UNIVERSITY OF GRONINGEN
Faculty of Economics and Business
MSc Business Administration, Specialization Finance
RECENT EVIDENCE ON THE EXISTENCE OF
STOCK MARKET SEASONALITIES IN THE
EUROPEAN UNION COUNTRIES
Supervising professor:
Prof. Dr. Lammertjan Dam
Author:
Andreas Papazian (s1951661)
2011-[2]
Abstract:
This paper examines four seasonal anomalies in stock market returns, across 27 European
countries, from 2000-2010: day of the week, month of the year, turn of the month and holiday
effect. The main research question is whether the previously documented patterns still exist
across this more recent time frame, or these seasonal effects have perished over time? Using
a large time series dataset within an OLS framework, I find evidence of a strong seasonality
in stock market returns in most of the countries studied, opposing the efficient market
hypothesis. Some of the seasonal patterns strongly persisted over time, i.e. holiday effect and
turn of the month effect, while those which have been intensively acknowledged in literature,
i.e. Monday effect and January effect, appear to have palled into insignificance. At the same
time, I also find that new and less documented stock market seasonalities are present
throughout the time period studied, i.e. April effect and September effect.
[3]
Table of contents
1. Introduction
2. Theoretical framework
2.1.
Efficient Market hypothesis
2.2.
The day of the week effect
2.3.
The month of the year effect
2.4.
The turn of the month effect
2.5.
The holiday effect
3. Research design
3.1.
Methodology
A. The day of the week effect and month of the year effect
B. The turn of the month effect
C. The holiday effect
3.2.
Data collection
4. Empirical results
4.1.
The day of the week effect
4.2.
The month of the year effect
4.3.
The turn of the month effect
4.4.
The holiday effect
[4]
1. Introduction
The evidence of a growing number of anomalies in the stock market returns casts doubt on the widely accepted efficient market hypothesis (EMH), initially developed by Eugene Fama in the 1960s. The efficient market hypothesis states that at any given time, asset prices fully reflect all available information. Therefore, past price movements cannot be used to predict future price movements as they follow an unpredictable “random walk”. Since the 1960s, numerous studies have examined the major equity markets to see whether stock market returns can be predicted through the existence of any seasonal patterns.
These studies bring important evidence on the existence of stock market seasonalities, although few of them address these issues on a large international scale. Moreover, there is also some recent evidence on the diminishing effect of these seasonalities (Chong, Hudson, Keasey & Littler, 2005), although further investigation, using more recent data from a large number of countries, needs to be provided in order to have a clearer image on these findings.
The diminishing effect of the seasonal anomalies might have been caused by the extensive academic attention during the last decades, which have led to intensively exploit these patterns (Moosa, 2007). Moreover, the periods of underperformance and losses which have occurred across the last decades might have incentivized investors to abandon these long term strategies and develop new ones.
Therefore, my thesis is primarily concerned with this documented transitory nature of seasonalities. The main research question can be stated as follows: Is there still evidence of the stock
market seasonalities at an international scale during recent years, or these effects diminished over time?
In addressing this issue, I examine the stock market seasonalities in 27 European countries, members of the European Union, using a large set of time series data from 2000-2010.
The present research brings some important contributions to the related academic literature. In answering the research question, I test for the presence of the stock market anomalies on a larger international scale compared to prior studies, using a more recent database. Moreover, I investigate whether the previously reported anomalies have persisted over time or there are any other uncovered new patterns which have emerged in the recent years. By comparing the countries in my dataset, I show whether the variables which contribute to an anomaly in, say, Italy, are also important in other countries. Hence, this is of great importance, since studies which focus on only few countries may bring an incorrect interpretation on the results.
[5]
& Wei Xu, 2006) and the holiday effect (Lakonishok & Smidt, 1988; Ariel, 1990; Meneu & Pardo, 2002).
My results bring interesting international evidence on the behavior of the documented patterns across a more recent time frame, showing that some of these patterns have persisted, others have diminished, while at the same time new ones have appeared to have emerged.
Thus, I find evidence that the intensively documented Monday effect seems to have paled into insignificance over the recent years. I find no strong negative returns on Mondays, unlike most previous findings. On the other hand, Monday returns are strongly influenced by the mean returns of the previous week, i.e. twist of the Monday effect, seemingly explained by the fact that weekends do not bring any additional information on the market. At the same time, other two strong patterns have emerged: negative Tuesday and positive Friday returns, conclusively bringing an international confirmation to the studies which have previously acknowledged them as local seasonalities (Agrawal & Tandon, 1994; Jaffe & Westerfield, 1985).
Another striking result is the absence of the most prevalent monthly pattern, i.e. January effect, in several countries. Instead, I now find a significant April effect, for which there is yet little evidence provided. Moreover, the results casts doubt on the tax-loss selling hypothesis (Reinganum, 1983), used as the main explanation for the January effect, as most of the countries studied exhibit a tax year ending in December.
However, other seasonal patterns seem to be well preserved across time. I find strong evidence on the existence of large returns before holidays, i.e. holiday effect, which are consistent with most of the previous studies. The results also show supporting evidence for a significant turn of the month effect. In most of the countries studied, abnormally large returns occur around the turn of the month interval, in particular the last and the first day of the month exhibiting the highest mean returns.
The remainder of this paper is organized as follows. Section 2 gives an overview of theoretical and empirical studies on this topic and raises the hypothesis to be tested. Section 3 contains the methodology used and the data description. Section 4 presents the results and section 5 concludes.
2. Theoretical framework
2.1. Efficient Market Hypothesis (EMH)
The Efficient Market Theory was developed by Fama (1960) and states that stock market returns are random and do not have any periodic pattern and thus, no excess return can be earned by investors.
[6]
1) the Weak Form of the EMH, in which stock prices reflect all the information from the historical prices; excess returns cannot be earned by applying the “pattern” of historical prices, since there is no pattern;
2) the Semistrong Form of the EMH, in which share prices reflect all information that is publicly available in the market, thus the prices adjust to new information rapidly and correctly and investors cannot obtain private benefits by trading on that information;
3) the Strong Form of the EMH, in which stock prices include all available information, both public and private, so no one can earn excess returns.
The EMH is associated with the idea of a “random walk,” which is a term heavily used in the finance literature to characterize a price series where all subsequent price changes represent random departures from previous prices. Consequently, prices fully reflect all the available information and even uninformed investors buying a diversified portfolio will attain a rate of return as generous as that achieved by the skilled traders.
However, researchers have long been asking themselves if the market efficiency is per se testable. Fama (1970) says that we can only test whether information is properly reflected in prices in the context of a pricing model that defines the meaning of “properly”. The caveat is that any anomalous evidence on the behavior of the returns can be due equally to market inefficiency or because a bad model of market equilibrium has been assumed.
Event study literature indicates that on average stock prices adjust quickly to information about investment decisions, dividend changes, and changes in capital structure and corporate-control transactions. Hence, prices adjust efficiently to firm-specific information.
On the other hand, the new research on private information states that there are insiders who possess private information which leads to abnormal returns (Jaffe, 1974(b)).
Interpreting the abnormal returns is subject to joint hypotheses issues, because these returns may be a result of market inefficiency, a bad model of market equilibrium or problems in the way the model is implemented. This joint hypothesis problem means that market efficiency as such can never be rejected.
This paper examines the attacks on the EMH and the belief that stock prices are partially predictable. One pattern observed on stock markets which opposes the EMH is the calendar effect, i.e. stock market seasonalities. Calendar effects may be defined as the tendency of financial asset returns to display systematic patterns at certain times of the day, week, month or year (Brooks, 2008). As a result, investors could develop trading strategies and gain abnormal profits based on such patterns.
2.2. Day of the week effect
Numerous studies have documented that average returns are significantly negative on Mondays and are abnormally large on Fridays. These findings contradict the calendar time
[7]
equal to three times that for the other weekdays. Moreover, the explanation for the day of the week effect opposes the efficient market hypothesis.
French (1980) provides a justification for the negative Monday returns and finds that they are caused by a week-end effect and not by a general “closed market effect”. However, the study is restricted to the U.S. stock market, for the time period from 1953 to 1977. The findings contradict the
trading time hypothesis, in which the expected return of the portfolio is the same for each trading day,
as well as the calendar time hypothesis. Therefore, an investor can profit from the market inefficiency by delaying the purchases until Monday and executing the sales on Fridays. Nevertheless, this advantage cannot be a permanent practice because the transaction costs can strongly affect the returns.
An interesting study about the day of the week anomaly is put forward in Jaffe and Westerfield’s (1985) paper. They find significantly lower mean returns on Tuesdays in Japanese and Australian stock markets and argue whether these returns are merely a reflection of a world-wide Monday effect or a local Tuesday return seasonality. Computing the correlation with the returns from foreign stock markets they do conclude there is no possible influence between them.
The diminishing Monday effect in the 80’s is also mentioned by Smirlock and Starks (1986). In addition to other studies, they check whether the previous week’s returns have any influence on Monday and Tuesday returns and find that if the market declined the previous week, this causes a negative Monday effect. Yet, the previous week’s market returns do not seem to affect Tuesday’s returns.
Lakonishok and Smidt (1988) find evidence of a higher rate of return on the last trading day of the week, whether last day is Friday or Saturday. Even when Friday is not the last trading day of the week, it still has a relatively high rate of return, possibly because Saturday was a short trading day. They also adjust the returns for dividends and conclude that the adjustment does not lead to any changes in the conclusions about the return seasonality and further research is needed in order to test other possible explanations for the seasonality.
A different possible explanation of Monday and Friday anomalous returns is provided by Fortune (1991). He suggests that companies and governments make public good news on weekdays, when markets are open and announce bad news mainly after the close on Friday, as investors cannot react until the Monday opening.
[8]
Caro (2006) tests for the day of the week effect in 12 countries, for a more recent time frame, from 1997 to 2004. They do not find evidence of a day of the week effect in most European markets since the average return for each day does not differ significantly from the other days of the week. However, the study reveals a possible seasonality in conditional volatility. Monday’s and Thursday’s, as well as Tuesday’s and Friday’s volatility is higher than the Wednesday’s volatility, thus these days appear to be more uncertain. Yet, the seasonality in volatility cannot be attributed to seasonality in returns for the up-mentioned days of the week, in most European markets.
Galai, Kedar-Levy and Schreiber (2008) check for any influence of the outliers on the stock return seasonalities. They stumble on a severe influence of a small number of outliers in the empirical estimation of the day of the week effect in their U.S. daily data over the 1980-2002 time frame. Consequently, the standard deviations of daily returns across weekdays declined because of this trimming.
Prokop (2010), analyzing the main stock indices from the mid 1969s to December 2007 from Germany and U.S., concludes that the day of the week effect seems to have faded over time. Moreover, he finds evidence of the reverse effect during the 1990s, one possible explanation for it being that market anomalies have vanished over time mainly due to the trading strategies developed, as a consequence of the massive academic research being published.
In order to address the main research question and based on the previous research conducted, my thesis will test several hypotheses for each of the stock market seasonalities previously discovered.
The following hypotheses concerning the day of the week effect are therefore examined: H0: Monday return is negative and lower compared to the other days of the week; H1: Tuesday return is negative;
H2: Friday return is positive and larger compared to the other days of the week;
H3: Monday return is larger when the mean return on the previous week was positive than when it was negative;
2.3. Month of the year effect
The month of the year effect refers to a seasonal pattern of stock returns occurring in a particular month of the year. One of the most prevalent monthly patterns is the so-called „January effect”, showing positive and larger returns in January, relative to the other months.
[9]
The research by Brown, Keim, Kleidon and Marsh (1983) reveals that during the period from 1931 to 1979 there is evidence of a significant January seasonal. Moreover, they emphasize that the magnitude of the abnormal January returns is not sensitive to variations in the tax rate. Therefore, tax-loss selling hypothesis fails to explain the month of the year effect in Australian stock market returns. Gultekin and Gultekin (1983) find December-January tax year seasonality for most of the countries studied and a March-April tax seasonal in the UK.
Cadsby (1989) brings evidence and firstly documents the existence of the so called Mark Twain effect, which consists in significantly lower average returns on October compared to the rest of the months.
Seyhun’s (1993) extensive study indicates that, for the period 1926 to 1991, January returns were too high to be equilibrium returns. The possible explanations provided for the anomalous returns are tax-loss selling or portfolio rebalancing effects, which are inconsistent with market efficiency.
Agrawal and Tandon’s (1994) evidence shows large and positive January returns and low returns in December in most of the 18 countries other than U.S. studied. The evidence from the eighteen countries examined supports the tax-loss selling hypothesis in most countries with a December tax-year and the April seasonality in the UK, which has a March tax-year. The evidence presented also confirms that the month of the year effect is not confined to small firms as in the U.S.
Moosa (2007) investigates the month of the year effect on the U.S. stock market, using monthly data of Dow Jones Industrial Average over the period 1970 to 2005. There is robust evidence showing that the positive January effect has paled into insignificance while negative July effect has become prominent. The economic reasoning for the diminishing January effect would be that traders become aware of financially significant anomalies, intensively exploiting them. The growing July effect might be explained by the selling pressure associated with the summer holiday season in the northern hemisphere, when investors sell stocks, raising cash to finance their holidays. Moreover, fund managers cautiously reduce their market risk while they are not closely watching their portfolios during their vacations.
The influence of the outliers in the mean rates of return severely affects the empirical results of the month of the year effect as well, by the fact that January’s mean returns turn from insignificant to significantly positive after controlling for outliers, as showed by Galai, Kedar-Levy and Schreiber (2008).
[10]
The present study will test the following hypotheses concerning the month of the year effect: H4: January return is positive and larger relative to December;
H5: December return is lower compared to the monthly average;
H6: October return is negative and lower compared to the monthly average.
2.4. Turn of the month effect
Another seasonal anomaly which has been thoroughly studied is the turn of the month effect, which consists in large, positive mean stock returns during the last few days of the month and the first days from the beginning of the following month.
Lakonishok and Smidt (1988) investigate for the turn of the month effect using the returns of the Dow Jones Industrial Average over the years 1897-1986. The turn of the month effect is studied using a four-day interval, between the last trading day of the month and the next three days of the following month and the findings show that, on average, the four-day return exceeds the return over the entire month. The results are, in general, consistent across the major sub-periods of the 90 years of U.S. daily data.
Ogden (1990) explains the turn of the month anomalies by the fact that at the end of the month investors receive the compensations from employment, dividends and interest and they usually invest it at the beginning of the following month, equity prices being pushed upwards.
Agrawal and Tandon (1994) find large returns relative to an average day and significantly positive on the last and first trading day of the month in most of the countries studied other than U.S., for the period from 1971 to 1987. The results reveal a strong and international evidence of the turn of the month effect.
Kunkel, Compton and Beyer (2003) also bring up for discussion a four-day turn of the month interval, i.e. TOM, and investigate whether the returns during these four days are larger. This approach, firstly employed by Lakonishok and Smidt (1988), is different from most of the studies which examined individual returns across the turn of the month period and more conclusive. They find persistent TOM returns in a large number of countries.
McConnel and Wei Xu (2006) combine their findings with Lakonishok and Smidt (1988) and conclude that over 109–year interval of 1897-2005, on average, all of the positive return to equities in the U.S. occurs during the turn of the month interval, meaning that over the other 16 trading days of the month investors receive no reward for bearing the market risk.
Considering the related academic research, I therefore point out the hypothesis which will be tested further on:
H7: Returns are unusually large across the turn of the month interval;
[11]
2.5. The holiday effect
The holiday effect refers to the tendency of the stock market to outperform prior to holidays. Partitioning the second half of December into three intervals, Lakonishok and Smidt (1988) find insignificant returns for the pre-Christmas period from mid-December to two trading days before Christmas, fairly large returns for the inter-holiday period from the first trading day after Christmas to two trading days before New Year, and exceptionally high returns for each of the two pre-holiday trading days preceding the Christmas and the New Year holidays. This anomalous behavior is attributed to the holiday effect.
Ariel (1990), using the U.S. data over the 1963-1982 period, stresses that on average the pre-holiday return is a factor of nine to fourteen times the return accruing on non-pre-pre-holidays. The paper excludes from possible explanations of the holiday effect the influence of other calendar anomalies. Therefore, the anomalous pre-holiday returns are not a manifestation of the January effect, weekend effect or small firm effect.
Kim and Park’s (1994) research is based on the U.S., Japan and UK stock markets and concludes that the holiday effect exists in all the studied countries and the effect in the U.K. and Japanese stock markets is independent of the holiday effect in the U.S.. Moreover, the results show that the holiday effect in Japan is not a closed-market effect. The empirical research reveals no relationship between the holiday effect and firm size.
Meneu and Pardo (2002) study the pre-holiday effect in Spain and observe that despite the specific higher returns on the pre-holiday trading days, the variance of the returns is much larger than for all other days. The average pre-holiday return is eighteen times larger than on non-holiday trading days. They also find that the pre-holiday effect still exists after taking into account the turn of the year effect, January effect, and day of the week effect by running separate regressions, using one dummy for the pre-holiday returns and an additional dummy for the yielding effect.
Schwert (2003) uncovers the vanishing effect of seasonalities after papers in which they were evidenced got published. The activities of practitioners who implement strategies to take advantage of the anomalous behavior might have caused the anomalies to vanish. However, he argues that further investigation needs to be provided in order to give more insight on the causes of these puzzling results. Meneu and Pardo (2004) investigate for the manifestation of the holiday effect on the Spanish Stock Exchange over the 1990s and find a strong evidence of abnormal returns just prior to holidays. These returns seem not to be related to other calendar anomalies such as Friday, January and turn of the year effects. A possible explanation provided is the predisposition of small investors to buy on pre-holidays, which leads to an increase in the average size of bid orders during these days.
[12]
are rather too historic. As a consequence, they act upon some information which becomes to any further extent non-valid. Thus, exploiting the anomaly which no longer exists may lead to a new anomaly, which is the reverse of the first.
In this study, I investigate whether the results are consistent with the following hypotheses, which would confirm the existence of holiday seasonality in stock market returns:
H9: Pre-holiday period returns are large and positive;
H10: The return on the last trading days before holidays is larger than the return on the pre-holiday period.
3. Research design
3.1. Methodology
A) Day of the week effect and month of the year effect
This section describes the methods used to test whether the stock market seasonalities still exist in recent times internationally.
Daily rate of return is computed as the percentage logarithmic change in the value of index compared to previous day’s closing value:
Rt = (LnPt– LnPt-1) X 100 (1)
where Rt is the continuously compounded rate of return in period t, Pt is the closing price in period t,
Pt-1 is the preceding closing price in period t-1 and Ln is the natural logarithm. All returns are in
percentage.
The most commonly used model for studying the seasonal behavior of capital markets is the dummy variable regression model. Following Agrawal and Tandon (1994), I test for the day of the week effect for each country, using the regression:
Rt = α1DDMON,t+ α2DDTUE,t+ α3DDWED,t+ α4DDTHU,t+ α5DDFRI,t+ εt (2)
where Rt is the daily return on day t, DMON,t through DFRI,t are the dummy variables from Monday
through Friday, αD is the expected daily mean stock return from Monday through Friday, εt is the
random error term.
Similar to most other studies, I expect the coefficients to be negative for Mondays and positive for Fridays, thus denoting a strong seasonality in stock returns across the days of the week, contradicting the trading time hypothesis. Moreover, some studies acknowledge for the emergence of a different pattern, such as strong negative Tuesday effect.
[13]
the mean return on the previous week is positive (RPW > 0) and the second one all Monday returns which follow negative mean returns on the previous week (RPW < 0). The following regression is tested:
Rt = α1PDNEG,t+ α2PDPOS,t+ εt (3)
where Rt is the return on day t, DNEG,t is the mean return on Monday with negative returns in the previous week, DPOS,t is the mean return on Monday with positive returns in the previous week and εt is
the random error term.
It has been shown that Monday return is strongly influenced by the return over the previous week, being larger when the previous week’s return is positive than when it is negative.
Secondly, I examine for the existence of monthly seasonalities, using the following dummy regression:
Rt = α1M DJAN,t + α2M DFEB,t + α3M DMAR,t + α4M DAPR,t + α5M DMAY,t + α6M DJUN,t + α7M DJUL,t + α8M DAbUG,t + α9M
DSEP,t + α10M DOCT,t + α11M DNOV,t + α12M DDEC,t + εt (4)
where Rt is the monthly return on month t, DJAN,t through DDEC,t are the dummy variables from January
through December, αM is the expected monthly mean stock return for January through December and εt
is the random error term. The monthly return is computed using the end of month closing prices. As reported by earlier studies, the January return is expected to be larger, compared to the return over the rest of the months. In addition, I expect to find lower December returns related to January and negative and low returns in October. However, other uncovered patterns may have emerged in recent times.
B) Turn of the month effect
To study the turn of the month effect, I use an eight-day interval, as suggested by Agrawal and Tandon (1994) and Lakonishok and Smidt (1988). Studies which employ an extended turn of the month interval do not find any significant effect during the extra days. I thus apply the eight-day interval methodology extensively to the last decade across the 27 European stock markets, using the following regression:
Rt = β-4D-4,t + β-3D-3,t+ β-2D-2,t + β-1D-1,t + β+1D+1,t + β+2D+2,t + β+3D+3,t + β+4D+4,t + εt (5)
where Rt is the return on day t, Di,t are binary dummy variables for the last and the first four trading
days of each month, the coefficients β-4 to β+4 are the mean returns for the eight trading days and εt is
the error term.
[14]
Rt= α + βTOMDTOM,t + εt (6)
where Rt is the return on day t, α is the intercept representing the mean return for the rest-of-the-month
period (ROM), DTOM is a binary dummy variable for the four-day turn of the month interval (TOM),
the coefficient βTOM is the mean difference between the mean TOM return and the mean ROM return
and εt is the error term.
I expect to find significantly positive and larger returns over the TOM interval, compared to an average return across the rest of the month (ROM) and large and positive returns on the last and first trading day of the month.
C) The holiday effect
Next, I examine the holiday effect, partitioning the end of December into three data series as follows: (1) the pre-holiday period, from mid-December to two trading days prior Christmas and from the first trading day after Christmas to two trading days prior New Year, (2) the pre-holiday days, as the last trading days prior Christmas and New-Year and (3) all the other days, except holidays. The methodology is similar to Agrawal and Tandon (1994).
Henceforth, an additional dummy regression is estimated:
Rt = γDper,t + δDday,t + θDother,t + εt (7)
where Rt is the return on day t, Dper is the dummy for the period up to two trading days before
holidays, Dday is the dummy for the last trading days before holidays, Dother is the dummy for ordinary
days, other than pre-holidays and εt is the random error term, which is assumed to be independently
and identically distributed as normal distribution.
The regression investigates whether the pre-holiday days exhibit larger mean returns, compared to all the other days. Similar to other studies, I expect to find positive returns on the last trading days before holidays, and larger compared to the pre-holiday period, which is also expected to exhibit positive returns.
3.2. Data collection
The data consists of series of daily returns of the corresponding composite stock market indices from the following 27 European Union countries: Austria, Belgium, Germany, Finland, France, Denmark, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain, Sweden, United Kingdom, Cyprus, Czech Republic, Estonia, Hungary, Latvia, Lithuania, Malta, Poland, Slovakia, Slovenia, Romania and Bulgaria. Each of the indices is calculated based on stock prices denominated in local currency.
[15]
There are three exceptions regarding the availability of data for Bulgaria, Cyprus and Greece for which the number of observations is lower due to the late starting dates of the three official stock indices. The total number of observations is 70.471, comprising all the countries, over ten years of daily data. Table 1 presents the indices symbols and a short description for each of the stock indices used.
Figure 1 shows the overall performance of the 27 stock indices across time, from January 2000 through January 2010. All the stock prices are expressed in Euros for the Eurozone countries and in the national currency of each country for the countries which are not yet members of the Eurozone. Therefore, the observed amplitude dissimilarities of each index dynamics are also caused by the conversion differences of each currency reported to euro, next to the market capitalization of each index considered.
As depicted in the chart, most of the stock indices follow a similar trend, with the two outstanding recent bubbles, i.e. dot.com and housing, being obvious. The overall performance increases dramatically, reaching a peak in 2007, followed by a sudden drop, explained by the recent economic crisis.
Table 2 comprises the summary statistics of the 27 cumulative stock indices data for each the day of the week, month of the year, turn of the month and pre-holiday returns, over 1st of January 2000 through 1st of January 2010. Inspection of the means conveys that the expected return is not constant across the effects studied. The mean return for Monday and Tuesday is negative, compared to the positive return on the last trading days of the week.
Moreover, positive January and large and positive April returns are noticed, as well as larger pre-holiday returns compared to the returns on non-holiday days. In contrast to previously documented January effect, the highest mean return occurs on April.
This may be supported by the tax loss selling hypothesis in countries with fiscal year ending in April. However, most of the countries studied exhibit a calendar tax-year, ending in December. Furthermore, the four day turn of the month return is positive and larger than an average return for the rest of the month.
However, the seasonal effects in stock market returns are investigated in depth further on and the results are presented in Section 4.
4. Empirical results
[16]
Table 1 Short description of the stock market composite indices
1
as in www.Bloomberg.com
Figure 1 Stock indices performance
COUNTRY
STOCK MARKET INDEX SYMBOL1
STOCK MARKET INDEX PROFILE
Austria ATX The Austrian Traded Index is a index of the most traded stocks on the Vienna Stock Exchange Belgium BEL20 The BEL 20 consists in 20 most liquid Belgian stocks traded on the Brussels Stock Exchange Bulgaria SOFIX The SOFIX Index comprises the most liquid companies listed on the Bulgarian Stock Exchange Cyprus CYSMMAPA Cyprus General Market Index (CSE)
Czech
Republic PX The official index of the Prague Stock Exchange
Denmark KAX The OMX Copenhagen Index includes all the stocks traded on the Copenhagen Stock Exchange Estonia TALSE OMX Tallinn includes all the shares listed on the Tallinn Stock Exchange
Finland HEX Consists in all the stocks traded on the Helsinki Stock Exchange France CAC The CAC-40 Index comprises 40 companies listed on the Paris Bourse Germany DAX30 The index of 30 selected stocks traded on the Frankfurt Stock Exchange Greece ASE Stocks listed on the Athens Stock Exchange
Hungary BUX Tracks the traded shares on the Budapest Stock Exchange
Ireland ISEQ Consists in all official listed equities in the Irish Stock Exchange and excludes UK registered companies Italy FTSEMIB Embodies the 40 most liquid and capitalized stocks listed on the Borsa Italiana
Latvia RIGSE OMX Riga index incorporates all the shares listed on the Riga Stock Exchange in Latvia Lithuania VILSE OMX Vilnius includes all the shares listed on the Vilnius Stock Exchange
Luxembourg LUXXX Includes the most capitalized and liquid Luxembourg stocks
Malta MALTEX The Malta Stock Exchange (MSE) index encompasses all shares traded on the Stock Exchange of Malta Netherlands AEX Is the leading Dutch index of the stocks traded on the Amsterdam Exchange
Poland WIG Includes all companies listed on the main Polish market, excluding foreign companies and investment funds Portugal PSI20 Consists in the top 20 stocks listed on the Lisbon Stock Exchange
Romania BET Comprises the most liquid 10 stocks listed on the Bucharest Stock Exchange Slovakia SAX The official stock index for the Bratislava Stock Exchange
Slovenia SVSM LJSE Composite Index is the Ljubljana Stock Exchange total market index Spain IBEX Comprises the 35 most liquid stocks traded on the Continuous market
Sweden OMX Contains 30 stocks with the largest trading volume on the Stockholm Stock Exchange United
Kingdom UKX
[17]
The persistence of these anomalies will suggest that stock market returns are not entirely random. This means that there is evidence of seasonal patterns, some which have persisted over decades or of other new undocumented patterns which have emerged more recently. These patterns show that stock market returns can be predicted to some extent, thus investors may take advantage of this predictability and have the opportunity to earn abnormal returns by timing their investments. Since my sample contains more recent data and a larger number of countries to be investigated, the findings presented here bring further evidence to support the persistence of some seasonalities in stock returns.
Table 2 Summary statistics of the stock indices returns
All abbreviations used in the table are explained as follows: Pre-per = pre-holiday period (Christmas and New Year), Pre-day = pre-holiday trading days (Christmas and New Year), Ordinary = trading days other than pre-holiday, Days (-4) through (+4) = the return in days around the turn-of-the-month, TOM = the four day turn-of-the-month return, ROM = the rest-of-the-month return, Std. Dev. = standard deviation, Prob. = probability, Obs. = observations.
Mean Maximum Minimum Std. Dev. Skewness Kurtosis Prob. Obs.
Monday -0.014 14.931 -18.955 1.690 -0.164 12.611 0.000 12,545 Tuesday -0.012 16.87 -17.172 1.494 -0.081 10.844 0.000 12,969 Wednesday -0.025 13.178 -20.899 1.528 -0.568 12.637 0.000 13,049 Thursday 0.041 14.563 -17.404 1.541 -0.284 11.539 0.000 12,930 Friday 0.069 11.093 -13.824 1.466 -0.332 11.969 0.000 12,766 January 0.171 29.953 -27.080 8.456 -0.315 4.221 0.000 255 February -0.494 29.079 -31.322 7.686 -0.091 4.923 0.000 258 March 0.217 22.742 -18.369 5.891 0.274 4.812 0.000 258 April 2.701 29.769 -17.870 6.968 0.783 4.579 0.000 258 May 0.363 36.377 -13.671 5.766 0.274 8.078 0.000 258 June -0.979 35.039 -18.264 6.285 0.095 7.015 0.000 258 July 0.821 29.236 -18.997 6.438 -0.003 4.427 0.000 258 August 1.533 29.286 -22.339 5.507 0.478 7.960 0.000 258 September -2.192 17.586 -33.942 9.324 -1.088 4.147 0.000 258 October -0.592 18.966 -46.036 9.808 -1.823 7.422 0.000 259 November 22.381 -30.370 7.159 -0.754 4.748 0.000 260 December 1.371 19.605 -17.102 5.052 -0.401 5.341 0.000 260 Pre-day 0.305 8.387 -3.919 1.591 1.597 12.988 0.000 494 Pre-per 0.077 7.317 -8.925 1.026 -0.342 9.148 0.000 1,849 Ordinary 0.007 42.865 -46.226 1.207 -0.287 36.821 0.000 61,922 (-4) 0.013 10.685 -11.901 1.359 0.198 10.885 0.000 3,095 (-3) 0.109 10.898 -10.931 1.494 0.692 10.545 0.000 3,090 (-2) 0.082 7.470 -10.212 1.483 -0.844 9.098 0.000 3,094 (-1) 0.192 11.880 -13.015 1.418 -0.088 10.864 0.000 3,096 (+1) 0.230 12.019 -11.761 1.588 -0.059 9.545 0.000 3,094 (+2) 0.071 8.828 -11.225 1.584 -0.076 8.043 0.000 3,094 (+3) 0.041 8.601 -13.116 1.464 -0.334 9.718 0.000 3,093 (+4) -0.012 7.260 -10.434 1.552 -1.249 11.281 0.000 3,091 TOM 0.133 12.019 -13.116 1.517 -0.130 9.458 0.000 12,377 ROM -0.016 12.123 -13.515 1.531 -0.227 10.385 0.000 51,817
[18]
4.1. The day of the week effect
As expected, the results presented in Table 3 show a strong seasonality in daily returns. From the pooled regression, I find negative returns on Monday, Tuesday and Thursday and positive on Wednesday and Friday. However, from all these, only the Tuesday, Wednesday and Friday returns are statistically significant.
The Monday return from the pooled regression appears to be negative, but insignificant. Contrary to the strong Monday effect previously reported, the results here are somehow puzzling, especially considering that in the country regressions, these returns are found to be statistically significant in only four of the 27 countries examined. These four statistically significant results are mixed, with most of them being positive and not negative as it was expected. Ireland is the only country from the sample where the negative Monday return is significant. However, other 14 countries exhibit negative Monday returns, and only ten of these returns are lower than the daily average. Therefore, the persistence of the Monday effect internationally cannot be categorically confirmed and H0 can neither be rejected, nor supported.
In Columns 9 and 10 of Table 3, I present the results for the second regression used to investigate the dependence of the Monday returns over the mean returns of the previous week. I find that the Monday return is higher when the mean return over the previous week was positive as compared to when it was negative in all the countries, except Germany. Although no clear explanation has been yet provided for this twist of the Monday effect, similar to Jaffe et al. (1989), I also find evidence of its existence over last decade. Hence, in 25 of the countries, the Monday returns are negative when the market has fallen in the previous week, while in 22 of the countries Monday returns following a week with positive average returns, are positive. Moreover, the results from the pooled regression are extremely significant at a 1% significance level. Therefore, Monday returns are strongly influenced by the previous week’s returns, supporting H3. A possible reasoning for these results could be that weekends do not bring any additional information on the market and investors’ intentions are not yet clear in the first trading day of the week.
[19]
effect do not support this hypothesis, considering the strong dependency of the Monday returns by the previous week returns which has been previously demonstrated. Assuming that there is an immediate reaction of investors to the negative information released over weekends, i.e. Monday open, Monday returns would be mostly negative, independently of the previous week’s return.
Therefore, there are other uncovered explanations for the presence of Friday effect. For example, Agrawal and Tandon (1994) demonstrate that Tuesday return remains negative, even when the mean return of the previous week is positive. Hence, further research needs to be done in order to bring further evidence for the information release hypothesis and see whether investors may have a late reaction, i.e. on Tuesdays, to the negative information released during weekends.
The results for the day of the week effect reported here oppose the findings of Sales and Caro (2006), which show that returns do not differ across the days of the week. In contrast, the results are in line with Agrawal and Tandon’s (1994), which bring evidence of a strong day of the week effect in the 1970’s and 1980’s. Moreover, in addition to negative Monday returns, negative Tuesday and positive Friday returns are also found in previous empirical studies (Fortune, 1992; Agrawal & Tandon, 1994; Sales & Caro, 2006).
Furthermore, evidence provided here shows that the existing patterns in daily returns are inconsistent with the efficient market hypothesis. Even if the profits of any investor aware of the market inefficiency are limited due to transaction costs, benefits may still arise. For example, investors may delay their purchases planned for Friday, over the weekend, until the beginning of the upcoming week and sales planned for Monday or Tuesday, for the prior Friday.
This is how the private information about seasonalities in stock returns may induce speculative behavior on the stock market returns. Moreover, my results bring no support for the calendar time hypothesis, which holds that Monday mean returns are three times the mean of the other days of the week.
However, though not statistically significant, negative Monday returns are reported in some of the countries studied. Compared to the statistically significant Monday seasonality found in most of the previous studies, the insignificant and isolated Monday effect found here for a more recent time period may be explained by the fact this effect perished over time on an international scale, mostly due to an extensive academic attention in the last decades. In a recent study, Prokop (2010) also notes a vanishing Monday effect in the recent years, supporting my findings for a more recent time frame.
4.2. The month of the year effect
[20]
Table 3 The day of the week effect for daily prices returns
The table presents the results of the regressions: a) Rt = α1DDMON,t+ α2DDTUE,t+ α3DDWED,t+ α4DDTHU,t+ α5DDFRI,t+ εt, where Rt is
the daily return on day t, DMON,t through DFRI,t are the dummy variables from Monday through Friday, αD is the expected daily
mean stock return from Monday through Friday, εt is the random error term and b) Rt = α1PDNEG,t+ α2PDPOS,t+ εt, where Rt is the
return on day t, α1P is the mean return on Monday with negative returns in the previous week, α2P is the mean return on Monday
with positive returns in the previous week and εt is the random error term. Average is the average expected mean return across
the days of the week. All estimates are made using OLS, applying White’s (1980) heteroskedasticity consistent standard errors. Rates of return are in percentage.
[21]
Unexpectedly, the results from the pooled regression reveal a significantly larger and positive mean return in April compared to the rest of the months. Moreover, the mean return in April is larger than the monthly mean return in all the countries, except Slovakia and Estonia. In 25 of the countries, the April mean returns are positive and in 12 of them are positive and statistically significant. Therefore, the results uncover a little previously documented pattern in stock returns, which consists in the existence of a strong April effect internationally. The April effect was also reported by Agrawal and Tandon (1994), covering a sample of 18 countries, 10 of which were from Europe.
Also of interest are the results for the negative September return, which occurs in all the countries, except Romania, Bulgaria, Slovakia and Lithuania. In the pooled regression as well as in 10 of the countries, this effect is statistically different from 0. The mean return in September is lower than the monthly average in 24 of the countries studied. Evidence on September effect has been previously provided by Lakonishok and Smidt (1988). Although possible causes of the September negative returns were previously examined, no evidence has been yet provided. Mark Hulbert, the founder of the rating agency Hulbert's Financial Digest, declared in one of his recent articles:
“After failing for several years to find a plausible explanation for why September is bad for stocks, I eventually gave up.”
Even though not as strong as the previous two patterns mentioned, I also find mostly positive mean returns on December. In 22 countries, December returns exceed the average return on the other months.
The results contradict Agrawal and Tandon (1994), which report low December returns compared to the monthly average and show no evidence in support of H5.
Negative September and positive December returns were also documented by Seyhun (1993) using 65 years of data and the results are consistent across time.
Furthermore, I find some evidence of the so-called Mark Twain effect, firstly documented by Cadsby (1989), which stresses that the return in October is significantly lower than in the rest of the year. The return on this month is negative and significantly different from zero when running the pooled regression. In addition, October generates lower returns compared to the other months in 18 of the countries studied, although the results are mostly not statistically significant. Hence, I find some evidence supporting H6.
[22]
Table 4 The month of the year effect for monthly pattern
The table presents the results of the regression: Rt = α1M DJAN,t+ α2MDFEB,t+ α3MDMAR,t+ α4MDAPR,t+ α5MDMAY,t+ α6M DJUN,t+ α7MDJUL,t+ α8MDAUG,t+ α9MDSEP,t+ α10MDOCT,t+
+α11MDNOV,t+ α12MDDEC,t+ εt, where Rt is the monthly return on month t, DJAN,t through DDEC,t are the dummy variables from January through December, αM is the expected
monthly mean stock return for January through December and εt is the random error term. Average is the average expected mean return across the months of the year. All estimates
are made using OLS, applying White’s (1980) heteroskedasticity consistent standard errors. Rates of return are in percentage.
(1) (2) January (3) February (4) March (5) April (6) May (7) June (8) July (9) August (10) September (11) October (12) November (13) December (14) Average (15) Pooled regression 0.171 -0.494 0.217 2.701** 0.363 -0,979*** 0.821** 1.667 -2.192*** -0.693* 0.000 1.371*** 0.358 AEX Netherlands -2.150 -1.676 -0.924 3.139* 0.000 -1.148 -1.217 0.948 -5.504*** 0.354 0.213 1.022 -0.579 ASE Greece -1.942 -4.436* 1.072 4.961** 1.974 -6.389*** 0.622 1.028 -1.102 -5.454** -6.888*** -0.571 -1.427 ATX Austria 0.477 0.434 2.994 3.816* 0.966 0.370 -0.490 1.164 -2.762 -2.810 0.151 3.031* 0.612 BEL20 Belgium -2.341 -1.674 -0.177 3.566** -0.381 -1.539 -0.098 2.035 -2.866* -1.116 -0.767 2.507 -0.238 BET Romania 6.057* 0.392 -1.658 4.328 2.923 2.022 4.860 0.450 0.515 -0.283 0.844 3.023 1.956 BUX Hungary 1.895 -0.552 -0.316 3.648* 1.159 -1.805 3,868* 2.170 -0.860 -2.118 -0.761 2.455 0.732 CAC France -2.318 -1.517 0.460 3.795** 0.010 -2.248 -0.417 0.213 -4.269*** 1.095 -0.156 1.208 -0.345 CYSMMAPA Cyprus 2.095 -5.631 1.646 9.973* 8.086 -5.064 3.851 1.304 -0.351 1.398 -7.340 0.896 0.905 DAX30 Germany -1.915 -1.816 -0.032 4.588** 0.156 -1.118 0.010 -1.027 -5.406*** 1.963 1.146 1.899 -0.129 FTSEMIB Italy -1.522 -0.947 -1.255 3.942** -1.093 -2.459 -0.871 1.097 -4.429** 0.288 1.336 -0.144 -0.505 HEX Finland -2.324 -2.974 0.261 3.412 -1.837 -3.946* -3.269 0.515 -2.376 3.670 1.603 -0.881 -0.679 IBEX Spain -1.439 0.958 -0.598 2.487 -0.482 -2.071 -0.026 0.699 -2.142 1.388 0.921 0.558 0.021 ISEQ Ireland -0.534 -1.725 1.216 2.626 0.444 -1.952 -3.410* 2.910 -3.878* -0.706 -1.447 1.228 -0.436 KAX Denmark 0.935 0.287 0.135 1.803 2.528 -0.554 0.516 2.807* -3.504** -0.977 -1.307 0.757 0.286 LUXXX Luxembourg 0.853 1.937 -1.769 2.639 2.092 0.857 -0.327 -0.954 -7.120*** -4.553* 1.980 3.057 -0.c109 MALTEX Malta 2.870* -0.120 0.583 -1.703 -2.397* -0.726 2.219 -0.995 -1.004 -0.543 1.039 1.317 0.045 OMXS Sweden -1.500 1.514 -0.961 3,520* -1.533 -2.171 0.694 -0.406 -4.124** 1.031 1.579 0.047 -0.192 PSI20 Portugal 0.570 0.704 -0.816 0.507 -0.957 -2.802* -1.532 0.368 -1.356 -0.928 1.608 1.176 -0.288 PX Czech Republic 1.488 0.984 2.239 2.078 -0.143 -2.491 3.392 1.769 -2.609 -0.641 -0.877 3.058 0.687 RIGSE Latvia -0.790 -2.790 2.191 2.276 -2.310 2.561 5.251** 4.839** -1.894 -0.476 -0.304 0.569 0.760 SAX Slovakia -0.426 2.583 1.192 -0.515 0.304 -0.202 1.769 3.810** 1.591 -0.537 2.513 0.342 1.035 SOFIX Bulgaria -0.390 2.041 -4.161 2.790 1.464 4.232 4.262 4.717 2.612 -2.880 -1.236 1.538 1.249 SVSM Slovenia 2.388 -3.181* -0.449 3.084* 1.642 0.253 2.091 3,070* -0.519 -0.776 -0.519 -0.799 0.524 TALSE Estonia 4,035* 1.303 2.167 0.919 0.474 -0.220 -1.475 4.853* -0.403 -3.479 0.231 3.300 0.975 UKX United Kingdom -3.474*** -0.885 0.259 2.664** 0.210 -1.847 -0.322 1.141 -2.905** 0.964 0.167 1.556 -0.206 VILSE Lithuania 2.813 -0.301 1.810 1.053 0.474 0.415 1.438 4.921* 2.239 -5.281** -1.045 1.161 0.808 WIG Poland 0.799 -0.921 1.546 2.530 0.966 -1.163 2.506 1.310 -2.708 0.451 0.260 2.357 0.661
[23]
Overall, the results show no strong tendency of stock prices to rise in January, relative to December, H4 being rejected in most of the countries. However, the hypothesis is strongly supported in Romania, Malta and Estonia. Therefore, the January effect seems to have vanished in most of the countries. Previous research, used as a benchmark for the present study, reports a strong January effect (Rozeff and Kinney, 1976; Reiganum, 1982; Seyhun, 1993; Agrawal & Tandon, 1994). Nevertheless, most of these academic studies rely on earlier data. Similar to my findings, Mehdian and Perry (2002) acknowledge no significant evidence for the January effect in recent years and no statistical support for the tax loss selling rationale. Compared to earlier data, they conclude that this effect previously existed, but diminished in time. Their evidence from the U.S. is now further confirmed for European countries.
I assume that one possible explanation for this palled and mostly reverse effect is the extensive academic attention during the last decades, particularly on the January effect, which has led to an intensively exploited pattern. In addition, information system has become more transparent and investors can obtain public information more easily.
In spite of the extensive research, there is no consensus as to why these anomalies started to exist as well as in identifying the main factors which caused these anomalies to change over time. However, the results here put on question the underlying possible explanations of the January effect mentioned on previous studies, particularly the tax-induced selling at the end of December (Reinganum, 1983). Of the 27 countries examined, all have tax years ending in December, except United Kingdom, which has a tax year ending in March. Moreover, examining the countries with the highest capital gains tax, i.e. over 20%, where the tax induced selling effect should be more pronounced, most of them exhibit negative instead of positive mean January returns, i.e. Netherlands, France, Finland, Ireland, Sweden, Latvia. Jones et al. (1987) also contradict the tax-loss-selling rationale, documenting that the January effect existed prior to income taxation.
Moreover, taking into account that across time the fiscal year interval remained unchanged in most of the countries studied, there should be other influences that caused the January effect to perish and counteract the underlying forces which initially determined it. However, tax issues have become more important on the political agenda, following the recent financial crisis. Therefore, governments imposed new tax rules starting the end of 2007, which distort investors’ behavior and have unintended consequences, not yet documented.
[24]
changes might have led to new emerging patterns. Changing trends in month of the year seasonalities were also mentioned in other studies (Barone & Cyr, 1994) and some researchers argue that their effect might be transitory in nature and vanishing over time.
4.3. The turn of the month effect
As depicted in Table 5, the results reveal a strong international turn of the month effect. This can be observed by looking at the returns across the trading days close to the turn of the month, i.e. the four trading days before and after the turn of the month. Results show that most of these returns are positive and higher compared to the rest of the month returns.
On day -1 all the countries exhibit positive returns and twelve of these returns are statistically significant. I also find a strong seasonality on day +1, returns on this day being unusually large compared to the returns occurred in the rest of the month and positive in all the countries except Greece, Malta, Bulgaria and Lithuania. In 18 of the countries, returns on the first day of the month are large and statistically significant. Therefore, trading days -1 and +1 experience the highest mean returns, the results being in line with Agrawal and Tandon (1994) and supporting H8.
I also look at the four-day TOM interval (day -1 to day +3) which was found as more meaningful in previous studies. The results for this regression are shown in the last two columns of Table 5. The evidence here is even stronger. I expected to find significantly positive and larger TOM returns compared to the ROM period. In 16 of the countries, negative ROM returns are found, while 11 countries have positive ROM returns. Moreover, the mean TOM return is higher than the mean ROM return in all the countries, except Malta and Lithuania and their difference, βTOM is different from zero in 19 countries,
bringing consistent evidence in support of H7.
Moreover, the TOM returns account on average for 113% of the monthly return, with a range from 33% for Malta to 341% in Italy.
The results are in line with most of the previous studies, which acknowledged abnormally large returns around the turn of the month interval (Ariel, 1987; Lakonishok & Smidt, 1988; Agrawal & Tandon, 1994; Kunkel, Compton & Beyer, 2003).
[25]
Table 5 Regression results for turn-of-the-month effect
This table presents the results for the following regressions: a) Rt= β-4D-4,t + β-3D-3,t+ β-2D-2,t + β-1D-1,t + β+1D+1,t + β+2D+2,t +
+β+3D+3,t + β+4D+4,t + εt, where Rt is the return on day t, Di,t are binary dummy variables for the last and the first four trading days
of each month, the coefficients β-4 to β+4 are the mean returns for the eight trading days and εt is the error term and b) Rt = α +
βTOMDTOM,t + εt, where Rt is the return on day t, α is the intercept representing the mean return for the rest-of-the-month period
(ROM), DTOM is a binary dummy variable for the four-day turn-of-the month interval (TOM), the coefficient βTOM is the mean
difference between the mean TOM return and the mean ROM return and εt is the error term. All estimates are made using OLS,
applying White’s (1980) heteroskedasticity consistent standard errors. Rates of return are in percentage.
(1) β-4 (2) β-3 (3) β-2 (4) β-1 (5) β+1 (6) β+2 (7) β+3 (8) β+4 (9) α (10) βTOM (11) Pooled regression 0.010 0.107*** 0.086*** 0.194*** 0.231*** 0.072*** 0.045* -0.011 -0.016*** 0.151*** Netherlands 0.104 -0.026 0.047 0.153 0.234* -0.025 0.011 -0.107 -0.054* 0.147* Greece -0.093 -0.004 -0.094 0.472* -0.102 0.075 -0.199 -0.180 -0.033 0.095 Austria 0.006 0.248* 0.300** 0.228* 0.384*** 0.100 0.021 0.017 -0.007 0.191*** Belgium 0.067 -0.007 0.093 0.281** 0.319*** 0.053 0.042 0.064 -0.052* 0.225*** Romania 0.104 -0.144 0.245* 0.260* 0.559*** 0.279* 0.115 0.078 0.043 0.261*** Hungary 0.224* 0.158 0.177 0.021 0.337** 0.201 0.005 0.239 * 0.010 0.131* France 0.109 0.097 0.078 0.240* 0.242* -0.059 0.041 -0.111 -0.045 0.161** Cyprus -0.360* -0.107 0.292 0.231 0.318* 0.152 0.118 0.251 -0.002 0.205** Germany 0.096 0.139 0.054 0.136 0.290** -0.094 0.160 -0.173 -0.035 0.159** Italy 0.026 0.130 0.196* 0.071 0.188 -0.175 0.143 -0.080 -0.041 0.099 Finland -0.247 0.312* 0.127 0.427** 0.277* 0.154 -0.092 0.201 -0.081* 0.273*** Spain 0.015 0.184 0.051 0.146 0.263** 0.013 0.131 -0.036 -0.030 0.169** Ireland -0.059 0.052 0.052 0.181 0.392*** 0.127 -0.088 0.094 -0.058* 0.212*** Denmark -0.069 -0.068 0.214** 0.190* 0.350*** 0.157* -0.087 0.090 -0.018 0.171*** Luxemb ourg -0.231 0.101 0.273* 0.055 0.202 -0.281 0.053 0.026 -0.008 0.016 Malta 0.110* 0.033 0.024 0.145** -0.034 -0.029 -0.077 -0.065 0.002 -0.001 Sweden -0.064 0.239* 0.253* 0.131 0.430*** 0.067 0.021 0.026 -0.047 0.209*** Portugal -0.052 0.000 0.070 0.134 0.178* -0.002 0.071 0.111 -0.039* 0.135*** Czech Republic 0.028 -0.056 0.005 0.278** 0.333** -0.082 0.048 0.157 0.006 0.138* Latvia -0.082 -0.077 0.078 0.153 0.074 -0.047 0.110 0.184 0.030 0.044 Slovakia 0.047 -0.090 -0.002 0.231** 0.060 0.225** 0.096 0.158 0.025 0.127** Bulgaria 0.030 -0.011 0.385** 0.482*** -0.260 0.099 0.300* 0.164 0.032 0.121 Slovenia 0.040 0.082 0.135* 0.154* 0.059 0.179** 0.139* 0.058 0.000 0.133*** Estonia 0.054 0.039 0.092 0.151 0.274** 0.179* 0.019 0.110 0.020 0.136** United Kingdom 0.016 0.088 0.110 0.106 0.292* 0.093 0.026 0.023 -0.040* 0.169*** Lithuani a 0.029 0.078 0.149 0.126 -0.035 -0.055 0.045 0.160 * 0.040* -0.019 Poland 0.128 0.151 0.069 0.026 0.332*** 0.109 -0.055 0.026 0.014 0.088
[26]
could be relevant to my study, taking into consideration the standardized cash flows and payment systems across European Union member states.
Finally, since there is evidence of a strong turn of the month effect in most of the countries studied, it can be concluded that its causes are not primarily related to the local stock market practices and there are other forces which continue to contribute to its persistence on an international scale.
4.4. The holiday effect
The results for the holiday effect are summarized in Table 6. Column 3 in Table 6 reports average returns for the last days prior to Christmas and New Year’s Day. Column 4 reports returns for the pre-holiday period subsample, less the two-day pre-holiday return, i.e. from mid-December to New Year’s Day, excluding the last trading days before the two analyzed holidays. Lastly, in column 6 are reported the expected returns in all the other days of the year excluding the ones considered in columns 3 and 4.
As expected, the results of the pooled regression show a statistically significant tendency of the stock market returns to be larger during the period preceding holidays, compared to the expected returns on non-holiday days. All 27 countries exhibit large and positive returns across the pre-holiday period interval, except Finland. Therefore the results bring supporting evidence for H9 in most of the countries studied.
As getting closer to a holiday, i.e. the last trading day, the expected return becomes even larger in most of the countries studied. In 19 of these countries, the two-day preholiday return exceeds the pre-holiday period return, hence H10 cannot be rejected. The results of the pooled regression show that the two-day pre-holiday return is on average 23 times as high as the return occurring on an ordinary day. Moreover, the results on individual countries also show a higher return on pre-holiday period, which is a factor of 0.8 to 20 times larger than the average return on the other days.
The present findings are consistent with most of the previous academic research. Ariel (1990), Agrawal and Tandon (1994), Lakonishok and Smidt (1998) and Meneu and Pardo (2004) also find significantly higher returns before holidays, as compared to other days.
Chong, Hudson, Keasey and Littler (2005) argue that the holiday effect diminished in time in the U.S. and they question the existence of this palled effect internationally. Yet, the present evidence shows persistent holiday seasonality in most European countries, therefore invalidating their assumption.
[27]
Table 6 Regression results for holiday effect
This table presents the results of the regression: Rt= γDper,t+ δDday,t+ θDother,t+ εt, where Rt is the return on day t, Dper is the
binary dummy for the period to two trading days before holidays, Dday is the binary dummy for the last trading days
preceding holidays, Dother is the binary dummy for ordinary days, other than pre-holidays, γ is the expected mean return for
the period to two trading days before holidays, δ is the expected mean return in the last trading days preceding holidays and εt is the random error term, which is assumed to be independently and identically distributed as normal distribution. All estimates are made using OLS, applying White’s (1980) heteroskedasticity consistent standard errors. Rates of return are in percentage. (1) (2) Pre-day (3) Pre-period (4) Other (5) Pooled regression 0.305*** 0.078** 0.013** AEX Netherlands -0.066 0.148 -0.040 ASE Greece 0.051 0.209 -0.040 ATX Austria 0.326 0.141 0.048* BEL20 Belgium 0.169 0.273* -0.030 BET Romania 1.431*** 0.097 0.089*** BUX Hungary 0.114 0.114 0.033 CAC France 0.055 0.148 -0.022 CYSMMAPA Cyprus 0.053 0.423 0.021 DAX30 Germany 0.415 0.048 -0.011 FTSEMIB Italy 0.016 0.055 -0.028 HEX Finland 0.436 -0.026 -0.031 IBEX Spain 0.153 0.017 -0.001 ISEQ Ireland 0.084 0.076 -0.025 KAX Denmark 0.374* 0.022 0.010 LUXXX Luxembourg 0.314 0.249 -0.016 MALTEX Malta 0.354** 0.016 -0.001 OMXS Sweden 0.169 0.052 -0.008 PSI20 Portugal 0.192 0.105 -0.019 PX Czech Republic 0.262 0.116 0.029 RIGSE Latvia 0.434 0.079 0.035 SAX Slovakia 0.502 0.067 0.062*** SOFIX Bulgaria 1.228*** 0.081 0.057* SVSM Slovenia 0.047 0.083 0.026 TALSE Estonia 0.215 0.158 0.041* UKX United Kingdom 0.084 0.188 -0.017 VILSE Lithuania 0.603** 0.102 0.032 WIG Poland 0.248 0.058 0.031
[28]
6. Conclusions
This paper examines the existence of stock market seasonalities across a large sample of countries, during 2000-2010. The main research question is whether the previously documented patterns still exist across the recent years on an international scale, or these seasonal effects perished over time, since previous evidence suggests that stock market seasonalities might be transitory in nature. In addition, I investigate whether new uncovered patterns emerged and if these new patterns exist internationally, i.e. not localized effects resulted from local practices and institutions. In addressing these issues, I rely on a larger sample of countries over the last decade, thus aiming to provide more insight on the complexity of stock market seasonalities.
The findings show no evidence supporting the Monday effect strongly reported in prior studies. Nevertheless, there is evidence of negative Tuesday and positive Friday returns in most of the countries studied. As for the month of the year effect, the findings are even more unanticipated. I expected to find a strong January effect internationally. However, the hypothesis is supported only for Romania, Malta and Estonia. Moreover, other less reported monthly patterns emerged in recent years: large and positive mean returns in April and negative in September. Turn of the month effect and holiday effect are persistent across time and most countries, the evidence here adding strong international proof to previous findings.
Several important conclusions can be drawn from the findings presented in this thesis. Firstly, the results confirm a strong seasonality in stock market returns in most of the analyzed countries. Secondly, some of the previously documented patterns are still significant in the recent years, i.e. holiday effect and turn of the month effect, thus confirming their persistence over decades in most of the countries studied. Thirdly, other intensively predicted stock market effects have diminished in time, i.e. Monday effect and January effect. Some scholars argue that these diminishing patterns are a tendency of stock markets to become gradually more efficient. However, simultaneously to the vanishing effect of some patterns in stock market returns, I find new patterns emerging in recent years, for which there is yet little evidence provided, i.e. April effect, September effect, Friday effect.
[29]
Moreover, the present results bring no evidence supporting the calendar time hypothesis and tax loss selling hypothesis, which initially provided explanations for seasonal patterns which recently seem to have paled into insignificance. As for the new emerged patterns, they are induced by other underlying causes yet little documented, possibly arising from the recent social, economic and political environment.
