Momentum Effect: A case study of Baltic States Stock Market

Nijmegen School of Management Master Thesis

Momentum Effect: A case study of Baltic States stock market

ByTOMAS ŽEBELYS (4847385)

This thesis aims to investigate the existence of momentum effect in the Baltic States stock market. Following the methodology developed by Jegadeesh and Titman (1993) momentum portfolios were formed based on a sample period from 2000 to 2016. Also, this study examines whether the obtained returns could be explained as a compensation for risk exposures through Fama and French (1993) three factors model. The results indicate that returns of momentum portfolios are mainly positive, although statistically insignificant for most of the momentum trading strategies. Furthermore, the risk-based approach could not explain observed momentum returns.

Supervisor: Dr. Sascha Füllbrunn Department of Economics

Table of Contents

1. Introduction ... 3

2. Literature review ... 4

2.1. EMH and market anomalies ... 4

2.2. International studies of momentum effect ... 8

2.3. Risk-based explanations ... 13

2.4. Behavioural explanations ... 15

3. Methodology and data ... 17

3.1. Data ... 17

3.2. Methodology ... 18

3.3. Methodology used for Fama-French three factors model ... 21

4. Empirical results ... 22

4.1. Momentum returns ... 22

4.2. Results of the Fama-French three factors regressions ... 29

5. Conclusion and discussion ... 31

6. Bibliography ... 34

1. Introduction

Since the financial markets have become accessible to the public, all kinds of investors have been trying to exploit the abundance of financial opportunities and the wide range of financial products. Development of the financial markets over the time has created attractive conditions for making financial decisions and facilitated the construction of investment portfolios that can easily replicate the principles of Modern Portfolio Theory. However, some investors are constantly trying to outperform the market and generate excess returns using their knowledge, various investment techniques, and models. Therefore, interest in trading strategies which have the potential to earn abnormal profits based on publicly available information has increased.

The proponents of the traditional finance theory assert that it is worthless to come up with such strategies due to market efficiency. Notwithstanding, Jegadeesh and Titman (1993) discovered that trading strategies of buying stocks which performed well and simultaneously short selling stocks which performed poorly over the past 3 to 12 months generate positive abnormal returns over holding periods between 1 to 4 quarters. This phenomenon of the continuation in stock returns is called momentum effect. Subsequent studies in this field supported the findings and provided strong evidence of the momentum effect in different financial markets across the world.

The excess returns of momentum trading strategies strongly contradict Efficient Market Hypothesis documented by Fama (1970) as publicly available historical stock prices are the only information required to form momentum portfolios. Numerous of researches have been conducted in order to find the explanation of this anomaly. Proponents of the market efficiency theory claim that it is the compensation for bearing risk while others from the behavioural perspective argue that momentum profits occur due to cognitive biases of investors. However, the consensus still has not been reached.

Most of the studies related to the presence of momentum effect have been concentrated primarily on the developed markets rather than emerging ones. The Baltic States are even more neglected as only a few references to the momentum effect and momentum trading strategies in the Baltic States stock market could be found (Avižinis & Pajuste, 2007; Stankevičienė & Gembickaja, 2012). Additionally, the Baltic stock market still faces a lack of new listings and low turnover. However, it has performed quite well with increasing OMX Baltic benchmark index and liquidity levels over the years. These features make it attractive for research purposes, and along with the lack of academic studies in the direction of momentum in the Baltic countries induced the motivation to embark upon this study.

The purpose of this thesis is to conduct a similar study as Jegadeesh and Titman (1993) and investigate the presence of momentum effect in the Baltic States stock market using new and more up to date data (2000 – 2016). In case of a positive returns of momentum portfolios, it is not possible to conclude that momentum strategies generate abnormal returns. Thus, this study also implements the most widely used Fama-French (1993) three-factors asset pricing model since risk-based approach might be able to explain the observed momentum return patterns in the Baltic States stock market. The returns of momentum portfolios will be risk-adjusted by regressing them on the Fama-French three factors and the results of time-series regressions might provide more informative assessment whether the returns obtained by momentum trading strategies are just a compensation for bearing risks or are they a result of potential market inefficiencies. The obtained results disclose that momentum returns are mainly positive in the Baltic States stock market for the analysed period, although insignificant for most of the momentum investing strategies. Furthermore, Fama-French three factor model could not explain observed momentum returns.

The study is structured as follows: section 2 presents the theoretical framework of the momentum effect, section 3 describes the data and methodology used in this study, section 4 presents the empirical analysis and results of the study, and section 5 provides the conclusions.

2. Literature review

2.1. EMH and market anomalies

The Efficient Market Hypothesis (EMH) introduced by Fama (1970) states that security prices fully reflect all available and relevant information. In general, the theory of market efficiency implies that investors cannot outperform the market and make positive abnormal profits by trading based on public information, which includes historical prices. It is mostly because stock prices at any point in time accurately incorporate news and reflect all available and known information.

Also, Fama (1970) asserted sufficient although not necessary conditions in regards to market be efficient. Firstly, there could be no transaction costs related to trading securities. Secondly, all information is free of charge and available to all participants of the market. Lastly, assessment and implication of all available information are homogenous by all market participants. And by taking into account all the conditions markets become frictionless. However, it is not likely that these conditions could be fully met in practice. Therefore, Fama

(1970) distinguished different levels of market efficiency, namely weak form, semi-strong form, and strong form. These forms are categorized by degree of available information.

Nevertheless, researchers in different markets across the world have identified the existence of various patterns in average stock returns which contradict the notion of EMH. These patterns have come to be known as “anomalies” since they could not be explained by traditional financial models, i.e. Capital Asset Pricing Model (henceforth CAPM) developed by Sharpe (1964) and Lintner (1965). Concretely, CAPM is based on a positive linear relationship between the expected return on a security and security’s beta, which denotes the measure of systematic risk of a security relative to the market as a whole. Following this model, there is no indication of market inefficiency if the return on the security conforms this linear relationship.

The theory of EMH and explanatory power of CAPM came into question when researchers have identified various deviations from this linear risk-return relationship. For instance, Banz (1981) and Reinganum (1981) documented size effect and showed that smaller firms (by market value of the common equity) on average outperform larger firms more than is predicted by the CAPM. Particularly, small stocks generate higher average returns and large stocks provide quite lower average returns given their estimates of CAPM betas. In a similar vein, other researchers documented value versus growth effect, which implies that different investment strategies’ portfolios based on various scaled-price ratios, such as earnings-to-price ratio, the book-to-market ratio, past sales growth, the cash-flow-to-price ratio and/or the dividend-to-price ratio manage to generate higher absolute and risk adjusted returns (Basu, 1977, 1983; Rosenberg, Reid & Lanstein, 1985; Lakonishok, Shleifer & Vishny, 1994).

I concentrate and elaborate on more recent evidence by Fama and French (1992) study, where they compared the performance of portfolios that were formed based on size and book-to-market ratio. Each year from 1963 to 1990 they ranked all the stocks traded on the New York Stock Exchange (NYSE), American Stock Exchange (AMEX) and NASDAQ into deciles based on companies’ market capitalization and observed the monthly average returns of each portfolio over the next year. They found that the smallest stocks decile portfolio on average outperforms the portfolio of the largest stocks decile by 0.74 percent per month. In the same study, Fama and French (1992) also grouped stocks into deciles based on their book-to-market ratio and measured the average returns of each decile portfolio over the following year. They found that highest book-to-market decile portfolio consisting of so called value stocks generates 1.53 percent higher returns on average than lowest book-to-market decile portfolio consisting of so-called growth stocks. Additionally, Fama and French (1992) repeated

calculations with a ranking measure of earnings-to-price ratio and observed that the average returns difference between the two extreme decile portfolios is 0.68 percent per month. They concluded that these differences are much higher than could be explained through differences in CAPM betas between the highest and lowest decile portfolios in all the cases.

Furthermore, De Bondt and Thaler (1985) presented the long-term reversals in stock returns. They found that “contrarian” investment strategies that are based on past stock returns (3 to 5 years) generate abnormal returns in the long-term (3 to 5 years) horizons. In their study, De Bondt and Thaler (1985) ranked all the stocks listed on NYSE from 1926 to 1982 based on their past 3-year cumulative returns and implemented the strategy of buying Losers portfolios that consist of stocks with low past returns and selling Winners portfolios that consist of stocks with high past returns. They observed that the Losers portfolios on average outperform the market by 19.6 percent in subsequent 3 to 5-year period after the portfolio formation. In contrast, the Winners portfolios on average generate 5 percent less than the market over the same holding period. They concluded that even though the Winners portfolios have a more systematic risk, the difference of about 25 percent between the Losers and the Winners portfolios is consistent, with overreaction hypothesis which implies that people tend to overreact to unexpected news events.

The momentum effect is also one of those financial markets anomalies widely examined in the context of market efficiency, traditional and behavioural finance theory for over more than two decades. In a similar way to long-term reversals and contrarian investment strategies, momentum also refers to the anticipation of return patterns based on a historical information and employment of momentum trading strategies allows for investors to exploit momentum effect. Most of the studies investigating the momentum anomaly are based on a pioneering study by Jegadeesh and Titman (1993) who presented ground breaking findings in this field.

Jegadeesh and Titman (1993) provided a clear evidence of profitable and significant relative strength strategies based on the simplest form of information, namely historical stock prices that are available to all market participants. Particularly, the authors in their study ranked the stocks listed on the NYSE and AMEX over the time span of 1965-1989 in ascending order, into (ten) deciles based on stocks’ past performance in previous 3, 6, 9 or 12 months. Then, in each month, momentum strategies buy the bottom decile Winners portfolio while short sell the top decile Losers portfolio at the same time and hold this long-short or also called zero-cost portfolio for the following period of time varying from 1 to 4 quarters. They referred to this as a J-month/K-month strategy where J-month denotes the formation and K-month denotes the holding period. In order to increase the power of empirical tests, Jagadeesh and Titman (1993)

used the overlapping portfolios technique. Additionally, they also performed tests with skipping a week between the formation and holding periods to ensure that the lagged reaction effect, the bid-ask spread or the price pressure documented by Jegadeesh (1990) and Lehman (1990) would not affect the results of the study.

Jegadeesh and Titman (1993) concluded significantly positive returns for all zero-cost portfolios except 3-month/3-month strategy without skipping a week. The strategy based on previous 12-month stocks returns and 3 months of holding position is the most successful in their study. This momentum strategy gives a return of 1.31 percent and 1.49 percent per month without and with one week lag between the formation and holding periods respectively. The strategies based on 6-month formation regardless of the holding time yield approximately 1 percent returns per month. They noted that excess returns are mostly from the long rather than short position of the transaction. To examine whether the returns of momentum trading strategies imply market inefficiency, Jegadeesh and Titman (1993) used one factor model benchmark, i.e. CAPM. Using this model only systematic risk is rewarded since unsystematic risk could be diversified away and if momentum returns are attributed to compensation for it, there would be no indication of markets’ inefficiency. However, they pointed out that neither lead-lag effects, nor systematic risk could explain the profitability of these strategies.

While some of the market participants well assessed the concept of momentum strategies introduced by Jegadeesh and Titman (1993) others criticized the strong evidence against the EMH. Arguing that the excess returns of momentum strategies are due to specific sample selections and data mining. In a follow-up study, Jegadeesh and Titman (2001) responded to criticism of data mining. Authors replicated their original research, including additional data of NYSE, AMEX and NASDAQ stocks listed between 1990 and 1998. Contrastingly to former study, they now controlled for small and low priced stocks. Jegadeesh and Titman (2001) investigated sub-periods returns and concluded that momentum strategies remain profitable in all sub-periods. Winners portfolios on average yield 0.56 percent per month more than equal-weighted index while Losers portfolios on average underperform the index by 0.67 percent per month. New findings suggest that Winners and Losers portfolios contribute about equally to momentum profits. These results also assure that findings in Jegadeesh and Titman (1993) study did not occur by chance.

The persistence of momentum effect is especially noteworthy, because in that time period of Jegadeesh and Titman (2001) study other documented anomalies such as a superior performance of small firms and value stocks over growth stocks disappeared or at least attenuated in the following studies and subsequent sample periods. For instance, the monthly

average size premium in the sample period precedent of the Banz (1981) study from 1965 to 1981 is 0.53 percent with t-stat of 2.34. However, in the subsequent sample period between 1982 and 2002 the average size premium is only 0.2 percent with t-stat of 0.67 (Schwert, 2003). Likewise, monthly average returns of a value premium in the period between 1994 and 2002 which is subsequent to sample period used in Fama and French (1992) study is -0.22 percent and not statistically significant with t-stat of -0.59 (Schwert, 2003). Although, there are other studies which reported results in favour of value effect. For instance, Fama and French (1998) reported that this phenomenon is still present not only in the U.S. but also internationally over the period between 1975 and 1995. Additionally, Davis, Fama and French (2000) analysed sample period from 1929 to 1997 and reported that the value premium is robust and they also showed that the value premium is observed in similar magnitude over the sample period from 1929 to 1963 which is precedent to Fama and French (1992) post-1963 evidence. They also concluded that in the earlier sample period size phenomenon is subsumed by the value effect. Other documented anomalies such as long-term reversals also have been explained after the introduction of Fama-French three factors model.

2.2. International studies of momentum effect

The phenomenon of momentum effect has motivated further exploration in this area. Therefore, the considerable amount of studies has been conducted in support of Jegadeesh and Titman (1993) research. In line with the methodology introduced by Jegadeesh and Titman (1993) various of authors have examined and found the presence of momentum effect not only in the U.S. but also in different financial markets.

Table 1 presents the several studies related to momentum effect and momentum returns in the U.S. stock markets. Column 1 shows the information of published studies, columns 2 and 3 report the information about the sample periods and stocks used in the papers. Column 4 informs about the size (percentage) of top and bottom stocks’ quantiles which are assigned to form Winners and Losers portfolios respectively since the momentum strategies do not consider stocks in-between top and bottom quantiles. Also, all presented studies use equally weighted stocks approach within Winners and Losers portfolios. Column 5 shows the length of the portfolio formation, skipping and holding periods. For example, 12/0.25/3 means that stocks are selected based on their previous 12-month returns and then are held for 3-month period. The number in the middle (0.25 month = 1 week) implies the skipping period between formation and investment periods. The last column reports the average raw returns per month

of momentum portfolios. In case, the study investigates several momentum strategies the most significant one is presented.

TABLE 1.OVERVIEW OF MOMENTUM RETURNS DOCUMENTED FOR THE U.S. STOCK MARKET

Study Period Data Portfolio

formation

Strategies Return per month Jegadeesh and Titman (1993) 1965-1989 NYSE/AMEX stocks 10% 12/0.25/3 1.49% Maskowitz and Grinblatt (1999) 1963-1995 NYSE/AMEX/NASDAQ stocks 30% 6/0/6 0.43% Hong et al. (2000) 1980-1996 NYSE/AMEX/NASDAQ stocks 30% 6/0/6 0.53% Lee and Swaminathan (2000) 1965-1995 NYSE/AMEX stocks 10% 12/0/3 1.54% Jegadeesh and Titman (2001) 1965-1998 NYSE/AMEX/NASDAQ stocks 10% 6/0/6 1.23% Grundy and Martin (2001) 1966-1995 NYSE/AMEX stocks 10% 6/1/1 1.34% Chordia and Shivakumar (2002) 1951-1963

1963-1994

NYSE/AMEX stocks 10% 6/0/6 0.83% 0.73% Griffin et al. (2003) 1926-2000 NYSE/AMEX stocks 20% 6/1/6 0.59% Avramov et al. (2007) 1985-2003 NYSE/AMEX/NASDAQ stocks 10% 6/1/6 1.49%

Chui et al. (2010) 1980-2003 NYSE/AMEX 1/3 6/1/6 0.79%

Fama and French (2012) 1989-2011 NYSE/AMEX/NASDAQ stocks 30% 12/1/1 0.64% Asness et al. (2013) 1972-2011 NYSE/AMEX/NASDAQ stocks 30% 12/1/1 0.5%

Table 1 reports that momentum returns in each study are positive and statistically significant at conventional levels during different sample periods ranging from 1926 to 2011. These findings provide strong evidence that momentum effect sill persists and profitability of momentum trading strategies even long after their discovery in 1993 still remains in the U.S. stock market. It is also noteworthy that all studies are similar in regards to applied methodology. It has become a common practice that researchers in their studies form zero-cost portfolios while buying past “winners” and selling past “losers” stocks’ quantiles. Furthermore, it is common to use the overlapping portfolios formation technique employing monthly stock prices data. All the studies in Table 1 use similar data sets and returns obtained by momentum investing strategies are estimated before trading costs. Moreover, most studies use data screening and adjustment process such as excluding stocks with low price or small market value in order to avoid that results might be driven by illiquid or low priced stocks. The similarity of applied methodology among the studies implies that there is consensus how to construct momentum portfolios and measure the returns obtained by momentum trading strategies. Thus, it needs to be considered that the evidence of the momentum essentially depends on the correctness of this methodology.

Furthermore, it could be noticed that momentum profits substantially differ in their magnitude across the studies presented in Table 1. While some of them document monthly

returns well above 1 percent on average, others report considerably weaker returns obtained by momentum investing strategies. At least to some extent, these differences could be attributed to characteristics of momentum portfolios formation. The studies notifying average monthly returns above 1 percent form Winners and Losers portfolios from the most extreme top and bottom 10 percent past performing stocks, whereas strategies which assign 20 percent or 30 percent top and bottom of stocks generate lower returns. Hence, it has become a common view that momentum portfolios constructed with larger quantiles are less profitable than portfolios formed from stocks with the most extreme prior performance.

An overview of studies investigating momentum returns outside the U.S. market is presented in Table 2. It could be noticed that the momentum is investigated in Europe, American countries outside the U.S., Asia and Africa including developed and emerging markets. I further elaborate on several of these studies.

TABLE 2.OVERVIEW OF MOMENTUM RETURNS DOCUMENTED FOR THE INTERNATIONAL STOCK MARKETS

Study Country Period Portfolio

formation

Strategies Return per month Rouwenhorst (1998) Europe (12 countries) 1978-1995 10% 9/1/6 1.30% Rouwenhorst (1999) Emerging markets (20 countries) 1982-1997 30% 6/1/6 0.39% Chui et al. (2000) Asia (8 countries) (not significant) 1975-2000 30% 6/1/6 0.38% Griffin et al. (2003) International (39 countries)

Africa

Americas (excl. U.S.) Asia (not significant) Europe 1926-2000 20% 6/1/6 1.63% 0.78% 0.32% 0.77% Doukas and McKnight (2005) Europe (13 countries) 1988-2001 20%

30%

6/0/6 0.89% 0.73% Hart et al. (2005) Emerging markets (32 countries) 1988-2004 15% 6/0/6 0.74% Chui et al. (2010) International (41 countries) 1980-2003 1/3 6/1/6 0.72% Fama and French (2012) Europe

Asia

1990-2011 30% 12/1/1 0.92% 0.69% Asness et al. (2013) Global stocks

U.K. Europe

Japan (not significant)

1972-2011 30% 12/1/1 0.46% 0.5% 0.68% 0.15% Cakici et al. (2013) Emerging markets (18 countries) 1990-2011 30% 12/1/1 0.86%

Table 2 presents the several studies related to momentum effect and momentum returns outside the U.S. Column 1 shows the information of published studies, columns 2 and 3 report the information about the examined regions and sample periods used in the papers respectively. Column 4 informs about the size of top and bottom stocks’ quantiles which are assigned to form Winners and Losers portfolios. Column 5 shows the length of the portfolio formation, skipping and holding periods. Column 6 reports the average raw returns per month of momentum portfolios. In case study investigates several momentum strategies the most significant one is presented.

Rouwenhorst (1998) study of the momentum effect from an international perspective is one of the earliest to support significant findings in the U.S. stock market. In his study, the author examined 12 (developed) European countries within the time frame of 1978 – 1995 using very similar methodology developed by Jegadeesh and Titman (1993). Rouwenhorst (1998) found that internationally diversified portfolios of medium term past “winners” outperform medium term past Losers portfolio approximately by 1% after adjusting for risk in all 12 European stock markets. Furthermore, Rouwenhorst (1998) also observed that returns are persistent when author controls for market risk or size factors and small stocks exhibit stronger momentum effect than large stocks. This evidence from European countries is consistent with results from the U.S. stock market and confirms that there is a common factor which may drive the returns of momentum trading strategies.

Rouwenhorst (1999) also analysed the momentum effect in 20 emerging markets, based on 1982 – 1997 data. The author concluded that portfolios formed from the top 30 percent of past “winners” and the bottom 30 percent of past “losers” stocks generate on average 0.39 percent returns per month across all 20 markets. Despite the fact that these returns are markedly lower compared to documented findings for the U.S. and developed European markets, according to Rouwenhorst (1999) returns in emerging and developed markets are driven by the qualitatively similar factors. In addition, the study shows that large stocks and growth stocks are outperformed by small stocks and value stocks respectively. Also, it is observed that share turnover has no direct relationship with expected return, although it is strongly cross-sectional correlated with other return factors.

Chui et al. (2000) examined momentum in eight Asian countries and found that the profits of aggregate momentum portfolios on average earn about 0.38 percent per month, however, results showed no significance, except for Hong Kong where they found significantly positive momentum returns. The authors explained that insignificance occurs due to the high volatility of momentum returns in the period of 1997 financial crisis. When authors excluded Japan and the period of crisis from the sample the average momentum returns in the pre-crisis period of 1975 – 1997 became significantly positive in half of the observed countries, but negative after crisis period. Chui et al. (2000) findings suggest that the profitability of momentum strategies is also present in Asia at least to some extent, except for Japan, Indonesia, and Korea. It also suggests that market states might have an influence on momentum effect. It is worth to mention this evidence about the momentum effect in Asia, especially Japan because it is one of only a few countries in the context of international momentum literature where momentum strategies are not remunerative.

Griffin et al. (2003) investigated the momentum effect globally and analysed the relationship between momentum profits and macroeconomic determinants. The authors used data ranging between 1926 and 2000 from 39 countries all over the world. Griffin et al. (2003) found that the zero-cost portfolios of 6-month formation and 6-month holding period on average yield 0.59 percent, 0.77 percent, 0.78 percent, 1.63 percent and 0.32 percent per month for U.S., Europe, Americas (excluding U.S.), Africa and Asia respectively. They noted that these momentum profits are highly statistically significant across the different regions except for Asia. However, they concluded that momentum returns can be explained neither by macroeconomic factors nor business cycle risk. Griffin et al. (2003) presented similar results as to those documented by Chui et al. (200) that momentum returns in Asia do not show any exceptional significance. However, contrary to the results of Chui et al. (2000), in Griffin et al. (2003) study even after removing Japan from the sample, it did not relevantly change the findings and returns obtained by momentum trading strategies are robust during different macroeconomic states.

The momentum effect in the Baltic States stock market has been examined only by a few researchers. For instance, Maniušis and Urba (2007) analysed momentum in the Baltic countries market based on dividend adjusted weekly stock prices data from 2000 to 2006 excluding delisted stocks. Using a methodology closely in line with Jegadeesh and Titman (1993) study, they presented positive and statistically significant monthly average returns for all used combinations of momentum strategies. In their study, the most profitable strategy is found to be 12-month formation and 3-month holding periods generating approximately 0.5 percent return per month. Another study by Avižinis and Pajuste (2007) examined the momentum effect in the Baltic States, Slovenia, Poland, Croatia, and Hungary between the time period of 2002 and 2006 as well using the methodology based on Jegadeesh and Titman (1993). They found that strategy of 6-month formation and 6-month holding periods is the most prominent yielding 3 percent per month on average for Winners-minus-Losers portfolio. Stankevičienė and Gembickaja (2012) also investigated market anomalies such as contrarian and momentum in the Baltic countries. Conversely to previously mentioned studies (Maniušis & Urba, 2007; Avižinis & Pajuste, 2007), Stankevičienė and Gembickaja (2012) concluded that neither momentum nor contrarian anomalies based on 2000-2009 data have tendency to appear in the Baltic States market since momentum as well as contrarian strategies have no consistency in returns.

To sum up, there is a consensus in the literature and among researchers that momentum effect and profitability of momentum trading strategies are present across international equity

markets. Considering the persistence of the momentum effect, various of researchers from the risk and behavioural based perspectives have sought to understand the causes of this anomaly. I will review few attempts to explain the momentum effect.

2.3. Risk-based explanations

The CAPM has received a lot of criticism among many researchers over the years because of the failure to explain anomalous behaviour in stock returns. Thus, researchers have started to formulate new models capable to explain various phenomena which contradict EMH.

As already mentioned earlier, Fama and French (1992) examined a set of anomalies and identified that two easily measured variables, book-to-market equity, and size (market value of equity) provide a great explanatory power in describing the cross-sectional variation in stock returns. In turn, Fama and French (1993) further elaborated on their previous findings and presented three factors model which encompasses two additional risk factors to the market risk factor in the CAPM, namely size (SMB) and value (HML) measured by the book-to-market ratio. Fama and French (1993) suggested that size and value variables might not be clear risk factors themselves, but they might be proxies which represent the influence of common risk factors omitted from the CAPM. The intuition behind SMB and HML factors is that small firms are associated with higher risk because they are more exposed to changes of market conditions and stocks with high book-to-market ratio carry more risk because value firms are more exposed to risks related to financial distress. Apparently, this model has performed better in explaining stock returns than CAPM and has quickly become one of the most prominent and commonly employed asset pricing model. The Fama-French three-factor model is expressed as follows:

𝑅_" − 𝑅_% = a_" + b_" 𝑅₍ − 𝑅_% + 𝑠_"𝑆𝑀𝐵 + ℎ_"𝐻𝑀𝐿 +e_" (1)

Where 𝑅_" − 𝑅_% is return of the portfolio i in excess of a risk free rate, 𝑅₀− 𝑅_% is an excess return of the market, SMB (small minus big) indicates the size premium by capturing a return difference of a small stocks portfolio over a portfolio composed of large stocks, HML (high minus low) indicates a value premium by capturing the return difference of a high book-to-market stocks portfolio over the portfolio composed of low book to book-to-market stocks.

In order to estimate the factor loadings that are interpreted as risk-factor sensitivities (slopes of the time-series regression) the excess return of the portfolio is regressed on the excess return of the market portfolio, size (SMB) and book-to-market (HML) factors portfolios. The

intercept a₁ (alpha) which is also interpreted as the return of the portfolio adjusted for the risk should be zero if model is able to explain returns on securities (Fama & French, 1993).

Fama and French (1996) documented that their three-factor model allows to explain not only most of the cross-sectional variation in average stock returns but also most of the anomalies that have “plagued” the CAPM. Specifically, they found that their model can explain the returns obtained when portfolios are constructed on earnings/price, cash flow/price, and sales growth. More importantly, the Fama-French three factor model provides an explanation of long-term reversals presented by De Bondt and Thaler (1985). However, Fama and French (1996) admitted that their three-factor model fails to capture momentum effect.

Jegadeesh and Titman (2001) also examined momentum returns after risk adjustment by the Fama-French three factors model. Particularly, they regressed monthly average returns of 6-month/6-month strategy’s momentum portfolios on the Fama-French three factors in order to estimate Fama-French alpha and momentum portfolios sensitivities to the Fama-French factors. The time-series regression disclosed that alpha for the zero-cost portfolio is positive and statistically significant. More importantly, they reported that alpha is 1.36 percent and it is even higher than the respective raw return of 1.23 percent. Jegadeesh and Titman (2001) noted that this difference appears because Losers portfolios are more sensitive to the Fama-French three factors.

Consistently with Fama and French (1996) and Jegadeesh and Titman (1993, 2001) findings, Grundy and Martin (2001) also found that risk adjusted momentum returns either by the CAPM or by the Fama-French three factors do not reduce the returns, instead they are higher or very similar to raw returns. However, they noted that even though momentum returns are robust after the adjustment by the Fama-French three factors model provide better explanatory power than the CAPM. Additionally, Grundy and Martin (2001) documented that cross-sectional differences in expected returns and industry effects also could not explain the profitability of momentum strategies.

The results of these studies indicate that momentum returns are robust and persist after the adjustments for risk, suggesting that the U.S. market is not entirely efficient. Although, the proponents of the EMH theory argue that even though anomalies are usually attributed as evidence of market inefficiency they could also be the result of pricing model misspecification. Since attempts to test for market (in)efficiency involve the joint hypothesis problem. Meaning that one must assume an equilibrium asset pricing model for expected returns to define the real returns on the assets. Therefore, it is incorrect to reject rational models because anomalies could

reflect either market inefficiency, an incorrect equilibrium model or both (Durlauf & Blume, 2008).

2.4. Behavioural explanations

The persistence of the momentum effect to risk-based explanations has encouraged to search for alternative ways to explain the causes of this anomaly. Researchers have discovered that people tend to rely on particular heuristics in decision making and judgment. Therefore, various of authors have presented behavioural models based on ideas that investors are not fully rational and “suffer” from different cognitive biases. Under this relatively young approach in the field of finance momentum effect is not an exception and proponents of behavioural perspective argue that momentum profits potentially arise due to the irrationality of investors and psychological biases of information processing.

Daniel, Hirshleifer, and Subramanyam (1998) developed a model based on overconfidence and self-attribution bias which implies that individuals attribute success to their own ability and attribute failure to chances or bad luck. As a consequence of these two psychological biases stock prices overreact to private and underreact to public information. The authors argue that investors overestimate their abilities to analyse information and thus underestimate errors. When investors receive the private signal that supports their forecasts, investors become even more overconfident and as a result, they push assets prices above their fundamental value. Afterward, when public information confirms their private signal investors “suffering” from self-attribution bias, on average, reinforce their overconfidence and this leads to further overreaction, hence the momentum effect. Whereas unfavourable news signals have a little impact to reduce the confidence. Momentum profits are reversed when prices are drawn back to their fundamentals by subsequently revealed public information.

Barberis, Schleifer, and Vishny (1998) presented an alternative behavioural theory to explain the momentum effect and long term reversals. Barberis et al. (1998) model is based on conservatism and the representativeness heuristics related to the underreaction and overreaction respectively. According to the authors, conservatism implies the tendency that investors in the face of new information update their beliefs too slowly. Therefore, when firms announce positive earnings prices rise too little as investors react insufficiently due to conservatism bias. By reason of this initial underreaction, prices are too low compared to their fundamental value, which gives higher average returns in the following periods and in turn generates momentum. Since the momentum effect and long-term reversals are interconnected

Barberis et al. (1998) in their model also argue that the representativeness bias that refer to the tendency that individuals identify certain samples by the properties of the parent population causes long-term reversals. As a consequence of representativeness heuristics, investors might mistakenly anticipate that companies exhibiting consistent and surprisingly good earnings will continue performing well in the future and thus overreact by pushing prices up too high. As a result of too high prices, following returns on average will be too low and hence prices revert to their fundamentals.

Hong and Stein (1999) introduced a different approach without any behavioural biases. Instead, the authors consider two groups of agents with bounded rationality, namely “news watchers” who are able to process only private information and “momentum traders” who ignore the fundamental news and trade based only on historical prices. Hong and Stein (1999) argue that new private information transmits with delay among the news watchers and this, in turn, causes the underreaction. Therefore, prices only gradually reflect the revealed information. Then, the momentum traders observe changes in prices and try to exploit this underreaction. They extrapolate and create excessive momentum by pushing prices above their fundamentals causing the overreaction which is reversed eventually.

The more recent study by Hur and Singh (2016) provided the first direct evidence supporting the assumptions of Jegadeesh and Titman (1993) and behavioural models that the momentum effect is caused by the combination of investors’ underreaction and overreaction to information. The authors assert that slow incorporation of firm-specific information on stock prices and the speed which corrects the errors in pricing play the main role in the returns behaviour of momentum strategies. Hur and Singh (2016) argue that momentum profits could be explained by both underreaction and delayed overreaction to information, although underreaction appears to be more causal in their model than overreaction. The findings of this study support the idea that behavioural approach is capable of providing the explanation of the momentum effect.

Indeed, there are rich and extensive literature documenting existence of the momentum effect across the world and various of studies have observed persistent abnormal momentum returns in international stock markets. It also seems that the explanation of momentum effect is more consistent with a behavioural based approach. However, as already mentioned earlier, studies investigating the momentum effect have been concentrated mainly in the developed rather than emerging markets. The Baltic States are not the exception and also face a limited number of studies in this area. It is also noteworthy that even the existing literature is not unified about the presence of momentum effect in the Baltic countries. While Avižinis and Pajuste (2007)

and Maniušis and Urba (2007) reported the statistically significant momentum returns, Stankevičienė and Gembickaja (2012) argue that the momentum returns have no consistency in the Baltic States stock market. Furthermore, Avižinis and Pajuste (2007) as well as Maniušis and Urba (2007) in their studies examined whether momentum returns could be explained by the CAPM. However, both studies concluded that this model could not explain profitability of momentum investing strategies. Seemingly, there are induced doubts and a gap in the academic literature related to the presence of the momentum effect. Hence, this study aims to explore the existence of the momentum effect in the Baltic States stock market following the methodology of Jegadeesh and Titman (1993). Furthermore, I will also use the Fama and French (1993) three factors model since none of the above-mentioned studies investigated returns of the momentum portfolios in the context of this more advanced model, especially when it comes to the risk adjustment. In turn, this will allow to fill the gap in the academic literature related to the momentum effect existence in the Baltic States stock market and contribute to the existing empirical literature with the new insights about risk-reward based approach in order to explain momentum returns.

3. Methodology and data

3.1. Data

The data used in this study consists of all stocks listed anytime during the sample cover period from the 1st of January, 2000 to 31st of December, 2016 on the Nasdaq Baltic Stock Exchange market (Tallinn, Riga, Vilnius). Since it is plausible that exclusion of delisted stocks could make the results of the study apt to survivorship bias, the sample also includes delisted stocks that were traded in the Baltic Stock market during the observation period. However, some of the delisted stocks with not available data or insufficient data points were removed from the sample. In addition, the Baltic Stock market consists of a relatively small number of stocks, hence I do not use screening factors such as low stock price, market capitalization or liquidity to exclude companies from the sample. Therefore, the total number of 90 companies: 68 listed and 22 delisted companies were included. This study uses broader and more up to date data than other studies which examined the momentum effect in the Baltic countries. In comparison, Maniušis and Urba (2007) used data set consisting of 71 stocks ranging from 2000 to 2006. Avižinis and Pajuste (2007) use even shorter sample period from 2002 to 2006 and Stankevičienė and Gembickaja (2012) used data set from 2000 to 2009. Former studies do not explicitly state the number of stocks used in the studies.

Historical daily stock closing prices adjusted for corporate actions and cash dividends, as well as the market capitalization and book-to-market values, were obtained from the Thomson Reuters Eikon database. However, different kind of data for some companies was not available or missing observations. Therefore, required information was manually collected from the OMX NASDAQ web page (2017). Moreover, daily quotes of the OMX Baltic Benchmark General Index (BBGI) for analysed time period were also retrieved from the OMX NASDAQ web page (2017). As a proxy for the risk-free rate in this study, I use 10-year German Government bond yield, which was obtained from the Deutsche Bundesbank (2017). It is mostly because high-rated sovereign bonds are usually and conventionally considered as free rates. German Government bond yields have been typically employed to represent risk-free rates of euro area for a long period of time due to their high liquidity and low credit risk (ECB, 2014). Additionally, according to recent study by Hooijman (2016) where the author compared risk-free rates based on criteria such as Availability, Intelligibility and Consistency, Hooijman (2016) concluded that the German government bond yields and the Overnight Index Swaps are one of the most suitable proxies as risk-free rates for valuation purposes. All the gathered data for the Baltic States are denominated in euro currency.

3.2. Methodology

The methodology of this study is closely in line with the paper introduced by Jegadeesh and Titman (1993). As previously stated, the formation of momentum portfolio is based on the selection of stocks according to their past performance in previous 3, 6, 9 or 12 months (J months) and each portfolio is held for 3, 6, 9, or 12 months (K months). Thus, this allows to analyse 16 different combinations of J-month/K-month strategies as shown in Table 3.

TABLE 3.OVERVIEW OF MOMENTUM STRATEGIES Formation period (J) H ol di n g pe ri od (K)

3 months 6 months 9 months 12 months

3 months 3/3 6/3 9/3 12/3

6 months 3/6 6/6 9/6 12/6

9 months 3/9 6/9 9/9 12/9

In order to increase the power of statistical tests, analysed strategies in this study include portfolios with overlapping holding periods. Therefore, portfolios are constructed each month, meaning that in any given month t, various J-month/K-month strategies hold series of portfolios, which were selected in the current month and in the previous K – 1 months. This results in revising the holdings every month. While with the non-overlapping portfolios investor holds only one portfolio at a time, giving less number of observations. However, this approach is often considered for private investors as it results in lower transaction costs due to less trading. Table 4 illustrates an example of 3/3 strategy with the non-overlapping and overlapping investment techniques. Additionally, following the methodology of Jegadeesh and Titman (1993) I also skip a week between portfolio ranking and holding periods in order to avoid microstructure distortions such as a lagged reaction effects, a bid-ask spread or a price pressure.

TABLE 4.THE EXAMPLE OF 3/3 STRATEGY WITH NON-OVERLAPPING AND OVERLAPPING HOLDING PERIODS

Month Non-overlapping Overlapping

January F orm at ion F orm at ion February F orm at ion March F orm at ion April H ol di ng F orm at ion H ol di ng F orm at ion May H ol di ng F orm at ion June H ol di ng F orm at ion July H ol di ng F orm at ion H ol di ng F orm at ion August H ol di ng September H ol di ng October H ol di ng F orm at ion H ol di ng November December

Moreover, the basic idea is then to form equally weighted zero-cost portfolios in each month, while buying the top decile Winners portfolios and short selling the bottom decile Losers portfolios in order to finance the purchase. These portfolios are then held for the K-month

period. More specifically, construction of the momentum portfolios starts by ranking the stocks each month based on their monthly returns in the past J-month. Since it is not explicitly stated which method are used to calculate monthly returns in Jegadeesh and Titman (1993) study, I use continuously compounded method which is the most common one when analysing stock returns (Campbell et al. 1997). Hence, for each company continuously compounded monthly returns are calculated according to the formula:

𝑅_".3 = 𝑙𝑛 67,9

67,9:;

where 𝑃_",3 is the closing price for the calendar date t for the company i and 𝑃_",3 is the closing price for the previous calendar date t for the company i. All the stock prices are already adjusted for dividends. Then, cumulative returns over the observed time period of J-month are calculated as follows:

𝐶𝑅",3 𝐽 = ?3@A𝑅",3 (3)

As proposed by Jegadeesh and Titman (1993), I also use 10 percent top and bottom breakpoints. Thus, based on their cumulative returns stocks are ranked into deciles, where the top 10 percent consists of past wining stocks (“winners”) and bottom 10 percent of past losing stocks (“losers”). In turn, this allows to calculate the return of the equally weighted Winners and Losers portfolios which consist of an equal number of N stocks:

𝐶𝑅_6,3 = _BA B 𝐶𝑅_",3(𝐽)

3@A (4)

Furthermore, the return of zero-cost momentum strategy portfolio denoted 𝐶𝑅_(,3 (𝐽) is then calculated by subtracting the return of Loser portfolio denoted 𝐶𝑅_E,3 (𝐽) from return of Winner portfolio denoted 𝐶𝑅_F,3 (𝐽):

𝐶𝑅(,3 𝐽 = 𝐶𝑅F,3 𝐽 − 𝐶𝑅E,3 𝐽

The monthly return averages of all momentum strategies are then calculated using the average of all zero-cost equally weighted portfolios constructed during the analysed period of time and divide it by the respective length of holding period:

𝑀𝑅 = _GA H 𝐶𝑅_(,3(𝐽)

3@A (6)

Also, following the Rouwenhorst (1999) study in emerging markets, when portfolios are constructed from 30 percent of past winning and past losing stocks respectively, I also perform the calculations with increased size of the portfolios using 30 percent top and bottom breakpoints. All the computations are carried out using STATA software and are completed for all combinations of momentum strategies and portfolios’ sizes.

3.3. Methodology used for Fama-French three factors model

In order to examine whether risk-based model is able to justify the observed momentum return patterns in the Baltic States stock market as a compensation for bearing additional risks or they imply market inefficiency, I employ the Fama-French three factors model. Therefore, following the methodology of Fama and French (1993, 1996) all stocks based on their market capitalization are divided into two groups. Stocks above the median are allocated to Big (B) group and stocks below the median to Small (S) group. Afterward, stocks are sorted into three groups according to their book-to-market ratio. 30 percent of stocks with the lowest ratio and 30 percent of stocks having the highest ratio are allocated to groups Low (L) and High (H) respectively. The rest 40 percent is allocated to group Medium (M). As a result of these intersections, six portfolios are created as illustrated in Table 5.

TABLE 5.ILLUSTRATION OF 6 PORTFOLIOS CONSTRUCTED ON SIZE AND BOOK-TO-MARKET

30% 40% 30%

B/M Size

H M L

50% S Small Value Small Neutral Small Growth

50% B Big Value Big Neutral Big Growth

Then, SMB (small minus big) and HML (high minus low) factors are computed according to formulas (7) and (8) respectively:

𝑆𝑀𝐵_G,3 = A_I 𝑆𝑚𝑎𝑙𝑙 𝑉𝑎𝑙𝑢𝑒_G,3+ 𝑆𝑚𝑎𝑙𝑙 𝑁𝑒𝑢𝑡𝑟𝑎𝑙_G,3+ 𝑆𝑚𝑎𝑙𝑙 𝐺𝑟𝑜𝑤𝑡ℎ_G,3 − A_I 𝐵𝑖𝑔 𝑉𝑎𝑙𝑢𝑒_G,3+ 𝐵𝑖𝑔 𝑁𝑒𝑢𝑡𝑟𝑎𝑙_G,3+ 𝐵𝑖𝑔 𝐺𝑟𝑜𝑤𝑡ℎ_G,3 (7)

where 𝑆𝑀𝐵_G,3is the difference between average returns of three small stocks and three big stocks portfolios.

𝐻𝑀𝐿_G,3 = A_W 𝑆𝑚𝑎𝑙𝑙 𝑉𝑎𝑙𝑢𝑒_G,3+ 𝐵𝑖𝑔 𝑉𝑎𝑙𝑢𝑒_G,3 − A_W 𝑆𝑚𝑎𝑙𝑙 𝐺𝑟𝑜𝑤𝑡ℎ_G,3+ 𝐵𝑖𝑔 𝐺𝑟𝑜𝑤𝑡ℎ_G,3 (8)

where 𝐻𝑀𝐿_G,3 is the difference between average returns of two high B/M (value) portfolios and two low B/M (growth) portfolios.

Then, to examine whether the momentum returns remain after risk adjustment, I will regress the returns obtained by momentum trading strategies on the Fama-French three factors model. Particularly, if the model is able to explain momentum returns the alpha (intercept) of time-series regressions should be indistinguishable from zero and insignificant. Furthermore, factor sensitivities (slopes of the time series regression) and R2 _{values will show whether risk factors} related to systematic risk, book-to-market ratio and size capture momentum returns.

4. Empirical results

4.1. Momentum returns

In this section, the study investigates the presence of momentum effect in the Baltic States stock market. Therefore, this section presents the returns of the different momentum strategies described above based on 16 years of data. All stocks with available historical stock prices data in the J months before the formation of portfolios time are included in the sample. In line with the base study of Jegadeesh and Titman (1993), Table 6 shows the decomposed average returns for the Winners, Losers and zero-cost portfolios with overlapping holding periods without skipping a week as well as with one week gap between the portfolio formation and holding periods.

TABLE 6.RAW RETURNS OF MOMENTUM PORTFOLIOS Panel A

Portfolios formed without a week gap

Panel B Portfolios formed with a week gap

J K= 3 6 9 12 K= 3 6 9 12 3 Winners 0.0091 0.0099* 0.0099* 0.0080* 0.0077 0.0061 0.0074 0.0068 (1.5246) (1.9333) (2.1072) (1.7995) (1.2467) (1.2114) (1.4913) (1.3968) 3 Losers 0.0069 0.0060 0.0055 0.0052 0.0058 0.0057 0.0056 0.0055 (1.2267) (1.1205) (0.8378) (0.9619) (1.5077) (1.4294) (1.3627) (1.3306) 3 zero-cost 0.0021 0.0038 0.0043 0.0028 0.0019 0.0004 0.0018 0.0013 (0.3519) (0.9260) (1.4372) (0.8187) (0.3367) (0.0261) (0.4677) (0.3410) 6 Winners 0.0061 0.0076 0.0099** 0.0056 0.0066 0.0093 0.0085* 0.0074 (0.9688) (1.4102) (2.0132) (1.1777) (1.0752) (1.7401) (1.7068) (1.5177) 6 Losers 0.0052 0.0040 0.0083 0.0054 0.0058 0.0063 0.0052 0.0047 (0.8149) (0.5858) (1.2642) (0.8424) (1.4723) (1.5877) (1.3235) (1.1723) 6 zero-cost 0.0009 0.0036 0.0016 0.0002 0.0008 0.0030 0.0033 0.0027 (0.1339) (0.5727) (0.2762) (0.0134) (0.0819) (0.7305) (0.9169) (0.8060) 9 Winners 0.0091 0.0092* 0.0083* 0.0053 0.0070 0.0111** 0.0103** 0.0066 (1.5946) (1.7850) (1.6657) (1.1341) (1.2614) (2.0866) (2.0580) (1.3201) 9 Losers 0.0045 0.0057 0.0067 0.0069 0.0055 0.0065* 0.0066* 0.0051 (0.5840) (0.7511) (0.9726) (0.9726) (1.4286) (1.6714) (1.6653) (1.2768) 9 zero-cost 0.0045 0.0034 0.0015 -0.0016 0.0015 0.0046 0.0032 0.0015 (0.5789) (0.4628) (0.2129) (-0.2617) (0.2623) (0.4628) (0.9263) (0.3154) 12 Winners 0.0069 0.0092* 0.0062 0.0059 0.0080 0.0081 0.0065 0.0027 (1.2839) (1.8128) (1.0994) (1.0743) (1.4016) (1.5052) (1.2510) (0.5418) 12 Losers 0.0101 0.0081 0.0097 0.0097 0.0062 0.0058 0.0059 0.0048 (1.3025) (1.0441) (1.2945) (1.3718) (1.4784) (1.4631) (1.3487) (1.0661) 12 zero-cost -0.0032 0.0010 -0.0035 -0.0038 0.0018 0.0023 0.0006 -0.0021 (-0.4099) (0.1366) (-0.6487) (-0.7363) (0.4382) (0.4780) (0.1721) (-0.6785) Notes: This is an overview of decomposed momentum portfolios formed with different investment strategies from

the top 10 percent and bottom 10 percent stocks with overlapping holding periods. The momentum portfolios in Panel A are formed immediately after the formation period. The momentum portfolios in Panel B are formed with a week gap between the formation period and the holding period. The t-statistics are reported in parentheses. *** Significant the at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level. Source: Author calculations.

Panel A shows the results of portfolios constructed immediately after the formation period. The results demonstrate that all momentum portfolios show statistically insignificant returns. Although, we can see in Panel A that 12 out of 16 zero-cost portfolios generate on average positive returns in the Baltic States stock market. Presumably, these returns are mainly driven by the long position in Winners portfolios rather than short side Losers portfolios. Jegadeesh and Titman (1993) also noted that the most of the excess momentum returns come from the Winners side of the transaction. Momentum profits would be increased if Losers portfolios

give relatively small or even negative returns, however the results do not show negative returns for the sell side portfolios neither in Panel A nor in Panel B. The most successful zero-cost portfolios are based on 3-month formation and 6-month and 9-month holding periods, also strategies of 6/6 and 9/3 show similar results. The average monthly returns for these strategies vary from 0.36 percent to 0.45 percent.

Looking closer at Table 6 and Figure 1 which illustrates the dynamics of percentage returns of decomposed portfolios, it could be seen that all Winners portfolios provide positive returns and some of them are statistically significant. For instance, Winners portfolios based on strategies of 3/9 or 6/9 generate almost 1 percent per month on average and are significantly different from zero. In comparison, it could be noticed that average returns for Losers portfolios vary in greater amplitude than Winners portfolios, from 0.4 percent to more than 1 percent per month, although none of the sell side portfolios have generated statistically significant returns since individual t-statistics are insufficiently low for all Losers portfolios.

FIGURE 1.GRAPHICAL OVERVIEW OF MOMENTUM PORTFOLIOS RETURNS IN PANEL A

To put it into perspective, Jegadeesh and Titman (1993) reported statistically significant returns of momentum portfolios in a U.S. stock market way higher than observed returns in this study. For instance, returns for momentum portfolios constructed immediately after the formation period in Jegadeesh and Titman (1993) study vary from 0.58 percent to 1.31 percent (excluding the most extreme and insignificant 3-month/3-month case). In addition, Winners

-0.60% -0.40% -0.20% 0.00% 0.20% 0.40% 0.60% 0.80% 1.00% 1.20% 3x3 3x6 3x9 3x12 6x3 6x6 6x9 6x12 9x3 9x6 9x9 9x12 12x3 12x6 12x9 12x12 A ve ra ge M ont hl y R et ur n Portfolio Strategies

Momentum portfolios without one week gap

fluctuate from 1.4 percent to 1.92 percent and Losers vary from 0.6 percent to 1.08 percent per month. In comparison, zero-cost and Winners portfolios in Jegadeesh and Titman (1993) study provide about two times higher returns than in this study whereas Losers portfolios give similar results.

Panel B shows the average returns of portfolios constructed with a one week gap after the formation period. Likewise, the results in panel A, none of the momentum portfolios generate statistically significant returns, however, we can see from the Table 6 that results are different to some extent. The findings show that equally weighted portfolios formed from the top decile winners’ stocks on average outperform the Losers portfolios in all the cases, except for the strategy based on the 12-month formation period followed by the 12-month holding period with a week gap between these periods. On the one hand, skipping a week diminishes the average monthly returns of all zero-cost portfolios based on 3-month formation period regardless of the holding period. On the other hand, returns of the zero-cost portfolios based on the longer formation periods increase notably and now 15 out of 16 portfolios generate positive returns. As we can see from the Table 6 and Figure 2 the most successful strategies select stocks based on a 6-month and 9–month formation and hold them for both 6-month and 9-month periods respectively. The most profitable is a 9-month formation and a 6-month holding strategy which on average yields 0.46 percent per month. It is a very similar result comparing to Maniušis and Urba (2007) recorded most prominent return in their study of 0.5 percent per month for 12/3 strategy when the portfolios are formed from the top and bottom 10 percent stocks with a week gap. Although, it is nothing similar to Avižinis and Pajuste (2007) study, where they reported 3 percent average return per month for a 6/6 strategy zero-cost portfolio with the same formation technique.

From the decomposed portfolio point of view, Winners portfolios as in the previous case provide positive returns for all strategies and some of them, for instance, 9/6 and 9/9 appears to be statistically significant. The Figure 2 shows that the average monthly returns for Winners portfolios vary from 0.61 percent to 1.11 percent (excluding the extreme case of 12/12 strategy). Delaying the portfolio formation, most notably diminishes the Winners portfolios returns based on 3 months ranking period, whereas returns of Losers portfolios, especially decrease for strategies based on 12-month formation. Also in this case, Losers portfolios based on 9-month ranking and 6-month, as well as 9-month investment periods show slight significance at the 10 percent level. The average monthly returns for Losers portfolios vary from 0.47 percent to 0.66 percent. It seems that in both cases the long side of the zero-cost

portfolios is relatively successful. However, the high returns of sell side portfolios diminish returns for Winners-minus-Losers portfolios.

Now comparing returns obtained by momentum trading strategies in this study with developed European markets analysed in Rouwenhorst (1998) research, the magnitude of the returns, again, differs quite strongly. For instance, Rouwenhorst (1998) reported returns for momentum portfolios formed in line with the methodology of Jegadeesh and Titman (1993), although skipping a month between formation and holding periods instead of a week, vary from 0.51 percent to 1.45 percent per month. Returns for the Winners portfolios fluctuate from 1.76 percent to 2.09 percent and monthly returns for the Losers portfolios vary from 0.64 percent to 1.25 percent and also all portfolios show high statistical significance in Rouwenhorst (1998) study.

FIGURE 2.GRAPHICAL OVERVIEW OF MOMENTUM PORTFOLIOS RETURNS IN PANEL B

Indeed, the returns of momentum portfolios in this study are considerably lower than those documented by Jegadeesh and Titman (1993, 2001), Rouwenhorst (1998) or Griffin et al. (2003) for developed markets. The average monthly returns and statistical insignificance seem more likely to study by Chui et al. (2000) in the Asian countries’ markets since they also reported very low (0.38 percent per month on average for 6/6 strategy portfolios) and insignificant returns. However, it could be argued that the insignificance of the results in this study potentially might be caused by the relatively small number of companies in the Baltic States stock market and hence, a small number of stocks in the momentum portfolios. As a

-0.40% -0.20% 0.00% 0.20% 0.40% 0.60% 0.80% 1.00% 1.20% 3x3 3x6 3x9 3x12 6x3 6x6 6x9 6x12 9x3 9x6 9x9 9x12 12x3 12x6 12x9 12x12 A ve ra ge M ont hl y R et ur n Portfolio Strategies

Momentum portfolios with one week gap

result of both Winners and Losers small portfolios’ size, it is possible that portfolios are not quite well diversified and even the single stock might influence portfolio performance and also a large standard error in statistical tests could be caused.

Therefore, as it was already stated in methodology part, I also perform analysis with increased size of portfolios as in Rouwenhorst (1999) study, where author analysed momentum in emerging markets and formed Winners and Losers portfolios from the top and bottom 30 percent of stocks respectively.

TABLE 7.RAW RETURNS OF MOMENTUM PORTFOLIOS WITH INCREASED PORTFOLIO SIZE

J K= 3 6 9 12 3 Winners 0.0086* 0.0088** 0.0098*** 0.0082** (1.9600) (2.1474) (2.6047) (1.8059) 3 Losers 0.0059 0.0066* 0.0085** 0.0074* (1.5312) (1.6802) (2.3136) (1.8203) 3 zero-cost 0.0024 0.0021* 0.0013 0.0008 (1.3864) (1.7216) (1.1655) (0.7434) 6 Winners 0.0078* 0.0094** 0.0094** 0.0080 (1.0752) (2.2486) (2.2836) (1.9060) 6 Winners 0.0062 0.0066* 0.0064* 0.0055 (1.4723) (1.7194) (1.6562) (1.3993) 6 zero-cost 0.0014 0.0028* 0.0030* 0.0025 (0.1339) (1.6673) (1.8278) (1.5710) 9 Winners 0.0093** 0.0111*** 0.0095** 0.0081* (2.1351) (2.6300) (2.2382) (1.8743) 9 Losers 0.0078** 0.0079** 0.0065* 0.0057 (2.0423) (2.0643) (1.6605) (1.4268) 9 zero-cost 0.0015 0.0032* 0.0030* 0.0024 (0.5934) (1.8286) (1.6636) (1.3801) 12 Winners 0.0091* 0.0089** 0.0098** 0.0073* (2.1294) (2.1111) (2.4584) (1.6949) 12 Losers 0.0064** 0.0063 0.0070* 0.0060 (1.6981) (1.6286) (1.9461) (1.5040) 12 zero-cost 0.0027 0.0026 0.0028 0.0013 (1.3023) (1.3113) (1.4416) (0.6750)

Notes: This is an overview of decomposed momentum portfolios formed with different investment strategies from the top and bottom 30 percent stocks with overlapping holding periods and skipping a week between the formation period and the holding period. The t-statistics are reported in parentheses. *** Significant the at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level. Source: Author calculations.

Table 7 presents the results of such technique in the Baltic stock market. As in the previous cases in order to increase the number of observations and power of statistical tests I use overlapping holding periods. Also, I skip a week between the formation and holding periods.

As stated in the Rouwenhorst (1999) study this kind of sorting somewhat attenuates the overall momentum returns. However, now all the zero-cost portfolios provide positive returns. Winners outperform Losers portfolios in all the cases. We can now observe that some of the momentum portfolios using the top and bottom 30 percent breakpoints show slight statistical significance at 10 percent level. Specifically, Winners-minus-Losers portfolios of 3/6, 6/6, 6/9, 9/6 and 9/9 strategies. The most successful zero-cost portfolio strategy as in the previous case when portfolios are formed with the top and bottom 10 percent breakpoints selects stocks based on 9 months past returns and then holds the position for 6 months, skipping a week in-between ranking and investment periods. Although, in this case, 9/6 strategy gives lower average monthly returns reaching 0.32 percent compared to 0.46 percent, however, now its t-stat is 1.83. This strategy’s returns are very similar to average returns of 0.39 percent per month for 6/6 strategy documented by Rouwenhorst (1999) for the emerging markets. The dynamics of decomposed portfolios in Figure 3 show that average monthly returns for Winners and Losers portfolios now vary in narrower amplitude than in the previous cases. They now fluctuate from 0.73 to 1.11 and from 0.55 to 0.85 for the Winners and Losers portfolios respectively. Also, a higher number of separate Winners and Losers portfolios show a strong statistical significance.

FIGURE 3.GRAPHICAL OVERVIEW OF MOMENTUM PORTFOLIOS RETURNS IN TABLE 7

0.00% 0.20% 0.40% 0.60% 0.80% 1.00% 1.20% 3x3 3x6 3x9 3x12 6x3 6x6 6x9 6x12 9x3 9x6 9x9 9x12 12x3 12x6 12x9 12x12 A ve ra ge M ont hl y R et ur n Portfolio Strategies

Momentum portfolios with one week gap