbell, Ramadorai & Schwartz, 2009).
However the persistence of these two anomalies in empirical studies are not only a clear violation of the Efficient Market Hypothesis, it also shows that the Capital Asset Pricing Model is unable to explain the behavior of stock prices. Proponents of the Efficient
Market Hypothesis and the classical financial theories, on the other hand, argue that, al-though the movement of the stock market is not a perfect random walk, researchers must not confuse statistical significance with economic significance. This means, for example, that investors are unlikely to create excess returns with a momentum strategy, because the statistical dependencies that result in momentum profits are very small. Furthermore a strategy based on momentum will not even outperform a buy-and-hold strategy, be-cause of the high transaction costs associated with such a strategy (Malkiel, 2003).
On the other hand the existence of momentum profits over the past decades has moti-vated researchers to look for explanations in other fields of finance to explain this ongoing deviation from the Efficient Market Hypothesis. One explanation is that momentum is caused by serial correlation of individual stock returns and that this serial correlation is caused by an underreaction or a delayed overreaction of investors to information. Where this under- or overreaction could be explained by a combination of theories from the field of behavioral finance, such as conservatism, over-confidence and/or self-attribution bias (Jegadeesh & Titman, 2001).
The previously mentioned serial correlation is especially strong in returns and rental growth of real estate (Case & Shiller,1989; An, Deng & Fisher, 2011). These character-istics make REITs, real estate investment trusts, ideal stocks to test earnings and price momentum strategies. Because, for price momentum strong returns serial correlation is a driver of momentum profits and for earnings momentum the driver is rental growth, due to its close relation to earnings growth. However, although there is a large amount of research done on price momentum within the REIT industry (Graff & Young, 1997;
Stevenson,2002; Chui et al.,2003a,b;Hung & Glascock, 2008,2010;Derwall et al.,2009;
Goebel et al., 2012; Feng et al., 2013), there are only a few studies done on the perfor-mance of earnings momentum within the REIT class (Price, Gatzlaff & Sirmans, 2012;
Feng, Price & Sirmans, 2013). Furthermore no study has been done on earnings and price momentum using European REIT returns, where the existence of momentum prof-its could be affected by the difference in behavior of investors in individualistic cultures compared to the US (Daniel et al., 1998). Finally the financial crises offers an unique opportunity to test the performance of these strategies during an economic downturn, which can have a severe negative effect on the performance of momentum strategies ( Je-gadeesh & Titman,2011).
The main question I want to answer here is if earnings and price momentum are related at the inter-industry level. While earnings and price momentum are mostly seen
as two different anomalies, they could actually be related. Because, the existence of mo-mentum can be explained by an under- or overreaction by investors to information and drift can be explained as an underreaction by investors to specific information, i.e. an earnings announcement. So the drift effect could completely or partially be captured by the price momentum effect. This is found byChan, Jegadeesh & Lakonishok(1996), they show that returns of a momentum strategy can partially be predicted by the returns of a drift strategy and vice versa, but they do not have the same explanatory power for future returns. Chordia & Shivakumar (2006) confirm these findings and extend the work of
Chan et al. (1996) by investigating if the systematic, or macro-level, component of the earnings drift is interrelating with price momentum. Their results show that this is indeed the case and that the predicitive ability of price momentum is primarily subsumed by the systematic component of earnings drift. However Chordia & Shivakumar (2006) use diversified drift portfolios which only reflect systematic information, so it is not clear if the relation between earnings and price momentum holds when controlling for systematic information. By focusing on REITs only it is possible to control for the systematic effects, because they operate in the same industry and are affected in the same way by macro economic variables. Feng et al. (2013) are the first to control for the systematic effects by using a sample of US REIT’s. They empirically show that drift and momentum are negatively related and that the relation between the two anomalies can not be explained by different reactions to systematic effects, which contradicts the findings of previous re-search. Here this investigation will be continued by testing the relation between earnings and price momentum strategies on North American and European REIT’s to see if the results still hold and are similar across different economies.
In this thesis I will investigate if earnings and price momentum continue to exist and examine the relation between the two anomalies within an industry. The industry that will be used is a sample of European equity REITs, real estate investment trusts, over the period January 2004 to January 2014. Moreover REITs are publicly traded companies that own and manage investment grade commercial real estate. The reason to look within an industry is to control for systematic effects influencing earnings and price momentum. The equity REIT industry has been proven to be a particular good industry for this, because, first of all, it is a relatively large and homogenous industry group. Fur-ther it is easy to determine which firms are part of the REIT industry (Chui, Titman & Wei, 2003b). Moreover, within an industry, there is a lower possibility that differences in risk, transparency and growth potential will have an unexpected influence on the results.
To investigate if price momentum and earnings momentum exist the following portfolios are constructed: All REITs in the sample are sorted in 10 percentile portfolios based on there price momentum, measured as the geometric mean over the past 3-, 6-, 9- or 12-months or based on their standardized unexpected earnings. Further a two way sort will be used to create 9 percentile portfolios based on earnings and price momentum to test a strategy based on a combination of the two forms of momentum. Then the returns on these portfolios are calculated over the next 3-, 6-, 9- or 12-months based on two different measures. The first measure is a buy-and-hold strategy following Chan et al.
(1996) the other measure is a rolling strategy following Chordia & Shivakumar (2006). Furthermore theFama & French(1993) multi-factor CAPM will be used to investigate if earnings momentum is able to capture price momentum, or vice versa, and finally Fama & Macbeth (1973) cross sectional regressions will be used to test if price and/or earnings momentum is priced.
In the next section, section 2, the previous literature on earnings and price momen-tum will be discussed. Section 3 will present the methods used and the hypotheses that will be tested. Section 4 describes the data used and section 5 shows the results of the main analyses. Section 6 reports the robustness checks and section 7 states the conclusion and discussion.
Literature Review
According to the Efficient Market Hypothesis (EMH) it is not possible for an investor to outperform the stock market, because, if the stock market is informationally efficient, there are no positive net present value companies for the investor to invest in. This makes it impossible for an investor to outperform the average market return on a risk adjusted basis. However as a large amount of research has shown over the past approximately 50 years, since Fama (1965) introduced the efficient market hypothesis, the EMH does not hold in reality. One of the anomalies that contradicts the EMH is the possibility of price momentum to predict future stock returns.
Such a price momentum strategy is a trading rule that states that an investor can re-alize significant abnormal returns by buying stocks that have performed the best over the past months and selling the stocks that have performed the worst over the past months. This trading rule was first documented by Levy (1967), however Jensen & Benington
(1970), among others, where skeptical about his findings. They suspected that Levy was actually data fishing, because he tested 68 different strategies before finding the price momentum strategy. Thus the focus in the literature continued to focus on contrarian strategies, a strategy where an investors sells stocks that have been performing good and buys stocks that have been performing bad. For example,De Bondt & Thaler(1985) and
Schiereck, De Bondt & Weber (1999) show that contrarian strategies are profitable for the long run andLehmann(1990) shows that such a strategy is also profitable in the very short run, in their case one week. Then in 1993 Jegadeesh & Titman find that between the long run and short run there is actually a possibility to create abnormal returns with a price momentum strategy, even when they take into account the short run reversals found by Lehmann. They test 16 different strategies by first creating portfolios based on their performance over the past 3-, 6-, 9- or 12-months. Secondly they sort the stocks based on their past performance and divide them in ten equally weighted percentile portfolios,
where the top decile portfolio is the loser portfolio and the bottom decile portfolio is the winner portfolio. Following the momentum strategy an investor buys the winner portfolio and sells the loser portfolio and holds it for 3-, 6-, 9- or 12-months. They show that these strategies are profitable, especially the winner portfolio based on the returns over the past 12-months and a holding period of 3-months without skipping a week between portfolio formation and holding period. This specific strategy resulted in a monthly return of 1.31 percent for the specific sample period. In 2001 Jegadeesh & Titmanenforce their results by showing that momentum profits continued to exists during the 1990s, countering the argument that momentum is due to a data fishing bias. Furthermore other researchers show that momentum is also profitable outside the US, so doesRouwenhorst(1998) show that momentum strategies are profitable in Europe and do Griffin et al.(2003) andChui et al.(2010) show that, although there are some exceptions in Asia, momentum strategies are profitable in most large markets. FinallyJegadeesh & Titman(1993,2011) show that there are two situations in which momentum strategies do not results in positive profits. The first is a form of seasonality that is unique to momentum strategies. In January momentum strategies earn negative returns, while in all the other months the returns are positive. This is significantly different to other well-known strategies where January is actually the best month, such as the size and book-to-market effect. The second is the performance of momentum strategies around an economic downturn. The reason for the negative returns during a financial downturn is because betas tend to be low for winning stocks and high for losing stocks after a crisis period. Thus, when the market recovers shortly after a crisis the losing stocks will outperform the winning stocks, because of their higher betas.
The other anomaly in the stock market that contradicts the EMH is called earnings momentum. For a strategy based on earnings momentum two main measures of earnings momentum are used in the literature. The first measure is a seasonal random walk mea-sure called SU E. The standardized unexpected earnings (SU E) meamea-sure is defined as the difference between quarterly earnings and expected quarterly earnings, standardized by the standard deviation of quarterly earnings. Chan et al. (1996) test an earnings momentum strategy over the period 1977 to 1993, as with a price momentum strategy all stocks are ranked by there past performance based on SU E and assigned to one of ten percentile portfolios. Their results show that the winner minus loser portfolio with a holding period of six months earns a return of 6.8 percent. However the spread between the lowest and the highest portfolio is only slightly higher a year after portfolio formation
(7.5 percent), is strongly reduced two years after portfolio formation (1.1 percent) and becomes negative three years after portfolio formation (-0.6 percent). These results are inline with earlier findings by Bernard & Thomas(1989) who also showed that a positive return of 4.2 percent could be earned by taking a long position in the highest portfolio and a short position in lowest portfolio with a holding period of 50 days after portfolio formation. More recently Chordia & Shivakumar(2006) show that over the period 1972 to 1999 and the sub periods 1972-1979, 1980-1989 and 1990-1999 significantly positive returns could have been earned based on earnings momentum and a holding period of six months.
The other measure for earning momentum is based on analyst forecasts of earnings, where the changes in these forecasts are used to measure earnings momentum, this mea-sure is also used byChan et al.(1996). They tested a earnings momentum strategy based on earnings forecasts over the same period as before (1977-1993). Again they use a six months holding period and they show that the portfolio with the best analyst revisions, the Up revision portfolio, outperforms the Down revision portfolio by 7.7 percent. Others find similar results over different sample periods, for exampleGivoly & Lakonishok(1979) find a 5 percent difference between the Up and Down revision portfolio over the period 1967 to 1974, which are inline with the results of Stickel (1991) for the period 1981 to 1984.
The former anomaly of the EMH, earnings momentum, is based on the fundamentals of a firm, while price momentum is based on past returns. However these returns can partly or completely be driven by the underlying fundamentals. So this implicates that price and earnings momentum are correlated, which results in the possibility that both effects are part of the same anomaly instead of two separate anomalies. To investigate the interrelation between the two effects Chan et al. (1996) use a two way sort based on price and earnings momentum. First they sort the stocks in three equally weighted portfolios based on the past six month returns. Than they create three other equally weighted portfolios sorted by a stocks SU E and finally they combine the two portfolios creating nine equally weighted earnings and price momentum portfolios. Their results show that a strategy using the two-way portfolio sort earns a return of 13.6 percent in the first six months and 25.7 percent in the first year. Furthermore the highest portfolio outperforms the lowest portfolio by 8.1 percent and 11.5 percent in the first six months and first year respectively. However the most interesting conclusion is that price mo-mentum not completely subsumes earnings momo-mentum, or vice versa, and this implies
that both strategies make use of different pieces of information. Chordia & Shivakumar
(2006) further investigate this relation and show, as many authors did before, that both earnings, measured with SU E, and price momentum can independently explain future returns. But Chordia & Shivakumar do not use the two-way portfolio sort to study the relation between earnings and price momentum, they use asset pricing tests.
Before continuing with the results of Chordia & Shivakumar(2006) first the use of asset pricing models with respect to momentum will be discussed. One of these asset pric-ing model used is the original one-factor CAPM, capital asset pricpric-ing model, of Sharpe
(1964) and Lintner (1965) that states that the cross-sectional differences in average re-turns of securities can be explained by market risk. Jegadeesh & Titman(1993) adjust for risk using the one-factor CAPM and show that momentum profits can not be explained by the cross sectional differences in risk due to the significant positive alphas they find, instead of alphas close to zero. However Fama & French (1992) and Jegadeesh (1990) show that market risk alone can not explain these cross-sectional differences and they show that the explanatory power of the CAPM can be improved by adding firm specific factors. Extending the CAPM with a size, small-minus-big (SM B), and book-to-market, high-minus-low (HM L), variable creates a three-factor CAPM, which is able to better capture the cross sectional differences in average returns affiliated with the idiosyncratic part of equities Fama & French (1993, 1996). The predictive ability of the two addi-tional firm-specific factors is also seen as support against the efficient market hypothesis
Avramov & Chordia(2006). Furthermore the three-factor CAPM is unable to explain the returns created by momentum. This is shown by Fama & French(1996),Grundy & Mar-tin (2001) andJegadeesh & Titman(2001), all three studies draw the same conclusion as
Jegadeesh & Titman (1993), due to the significant positive alphas they find, momentum profits can not be explained by the cross sectional differences in risk. Although Fama & French(1996) do state that the three-factor model is able to capture all anomalies related to CAPM, excluding momentum.
By extending the previous three-factor model with a fourth-factor Chordia & Shiv-akumar (2006) try to see if this fourth-factor is able to capture the momentum anomaly. The fourth-factor is a winner minus loser (subtracting the lowest from the highest port-folio) earnings or price momentum portfolio, followingCarhart (1997), called P M N and W M L respectively. Furthermore the P M N and W M L portfolio are well diversified so their returns capture only systematic information. This is useful, because based on the
findings of Avramov & Chordia (2006) andChordia & Shivakumar (2002) price momen-tum is not related to firm-specific news, but to macroeconomic variables. Continuing to the results of Chordia & Shivakumar, they show that the systematic component of earnings momentum captures price momentum, contradicting Chan et al. (1996). More-over, because price momentum is captured by P M N and P M N is closely related to the macroeconomy it is unlikley that price momentum can be explained by the idiosyncratic part of returns.
Furthermore on the one hand Chordia & Shivakumar (2006) argue that the main macroeconomic link between price and earnings momentum is inflation, while on the other hand Sadka (2006) argues that systematic liquidity is the main macroeconomic link. However first the possibility has to be excluded that the link between the two is not completely systematic in nature, therefore some of the literature solely focuses on real estate investment trusts, REIT’s. The reason for this is that with REIT’s it is possible to control for different reactions to systematic effects, because it is a relatively homogenous industry. Further the strong autocorrelation in real estate returns Case & Shiller(1989) and rental growthAn et al.(2011) increases the possibility of profitable earnings and price momentum strategies. Finally both anomalies are present within the REIT industry.
There has been a lot of attention for price momentum within the REIT industry. For example Chui et al. (2003a) show that momentum is a important determinant of future returns, just likeChui et al. (2003b) and Derwall et al.(2009). FurtherHung & Glascock
(2008) find that the momentum returns are higher when the entire market is up. Next
Goebel et al.(2012) show that a price momentum strategy is still profitable, their sample from 1993 to 2009 also resulted in positive returns. Furthermore Chui et al.(2003a) and
Chui et al.(2003b) also show that a price momentum strategy applied on REIT’s results in higher retuns compared to the same strategy applied on non-REIT’s. However there is far less literature on earnings momentum in the REIT literature, even though there is a close relation between earnings and rental growth. According to Feng et al. (2013) there is only one study on earnings momentum within the REIT industry. This study by Price et al. (2012) over the period 1982 to 2008 results in a return of 20 percent per year following an earnings momentum strategy based on SU E, while the same strategy on non-REIT’s resulted in a return of only 12 percent per year. The results ofFeng et al.
(2013) are inline with the previous results, there earnings momentum strategy, which is also based on the SU E measure, results in a return on the winner minus loser portfolio of 8.7 percent a year, where the winner portfolio had a return of around 20 percent per year.
However the most interesting thing found by Feng et al.(2013) is the fact that earnings and price momentum is still negatively related after controlling for systematic effects. Thus the connection between the two forms of momentum is not primarily systematic in nature.
The previous discussion has focused on the existence of momentum profits, however it did not provide a intuitive explanation of why momentum profits exist. One of these explanation of momentum profits is that there is a delayed reaction to information, how-ever Jegadeesh & Titman (1993) argue that there are also other factors contributing to momentum profits, that is the cross sectional dispersion in expected returns or the serial correlation in factor returns. The former contributes to momentum profits, because when a security has relatively high returns in the previous period one can expect the security to have above normal returns in the following period, however only if a component of realized returns is related to expected returns. However, as discussed before, Fama & French (1996), Grundy & Martin (2001) and Jegadeesh & Titman (1993, 2001) show that differences in risk across stocks can not explain momentum profits. Furthermore
Jegadeesh & Titman (1993) also find that the serial correlation in factor returns is not an explanation of momentum profits, because the serial correlation in factor returns is unlikely to have a positive relation with momentum profits.
Thus, because risk-based models are not able to explain the momentum effect, re-searchers have turned towards the field of behavioral finance to search for explanations. The models assume that momentum is caused by serial correlation of individual stock returns. However the problem with this assumption is that it is unclear if this serial corre-lation is caused by an underreaction or a delayed overreaction of investors to information. According to the efficient market hypotheses if public information about a stock arrives, this information is immediately incorporated into the stocks price. However in the case of an underreaction information is not immediately incorporated to its full extend into the stocks price. Eventually the information will be fully incorporated into the stocks price. In terms of momentum profits this means that a winner stock will have positive returns during the holding period, as the information is slowly incorporated into the stock price, and normal returns during the following period. So what reason does an investor have to underreact to information?
One explanation could be conservatism, a theory that comes from psychology, which was reported by, among other psychologists, Edwards (1968). In behavioral finance the conservatism bias means that investors react slowly to new public information like an
earnings announcement. However it is unclear why investors do not adjust their expec-tations completely to this information. Barberis et al. (1998) show that in the case of, for example, an earnings announcement, investors expect the news to have a large tem-porary component and they do not want to completely ignore their previous estimates of earnings. This causes an underreaction to new information.
An other reason that investors underreact to information found in the literature is because of the so-called disposition effect. The name for this effect was introduced by
Shefrin & Statman (1985) and it means that investors have the tendency to hold on to losing stocks for too long and sell winning stocks too early. Researchers have found this effect in experimental markets as well as in financial markets, the latter also includes real estate markets.
An other reason for the under reaction of the stock price to new information is that investors use a certain reference price level as an anchor level to evaluate the impact of new information. One of these reference points, studied byGeorge & Hwang (2004), is if a stock is reaching its highest point in the past 52 weeks, also known as the 52-week high and low prices. The stocks reaching their 52-week high or low are commonly available in newspapers that report stock prices, so this information is available to all investors on the financial markets. George & Hwang find that the more a stock price is near the 52-week high, or anchor price level, the less an investor is willing to buy or hold this stock. Similar, for a stock that is close to its 52-week low, an investor is less willing to sell or hold this stock. So, when new, for example, positive information arrives about a stock that is close to its 52-week high, investors are less likely to buy or hold this stock. This in turn reduces the effect of the information on the stock price, causing an underreaction to the new information in the short run. However in the longer run the price will go up to the level the information implies, resulting in possible momentum profits.
On the other hand momentum profits could also be a result of a delayed overreaction to new information by investors. One example why investors would overreact is called the representativeness heuristic. In finance representativeness is the tendency of investors to assign a stock to a certain representative group based on a consistent stream of in-formation without considering the effects of probability. Thus an investor might observe that the earnings of a certain company have consistently been growing over the past and based on this information he defines the companys stock as a growth stock. However in the process the investors ignores the fact that the probability that this trend, that he has observed, will continue is very small as well as the fact that it is very unlikely that a
company will just keep on growing Barberis et al. (1998).
Another example why investors could overreact to information is because of investor overconfidence. An investor gets more and more confident about his skill to select winning stocks when public information confirms his private information. However his confidence does not decrease in the same extent when public information contradicts his private information. Because of this investors belief that their previous winner stocks where due to their stock selection skills and that their previous loser stocks are just due to bad luck, this is called self-attribution bias. Due to this self-attribution investors become overcon-fident about their stock selection skills and overestimate the precision of the information about their winning stocks. The combination of over confidence in their skills and in-formation lead investors to drive up the prices of their winners above their fundamental values Daniel et al. (1998).
Finally these behavioral theories could also provide an explanation if large differ-ences are found between the performances of the momentum strategies in Europe and North America. So does Chui et al. (2010) argue that differences in momentum profits between countries arise, because some, but not all, countries are affected by behavioral biases that cause an under- or overreactions to information. The overconfidence and/or self-attribution bias, for example, is more profound in individualistic countries like the US, while in Europe certain countries are less individualistic reducing the behavior lead-ing to momentum profits. An other explanation for the differences that could arise is that investors in different economies react differently to common information. For example different legal constraints between the US and Europe could cause different reactions to information. However these differences could also be caused by differences in stock market participation between the US and Europe. Moreover the difference in size of the REIT’s in the US and Europe could also lead to differences in momentum profits. Due to the fact that big firms lead small firms, which becomes more important within an industry (Hou,2007).
Methodology
This section describes the hypotheses and methods that will be used to examine the relationship between earnings and price momentum for European (EU) and North Amer-ican (US) REIT’s during the period January 2004 to January 2014. The first subsection describes the methods used to analyse the performances of a portfolio based on a price and/or earnings momentum strategy and the second subsection describes the factor mod-els. Both methods are used, because on the one hand portfolio performance tests enable direct comparison between return patterns (Daniel et al.,1997,1998), on the other hand they do not account for risk premiums which factor models do (Fama & French, 1993;
Carhart, 1997). Thus, as Feng et al. (2013) I use both methods to avoid this problem.
3.1
Portfolio Analysis
The analysis of the relationship between price and earnings momentum starts by testing the two strategies separately. This is done to confirm that the two strategies work during the selected sample period and to get an estimation of the returns generated by the momentum strategies. Thus the first hypotheses that will be tested is if a price momentum strategy is able to predict future returns for both EU and US REIT’s.
H1.0 : A price momentum strategy is not able to predict future REIT
returns.
H1.1 : A price momentum strategy is able to predict future REIT
re-turns.
Based on previous literature on US REIT’s it is expected that the H1.0 will be rejected
for both EU and US REIT’s. As previously discussed Feng et al. (2013) show that 13
during the period 1993 to 2010 a price momentum strategy was able to predict future REIT returns, which in turn supports the findings of Chui et al. (2003a), who found a prevailing and strong price momentum effect for US REIT’s from 1990 to 1999. Further based onRouwenhorst(1998),Griffin et al. (2003) andChui et al.(2010), who show that for stocks price momentum strategies are profitable in Europe and most large economies, it is expected that the conclusions of Feng et al. (2013) and Chui et al. (2003a) will also hold for EU REIT’s
Next the hypotheses if an earnings momentum strategy is able to predict future REIT returns will be tested for both EU and US REIT’s.
H2.0 : An earnings momentum strategy is not able to predict future
REIT returns.
H2.1 : An earnings momentum strategy is able to predict future REIT
returns.
Based on the findings of Feng et al. (2013) on earnings momentum for US REIT’s it is expected that the H2.0 will be rejected for US REIT’s, because they found that, like a
price momentum strategy, an earnings momentum strategy resulted in significant posi-tive monthly returns. These findings are inline with the results of Price et al.(2012) on earnings momentum for US REIT’s. Further based on Chui et al. (2010) it is expected that the H2.0 will also be rejected for EU REIT’s.
To test the first two hypotheses two sets of portfolios are created based on earnings and price momentum. For both effects a measure is used to capture the momentum effect. For the first effect, earnings momentum, a seasonal random walk measure called SU E is used, following Chordia & Shivakumar (2006). SU E is calculated by first gen-erating the earnings per share variable by dividing the earnings in the current month by the number of shares outstanding in that same month. Next SU E is calculated as follows:
SU E = Ei,q− Ei,q−4 σi,j
(3.1) where Ei,q is the earnings per share i in the current quarter q and Ei,q−4 is the earnings
per share four quarters ago (q − 4). The part, Ei,q Ei,q−4, measures unexpected earnings
over the last eight quarters (j). The standard deviation is used to standardize the unex-pected earnings instead of the stock price, total assets, market capitalization or net sales, because these variables themselves could be estimators of expected returns. When these variables explain part of the cross sectional differences found they would bias the results
Chordia & Shivakumar(2006).
The second effect, price momentum, is measured by calculating the geometric mean of returns for each firm over the past j months, where j is 3-, 6-, 9- or 12-months, fol-lowing Feng et al. (2013). First returns are calculated by dividing the change in price between month t and t − 1 by the price in month t − 1 and then price momentum (M OM ) is calculated as follows: M OMi,j = "t−1 Y t−j (1 + ri,t) #1/j − 1 (3.2) where ri,t is the return on stock i in a given month t, which is between t − 1 and t − j.
These two measures of earnings and price momentum are than independently used to create equally weighted decile and tercile portfolios. First the decile portfolios are used to determine the returns based on an earnings or price momentum strategy sepa-rately. Secondly, the tercile portfolios are used to create 9 equally weighted portfolios, by combining the terciles based on SU E and M OM . These portfolios are used to examine whether the two anomalies are related within the REIT industry.
Next the monthly post portfolio formation returns are calculated. Two methods are used to calculate these returns. The first follows Chan, Jegadeesh & Lakonishok (1996) (CJ L) who uses a buy-and-hold strategy over the following k months after portfolios formation, where k is 3-, 6-, 9- or 12-months. The returns of the buy-and-hold strategy over these months are raised to the power of 1/k to compute monthly returns:
CJ Lz,t,k = Pi 1 (pi,t+k−pi,t) pi,t+k i + 1 1/k − 1 (3.3) where CJ Lz,t,k is the average monthly return on a certain portfolio z in month t with
a holding period k and pi,t is the price of a certain stock i within portfolio z in month
t. Further (pi,t+kp −pi,t)
i,t+k is the return on a certain stock i within portfolio z over holding
monthly return on a certain portfolio z.
The second measure followsJegadeesh & Titman(1993) (J T ), they employ a rolling technique where the returns for month t through t + k is a function of the past k ranking strategies, with only the weight of 1/k of the securities changing each month, and the rest being carried over from the previous month, where k is 3-, 6-, 9- or 12-months.
J Tz,t,k = t+k Y t Pi 1 (pi,t+1−pi,t) pi,t+1 i + 1 1/k − 1 (3.4) where J Tz,t,k is the average monthly return of portfolio z in month t to t + k. The part
Pi 1 (pi,t+k−pi,t) pi,t+k i + 1
is equal to the return on portfolio z in month t with a holding pe-riod of one month. Next the monthly returns of the total portfolio z in the current month is then calculated by taking the geometric mean of these monthly portfolio returns over the previous k months.
As a robustness check these returns are also calculated by skipping 1 month between portfolio formation and the holding period. This is done to control for the bid-ask bounce that could affect the returns of the earnings and price momentum strategies, because the bid-ask bounce can have a downward effect on the returns when they are measured over adjoining months (Chan et al., 1996).
Furthermore for each month, each method of return calculation and each form of momentum the difference between the return on the lowest percentile and highest per-centile is calculated (p10−p1) called HM L, the highest minus the lowest portfolio. Which
represents the returns of a long-short strategy with a long position in the High portfolio and a short position in the Low portfolio.
Next the previous described methods to calculate monthly portfolio returns are also used to compute the returns on the 9 portolios based on a combination of earnings and price momentum. With these returns the third hypotheses, if a combined strategy based on earnings and price momentum outperforms the two separate strategies, can be tested for both EU and US REIT’s.
H3.0 : Earnings and price momentum are not related at the industry
level.
Based on the research done byChordia & Shivakumar (2006) it is expected that the H3.0
can not be rejected. Because they found that for typical equities the ability of momentum strategies to predict future returns is subsumed by the systematic, or macro-level, part of earnings momentum. However REIT’s are part of the same industry and firms in the same industry are expected to have a comparable reaction to changes in macroeconomic variables (Mueller & Pauley, 1995). Thus, by using only US REIT’s it is possible to control for the systematic component of earnings momentum andFeng et al.(2013) show that in that case H3.0 can be rejected. Moreover here a difference between EU and US
REIT’s could arise, because EU REIT’s could be affected by country specific shocks. Finally as a robustness check the analysis is repeated for the crisis and non crisis, or normal, period, where the normal period is the period from January 2004 to September 2008, before the fall of the investment bank Lehman Brothers, and from January 2011 till January 2014. The crisis period is between September 2008 and January 2011. This is done to study the impact of the financial crisis on the performance of earnings and price momentum strategies, which has a significant effect on momentum returns as shown by
Jegadeesh & Titman(2011). Their results show a loss of 36.50 percent in 2009 caused by a momentum strategy based on a 6-month formation and holding period.
3.2
Asset Pricing Models
To further study the relationship between earnings and price momentum strategies on an inter-industry level the Fama-French three factor model will be used (and extended) to examine the relation between the two strategies. Moreover the model can be used to see if a component of earnings momentum explains part of or all the payoffs created by price momentum, or vice versa. By using a REIT sample, like Feng et al. (2013), to control for the systematic component of both forms of momentum it is possible to see if the relation between the two anomalies is only systematic of nature. Further The Fama and French factor model is used, because, asFama & French (1996) have shown, their three-factor CAPM is able to explain all CAPM-related anomalies except for momentum. By including the zero-investment P M N , positive minus negative, or W M L, winner minus loser, portfolios in the three-factor model it is possible to see if one of these variables can capture the momentum anomaly.
the Fama and French three factor model, model 3.5.
E(ri) − rf = βi[E(rmkt) − rf] + siE(SM B) + hiE(HM L) (3.5)
Where E(ri) − rf is excess return, E(rmkt) − rf is excess return on the market, SM B is
small minus big and HM L is high minus low. These factors together are the expected risk premia and βi, si and hi are the slopes in the time series regressions:
ri− rf = αi+ βi[rmkt− rf] + siSM B + hiHM L + i (3.6)
Where αi is the slope and i is the error of the time series regression. Moreover this
and the following regression are done for each price momentum portfolio. Further the Hausman test is used to show if the time series regressions should be run with fixed or random effects. Based onChordia & Shivakumar(2006) for stocks andFeng et al.(2013) for US REIT’s the αi’s should monotonically increase from the lowest to the highest
portfolio, indicating that there is price momentum on a risk adjusted basis rejecting H2.0
Next the model is extended by including P M N :
ri− rf = αi+ βi[rmkt− rf] + siSM B + hiHM L + piP M N + i (3.7)
By comparing the differences in the αi’s of the two time series regressions it is possible
to see if earnings momentum captures the impact of price momentum on returns. Based on the results of Feng et al. (2013) it is expected that H3.0 will be rejected. They find
that after including P M N the αi’s stop to increase monotonically from the lowest to the
highest portfolio. Moreover they find that the difference between the intercept of the lowest and highest portfolio is zero. Both indicating that earnings momentum is related to price momentum.
To test if the opposite is true, price momentum captures the impact of earnings momentum, the regression of model 3.6 is repeated, but now for each price momentum portfolio. Based onChordia & Shivakumar(2006) for stocks andFeng et al.(2013) for US REIT’s the αi’s should monotonically increase from the lowest to the highest portfolio,
indicating that there is earnings momentum on a risk adjusted basis rejecting H2.0. Next
the model is extended to a four-factor model by including W M L (Carhart, 1997): ri− rf = αi+ βi[rmkt− rf] + siSM B + hiHM L + wiW M L + i (3.8)
Again the differences in αi’s of the two time series regressions are compared to see if price
momentum captures the impact of earnings momentum. Based on Feng et al. (2013) H3.0 can not be rejected, because there results show that the αi’s continue to increase
monotonically from the lowest to the highest portfolio. Further they show that the effect of W M L is small and negative, indicating a modest negative relation between earning and price momentum. Finally as a robustness check the regressions are repeated for the crisis period and the non-crisis period.
As a last test, following Chan et al. (1996), the Fama & Macbeth (1974) cross-sectional regressions are used to see if earnings and price momentum are priced at the firm level. The CJ L 6-month buy-and-hold return is used as the dependent variable and SU E and M OMj, from model 3.1 and 3.2, will be used, where j is 3-, 6-, 9- or
12-months, together and independently as independent variables. Further, like Chan et al.
(1996) and Feng et al. (2013), size, measured as the natural logarithm of a firms market capitalization, will be used as a catch all variable to account for other effects on the cross-section of returns. If there is a relation between the returns and SU E and M OM , than, for example, a significant positive coefficient would suggests that past returns or earning (or both) are related to higher returns in the following period. While an insignificant coefficient would suggests that there is no relation at all. Furthermore by including both SU E and M OM as independent variables it is possible to see how they are related by the way they effect each others coefficients.
Data
To test the hypotheses data on North American and European REIT’s is collected over the period January 2004 till January 2014. From this dataset two separate samples are created. The first sample includes only REIT’s that are located in North America, the US REIT’s sample, and the second sample includes only REIT’s located in Europe, the EU REIT’s sample. The two samples are used to compare the results for EU and US REIT’s and to compare the results for US REIT’s with previous literature on US REIT’s. For the US REIT’s dataset the following variables are retrieved from the CRSP Ziman REIT database on monthly prices; the ticker, primary exchange, REIT type, shares outstanding (recorded in thousands) and total return. The primary exchange variable is used to drop all REIT’s that are not traded on the NYSE, AMEX or NASDAQ following
Price et al. (2012). Next, based on the REIT type, all REIT’s are dropped that are not equity REIT’s. This is done to make sure the dataset consists of a homogenous industry group. Moreover most listed REIT’s are equity REIT’s (Chui et al., 2003a). Further prices that have a negative sign are multiplied by negative one, because if the price has a negative sign this means that CRSP has used a bid/ask average instead of the actual closing price, which wil not have a large impact on the analysis.
The data on earnings is retrieved from COMPUSTAT North America fundamentals quarterly and in this dataset the net income (DATA 69 ) and preferred dividend (DATA 24 ), which are both recorded in millions, are included to calculate the earnings per quarter. I did not use the basic earnings per share variable provided by COMPUSTAT, because to many variables were missing and this would make the analysis of earnings momentum irrelevant. Next the COMPUSTAT dataset is merged with the Ziman dataset and all REIT’s that have no net income or dividends data are dropped. To control for firms with some missing net income and/or preferred dividends values, both variables are first forward and then backward filled. Furthermore REIT’s that have been trading for
less than 2 years are dropped, because to calculate SU E at least two years of earnings data is required.
To check for outliers distribution plots are generated for the main variables; size, earnings per share, return, SU E, M OM3, M OM6, M OM9 and M OM12. From figure
A.1it can be seen that both earnings per share and SU E have at least one large outlier. Based on this observation the following tickers are dropped from the dataset; O, RSE, ALX and GYRO.
Finally the dataset with the monthly Fama & French (1993) factor data retrieved from the Kenneth R. French Data Library is merged with the prior dataset. For the SM B (small minus big), HM L (high minus low) and excess market return variables the Fama and French Benchmark Factors monthly file is used and for the risk free rate the Fama and French Factors file is used.
For the EU REIT’s dataset the historical values of the monthly prices, total return and shares outstanding are retrieved from SNL. All REIT’s are included that are part
Table 4.1:
Distribution of REIT’s in Europe
Country # % Belgium 10 13% Finland 1 1% France 18 23% Germany 3 4% Greece 1 1% Ireland 3 4% Italy 2 3% Netherlands 5 6% Spain 4 5% Turkey 7 9% United Kingdom 26 33% Total 80 100%
of the SNL Europe REIT index. Table4.1provides an overview of the distribution of REIT’s within the index over the different European countries. As can be seen from table 4.1, most of the REIT’s are located in the United Kingdom (33%), followed by France (23%) and Belgium (13%). For all the REIT’s in the index the data on quarterly net in-come and preferred dividend are retrieved from the income statements of the individual REIT’s, also provided by SNL. Further this dataset is merged with a dataset containing the monthly European
Fama & French (1993) factor data which is re-trieved from Bloomberg. The dataset includes the following factors; the risk-free rate, excess market return, SM B and HM L. Next all REIT’s are dropped from the dataset if they have les than 2 year of monthly data available. This is done for the same reason as for the US dataset. Also REIT’s that have no earnings and/or preferred dividend data are dropped from the dataset. Furthermore, because for the EU REIT’s the variables net income, preferred dividends and shares outstanding contain missing values all three variables are first forward and then backward filled. Finally the companies with the following tickers
are dropped from the dataset to control for outliers (figureA.3) in the return data; BYG, DGGYO, FMU, HAB and ICAD.
Table 4.2: Descriptive Statistics
This table provides descriptive statistics (mean, minimum, 25th percentile, median, 75th percentile, standard deviation, and number of monthly observations). Size is measured as a firm’s market capitalization, in thousands, calculated by multiplying the number of shares outstanding by the share price. Earnings per share represents the quarterly net in-come minus preferred dividend divided by the number of shares outstanding at the end of the quarter. Returns is the monthly returns, in decimal format. SU E denotes standardized unexpected earnings, calculated as the past 4-quarter minus the most recent earnings per share standardized by the standard deviation of earnings over the past eight quarters. M OMj is the geometric mean of a firms return over the prior j months,where j is 3-, 6-, 9- or 12-, in decimal format.
Mean Min p25 p50 p75 Max S.D. N
Panel A: US Sample
Size 2,793,129 2,478 558,652 1,407,340 2,987,677 55,200,000 4,425,176 11,655
Earnings per Share 0.179 -9.150 -0.013 0.162 0.373 6.351 0.754 11,583
Returns 0.010 -0.798 -0.035 0.013 0.058 2.364 0.111 11,655
Panel B: US Earnings and Price Momentum Portfolios
SU E 0.055 -4.899 -0.547 -0.003 0.647 5.387 1.252 8,702 M OM3 0.006 -0.624 -0.017 0.011 0.036 0.594 0.062 11,417 M OM6 0.005 -0.468 -0.009 0.011 0.027 0.313 0.047 11,160 M OM9 0.005 -0.375 -0.006 0.010 0.024 0.230 0.039 10,887 M OM12 0.004 -0.289 -0.005 0.009 0.022 0.198 0.035 10,599 Panel C: EU Sample Size 1,290,000 5,210 195,000 472,000 1,340,000 19,300,000 2,140,000 6,571
Earnings per Share 1.068 -8.342 0.012 0.262 1.363 27.132 3.253 6,529
Returns 0.008 -0.868 -0.035 0.008 0.060 0.822 0.119 6,554
Panel D: EU Earnings and Price Momentum Portfolios
SU E -0.019 -4.899 -0.745 -0.013 0.738 5.940 1.309 5,107
M OM3 0.003 -0.735 -0.021 0.008 0.038 0.414 0.076 6,448
M OM6 0.001 -0.566 -0.016 0.007 0.030 0.344 0.060 6,325
M OM9 0.000 -0.427 -0.016 0.007 0.027 0.275 0.052 6,201
M OM12 -0.001 -0.344 -0.015 0.006 0.024 0.155 0.046 6,075
Table 4.2 shows the descriptive statistics for the two separate datasets used for the main analysis. Panel A in table 4.2 shows for the US dataset the Size, measured as the market capitalization, in thousands, Earnings per Share, measured as the earnings divided by the number of shares outstanding, and Returns, which is the monthly return. The mean size in panel A is around 3.1 billion and the median, p50, is around 1.7 billion. This means that the distribution of firm size is positively skewed and most REIT’s in the dataset are relatively small firms. Panel A also shows that there are 11,655 observations from 131 REIT’s left after data adjustments. Further the earnings per share seem to be normally distributed, based on the mean of 0.179 and median of 0.162 that lie close together and this is confirmed by the distribution histogram of earnings per share in figure
monthly returns are also fairly normally distributed with a mean and median of 0.10 and 0.13 respectively. The characteristics of the dataset are inline with the dataset used by
Feng et al. (2013).
Next panel B of table 4.2 reports the descriptive statistics of the SU E and the four different M OM variables, where M OMj is the geometric mean based on the past
j months. As can be seen from this panel SU E has a mean of 0.055 and a median of -0.003 indicating that the distirbution of SU E is slightly negatively skewed, however the distribution histogram of SU E in figure A.2 shows that it is fairly normally distributed. Further all the M OM variables seem to be normally distributed, as the returns variable with which they are calculated. The mean and medians of the four measures, M OM3,
M OM6, M OM9 and M OM12, are very similar. The means are all equal or close to 0.005
and the medians are all close to 0.010. These statistics are also inline with the descriptive statistics of the SU E and M OM variable of Feng et al. (2013)
For the EU REIT’s sample the summary statistics of the same variables are shown in panel C and D of table 4.2. Starting with the firm size in panel C, the distribution of firm size is positively skewed, similar to the US dataset, however the average size of the European REIT’s is about half the size of the REIT’s in the US sample. Further, after data adjustments, the dataset contains 6571 observations for 75 different REIT’s. Next, the earnings per share are more widely spread compared to the US dataset, with a minimum of -8.342 and a maximum 27.132, and the distribution of earnings per share is negatively skewed with a mean of 1.068 and a median of 0.012. However the distibution histogram of earnings per share in figure A.4 indicates a normal distribution. Finally returns are also normally distributed with a mean and median of 0.008, which is also inline with the US dataset. Continuing with panel D, the SU E variable and all the M OM variables seem, like the US REIT’s sample, to be normally distributed based on the mean and median, which is -0.019 and -0.013 for the SU E variable and around 0.001 and 0.007 respectively for the four different M OM variables. The histograms in figure
Results
In this section the main results will be presented. Starting with the average monthly returns resulting from the price and earnings momentum strategies for both EU and US REIT’s over the period January 2004 to January 2014. Next the results will be discussed of price and earnings momentum in the light of asset pricing models.
5.1
Portfolio Analysis
Table5.1and5.2present the results of a price momentum strategy on US and EU REIT’s over the entire sample period. The returns are based on a j-month portfolio formation period and a k-month holding period after portfolio formation, where j and k are 3-, 6-, 9- or 12-months. Further no month is skipped between the formation and holding period. This results in 4x4 different price momentum strategies for both samples. A price mo-mentum strategy based on a j-months formation period and a k-months holding period is indicated as M OMj,k. Furthermore the two different measures of monthly return based
on the buy-and-hold strategy ofChan et al.(1996) and the rolling strategy ofChordia & Shivakumar (2006), indicated as CJ L and CS respectively, are presented for all strate-gies. Moreover all returns are presented per percentile, where Low is the portfolio based on the lowest and High is the portfolio based on the highest price momentum over the past j-months. Finally the value of the high minus low (HM L) return is given at the bottom of the panels, where an asterisk indicates significance at the 5 percent level.
Starting with the portfolio returns of a price momentum strategy with a formation period of 3-months in panel A of table5.1for the US sample. From panel A it can be seen that HM L portfolio returns based on the CJ L method are all positive and significant, ranging from 0.48 percent to 1.08 percent. Further the return increases as the holding
Table 5.1:
Monthly Price Momentum Portfolio Returns - US Sample
This table provides the mean returns of the four different M OM portfolios over the entire sample period. The returns are based on a j-month portfolio formation period and a k-month holding period after portfolio formation, where j and k are 3-, 6-, 9- or 12-months. Further no month is skipped between the formation and holding period. A price momentum strategy based on a j-months formation period and a k-months holding period is indicated as M OMj,k.
Furthermore the two different measures of monthly return based on the buy-and-hold strategy ofChan et al.(1996) and the rolling strategy ofChordia & Shivakumar(2006), are indicated as CJ L and CS respectively. Moreover all returns are presented per percentile, where Low is the portfolio based on the lowest and High is the portfolio based on the highest price momentum over the past j-months. Finally the value of the high minus low (HM L) return is given at bottom of the panels, where an asterisk indicates significance at the 5 percent level.
Deciles k = 3 k = 6 k = 9 k = 12 CJL CS CJL CS CJL CS CJL CS Panel A: M OM3,k Low 0.23% 1.24% -0.02% 1.03% -0.31% 0.90% -0.46% 0.84% 2 0.73% 1.26% 0.55% 1.15% 0.46% 1.05% 0.33% 0.99% 3 0.67% 0.60% 0.58% 0.48% 0.50% 0.36% 0.41% 0.30% 4 0.70% 0.72% 0.61% 0.62% 0.50% 0.54% 0.45% 0.50% 5 0.52% 0.45% 0.54% 0.36% 0.48% 0.29% 0.45% 0.25% 6 0.64% 0.67% 0.65% 0.59% 0.58% 0.48% 0.52% 0.44% 7 0.86% 0.75% 0.65% 0.66% 0.54% 0.60% 0.54% 0.55% 8 0.97% 1.07% 0.75% 0.97% 0.64% 0.89% 0.62% 0.86% 9 0.90% 0.68% 0.82% 0.59% 0.73% 0.51% 0.71% 0.48% High 0.71% 0.69% 0.63% 0.57% 0.59% 0.50% 0.63% 0.48% HM L 0.48%* -0.57%* 0.66%* -0.48%* 0.90%* -0.42%* 1.08%* -0.38%* Panel B: M OM6,k Low 0.25% 1.08% -0.32% 0.91% -0.57% 0.78% -0.67% 0.71% 2 0.47% 0.84% 0.34% 0.70% 0.25% 0.59% 0.22% 0.53% 3 0.45% 0.67% 0.36% 0.54% 0.33% 0.42% 0.23% 0.37% 4 0.45% 0.34% 0.45% 0.23% 0.36% 0.14% 0.26% 0.10% 5 0.75% 0.93% 0.57% 0.85% 0.48% 0.78% 0.49% 0.74% 6 0.76% 1.08% 0.66% 0.98% 0.57% 0.89% 0.52% 0.83% 7 0.84% 0.75% 0.69% 0.67% 0.59% 0.61% 0.56% 0.58% 8 0.95% 1.01% 0.79% 0.92% 0.76% 0.84% 0.65% 0.80% 9 0.76% 0.82% 0.81% 0.77% 0.82% 0.71% 0.76% 0.68% High 0.64% 0.57% 0.72% 0.44% 0.78% 0.36% 0.79% 0.32% HM L 0.36*% -0.54*% 1.02*% -0.51*% 1.34*% -0.46*% 1.44*% -0.42% Panel C: M OM9,k Low -0.20% 0.60% -0.60% 0.44% -0.74% 0.30% -0.82% 0.23% 2 0.63% 0.60% 0.42% 0.47% 0.26% 0.37% 0.26% 0.30% 3 0.32% 0.66% 0.20% 0.56% 0.14% 0.45% 0.17% 0.40% 4 0.39% 0.56% 0.31% 0.44% 0.31% 0.37% 0.26% 0.35% 5 0.47% 0.91% 0.52% 0.81% 0.45% 0.72% 0.38% 0.67% 6 0.65% 0.95% 0.64% 0.85% 0.55% 0.74% 0.43% 0.68% 7 0.69% 0.92% 0.74% 0.83% 0.60% 0.76% 0.50% 0.72% 8 0.91% 1.16% 0.83% 1.12% 0.73% 1.04% 0.67% 1.01% 9 0.80% 0.73% 0.93% 0.63% 0.83% 0.57% 0.76% 0.54% High 0.81% 0.97% 0.82% 0.85% 0.84% 0.76% 0.78% 0.72% HM L 0.99*% 0.34*% 1.39*% 0.38*% 1.55*% 0.43*% 1.56*% 0.47*% Panel D: M OM12,k Low -0.26% 0.47% -0.62% 0.29% -0.80% 0.14% -0.83% 0.06% 2 0.60% 1.27% 0.36% 1.15% 0.23% 1.03% 0.19% 0.97% 3 0.08% 0.00% -0.01% -0.12% 0.02% -0.23% 0.05% -0.27% 4 0.51% 0.98% 0.43% 0.87% 0.35% 0.79% 0.37% 0.77% 5 0.51% 0.50% 0.50% 0.40% 0.45% 0.32% 0.38% 0.27% 6 0.52% 0.51% 0.66% 0.42% 0.51% 0.31% 0.36% 0.25% 7 0.78% 0.99% 0.76% 0.89% 0.64% 0.82% 0.59% 0.79% 8 1.23% 0.98% 0.91% 0.91% 0.74% 0.84% 0.66% 0.81% 9 0.95% 1.19% 0.80% 1.10% 0.69% 1.02% 0.66% 0.97% High 0.94% 1.23% 0.93% 1.14% 0.87% 1.06% 0.78% 1.03% HM L 1.19*% 0.72*% 1.52*% 0.82*% 1.64*% 0.88*% 1.58*% 0.95*%
period k increases. However the HM L portfolio returns based on the CS method are all negative and significant, ranging from -0.57 percent to -0.38 percent. Furthermore the returns increase as the holding period k increases. Next when looking at the individual portfolio CJ L returns there is a large difference between the return of the Low and sec-ond portfolio, but there are only small differences between the other 9 portfolios and this is true for all the holding periods. For the CS returns this is similar, although here the large diffence is between the second and third portfolio.
Next the results for the HM L portfolios are very similar in panel B, where price momentum is based on a 6-month formation period, compared to panel A. Again the HM L portfolio CJ L returns are all positive and significant, ranging from 0.36 percent to 1.44 percent, and the CS returns are all negative and significant, ranging from -0.54 percent to -0.42 percent. Further the returns are still increasing as the holding period k increases. The main difference between panel A and B is found in the differences be-tween the returns of the individual portfolios. In panel B the CJ L returns increase as the portfolio moves from Low to High. Although it is not the case that the High portfolio always creates the highest return. Further for the individual portfolio CS returns there is no clear trend when moving from the lowest to the highest portfolio.
Continuing with panel C (M OM9,k) and D (M OM12,k) the results for the CJ L
re-turns are similar compared to panel B with even higher HM L rere-turns and overall an outperformance of the returns of the higher portfolios compared to the lower portfolios. Further the CS returns are all positive and significant, ranging from 0.34 percent to 0.47 percent in panel C and 0.72 percent to 0.95 percent in panel D. Further the returns based on CS method increases as k increases, however between k = 9 and k = 12 the CJ L returns in panel C are constant and in in panel D are decreasing. Finally almost no percentile portfolio in panel A, B, C and D resulted in negative returns.
Thus, based on the results of table 5.1 it is possible to reject H1.0 that price
mo-mentum is not able to predict future REIT returns. The increasing CJ L returns between the Low and High portfolio and the positive CJ L returns on the HM L portfolios clearly show the exsitence of price momentum in the US REIT’s sample, which was expected based on Jegadeesh & Titman(1993),Chan et al. (1996), Chordia & Shivakumar(2006) and Feng et al.(2013). Although the HM L portfolio CS returns in panel A and B could indicate that a contrarian strategy would actually be profitable, which would be inline with De Bondt & Thaler (1985), Lehmann (1990) and Schiereck et al. (1999). However as the formation period of the momentum portfolios increases the CS returns increase
Table 5.2:
Monthly Price Momentum Portfolio Returns - EU Sample
This table provides the mean returns of the four different M OM portfolios over the entire sample period. The returns are based on a j-month portfolio formation period and a k-month holding period after portfolio formation, where j and k are 3-, 6-, 9- or 12-months. Further no month is skipped between the formation and holding period. A price momentum strategy based on a j-months formation period and a k-months holding period is indicated as M OMj,k.
Furthermore the two different measures of monthly return based on the buy-and-hold strategy ofChan et al.(1996) and the rolling strategy ofChordia & Shivakumar(2006), are indicated as CJ L and CS respectively. Moreover all returns are presented per percentile, where low is the portfolio based on the lowest and high is the portfolio based on the highest price momentum over the past j-months. Finally the value of the high minus low (HM L) return is given at bottom of the panels, where an asterisk indicates significance at the 5 percent level.
Deciles k = 3 k = 6 k = 9 k = 12 CJL CS CJL CS CJL CS CJL CS Panel A: M OM3,k Low -0.26% 0.53% -0.35% 0.40% -0.34% 0.35% -0.59% 0.30% 2 -0.13% -0.48% -0.15% -0.54% -0.09% -0.55% -0.15% -0.54% 3 0.05% 0.49% -0.20% 0.42% -0.22% 0.38% -0.24% 0.34% 4 0.27% 0.58% 0.32% 0.54% 0.21% 0.50% 0.14% 0.44% 5 0.15% 0.79% 0.01% 0.67% 0.12% 0.66% 0.12% 0.65% 6 0.10% 0.18% 0.13% 0.15% 0.00% 0.17% -0.05% 0.21% 7 0.40% 0.58% 0.33% 0.52% 0.18% 0.45% 0.03% 0.42% 8 0.72% 0.44% 0.43% 0.37% 0.28% 0.34% 0.24% 0.32% 9 0.45% 0.10% 0.30% -0.02% 0.12% -0.03% 0.15% -0.02% High -0.32% -0.20% -0.78% -0.41% -0.95% -0.51% -0.83% -0.61% HM L -0.08% -0.75*% -0.46*% -0.85*% -0.65*% -0.92*% -0.31*% -0.98*% Panel B: M OM6,k Low -0.71% -0.16% -0.60% -0.28% -0.75% -0.31% -0.84% -0.36% 2 -0.09% -0.32% 0.02% -0.43% -0.03% -0.46% -0.11% -0.49% 3 0.15% 0.36% -0.14% 0.30% -0.18% 0.29% -0.17% 0.26% 4 -0.05% 0.00% -0.21% -0.11% -0.28% -0.20% -0.21% -0.26% 5 0.33% 0.92% 0.08% 0.87% -0.10% 0.86% -0.09% 0.87% 6 0.25% 0.44% 0.16% 0.37% 0.09% 0.34% 0.00% 0.35% 7 0.40% 0.54% 0.34% 0.54% 0.23% 0.53% 0.18% 0.55% 8 0.60% 0.87% 0.47% 0.75% 0.34% 0.74% 0.16% 0.68% 9 0.96% 1.31% 0.41% 1.16% 0.16% 1.12% 0.01% 1.13% High -0.68% -0.91% -0.97% -1.06% -0.87% -1.14% -0.68% -1.16% HM L 0.05% -0.74*% -0.36*% -0.83*% -0.14*% -0.90*% 0.11% -0.90*% Panel C: M OM9,k Low -0.20% -0.14% -0.49% -0.29% -0.64% -0.30% -0.86% -0.33% 2 -0.18% 0.07% -0.37% 0.03% -0.49% 0.02% -0.48% -0.01% 3 -0.53% -0.15% -0.50% -0.24% -0.28% -0.30% -0.25% -0.31% 4 0.56% 0.99% 0.23% 0.86% 0.13% 0.78% 0.07% 0.73% 5 0.11% 0.68% -0.27% 0.55% -0.27% 0.58% -0.17% 0.59% 6 0.34% 0.17% 0.28% 0.09% 0.02% 0.04% -0.11% 0.00% 7 0.62% 0.60% 0.53% 0.51% 0.35% 0.48% 0.25% 0.51% 8 0.26% 0.51% 0.01% 0.45% 0.01% 0.41% -0.04% 0.39% 9 0.06% 0.53% -0.07% 0.41% -0.09% 0.39% -0.14% 0.39% High -0.47% -0.13% -0.65% -0.32% -0.83% -0.37% -0.79% -0.42% HM L -0.30*% -0.03% -0.19*% -0.07% -0.23*% -0.14*% 0.01% -0.20*% Panel D: M OM12,k Low -0.73% -0.03% -0.64% -0.16% -0.76% -0.24% -0.86% -0.31% 2 -0.39% -0.07% -0.55% -0.15% -0.76% -0.14% -0.62% -0.14% 3 -0.60% -0.08% -0.49% -0.14% -0.49% -0.18% -0.49% -0.23% 4 0.11% 0.38% -0.15% 0.30% -0.15% 0.27% -0.18% 0.25% 5 0.18% 0.59% 0.01% 0.49% -0.06% 0.48% -0.07% 0.48% 6 0.17% 0.29% -0.03% 0.26% -0.10% 0.28% -0.11% 0.27% 7 0.68% 1.10% 0.46% 0.97% 0.38% 0.93% 0.26% 0.90% 8 0.25% 0.47% 0.16% 0.43% -0.05% 0.37% -0.17% 0.34% 9 0.48% 0.55% 0.24% 0.43% 0.23% 0.44% 0.06% 0.43% High -0.85% -0.22% -1.14% -0.49% -1.19% -0.60% -1.21% -0.65% HM L -0.12% -0.18% -0.52*% -0.31*% -0.46*% -0.37*% -0.40*% -0.38*%
