1
Momentum investing in the Netherlands
Tom Hayje
10025332
Bachelor Thesis
Subject : Finance
Faculty of Economics & Business
University Of Amsterdam
Supervisor : Pepijn Trietsch
21-02-2014
2
Table of contents:
1. Introduction
3
2. Literature review
4
2.1 Momentum research `
42.2 Momentum results
72.3Possible explanations offered
82.4Summary of momentum results
103. Data & Methodology
10
3.1 Data
103.2 Methodology
113.2.1 Portfolio formation method 11
3.2.2 Risk adjusted portfolios 12
4. Empirical results
13
4.1 T-test results
134.2 Individual strategies reviewed
154.2.1 strategy A, B and C 15 4.2.2 Strategy D, E and F 17
5. Conclusion
19
6. References
21
7. Appendix
22
3
1. Introduction
“Past performance is no guarantee of future results” is an often used disclaimer when dealing with financial services. It’s supposed to warn new potential investors that basing investment decisions on past performances might not always grant the expected results. Yet there are still investment strategies around which base their strategy on the past performance of stocks. For instance, contrarian and momentum strategies use the past returns of stocks to decide what stocks are included in their investment portfolio. Several papers (Jegadeesh, 1993) (Schierek, 1999) dealing with contrarian and momentum investing have been published to check the viability of these investment strategies. Momentum investing in particular has been the
subject of multiple papers with several showing that this form of investing does indeed grant a return that could be considered abnormal when compared to a benchmark. For instance, one paper stated that a monthly return of nearly one percent above a benchmark index was achieved in 1993 (Jegadeesh, 1993).
Momentum investing has been described as: the purchasing of stocks that have performed well and the sale of stocks that have performed poorly in the past (O’Donnell, 2007). Another paper characterized it as “buying winners and selling losers” (Jegadeesh, 1993). So
momentum investing can be considered investing in the stocks that are doing good, the ones that have been moving in the right direction and hoping that because of momentum will keep doing that in the foreseeable future and trying to capitalize on that. While the quote about past performance is not a scientific theory, an investment strategy on past performance should not be capable of achieving significant abnormal returns due to the efficient market hypothesis. The EMH says in its weak form that future prices cannot be predicted from the past and that technical analysis of the past cannot generate abnormal returns (O’Donnell, 2007).
Momentum investing generating excess returns is in violation of the efficient market hypothesis, yet this strategy has been around for almost twenty years.
There are several explanations offered up as to why momentum works while the EMH says it should not, a few are listed here:
Investors are reluctant and/or slow to update their portfolios when new information emerges and when they finally update they overcompensate (George, 2004)
Buying the stocks that performed relatively good in the past and selling those that performed relatively poor steers the value of these stocks away from their long-run value and causes a price overreaction (Jegadeesh, 1993)
Another theory put forth by those investigating momentum is that the “Bandwagon effect” will cause investors that see rising stock price to join in and invest along with it(O’Donnell, 2007). Ebenezer and Tian claim that momentum profits will be significant when the state of the market(UP or DOWN) will not change too often
Momentum investing has been shown to work in the United states multiple times by for instance :Jegadeesh & Titman(1993), Scowcroft(2004), George and Hwang(2004), on an international level by Rouwenhorst(1998), and in Germany by Schierek et al(1993). however two research papers in Ireland have shown that momentum does not work in
4 Since momentum investing apparently is not viable everywhere it could be interesting to find out whether or not it works in the Netherlands.
In this paper momentum investing will be tested in the Dutch stock market to see if the momentum strategy can be used to “beat” the market. The question that will be answered in this paper is: : Does momentum investing in the Dutch stock market grant a significant
abnormal return?
To try and answer this question this paper will do the following:
First a brief review of all the literature used in this paper will be given in paragraph 2 , this involves academic papers researching momentum in the US stock market and papers researching momentum in the European stock After the literature review, paragraph 3
establishes what kind of data is used and what kind of methods are used on the data in order to answer the research question..
This is followed up by paragraph 4 which depicts the empirical results of the tests and the results of the individual momentum strategies. In paragraph 5 the paper ends with a quick summary of what has been done in this paper, after which we answer the research question mentioned earlier and discuss the potential limitations of this research and suggest points that can be refined for further research.
2. Literature Review
In this section the literature used and referred to in this paper will be outlined briefly. This paragraph starts by reviewing the various momentum research already done, after which the results and conclusions are shortly described and afterwards some of the explanations for these results are mentioned.
2.1 Momentum Research.
This paragraph shortly reviews the kinds of research done on momentum investing. Momentum strategies came to prominence after the paper done by Jegadeesh & Titman showed that buying winners and selling losers granted them significant abnormal returns. In their 1993 paper they analyzed stock returns from the 1965 to 1989 period, ranked these stocks according to their past performance, divided them up into different portfolios where they were matched with similar performing stocks. Afterwards they took a long position in the stocks that had the best performance and a short position in the stocks that had the worst performance. Each portfolio would contain the equally weighted top and bottom ten percent of the past period (Jegadeesh, 1993). They used different strategies which varied in the amount of months on which performance was based and the amount of months the formed portfolios would be held, which added up to a total of sixteen strategies to be tested. These sixteen strategies are then tested a second time but this time with a week in between the end of the formation period and the start of the holding period. This was done to potentially avoid some of the lagged reaction effects, price pressure and bid-ask spread that was present in an earlier paper done by Jegadeesh (Jegadeesh, 1993). A strategy that is highlighted in this research paper is the six month formation and holding strategy which is analyzed specifically
5 and used to represent all the other strategies that had significant results.
In a follow-up paper by Jegadeesh & Titman from 2001 the researchers repeat their previous experiment on another time period and there are some alternative explanations offered up for the results obtained in their previous paper. After the 1993 paper there was some debate on whether or the results were proof of market inefficiency or the result of data mining
(Jegadeesh, 2001) Jegadeesh & Titman repeat their former experiments on the years between 1993 and 2001 to reinforce the validity of their last research paper.
Strategies similar to the ones done by Jegadeesh & Titman, which can be referred to as the “traditional momentum strategies”, are compared to the 52-week high momentum strategy in the paper done by Thomas J. George and Chuan-Yang Hwang.
In the 2004 paper, they test the strategy of taking a long position in stocks whose current price is close to the 52-week high, and take a short position in stocks that have a current price which is far from the 52-week high. The data used in this research is similar to the preceding papers which dealt with momentum investing, all stocks from CRSP from the 1963-2001 time period (George,2004).
The paper compares 3 different strategies: the Jegadeesh & Titman(1993) strategy, the
strategy employed by Moskowitz & Grinblatt (1999) which is similar to ones done Jegadeesh & Titman but invests in top 30% and sells bottom 30% of the stocks, and the paper’s own strategy the 52-week high.
Similar to other papers, the 6 month formation and holding strategy is the analyzed in particular.
Unlike other research that tests whether or not momentum profits are significant, Ebenezer and Tian analyze the effects that market conditions have on momentum profits in their paper “market dynamics and momentum profits”.
The sample period used for this paper contains CRSP stock returns from January 1927 up to and including December 2005.
The authors state that changing market conditions will result in lower momentum profits than when market conditions stay the same (Ebenezer,2007). Market conditions are defined as “UP” and as “DOWN”, a market is classified as “UP” when the one year return is non-negative and a market is classified as “DOWN” when the past one year return is non-negative (Ebenezer,2007).
The unique thing in this paper is that it does not use the standard 3/6/12 months for formation and holding, instead it has mixtures of 5,02 months holding period with a 9,48 months
formation period (Ebenezer, 2007).
Scowcroft & Sefton add to existing momentum literature by not only testing momentum strategies on an individual stock level but also test the validity of the investing method on an industry level.
In their paper Scowcroft & Sefton shortly summarize the earlier results and findings from other momentum papers. They then proceed to test momentum on the stock level and the industry level over two sample periods , 1992-2003 and 1980-2003. Also included in their
6 paper is an appendix on how to build a momentum portfolio which was based on Jegadeesh & Titman’s earlier method (Jegadeesh, 1993). This appendix serves as the foundation on which the momentum portfolios of this paper are built.
Rouwenhorst differentiates its paper from other momentum papers by writing about the results of testing momentum strategies in multiple European countries in his paper “International Momentum strategies”.
Rouwenhorst theorized that because past research has always used the same database, the results could have been caused by sample selection (Rouwenhorst, 1998). To test whether similar results would be obtained outside of the US, Rouwenhorst decided to employ momentum strategies on an international level. This would be done by including individual stocks from twelve different European countries in the test sample from 1980-1995. All of the returns would be converted to a single currency, which was the Deutschmark. Similar to previous momentum papers extra focus is given to the 6 month strategy.
Continuing the European momentum research,Schierek, De Bondt and Weber test the profitability of momentum and contrarian strategies in Germany in their 1999 paper. They employ a momentum strategy that is quite similar to the 1993 method of Jegadeesh & Titman. The difference being that their research uses the data from the Frankfurt Stock Exchange (FSE) over a period of 31 years (1961-1991). The authors deemed further research of momentum strategy in a different setting to be relevant because the setting and trading regulations are different between those in America and Germany, the FSE has no explicit ask-bid spreads and it is always in the interest of economic theory to test whether empirical results will be the same or similar in a different setting (Schierek,1999).
The paper shortly reviews the stock exchange that will be analyzed, describes both contrarian and momentum strategies and follows up with the results from testing. In this research the stocks were ranked based on their cumulative excess returns, with excess returns defined as market return minus a compounded market return (Schierek, 1999). Unlike other momentum strategies Schierek et al also added a one month formation and holding strategy.
More recent momentum research is described in the two papers by the O’Sullivans (2010) and O’Donnell and Bauer (2007) in which momentum strategies are tested in the Irish stock market.
Unlike the majority of momentum papers however O’Donnell and Bauer define “winners minus losers” as the practice of only buying the winners without short selling the losers and buying the losers without short selling the winners (O’Donnell, 2007). The winners minus losers which was defined in the 1993 paper (Jegadeesh, 1993) is referred to as simply ”Momentum” investing.
Two years later the O’Sullivans question the validity of the results obtained by O’Donnell and Bauer due to highly non-normally distribution and serially correlation of their results
(O’Sullivan, 2010) because of this they decide to retest momentum strategies in the Irish market.
7
2.2 Research results.
This paragraph shortly describes the results and conclusions from the research mentioned in the previous paragraph.
The results from testing in Jegadeesh & Titman’s research in 1993 showed that 31 of the 32 strategies had significant results, with the three month formation and holding strategy being the exception (Jegadeesh, 1993). However most of the positive abnormal returns seemed to dissipate after a while, this process is referred to as reversal (Jegadeesh, 1993). The paper concluded that buying winners and selling losers realized significant returns over a 24-year period. Other findings were that their results were in line with delayed price reactions with regard to firm-specific information and that this was not due to systematic risk and delayed stock price reactions to a common factor effect (Jegadeesh, 1993).
Eight years later the researchers found that the results were similar to those from the previous testing period (Jegadeesh,2001). Having momentum strategies yield excess abnormal returns after the publication of the original paper had positive effects for the validity of the
investment form since other well-known anomalies such as the small firm effect did not occur anymore in sample periods after their initial discovery. The consistent results of the
experiments in different setting rejects the idea of data-mining in the original research (Jegadeesh, 2001). Jegadeesh and Titman conclude that because of the similar results with their previous paper any doubts about biases and strategy viability can be lessened
The results of the strategy comparison done by George and Hwang are that all three
investment strategies yielded abnormal returns, however the 52-week high has returns that are almost double of the other two strategies (George,2004). Contrary to other methods for
momentum investing the 52-week high does not experience return reversals (George, 2004). This finding has implied that reversals are not related to the bias that affects the short-term pricing (George, 2004) Another finding shown by the empirical results is that using the 52-week method high as a basis of past returns improved the predictability of future returns (George,2004) Since the 52-week high is one of the statistics that is supplied most often with stock reporting, having this strategy grant significant profits is a sign of market inefficiency (George,2004).
Researching the effect of changing market conditions showed Ebenezer & Tian that after a market changed from UP to DOWN in the following period it caused a decline of momentum profits from 2,09 percent to -0,01 percent per month. While a market that changes from DOWN to UP will decrease the average monthly momentum profit from 3,53 percent to -2,54 percent. These results confirm the assertion made by the researchers that volatile market conditions will have a negative effect on average momentum profits (Ebenezer, 2007). Meanwhile Scowcroft & Sefton discovered that momentum investing is also a valid on an industry level (Scowcroft, 2004) and they added that price return momentum is primarily derived from industry momentum (Scowcroft, 2004).
8 Momentum was also shown to be present on an international level after Rouwenhorst
concluded that momentum profits were also significant when investing on an international, with medium term winners minus losers earning an abnormal return of one percent per month (Rouwenhorst, 1998).
Rouwenhorst noted that these momentum profits could not be caused by risk since controlling for risk only increased the abnormal performance of the strategy (Rouwenhorst, 1998).
Momentum occurred in each of the twelve countries that were included in the sample. Also, interesting to note is that the research showed that there was a commonality found in both the US momentum strategies and this European momentum strategy
(Rouwenhorst,1998). This could mean that there is a common cause for both these abnormal returns in momentum investing although Rouwenhorst did add that finding out the exact cause was beyond the scope of this research paper and a more detailed analysis should be left for future research.
Narrowing it done from Europe to Germany, the results of testing concluded that momentum strategies granted abnormal returns. Schierek et al stated that the exact reason/factor
responsible for these results could not be found in their research. Schierek et al also added that the costs for putting this investment into practice (transaction costs for example) are modest and that the results from these strategies are therefore worth considering by portfolio
managers (Schierek, 1999).
Unlike the previous results, The papers tested in Ireland found no evidence to support the claim that momentum based investing in the Irish stock market could yield abnormal returns. The results showed that the regular form of momentum investing has a lower average than the market mean return and a higher variance than the market so it would grant a lower return at a higher risk than the market (O’Donnell, 2007). “Winner minus loser” portfolios however do yield an significant abnormal return according to O’Donnell and Bauer. In the 2007 paper they conclude that while momentum strategies do not yield a significant result over the entire sample period, there is some proof of momentum profits during small periods in the sample (O’Donnell, 2007). The paper further states that their findings are in line with other less developed markets.
In the 2010 paper the results gave no indication of momentum investing (winners minus losers) being a profitable strategy when compared to the market index. The authors state that their results are qualitatively similar to the previous paper done on the Irish market
(O’Sullivan, 2010) but that their paper adds to the momentum literature by acknowledging the non-normality of stock returns and that return outliers detract from profitable momentum portfolios (O’Sullivan, 2010). They concluded that the Irish market was somewhat efficient (O’Sullivan, 2010).
2.3 Possible explanations offered.
This paragraph shortly summarizes some of the possible explanations offered by some researchers that could be used to explain the results described in the previous paragraph.
9 Jegadeesh & Titman discuss a few possible explanations for the returns that can be obtained by momentum investing.
In their 1993 paper a closer look is taken at a model used for determining stock returns, which offers up cross-sectional dispersion, positive serial correlation and average serial covariance of security returns components as perhaps causing the abnormal returns (Jegadeesh, 1993). Another possible explanation discussed is the fact that the winner portfolios that pick the stocks with higher risk benefit from the higher returns. The profitability of momentum
strategies being the result of market under reaction to firm specific information would indicate that the stock market would be inefficient (Jegadeesh, 1993).
The 2001 paper offers support for several behavior models which would indicate that momentum profits are due to lagged overreactions (Jegadeesh,2001).
They discuss a few of these behavioral models, for instance they mention a theory by Daniel et al from 1998 in which “self-attribution” bias among informed traders is mentioned as being a possible cause of momentum profits(Jegadeesh, 2001). This bias would occur after a
positive feedback from earlier investment, which causes informed traders to overestimate their ability to pick winners and losers resulting in pushing prices above fundamental values, this delayed overreaction could then lead to momentum profits (Jegadeesh, 2001).
Another behavior model that is referenced to is the one by Hong and Stein, this model suggests that there are two groups of traders who both have a different effect on stock prices and thus momentum profits. The “Informed traders” trade based on future signals observed while ignoring past trading history and because of a slowed reaction time to new information the effect of these investors is only partially incorporated in stock prices (Jegadeesh, 2001). The “other” traders tend to trade based on limited past history but ignore fundamental information, this causes the prices of past winners to rise above fundamental value (Jegadeesh, 2001). The delay in the reaction combined with the rising above fundamental value could then explain momentum profits according to Hong and Stein.
Jegadeesh and Titman conclude that the behavioral models discussed are at best only partially capable of explaining momentum profits due to the fact that these models are not always consistent with empirical results(Jegadeesh, 2001).
George and Hwang theorize that traders look at the 52-week high as an anchor and that they are reluctant to bid the stock price higher when the 52-week high is at a highpoint, and that traders are unwilling to sell when the stock is far below the 52-week high(George, 2004). Because of this “Anchor and Adjust” bias they state that the return predictability is not because of past returns but because of traders their slow reaction to price information regarding their 52-week high anchor (George, 2004).
The paper done by Ebenezer & Tian offers three models to explain why changing market conditions have a decreasing effect on momentum profits, the DHS model, the HS model and the SS model.
The DHS model is a variation of the “Self-attribution” discussed in the 2001 paper mentioned earlier (Jegadeesh, 2001), it theorizes that confirming news after a certain market condition, so a price appreciation of certain stocks after a UP market, will increase investor
10 overconfidence and non-confirming news, a price increase after a DOWN market will
decrease overconfidence (Ebenezer, 2007).
The HS model postulates that private information on stocks will spread out over time, resulting in a positive serial correlation between returns. Momentum traders will then start trading because of this positive correlation and their trading activity will then cause a delayed overreaction to the private information causing momentum profits. (Ebenezer, 2007).
The third model, the SS model, is based on the theory that growth options will increase return autocorrelation and momentum profits when markets continue to stay in the UP state
(Ebenezer, 2007).
The authors concluded that all three models were consistent with the fact that momentum profits were higher after the market continued in the UP state, but that only the DHS model was consistent when the market continued in the DOWN state (Ebenezer, 2007).
2.4 Summary of momentum results.
A table that summarizes the year of publication, the focus of the momentum strategies, indices used, sample periods and monthly profits for the 6 month formation and holding strategy is depicted below in the appendix under table 1. It can be noted that for almost all papers listed that there is some form of momentum profitability except for the two papers about the Irish stock market.
3. Data & Methodology
In this paragraph the data and methodology used for this research will be discussed. 3.1 Data
The data used for this research will be the stocks from the Dutch market. This paper defines the Dutch stock market as stocks that were part of either the AEX index, the AMX index and the AscX index during the 2000-2013 period. More specifically the data contains the returns of January 2000 up until January 2013, which adds up to 156 months of data. The list of companies that were part of these indices is provided by DataStream.
Similar to previous research, financial firms will be excluded from the data sample, due to the differences of leverage effects between financial and non-financial firms (Jegadeesh, 1993). Also excluded from the sample are stocks that do not have any data on the total returns after 30-8-2007,which brings the sample size down to 60 from 67. This is done in order to have a more usable sample size on which to make weighted indices
The sample period is from January 2000 up until January 2013, this is a smaller time period than in Jegadeesh & Titman’s first paper, but it is similar to their 2001 paper. Since the results of both papers had similar results and conclusions, the 13 year time period is deemed large enough for this study (Jegadeesh, 2001)
In this paper abnormal returns are defined as momentum returns, which are winner portfolios minus loser portfolios, subtracted by a benchmark return, so:
11 AR = Rm - Rb
The AEX index will be used as a benchmark return. The benchmark AEX index is provided by investing.com which lists the returns and relative changes of this index on its website. Another benchmark used for in this research is an equally weighted index(EWB) which is constructed by the taking the average of the monthly returns for the AEX, AMX and AscX for the sample period.
For the risk-adjusted portfolios constructed using the CAPM, abnormal return is defined as momentum portfolio return minus expected portfolio return:
AR = Rp – Rep
The risk free rate for the CAPM formula is defined as the average risk free return over a 10 year Dutch government bond.
Using the CAPM to adjust for risk has been used in other momentum literature, Jegadeesh & Titman used a variation of it in one of their papers (Jegadeesh, 1993) and it has been used in a recent paper by O’Donnell as well (O’Donnell, 2007)
3.2 Methodology
In this study a portfolio formation method that is almost identical to the Jegadeesh & Titman method will be used. It will be based on the one used in Scowcroft & Sefton’s 2004 paper “Understanding Momentum”. They analyzed momentum in a similar sample period as this research (11 years). Like Sefton & Scowcroft, this paper will use strategies that select stocks based on their past 3 to 12 months performance and employ the same holding periods that will also be from 3 to 12 months. The formation period of a portfolio in months is referred to as “J” months and the holding period of a portfolio in months is referred to as “K” months. For example when J = 6 and K = 6, then the portfolio will contain stocks that are based on performance from the past 6 months and will be held for a duration of 6 months. “
3.2.1 Portfolio formation method
The portfolio formation process that is based on the one by Scowcroft & Sefton (Scowcroft,2004) is described by these following steps:
1.The monthly stock returns from the 60 stocks included in the 13 year sample period are divided up into periods of either three, six or twelve months.
2. The monthly average returns of the three/six/twelve month periods are then ranked from high to low.
3. The companies of the twelve stocks that performed the best in the first J month period(top 20 percent) will then be listed as “winners” while the twelve stocks that have performed the worst in the first three/six/twelve month period will be listed as “losers”.
4. To determine the winner returns, the returns of the top twelve companies of the previous J period will be added up. And to determine the loser returns the returns of the bottom twelve companies of the previous J month period will be added up. Both the winner and loser portfolios are then divided by twelve to represent an equal weighted portfolio.
5. Momentum portfolio return is then determined by taking he return of the winner portfolio and subtracting it by the returns of the loser portfolio. This would represent momentum portfolio #1. It uses the first J months period as the formation period and the second period is used as the K months period.
12 6. Momentum return for the entire sample period is calculated by taking the average of the returns of all the momentum portfolios
This portfolio method is a buy/sell and hold method and has no overlapping portfolios (Jegadeesh, 2001).
The image below should illustrate the formation and holding period process:
To test whether or not momentum investing can grant investors abnormal returns a t-test will be performed on the average momentum returns and the average return from the benchmark index, in this case either the AEX index or the EWB(equally weighted benchmark) index. A two-sample t-test will be used for testing the results since the data sets are from two different samples. In the t-test an Alpha of 0,10 will be used to test the H0 hypothesis. The H0
hypothesis is: The average difference between momentum return and the benchmark return is 0, in technical terms: µ = 0. This would mean that investing using a momentum strategy would not grant significantly higher or lower return than the benchmark market index and as a result not grant any significant abnormal returns. The alternative hypothesis, H1, is that momentum strategies would grant a significant higher return than the benchmark index, in technical terms: µ > 0.
Potential outliers in the portfolio returns will be identified through the following method: First the median, the lower quartile and the upper quartile will be calculated. Afterwards the interquartile range will be calculated and this number will first be multiplied by 1,5 and then it is subtracted from the lower quartile and added to the upper quartile. Any number outside the range is considered a “mild” outlier. Extreme outliers can be identified by multiplying the interquartile range by 3 and adding this to the upper quartile and subtracting it from the lower quartile, any number outside this range is considered an extreme outlier.
3.2.2Risk adjusted portfolios
In the 1993 paper (Jegadeesh, 1993) a possible explanation for the high momentum returns was that the winner stocks picked the riskiest stocks and as a result gained higher returns. To compensate for this possibility and to test whether or not risk is significant for this research, risk adjusted portfolios will be constructed using the CAPM method. The CAPM can be defined by the following formula :
Rep = Rf + β*(Rm – Rf)
This formula states that the expected portfolio return equals the risk free rate plus the excess market return multiplied by the portfolio beta. The market return is defined as the average period return from the benchmark index, so for the first portfolio of the J =12, K = 12
Formation period = J Portfolio creation Holding period = K Portfolio Return
13 strategy, the market return is defined as the yearly average return that corresponds with the time period of the first portfolio. The risk free rate as mentioned earlier is the average return of a ten year government bond. The betas for this formula are constructed by the following formula:
β
= Covariance of individual stock with benchmark index / Benchmark index variance. This formula is used for each individual stock that is part of our sample. To get a portfolio beta, which consists of 24 stocks each (12 winners and 12 losers) the betas that are part of the momentum portfolio are added up and then divided by 24 to get an equal weighted beta for that particular portfolio. Expected portfolio return is then calculated by adding the multiplied excess market return, market return minus risk free rate) to the risk free rate. The abnormal return for the risk adjusted portfolios is then calculated by subtracting the expected portfolio return from the actual momentum portfolio returns.A step-by-step procedure of this procedure:
1. For the first momentum portfolio, look up all the companies that are either winners or losers in that particular portfolio.
2. Find all the corresponding betas and take the average of the 24 betas to get equal weighted portfolio beta
3. Calculate expected portfolio return using the CAPM formula.
4. Subtract expected return from actual winners minus losers portfolio return
4. Empirical results
In this paragraph the results of the t-tests compared to the benchmark will be explained, after which the results of the risk-adjusted CAPM momentum returns will be listed for comparison, this section closes with a quick review of the three individual strategies that are tested.
4.1 T-test results
This section deals with the results of the t-tests performed on the Momentum versus benchmark tests and the results of the CAPM momentum abnormal returns tests. Table 1
Looking at the data from the first table it can be seen that four out of six momentum strategies tested in this paper do not have a significant t-statistic when compared to the AEX as a
Momentum returns compared to Aex
Benchmark A B C D E F AEX
Average monthly return 0,541 0,290 0,809 3,127 0,380 1,876 0,176 Variance 11,482 5,515 5,078 5,699 7,021 8,48 36,931
T- statistical data 0,539 0,169 0,781 3,505 0,284 1,753 P(T<=t) one-sided 0,295 0,433 0,221 0,001 0,399 0,0478 Critical area T-test: one-sided 1,287 1,291 1,315 1,132 1,293 1,327
14 benchmark and that there is not enough evidence to support the alternative hypothesis. The specifics from each individual strategy are explained further on in this paper and the winner minus portfolio returns can be viewed in the appendix. Strategies D and F are both mixed strategies with a twelve month formation period and will be elaborated upon later in this paper.
The next table depicts the results from the t-test comparing the six momentum strategies against an equally weighted benchmark index :
Table 2
Momentum returns compared to
Equal weighted Benchmark A B C D E F EWB
Average monthly return 0,541 0,290 0,809 3,127 0,380 1,876 0,192 Variance 11,482 5,15 5,078 5,700 7,021 8,48 23,643
T- statistical data 0,570 0,161 0,8144 3,708 0,287 1,818 P(T<=t) one-sided 0,285 0,436 0,212 0,001 0,388 0,044 Critical area T-test: one-sided 1,289 1,295 1,325 1,328 1,297 1,336
Similar to the previous t-test only four out six strategies have significant results. All strategies however have a higher t-statistic when an equally weighted index is used as the benchmark. The third table depicts the results of t-tests performed on the abnormal returns of the various strategies when adjusted for risk using the CAPM method and using the AEX as a benchmark: Table 3
AEX CAPM momentum abnormal
returns A B C D E F
Average monthly abnormal return 0,4501 0,211 -0,0265 3,0758 0,301 1,746
Variance 11,681 5,79 7,871 5,765 7,511 8,382
T- statistical data 0,94 0,438 -0,032 4,437 0,549 2,089 P(T<=t) one-sided 0,176 0,332 0,487 0,00049 0,293 0,03 Critical area T-test: one-sided 1,298 1,317 1,363 1,363 1,317 1,363
In the table above it can be seen that all when compared to the AEX benchmark the t-statistics with the exception of strategy C are higher when adjusted for risk using the CAPM formula. The comparatively lower average abnormal return of strategy C is also displayed in the 2007 paper where the T-statistic from the normal strategy dropped from a 0,80 to a -1,51
(O’Donnell, 2007) after being adjusted for risk using the CAPM method. In the 2007 paper all the other strategies except strategy F also had a higher t-statistic after risk adjustment using
15 the CAPM (O’Donnell, 2007).
The following table on the next page depicts the results of t-testing the abnormal returns after adjusting for risk using the equal weighted benchmark (EWB) and using equally weighted betas
When the EWB is used as the benchmark along with the betas being equally weighted as well it can be seen that all six strategies have a higher t-statistic after being adjusted for risk. This differs from the results in the previous table, where strategy C had a considerably lower t-statistic after risk adjustment.
Table 2
EWB CAPM momentum abnormal
returns A B C D E F
Average monthly abnormal return 0,4513 0,204 0,647 2,97 0,294 1,719
Variance 11,774 5,56 5,474 5,929 7,045 8,969
T- statistical data 0,939 0,433 0,958 4,225 0,554 1,989 P(T<=t) one-sided 0,176 0,334 0,179 0,0007 0,292 0,036 Critical area T-test: one-sided 1,298 1,317 1,363 1,363 1,317 1,363
While adjusting for risk using the AEX and the EWB increased the t-statistics for the majority of the strategies tested in this research, the changes were not significant enough to change the significance of strategies that was determined without the risk adjustment. 4.2 Individual strategies reviewed.
This paragraph reviews the six momentum strategies.
4.2.1 Strategy A, B and C
This sub-paragraph describes the three momentum strategies that use the same amount of months for the formation and the holding periods.
Strategy A, J =6, K =6
The first strategy that is tested is the strategy in which the winner and loser portfolios are based on the past three months performance and will be held for the following three months after which the return can be measured.
Since there are 156 months of data to be used in this test, the sample period is divided up into 52 three month periods. Since the first period is used only as a formation period, the sample period contains 51 winner minus loser portfolios that are made for this particular strategy. The returns from this strategy can be considered quite volatile, with the maximum return being around 9,93 percent and the minimum return being approximately -7,69 percent, the
16 aforementioned returns are also considered mild outliers using the method that was
established in the methodology section of this paper.. These volatile returns are not without effect, The average return of 0,541 also has a relatively high variance of 11,482.
As mentioned earlier, the empirical data that is displayed above shows that the t-test of the momentum returns from the three months strategy does not grant significant abnormal returns when compared to the benchmark index. This is in line with some momentum literature as seen in the paper by Jegadeesh & Titman where the three month formation and holding period strategy was the only strategy in their research which did not yield significant results
(Jegadeesh,1993). However in the paper on international momentum done by Rouwenhorst it was significant at the five percent level (Rouwenhorst, 1998).
The returns for each individual portfolio from strategy A can be observed in table 2 from the appendix.
Strategy B , J=6 K=6
The second strategy that is tested is the six month formation and holding strategy. Instead of basing performance and holding the stocks for three months it is now six months instead. The sample period will now be divided up into periods of six months each, starting from January 2000 up to January 2013, that amounts to 26 periods of six months. These 26 periods can be used to make 25 winner – loser portfolios, because as in the first experiment the first period will be used as a formation period only.
These returns are less volatile than the returns from the previous test, with the maximum return being around the 6,023 percent and the minimum return being close to -5,178. Similar to the previous strategy both the maximum and minimum returns are considered outliers by the earlier mentioned method of multiplying the interquartile range by one and half.
As in the previous experiment the momentum strategy does not yield significant results, with a very low t-stat and a high P stat it can be stated that a six month formation and holding strategy does not grant excess abnormal returns when benchmarked to the AEX and the equally weighted benchmark.
Contrary to most momentum literature however, this strategy yields a lower t-statistic than the three month strategy. In Jegadeesh & Titman’s paper the 3 month strategy was not significant at the five percent level while the six month strategy was (Jegadeesh,1993) In Rouwenhorst’s paper on international momentum the three month strategy was significant, but the t-statistic was lower than the 6 month strategy (Rouwenhorst, 1998). However in the paper done by the O’Sullivans where neither the three and six month strategy yield significant results at the five and ten percent level in the, the three month strategy has a higher t-statistic and a lower p-value which indicates that the three months strategy can perform better than its six months equivalent.(O’Sullivan, 2010)
The returns for each individual portfolio from strategy B can be seen in table 3 from the appendix.
17 The third strategy tested is the strategy in which the winner and loser portfolios are based on the returns from the past twelve months and then held for another twelve months. Similar to the previous two experiments the data sample is be divided up into periods of J months, which for strategy C is twelve. Thus the sample is reduced to thirteen periods from which one will be used as a formation period only.
The returns from this strategy are less volatile than the returns from the previous two tests with a maximum return of approximately 3,54 and a minimum return of close to -4,14 percent. The sample contains only two negative returns, and both these negative returns can be considered mild outliers.
The 12 months strategy has a higher average return and a lower variance than both the previous strategies but it also does not have a high enough t-statistic to allow rejection of the H0 hypothesis.
This strategy not being significant has occurred before in momentum literature (O’Donnell, 2007) (O’Sullivan, 2010). Worth mentioning is the fact that this strategy yielded a higher t-statistic than the six months strategy which was not the case in several other papers dealing with momentum (Jegadeesh, 1993) (Scowcroft, 2004) (Schierek, 1999)
When excluding the outliers from the sample and apply the t-test to the remaining returns the t-statistic improves dramatically from 0,781 to 2,618 with a p value of 0,005, thus making it significant enough to reject the H0 hypothesis. However with outliers it is always up for debate whether or not it can be justified to exclude the outliers and in this case the outliers are the only negative returns in the sample so to exclude them might compromise the research. In the other two strategies excluding the outliers also improves the t-statistic for both tests but the increase is marginal at best and does not change the outcome of the significance.
The individual returns for each portfolio from strategy C can be seen in table 4 from the appendix.
4.2.2 Strategy D, E and F
This sub-paragraph reviews the three momentum strategies in which the amount of months for the formation period differs from the amount of months for the holding period. The individual portfolio returns from these three strategies can be seen in the appendix.
Strategy D, J = 12, K = 3
This strategy sorts the stocks into the winner and loser portfolios based on the past twelve month performance and then holds these portfolios for three months. As in the previous strategy the sample is divided up into thirteen formation periods.
This is the first strategy that has yielded significant results with an average return of 3,127 and a t-statistic of 3,505 and 3,708 depending on the benchmark. This means that this strategy is significant on the one percent level and thus for this particular strategy the H0 hypothesis can
18 be rejected based on the data from the t-test. This strategy producing high returns is not a deviation from other findings in momentum literature since this strategy also yielded relatively high t-statistic, for instance in the 1993 by Jegadeesh & Titman the t-statistic for this particular strategy was 3,74 (Jegadeesh, 1993) which is similar to the one in this research. And in the Irish stock market while this strategy did not yield significant results the resulting t-statistic was one of the highest from the research which shows the power of this particular strategy (O’Donnell, 2007).
The returns are not very volatile with only a variance of 5,700 which is relatively low when compared to the other strategies and its high average return. The maximum return is 5,891 while the minimum return and mild outlier is -2,941.
A possible explanation for this strategy yielding significant results unlike most of the other strategies are:
The returns are relatively higher because the winner and loser portfolios are only held for three months and thus the chance of return reversal, the dissipation of the positive abnormal returns, is lower (Jegadeesh, 1993). Another possible explanation is that during the relatively short holding period the probability that the state of the market conditions will change drastically is also lower, changing market conditions negatively affect momentum returns.(Ebenezer, 2007).
However, strategy A and E also have a three month formation period, so perhaps the reason for the significant results is the relatively lower variance which is present in all the strategies with a twelve month formation period.
Strategy E, J = 6, K =3
In this strategy the formation period for the portfolios is six months long and the holding period is 3 months long. The sample period is, similar to strategy B, 26 periods of six months of which 25 portfolios can be made.
This is the only mixed strategy that is tested in this paper that does not have a significantly high enough t-statistic to infer that the initial hypothesis can be rejected. This is due to its relatively low average return and its comparatively high variance. The returns are all
relatively low and thirteen of the 25 returns are negative which can explain the results. There are no outliers in this data sample.
In momentum literature this strategy has produced varying results. In 1998 and in 2004 the accompanying t-statistics for this strategy were 3,07 and 1,93 respectively(Rouwenhorst, 1998) (Scowcroft, 2004), which are both significant. This contrasts with the results from Germany(1993) and Ireland(2007) in which the t-statistics were -0,11 and 0,136 respectively (Schierek, 1999) (O’Donnell, 2007).
This strategy has a higher return that its J =6 , K = 6 counterpart and a lower return its J = 3, K = 3 counterpart. Similar to the other two mixed strategies this could be due to having a
19 shorter holding period, which could reduce the possibility of return reversals lowering the average returns (Jegadeesh, 1993)
Strategy F, J = 12, K = 6
The last strategy tested in this research paper uses a twelve month formation period to determine the winner and loser portfolio that will then be held for the next six months. The sample period is divided into twelve periods of twelve months as was the case with strategy C and D.
This is the second strategy that has significant results. The t-statistic is significant at the five percent level with an average return of 1,876 and a variance of 8,48. As a result of the low probability of obtaining a t-statistic this high it can be suggested that this particular strategy does indeed grant significant abnormal returns. The portfolio returns contain two results that can be considered mild outliers, -2,405 and -4, 529.
The results from this particular tests reaffirms the results from earlier research papers in which the twelve month formation period and six month holding period, in these papers the t-statistic ranged from 1,75 to 3,66 (Jegadeesh, 1993) (Rouwenhorst, 1998) (Schierek, 1998) (Scowcroft, 2004). In the Irish market however, as was the case with other momentum strategies, this particular strategy did not yield significant results (O’Donnell, 2007) (O’Sullivan, 2010).
The same reasoning that applied to strategy D with regards to having a higher average return than the J =12, K = 12 counterpart can be considered for this strategy. This is reaffirmed by the fact that this strategy has a lower average return than strategy D which could be explained by having a twice as big holding period than strategy D.
5. Conclusion
The purpose of this research paper was to answer the question: Does momentum investing in
the Dutch stock market grant a significant abnormal return?
Momentum can be defined as buying the stocks that performed best and selling those that performed the worst, abnormal return is defined as either the return from the momentum strategy subtracted by the market benchmark or the realized portfolio return subtracted by the expected portfolio return. The Dutch stock market are the stocks from the AEX, AMX and the AscX. Several explanations have been offered for the causes of momentum with some authors claiming that it is investors overreacting/underreacting to certain factors, or their
unwillingness to (timely) react or update their portfolios. Data used in this research was the percentages of stock returns from a 13 year period between 2000 and 2013.
To answer this question all the returns from the Dutch stock market are divided into periods of either three, six or twelve months and then ranked based on their total returns. Based on their past performance the stocks are put into either “winner” or “loser” portfolios. All the winner portfolio returns are added up and then subtracted by the sum of all the loser portfolio
20 returns to determine the winner minus loser momentum strategy return.
The momentum strategy returns are then included in a t-test along with the returns from the AEX benchmark index and an equally weighted benchmark(EWB) index to see whether or not the H0 hypothesis, that there is no difference between the momentum and benchmark return, can be rejected. To adjust for potential high risk, the CAPM formula was used to calculate expected portfolio returns and this was subtracted from the actual portfolio returns to get abnormal returns which were also t-tested to see if the H0 hypothesis could be rejected when momentum returns were adjusted for risk.
Based on the tests that are performed in this paper, in which only four out of six strategy did not manage to reject the H0 hypothesis, it can be concluded that based on the results from the t-tests performed that most of the momentum investing strategies do not grant a significant abnormal return in the Dutch stock market. This would mean that the significance of the returns from strategy D and F would be more of an exception than the rule.
Possible explanations for this result could be that the market conditions changed too much for momentum to generate abnormal returns (Ebenezer, 2007) or that the Dutch stock market is similar in efficiency to the Irish one (O’Sullivan, 2010).
Possible limitations of this research are the relatively low amount of momentum strategies to be tested, other momentum papers typically contain around 10-16 strategies, but due to time constraints this was not possible. Another limitation of this research was that there is no Fama & French data available on the Dutch stock market, so the three-factor model/four-factor model could not be used in this research.
A factor that could be added in future research on this particular market is the use of
overlapping portfolios, due to the complexity of these and time constraints they were beyond the scope of this paper.
21
6. References
Ebenezer, A. and Tian, G.Y. Market Dynamics and Momentum Profits (2007). 20th Australasian Finance & Banking Conference 2007 Paper.
George, T. J. and Yang Hwang C. The 52-Week High and Momentum Investing. (Oct., 2004) The Journal of Finance: Vol. 59, No. 5 contains: pp. 2145-2176 (32 pages)
Jegadeesh ,N. and Titman S. Returns to Buying Winners and Selling Losers: Implications for
Stock Market Efficiency (Mar., 1993)The Journal of Finance : Vol. 48, No. 1 contains: . pp.
65-91 (27 pages)
Jegadeesh, N. Titman, S. Profitability of Momentum Strategies: An Evaluation of Alternative
Explanations ( 2001) The Journal of finance VOI, IA1, N O 2.
O'Donnell , Dan J. and Baur, Dirk G., Momentum in the Irish Stock Market (November 2007) O’Sullivan, F. and O’Sullivan, N. The profitability of momentum trading strategies in the
Irish equity market (February 5, 2010). Irish Accounting Review, 2010, Vol 17, Issue 1,
55-68
Rouwenhorst, G. International Momentum Strategies, (February 1998) The Journal of Finance Volume 53, Issue 1, pages 267–284
Schiereck, D, De Bondt, W. and Weber, M. Contrarian and Momentum Strategies in
Germany (Nov. - Dec., 1999),: Financial Analysts Journal Vol. 55, No. 6, Behavioral Finance
pp. 104-116
Sefton, J. and Scowcroft, A, Understanding Momentum. Financial Analysts Journal, Vol. 61, No. 2, pp. 64-82, April 2005
22
7. Appendix.
7.1 Appendix table 1 Author/ Year Momentum Strategy Index Sample PeriodMonthly profits for the 6 month formation/holding strategy Jegadeesh & Titman 1993 Individual stock price NYSE & AMEX
1965-1989 Approximately 1 percent per month. Unprofitable under 1 month and 1 year.
Rouwenhorst 1998 Individual Stock Price 12 European countries
1978-1995 Approximately 1 percent per month. Jegadeesh & Titman 2001 Individual Stock price NYSE, AMEX & NASDAQ
1990 - 2001 Approximately 1 percent per month. Unprofitable under 1 month and 1 year.
George & Hwang 2004
52-week high CRSP 1963-2001 0,45 percent per month, without reversals.
Scowcroft & Sefton 2004
Industry level MSCI global
1980-2003 & 1992-2003
0,57 percent per month.
Ebenezer & Tian 2007
Market dynamic
CRSP 1927-2005 Does not use standard 6 month strategy. Schierek, De Bondt & Weber Individual stock price
FSE 1961-19991 0,58 percent per month.
O’Donnel & Bauer
Individual stock price
ISEQ 1984-2007 0,033 per month, not significant. O’Sullivan & O’Sullivan Alternative size momentum portfolios
ISEQ 1988-2006 No Profit given for 6 month but the 6 month strategy had a t-statistic of 0,431 which is not significant.
23
7.2 Appendix table 2.
J =3, K =3, Momentum, winners minus losers portfolio returns.
Return #1: -7,69581 Return #2: 9,93966094 Return #3: 0,662821 Return #4: 1,552269
Return #5: 6,22704556 Return #6 -2,6738889 Return #7: -1,40269 Return #8: 0,9448139
Return #9: 0,21948606 Return #10 1,85074817 Return #11: -3,55043 Return #12: -2,091321
Return #13: 3,85545879 Return #14: 2,3167362 Return #15: 1,134708 Return #16: 0,1623933
Return #17: 4,86236835 Return #18: 6,317999 Return #19: -1,05812 Return #20: 3,0976144
Return #21: 0,1311674 Return #22: -0,3767028 Return #23: -0,66199 Return #24 1,2848506
Return #25: -4,5252302 Return #26 0,24989141 Return #27 -0,53142 Return #28 1,3490979
Return #29 0,96155807 Return #30 1,18487254 Return #31 0,142598 Return #32 -1,219617
Return #33 0,39687043 Return #34 -4,7880485 Return #35 3,115621 Return #36 -5,535886
Return #37 5,23139879 Return #38 -2,2550404 Return #39 0,72399 Return #40 4,6546673
Return #41 1,51028686 Return #42 -4,0186675 Return #43 -3,55224 Return #44 2,7353121
Return #45 0,9172325 Return #46 6,45365147 Return #47 2,858329 Return #48 0,8943244
Return #49 1,87567576 Return #50 -4,3861075 Return #51 -1,88249
7.3 Appendix table 3.
J = 6, K = 6, Momentum, winners minus losers portfolio returns.
Return #1: 1,09734964 Return #2: 6,02334251 Return #3: -1,96934 Return #4: 0,8174271
Return #5: 0,32111097 Return #6 -3,2033972 Return #7: 1,205193 Return #8: 2,7243378
Return #9: 2,50218689 Return #10 0,44741057 Return #11: 1,723625 Return #12: 3,3275955
Return #13: -3,1160828 Return #14: 0,93750491 Return #15: -0,54589 Return #16: 0,440547
Return #17: -1,7095435 Return #18: -5,1772723 Return #19: 0,733994 Return #20: 0,1276455
Return #21: -0,9136498 Return #22: -0,8739231 Return #23: 2,478833 Return #24 -1,407449
Return #25: 1,26277153
7.4 Appendix table 4.
J = 12, K = 12, Momentum, winners minus losers portfolio returns.
Return #1: 3,54388789 Return #2: 2,15935171 Return #3: -4,14512 Return #4: 2,1773837
Return #5: 1,97217141 Return #6 1,06241347 Return #7: 0,025047 Return #8: 0,3926076
Return #9: -3,0187111 Return #10 1,76213923 Return #11: 2,030789 Return #12: 1,7492411
24
7.5 Appendix table 5.
J = 12, K = 3 Momentum, winners minus losers portfolio returns.
Return #1: 5,20672307 Return #2: 2,16702653 Return #3: 2,995462 Return #4: 2,699447
Return #5: 3,27393899 Return #6 4,35669405 Return #7: 4,778837 Return #8: -0,536234
Return #9: -2,1949322 Return #10 4,70839102 Return #11: 4,178525 Return #12: 5,8910844
7.6 Appendix table 6.
J = 6, K = 3 Momentum, winners minus losers portfolio returns.
Return #1: 3,15887633 Return #2: 6,4319711 Return #3: -1,33059 Return #4: -1,044048
Return #5: -1,8837538 Return #6 -1,8698613 Return #7: -0,08572 Return #8: 1,146724
Return #9: 3,33405666 Return #10 2,00454943 Return #11: 1,085217 Return #12: 5,4317119
Return #13: -3,0721823 Return #14: 2,33399404 Return #15: -0,94152 Return #16: -1,009553
Return #17: -2,3882527 Return #18: -3,4309801 Return #19: -2,65664 Return #20: -0,999638
Return #21: -2,4861368 Return #22: 1,86231619 Return #23: 1,796249 Return #24 2,1929223
Return #25: 1,92852477
7.7 Appendix table 7.
J = 12, K = 6 Momentum, winners minus losers portfolio returns.
Return #1: 5,0916536 Return #2: 3,7722999 Return #3: -4,52953 Return #4: 3,3753831
Return #5: 3,82449543 Return #6 3,71266095 Return #7: 3,617264 Return #8: 2,2696761
Return #9: -2,4050175 Return #10 2,45290598 Return #11: 2,10682 Return #12: -0,769048
