Calendar Effects in US Stock Markets

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 26

Calendar effects in US stock markets

A statistical research to the Turn-of-the-year, sell-in-May and weekend effect on


the NYSE, AMEX and NASDAQ

Joost Schoute, 30-6-2015


Student Number: 1975560
Thesis Master Finance
Vrije Universiteit Amsterdam
Thesis Supervisor: Dr. V. Atanasov
Correspondence: joost1schoute@hotmail.com
Abstract

In this thesis I study the turn-of-the-year and the sell-in-May effect on the NYSE, AMEX and
NASDAQ over the period June 1963 to December 2014. The weekend effect is studied over the period
2010-2014. Using OLS-regressions I find evidence for a positive January and sell-in-May effect. I do
not find evidence for the weekend effect or negative December returns. I provide a robustness check
over different periods and size and book-to-market deciles. My findings are that the turn-of-the-year
effect only holds true for small stocks and stocks with a high book-to-market ratio, and that the
weekend effect only holds true for small stocks.

2
1. Introduction
According to the efficient market hypothesis risk is priced. This means that if an investment contains
risk, the investor will be rewarded with a higher return than the risk free interest rate to justify the risk.
It also means that higher returns are not possible without any additional risk involved. If there would
be an investment opportunity with a higher return than the risk justifies, investors would invest more
in the project, lowering the returns until the risk-return ratio is in equilibrium again. If the return stays
higher than the risk justifies the investment should, according to the efficient market hypothesis,
contain some unobserved risk to justify the profit. If there is no additional risk whatsoever the
observation would contradict the efficient market hypothesis.
Returns can be predicted based on the date. The January effect is the effect that average
January returns are higher than the average returns during other months. Assuming that January does
not contain any additional risk compared to the other months this finding is in contradiction with the
efficient market hypothesis. The January effect is called a calendar anomaly. A calendar anomaly is a
return regularity based on a weekday, calendar month or a specific day during the year. This means
that the likelihood of higher or lower stock-returns increases on specific days. In this paper I
conducted a research on calendar anomalies. The calendar anomalies discussed in this paper are the
January effect, the sell-in-May effect and the weekend effect. Since the January effect is related to the
return regularities in December I have also examined the returns in December.
The January effect was first discovered by Rozeff and Kinney (1976) on the NYSE. The
weekend effect was addressed by French (1980). He found that average Monday returns are
significantly negative on the S&P index. Sell-in-May and go away is an old saying between stock
traders. The first ones to actually examine the anomaly were Bouman and Jacobsen (2002). They
found that winter returns (November-April) are significantly higher than summer returns (May-
October) in 36 out of 37 countries in their sample. Agrawal and Tandon (1994) conducted a research
on several calendar effects in 18 countries to examine if these anomalies also exist outside of the USA.
Some explanations of market anomalies are in line with the efficient market hypothesis.
Hudson et al. (2002) argue that academics are looking for general, long-term laws and statistical
significance and have less interest in whether their information is up-to date. Investors look for trends
and up-to datedness of their information which is crucial for them. These different objectives make it
hard for investors to turn academic knowledge into real profit. Zhang and Jacobsen (2012) showed that
investors need a long investment horizon to generate profit on taking a position against the calendar
anomalies. These limitations to generate profit could explain why the anomalies are not arbitrated
away. Another reason could be the existence of unobserved risk.
In this paper I conducted a research to examine if the January effect, the December effect and
the sell-in-May existed during the period June 1963 to December 2014 and the weekend effect during
the period 2010-2014 on the NYSE, AMEX and NASDAQ stock exchange. I also created different

3
sub-periods and size and value based portfolios as a robustness check. I examine the weekend effect
and the January effect according to the method of Agrawal and Tandon (1994). The sell-in-May effect
is examined according to a method described in section 4. Both of the methods use an ordinary OLS-
regression.
The result of the empirical analysis is that the January, December and sell-in-May effects are
positive and significant and that de weekend effect only holds true for small stocks. The remainder of
this thesis is organized as follows: I present the literature research in the next section. Section 3
contains the data description. Section 4 explains the methodology used to derive the results. Section 5
contains the result of the regression analysis and its interpretation. Section 6 concludes and provides
some recommendations for further research.

4
2 Literature Review

2.1 The January effect


The January effect was first discovered by Rozeff and Kinney (1976). Using stock returns of the
NYSE between 1904-74 they show that January had an average return of 3,48 percent in contrast to
the other months who had an average return of 0,42 percent (Rozeff and Kinney, 1976). For their
calculation Rozeff and Kinney (1976) have used an equally-weighted-index. If they would have used a
value-weighted-index they would not have found a January effect. This is because, according to Ritter
(1988), the January effect only holds true for small firms. Blume and Stambaugh (1983) were the first
ones to completely attribute the size effect to the January effect. According to their results the size
effect only takes place during the month January, contradicting the results of Fama and French (1992)
who found a size effect over the whole year.
The January effect is confirmed by many studies. Roll (1983) offered a tax-related
explanation. His theory states that investors tend to sell stocks that had a negative return over the past
year so they can deduct the losses from their official income. This causes the price to decline further.
After the losses are deducted, stock prices will rise to their initial equilibrium again. Reinganum
(1983) and Roll (1983) conclude that the January effect only holds true for small-capitalization
companies, especially during the first days of January. According to their research these returns cannot
be fully explained by tax-reasons. Their results could therefore challenge the efficient market
hypothesis.
However, Zhang and Jacobsen (2012) find that the January effect emerged first in the UK
around 1830. A capital gain tax was introduced in England in 1965, and the tax year has always started
in April instead of in January. Therefore they state that the January effect cannot be explained by tax
reasons, but is somehow related to Christmas holidays.
An explanation for the January effect could be the insider-trading/ information-release
hypothesis. This hypothesis states that management receives non-public information in the month
January. They can use this information to trade, creating a disadvantage for the public which does not
have this information. In order to compensate for the disadvantage the public demands a higher return.
The theory does not explain the reversion effect compared to the previous year and it does not explain
why this happens only to small firms. Besides that there is little empirical evidence to back this theory
(Ritter, 1988). Rozeff and Kenny (1976) found that the CAPM-beta shows seasonal patterns, meaning
that risk is higher priced during January. This theory however does not provide an explanation, it
provides an observation, namely higher stock returns during January.
Ritter (1988) states in his research that on the New York Stock Exchange (NYSE) between
1971-1985 the differences in returns between the smallest and largest decile was 8,17% on average.
His explanation is as follows: The January effect is indeed caused by tax motivated buy-sell behavior.

5
The reason that this only happens to small firms is because institutional investors, who mostly invest
in big firms, are much less affected by tax-incentives. Individual investors, who tend to invest in small
firms, are affected by tax-incentives. Individual investors tend to “park” their proceeds in their
brokerage account for a period of time before they re-invest it. This causes the stock prices of small
firms to decline in December and to rise in January. The empirical research of Ritter (1988) confirms
his theory and is cited by many researchers studying the January effect. D’Mello, Ferris, and Hwang
(2003) confirm the tax-related explanations of Ritter (1988). However, according to Haug and
Hirschey (2006) several tax regulations were reformed causing the tax-incentives for big institutions to
disappear 1986. This is not in accordance with the theory of Ritter (1988) and his empirical time
horizon. Haug and Hirschey (2006) introduced the window-dressing-hypothesis. According to the
window-dressing-hypothesis institutions do not sell losers for tax, rather than for reputational benefits.
Getting losing stocks off the books would improve the reputation of the company when it publishes its
end-of-the-year statement. The Tax-Reform-Act (TRA) of 1986 took away the tax-benefits
institutional investors could gain by selling losers in December. Haug and Hirschey (2006) examined
the period after the TRA to find proof for the window-dressing-hypothesis. Their results show that the
January effect still exists after the period 1986 for small companies. If the assumption of Ritter (1988)
concerning the individual investor and his preference for small stocks is true, the result of Haug and
Hirschey (2006) does not have to be true because of the window-dressing-hypothesis. This because
individual investors still have a tax-based incentive to sell losers in December and to buy them back in
January. Remarkable is that the results of Ritter (1988) only hold true for small firms and not, as Haug
and Hirschey (2006) argue, also for big firms until 1986.
In an efficient market, market-anomalies should be arbitrated away. Stephan et al. (2009)
therefore conducted a more recent research to the January effect. In their research they examined US
stock markets. They conclude that the January effect still exists on the NYSE but also appears in new
stock exchanges like the NASDAQ which emerged in the seventies. In contradiction to the efficient
market hypothesis the volume of trading is not higher during December and January, meaning that
traders are not attempting to arbitrate the anomaly away. This means that the January anomaly has not
been solved yet, and further research could provide new insights concerning this problem.
Agrawal and Tandon (1994), who’s method I am following, showed that returns in January are
large in most countries they have examined.

2.2 The December effect


According to Ritter (1988) the January effect is caused by individual investors who dump their losing
stocks in December causing the prices to decline. In January they ought to rise back to their initial
equilibrium. Agrawal and Tandon (1994) also find evidence for the tax-loss selling hypothesis.
There is another effect that could influence the returns during December. It is called the

6
holiday effect. The holiday effect is the effect that stock returns on the last day before a holiday are
higher than during other days. Since the second part of December has two holidays, namely the
Christmas holidays and New Year’s Eve, stock returns should be higher during December than during
other months. Roll (1983) observed unusual high stock returns on the last day of December.
Lakonishok and Smidt (1988) find that there is a large increase of stock returns on the DJIA. They
observe an increase of returns of 0,248 percent per day during the inter-holiday period and 0,386
percent per day on the two preholiday days over a period of 90 years. Agrawal and Tandon (1994) find
evidence of positive pre-Christmas returns in seven countries and in almost all of the countries large
positive inter-holiday returns. An explanation for this could be the short-selling theory (Ariel, 1990)
which is explained in section 2.5. Chong et al. (2005) find in their analysis that there is a holiday effect
in the US over the period January 1973 until July 2003. However, their results show that the effect has
been declining. The effect was reversed during the period 1994-2000 and has not been strong over the
period 2000-2003. An explanation Chong et al. (2005) offer is that traders trade too actively against
the anomaly causing a reversal effect.
The tax-related theories state that the December effect should be negative. The holiday related
theories state that the December effect should be positive. This paper examines whether the December
effect is positive or negative and/or significant.

2.4 Sell-in-May effect

Bouman and Jacobsen (2002) examined the sell-in-May effect, also called the “Halloween indicator”.
The sell-in-May effect is the effect that winter returns (November-April) are significantly higher than
summer returns (May-October). Their conclusion is that there exists a sell-in-May effect in 36 out of
37 countries in their sample. The effect is especially significant in European countries.
The sell-in-May effect is an interesting effect for a couple of reasons. Firstly because the effect
does not seem to be bound to a particular period or stock market. Secondly the anomaly is not
arbitrated away by investors after its discovery. In addition to this a simple investment strategy could
make a lot of profit with this anomaly. Besides this data snooping bias seems an unlikely explanation
for this anomaly which makes this anomaly a special one (Bouman and Jacobsen, 2002).
Swinkels and van Vliet (2012) examined five market anomalies. The Halloween effect,
January effect, turn-of-the-month effect, weekend effect and the holiday effect. Their conclusion is
that if one controls for the other seasonal effects only two seasonal effects seem to be significant.
According to their research only the turn-of-the-month and the sell-in-May effects are significant.
Zhang and Jacobsen (2012) argue that calendar effects in general can be the result of data
snooping. They considered that the statistical evidence that is delivered by scientists concerning
several calendar effects is being based on a too short sample period. They therefore examined UK
stock returns over a period of 300 years starting in 1693. They show that the January effect emerged

7
around 1830 coinciding with Christmas becoming a public holiday. In fact they also show that most
months had their 50 years of fame, emphasizing the importance of using long term datasets.
Remarkable is the persistence of the sell-in-May effect which turns out to be significant over the
whole period and in all the 50-year subsamples they used. Just like Bouman and Jacobsen (2002),
Zhang and Jacobsen (2012) developed a trading strategy using 100, 50 and 30 year horizons in which
they trade against the sell-in-May anomaly. Although their investment strategy seems successful, they
acknowledge the limitations for investors, since most investors have a shorter investment horizon.
Despite the fact that the anomaly is very old and a lot of research has been conducted on this
phenomenon, rational explanations are missing. Some researchers point out the issue of data snooping,
lack of economic significance or to the higher volatility of the markets during the summer. Bouman
and Jacobsen (2002) refute these arguments, leaving open more behavioral explanations. Bouman and
Jacobsen (2002) found a negative significant relation between the level of outbound traveling and
stock returns. It could be that investors need their money to go on vacation or maybe they do not want
to worry about their money while on vacation. These behavioral answers do not explain why this
anomaly is not arbitrated away.

2.5 The weekend effect

According to the weekend effect hypothesis, stock market returns tend to be higher on Fridays and
lower on Mondays. French (1980) found that returns on Monday were significantly negative during
the most five-year sub periods, as well as over the full period of the S&P composite portfolio from
1953 to 1977. This result is remarkable since according to the calendar time hypothesis the related
companies continue to add value over time. So the return of the stock should be three times the return
of another weekday. The effect seemed, especially for the larger firms, to disappear between 1980 and
1990 (Jones and Shemesh, 2010). Investors cannot trade during the weekend. However, they can
observe information which they incorporate in their investment decision. If one assumes that a trader
observes as much information during a weekend day as during a weekday, the volatility of Monday
price movements should be √3 times higher than a regular day (Fortune, 1999). Roll and French
(1986) observed that stock price volatilities are higher from the beginning to the end of a trading day,
than from the end to the beginning of the week, despite the fact that more time has passed. The results
of Fortune (1999) are in line with these results. Keim and Stambaugh (1984) show in their paper that
there are lower returns on Friday when it is possible to trade the next day. Agrawal and Tandon (1994)
show that the lowest and negative returns are on Mondays in nine and on Tuesday in eight countries.
They also find large and significant stock returns on Fridays. They also observe the highest return
variances on Monday and the lowest on Fridays.
Several explanations have been suggested and investigated. An explanation would be the delay

8
between trading and settlement in stocks. The settlement of a stock occurs three business days later
than the trade. Thus buyers on Monday and Tuesday pay during the same week, however buyers on
Wednesday, Thursday and Friday pay five days later enjoying two more days of interest free credit.
The Monday buyers should therefore be compensated with lower buy prices. Another explanation is
the dividend exclusion hypothesis. The dividend exclusion hypothesis states that ex-dividend dates
tend to cluster on Mondays, explaining a part of the price decline. The information release hypothesis
states that companies tend to announce good news during the week to push stock prices up, and bad
news during the weekend hoping that a part of the bad news will be forgotten by Monday (Fortune,
1999). According to Chen and Singal (2003) deeper research of these theories does not provide any
satisfying explanation. Chen and Singal (2003) argue that the weekend effect is caused by short
sellers. The theory states that since short selling contains theoretically an unlimited risk, because the
prices can move up, short sellers would like to cover their positions during the weekends since they
are unable to trade then. Indeed they find that the weekend effect is strongly observed for stocks with a
relatively high short interest rate. They also find that the weekend effect weakens after 1977 and
disappears after 1990 for stocks with put options. Chen and Singal (2003) argued that the elimination
of the weekend effect of large stocks is caused by options that were introduced during that time. Jones
and Shemesh (2010) argue that the introduction of options only shifts the risk to the writers of the
options. Therefore the weekend effect should be observable in the option market. Their research
covering all US listed options on equities, indexes and ETFs shows a weekend effect in the option
markets confirming the theory of Chen and Singal (2003). The limitation of the research of Jones and
Shemesh (2010) is that they do not observe a change in volume of options in a weekly pattern.

2.6 Criticism
The papers exploring and describing the existence of market anomalies have been criticized in several
papers. Keim (1989) has studied the bid-ask bias. When an OTC trading of a stock does not occur,
analysts tend to take the average, or a number consisting out of a combination, of the bid- and ask
price. For example, small firms tend to experience a high frequency of trading at the end of the year
and a low frequency of trading at the beginning of the year. Keim (1989) argues that the observed
price can either be the bid price, the ask price or an average of the two prices. The ratio of the bid-to-
ask price is at its highest point in December, the moment when the end-of-the-day price is at its closest
to the bid price, and at its lowest point in January. Since the bid-to-ask price ratio stays below one over
the whole period, there is a trading-pattern bias in the calculation of the returns of end-of-December
and end-of-January prices. This bias could explain 1,5% to 2,5% of the returns of the turn-of-the-year
effect. This effect could also explain a part of the weekend anomaly. Observed prices tend to be close
to the ask price on Friday and close to the bid price on Monday. Keim (1989) found in his research an
overall bias of 0,04%, on Monday a bias of -0,031% and on Friday a bias of 0,078%.

9
Another critique is the data snooping critique. In economics in general it is not possible to
isolate a phenomenon, generate a result or test a hypothesis. Most of the time a hypothesis is being
created with and tested on the available data. Calendar anomalies are intensively studied by
researchers for every month, week and day. The data snooping critique is based on the fact that the
seasonal anomalies are deducted from empirical data, translated into a hypothesis and then tested on
the same data again. Sullivan et al. (1998) concluded that many calendar anomalies are significant in
an in-sample test. However they turn out to be insignificant or even contradicting if they are being
tested out- of sample. In their paper they have used a bootstrap snooper to adjust the results for data
snooping. They have examined the DJIA for the period 1897-1996. Their conclusion is that in none of
the sub-samples, the full period nor in the out- of sample tests, can calendar anomalies outperform
their benchmark market index. They have not examined small and stocks with a high book to market
ratio in particular, which could influence their result. However, they do find robust evidence for the
size-January effect.

10
3. Data description:
The book equity values are downloaded from the COMPUSTAT database. Since COMPUSTAT has a
bias towards big firms prior to 1962 (Fama and French, 1995) I have selected the time horizon 1962-
2014. In COMPUSTAT I have selected all the data from 1960-12 until 2014-12 because the shares are
included in the regression once they are in COMPUSTAT for at least 2 years. This means that I have
excluded the first 2 years of observations in COMPUSTAT which is in line with Fama and French
(1995). The book value is measured on an annual basis. It is calculated according to the method used
by Fama and French (1993). I first took the total parent stockholders’ equity (SEQ). Then I added the
Deferred Taxes and Investment Tax Credit (TXDITC) where available, after which I subtracted the
preferred stock values. The preferred stock value consists of one of three variables. Wherever it was
available the Preferred Stock Redemption Value (PSTKRV) was used. If the PSTKRV was missing
the Preferred Stock Liquidating Value (PSTKL) was used instead. As a last option I used the
Preferred/Preference Stock (Capital) – Total value (PSTK) variable, otherwise a missing value is
noted. All financial companies with a GIC code of 40 are removed. This is all in line with the data
mining method of Fama and French (1993). Negative equity values are dropped. In order to combine
the data from the CRSP and COMPUSTAT database I use CUSIP codes. Sometimes a CUSIP code
larger than 1E+10 is observed in the COMPUSTAT database. Since the Wharton system does not
accept CUSIP codes larger than 1E+10 these CUSIP codes are removed leaving us with 9877 different
securities observed in the COMPUSTAT file.

The CRSP data ranges from 1-6-1963 until 31-12-2014. It contains the monthly returns from
all stocks which are left after cleaning up the COMPUSTAT data. In line with the method used by
Fama and French (1993) only stocks with ordinary common equity are included in the regression. So
stocks with another share code than 10 and 11 are dropped. In accordance with the method of Fama
and French (1995) I only included stocks which are traded on the NYSE, AMEX or NASDAQ.
The returns are multiplied by 100, the size is calculated by multiplying the price times the
shares outstanding. I divided the size by a million in order to have the size in the same units as the
equity. The book-to-market ratio is calculated by dividing the book equity by the market equity. Book-
to-market ratios higher than 10 or lower than 0,1 are removed.
In addition to the monthly returns I also used daily returns. These returns are from the period
2010-2014 and are cleaned in the same way as the monthly returns.

11
4. Method
The average of the monthly returns and its significance are calculated according to the method used by
Agrawal and Tandon (1994). This is done using an ordinary OLS-regression as described in formula 1.
The month variables are dummies. With this formula the average returns and their significance per
month are calculated. I check whether January returns are indeed higher than the returns in other
months and whether December returns are either negative or positive. In order to check whether the
regression has heteroscedasticity a White-test is performed. In case heteroscedasticity is observed I
have used robust standard errors with a bias correction of 1/(1-h)2 which is the best correction to
prevent heteroscedasticity bias. After the regression is performed a Kruskal-Wallist (KW) is carried
out. A KW-test is the non-parametric test of the hypothesis that the returns of the twelve months are
drawn from the same population.

Ri,t= β1JANt + β2FEBt + β3MARt + β4APRILt + β5MAYt + β6JUNEt + β7JULYt + β8AUGt + β9SEPt +
β10OCTt + β11NOVt + β12DECt + εit (1)

H0: β1= β2=…= β11= β12 H1: β1≠ β2 v β2≠ β3 v … v β10≠ β11 v β11≠ β12 α = 0.05

The KW-test shows whether the month of the year does influence the returns. The resulting
coefficients show the average return and its significance during that month. According to the literature
the β1 coefficient should be higher than the other coefficients. According to the tax-related theory the
β12 coefficient should be lower than most of the other coefficients or even negative. However, the
holiday effect theory states that the β12 coefficient should be higher than most of the other coefficients.
Whether these statements are true is shown in the next section.

The sell-in-May effect is calculated according to the following formula:

Ri,t= β0 + β1WINTERt + εit (2)

H0: β1=0 H1: β1>0 α = 0.05

The WINTERt dummy is 1 for the months November until April. With this formula the average
winter return can be calculated and it can be checked whether these returns are significantly different
from the returns over the year. In order to check whether the effect has changed over time the
regression described in formula 1 and 2 is repeated over the sub-period June 1963 until December
1989 and over the sub-period January 1990 until December 2014.

As a robustness check the observations are divided into small and big and into high book-to-
market and low book-to-market ratio quintiles. I then perform the regressions in model 1 until 3 only
using the observations in the upper and lower quintile.

12
The weekend effect is examined the same way as the January and December effect. The
formula used is described in model 3. The weekdays described in the formula are dummies. The
equality of the means across all days of the week is tested by an F-test.

Ri,t= β1MONt + β2TUEt + β3WEDt + β4THURSt + β5FRIt + εit (3)

H0: β1 = β2 = β3 = β4 = β5 H1: β1≠ β2 v β2≠ β3 v β3≠ β4 v β4 ≠ β5 α = 0.05

Just as the coefficients in model 1 the resulting coefficients show the average return and its
significance during that day. I check whether Friday returns are indeed higher and if Monday returns
are lower compared to the other days. Just as in the first two models the observations containing the
upper and lower book-to-market and size quintiles are used as a robustness check.

The next section shows the results of the regression described in this chapter. The following
section summarizes, concludes and provides some recommendations for further research.

13
5. Results

5.1 Benchmark results

Table 4 shows the result of the benchmark regressions. The KW-test is significant in all the
regressions, meaning that it is unlikely that the returns over the months would be drawn from the same
distribution. The H0 hypothesis is rejected at the 0,01% significance level. The conclusion is that
monthly calendar effects do exist.
January is, in line with the literature as well as with the research of Agrawal and Tandon
(1994), the month with the highest returns over the whole year. Interesting is that the January effect
seems to be stronger prior to 1990 which is in line with the theory of Haug and Hirschey (2006) who
claim that the tax-incentive for big institutions disappeared in 1986. Another explanation could be that
in the eighties a lot of research has been conducted to the January effect and one of the results could be
that the January effect is therefore arbitrated away.
December returns over the whole sample are not low which is not in line with the results of
Agrawal and Tandon (1994). Apparently the US market differs from the markets in the rest of the
world. This is also not in line with the tax-related theory of Ritter (1988), who claims that high
January returns are related to low December returns because investors tend to dump their losing stocks
in December. The positive December returns might be the result of the holiday effect, confirming the
results of Lakonishok and Smidt (1988) and Roll (1983), who observed unusual high stock returns
during the holidays in December. My results contradict the results of Chong et al. (2005), who
observed a declining holiday effect. If the positive December effect is indeed caused by the holiday
effect, this effect increased in the second period.
Winter effects are positive and significant in all three regressions. The winter returns have
decreased with 0,415%. This might be the result of arbitrageurs trading against the anomaly. In order
to make sure that the winter coefficient is not the result of the January effect, I did a robustness check
by performing the regression described in model 4.

Ri,t= β0 + β1JANt + β2WINTERt + εit (4)

The result is that a part of the sell-in-May effect is caused by the January effect, but even when
a January coefficient is introduced the sell-in-May effect stays significantly positive.
My results also show that the returns over the whole period from June until October are lower
than during other months. This might confirm the outbound traveling theory of Bouman and Jacobsen
(2002), which states that there is a negative correlation between the numbers of outbound travelers and
stock returns because investors tend to sell their stocks if they go on a holiday.
Another remarkable thing is the high significance of almost all of the coefficients. Of all of the
coefficients of the 3 regressions only the September coefficient is not significant. All the averages are

14
significant. The R-squared value is low. The high significance of the calendar dummies and the low R-
squared value mean that calendar effects are persistent, but one needs a long investment horizon to
generate profit on it. This is in line with the findings of Zhang and Jacobsen (2012). The low R-
squared value could explain why the anomalies are not arbitrated away.

5.2 Robustness checks

In this sub-chapter I discuss the results of the regressions performed over the 2 smallest and largest
deciles and over the 2 highest and lowest book-to-market deciles. I did this to examine whether the
calendar effects persist over different deciles.

The results of the robustness check over the period June 1963 – December 2014 is provided in
table 5. My results show a difference between small stocks and stocks with a high book-to-market
ratio compared to big stocks and stock with a low book-to-market ratio. From February until
December most of the coefficients of small stocks and stocks with a high book-to-market ratio are
significantly negative while the coefficients of big stocks and stocks with a low book-to-market ratio
are significantly positive.
The January returns for the high and small deciles are the highest and are accompanied by
negative December returns. The combination of high January returns with negative December returns
for the small and high deciles might confirm the tax-related hypothesis of Ritter (1988). The tax-
related hypothesis of Ritter (1988) claims that the January effect only holds true for small firms
because the tax-related incentives only apply to individual investors who invest mostly in small firms.
December returns are high and significant for big stocks and stocks with a low book-to-market ratio. It
might be that the tax-related hypothesis applies stronger for small firms and firms with a high book-to-
market ratio and that the holiday effect is stronger for big firms and firms with a low book-to-market
ratio. These findings imply that calendar effects do differ over different deciles.
The negative averages of the high and small deciles are remarkable. They seem to be
accompanied by high winter returns. The robustness check shows that the high winter returns are
mostly due to the January effect. After adjusting for the January effect more regular winter effects are
observed.

The results of the robustness check over the two sub-periods are shown in table 6 and 7. The
January coefficient for the high deciles has declined by more than 50%. This might be related to the
theory of Haug and Hirschey (2006), which claims that the tax-incentive for big institutions
disappeared after 1986. It might be that before 1986 big institutions had the incentive to dump stocks
with a high book-to-market ratio and rebuy them in January. If this incentive disappeared this might
explain the decline of the January coefficient. In order to prove this more research to the tax incentives
for big institutions and the relation with the book-to-market ratio should be conducted. Also the rise of
the December coefficient for the big stocks could be a result of the disappeared incentive. It could be

15
that institutions are not dumping their stocks during December anymore, raising December average
returns for big stocks.
The January coefficient for the small deciles remains high. This also confirms the theory of
Haug and Hirschey (2006) that the tax-incentive remains for individual investors who mostly invest in
small firms.

5.3 Daily returns

The results of the daily regression are provided in table 9. The p-value of the F-test is significant in
every regression, meaning that the day of the week does influence returns. Average Monday returns
are, in line with the results of Agrawal and Tandon (1994), the lowest in 4 and significant in 3
regressions. Monday returns are significantly negative on a 5% level over the whole regression, which
is in line with the results of French (1980). Volatilities on Mondays are the highest in all 5 regressions
which is partly in line with the theory expressed in Fortune (1999), which predicted higher price
volatilities during Mondays.
The Friday effect holds true for stocks with a low book-to-market ratio and for small stocks.
The average Monday returns are significantly negative over the whole regression, the regression with
the small stocks and the regression with stocks with a high book-to-market ratio. The only regression
where a weekend effect is observed is the regression with the small stocks. For big stocks there does
not seem to be any weekend effect. This is in line with the theory of Jones and Shemesh (2010), who
claim that the effect for the big firms disappeared between 1980 and 1990, and with the results of
Chen and Singal (2003), who argued that the elimination of the weekend effect of large stocks are
caused by options that were introduced during those days. My results show that the weekend effect
does not hold for big stocks during the period 2010 to 2014. This might be the result of options which
are mostly available for big stocks.
The highest returns in the overall regression and the regression with the big stocks are on
Tuesday. This is not in line with the results of Agrawal and Tandon (1994) where Tuesday returns are
the lowest in eight out of 17 countries. My results could be due to a reversal effect caused by the low
Monday returns. This might be related to the information release hypothesis of Fortune (1999). If
companies announce bad new during the weekend, stock market returns on Monday should be low.
Stock markets could overreact to bad news. The high Tuesday returns could be the results of prices
moving into their equilibrium state again.
Just as in the previous regressions the R-squared is low, meaning that one needs a long
investment horizon in order to generate profit from this investment strategy.

16
6. Conclusion
In this thesis I conducted a research to the turn-of-the-year, the Sell-in-May and the weekend. The
literature written about these anomalies predict high returns during January, especially for small firms.
My results confirm the literature except for big stocks, where December returns are higher than
January returns. The literature is contradicting about the returns during December. According to the
holiday effect hypothesis returns during December should be higher than usual, but according to the
tax-related hypothesis returns should be lower. According to my results December returns are high
except for small firms and firms with a high book-to-market ratio. The low December returns for the
small firms and firms with a high book-to-market ratio coincide with high January returns, confirming
the tax-related hypothesis. It might be that the contradicting hypotheses both hold for different deciles.
According to the Sell-in-May hypothesis winter returns should be significantly higher than
summer returns. Indeed the existing literature and my results confirm the existence of a winter
premium. My results show that the winter premium is especially significant for small stocks and
stocks with a high book-to-market ratio, but that this is mostly due to the January effect.
According to my results the January and sell-in-May effect are weaker in the period 1990-
2014 compared to the period 1963-1989. The declining January effect is large for stocks with a high
book-to-market ratio. These stocks show a declining January effect of more than 50%. The January
effect for small stocks remains high.
Previously done research suggests high returns on Friday and low returns on Monday. My
results confirm the existence of a weekend effect, but only for small stocks, confirming the theory that
the weekend effect only holds true for stocks without options. For the overall regression the lowest
returns are observed on Mondays and the highest on Tuesdays; contradicting the literature.
The calendar anomalies are significant but their predictive power is low. This could be the
reason why the anomalies are not arbitrated away. The predictive power of the anomalies is not high
enough for an investor with a limited investment time horizon to generate profit from it.
Further research could be conducted to the relation between small companies, companies with
a high book-to-market ratio and the turn-of-the-year effect in comparison with big companies and
companies with a low book-to-market ratio. The declining January effect for high book-to-market
stocks could also be further investigated. The relation between the weekend effect, small companies
and the availability of options could provide some new insights.

17
7. References
Agrawal, Anup, and Kishore Tandon. "Anomalies or illusions? Evidence from stock markets in
eighteen countries." Journal of international Money and Finance 13.1 (1994): 83-106.

Ariel, R.A., 1990. High stock returns before holidays: existence and evidence on possible causes.
Journal of Finance 45 (5), 1611-1626.

Bouman, Sven, and Ben Jacobsen. "The Halloween indicator," Sell-in-May and go away": Another
puzzle." American Economic Review (2002): 1618-1635.

Chen, Honghui, and Vijay Singal. "Role of speculative short sales in price formation: The case of the
weekend effect." The Journal of Finance 58.2 (2003): 685-706.

Chong, Ryan, et al. "Pre-holiday effects: International evidence on the decline and reversal of a stock
market anomaly." Journal of International Money and Finance 24.8 (2005): 1226-1236.

D’Mello, Ranjan, Stephen P. Ferris, and Chuan Yang Hwang. 2003. “The Tax-Loss Selling
Hypothesis, Market Liquidity, and Price Pressure around the Turn-of-the-Year.” Journal of Financial
Markets, vol. 6, no. 1 (January):73–98.

Easterday, Kathryn E., Pradyot K. Sen, and Jens A. Stephan. "The persistence of the small
firm/January effect: Is it consistent with investors’ learning and arbitrage efforts?." The Quarterly
Review of Economics and Finance 49.3 (2009): 1172-1193.

Fama, Eugene F., and Kenneth R. French. "Common risk factors in the returns on stocks and
bonds." Journal of financial economics 33.1 (1993): 3-56.

Fama, Eugene F., and Kenneth R. French. "Size and book‐to‐market factors in earnings and
returns." The Journal of Finance 50.1 (1995): 131-155.

Fama, Eugene F., and Kenneth R. French. "The cross‐section of expected stock returns." the Journal
of Finance 47.2 (1992): 427-465

Fortune, Peter. "Are stock returns different over weekends? A jump diffusion analysis of the weekend
effect." New England Economic Review 10 (1999): 3-19.

French, Kenneth R. "Stock returns and the weekend effect." Journal of financial economics 8.1
(1980): 55-69.

French, Kenneth R., and Richard Roll. "Stock return variances: The arrival of information and the
reaction of traders." Journal of financial economics 17.1 (1986): 5-26.

Haug, Mark, and Mark Hirschey. "The january effect." Financial Analysts Journal 62.5 (2006): 78-88.

18
Hudson, R.; Keasey, K. and Littler, K., (2002): “Why investors should be cautious of the academic
approach to testing for stock market anomalies”, Applied Financial Economics, 12, 681-686.

Jones, Christopher S., and Joshua Shemesh. "The weekend effect in equity option returns." AFA 2010
Atlanta Meetings Paper. 2010.

Keim, Donald B. "Trading patterns, bid-ask spreads, and estimated security returns: The case of
common stocks at calendar turning points." Journal of Financial Economics 25.1 (1989): 75-97.

Keim, Donald B., and Robert F. Stambaugh. "A further investigation of the weekend effect in stock
returns." The journal of finance 39.3 (1984): 819-835.

Lakonishok, Josef, and Seymour Smidt. "Are seasonal anomalies real? A ninety-year
perspective." Review of Financial Studies 1.4 (1988): 403-425.

Marshall Blume and Robert Stambaugh. "Biases in Computed Returns: An Application to the Size
Effect." Journal of Financial Economics 12 (November 1983), 387-404.

Reinganum, Marc R. 1983. “The Anomalous Stock Market Behavior of Small Firms in January:
Empirical Tests for TaxLoss Selling Effects.” Journal of Financial Economics, vol. 12 (June):89–104

Richard Roll. "Vas ist das? The Turn of the Year Effect and the Return Premia of Small Firms."
Journal of Portfolio Management 9 (Winter 1983), 18-28.

Ritter, Jay R. 1988. “The Buying and Selling Behavior of Individual Investors at the Turn of the
Year.” Journal of Finance, vol. 43, no. 3 (July):701–719.

Rozeff, Michael S., and William R. Kinney. "Capital market seasonality: The case of stock
returns." Journal of financial economics 3.4 (1976): 379-402.

Sullivan, Ryan, Allan Timmermann, and Halbert White. "Dangers of data-driven inference: the case of
calendar effects in stock returns." Department of Economics, UCSD (1998).

Swinkels, Laurens, and Pim van Vliet. "An anatomy of calendar effects."Journal of Asset
Management 13.4 (2012): 271-286.

Zhang, Cherry Y., and Ben Jacobsen. "Are monthly seasonals real? A three century
perspective." Review of Finance (2012): rfs035.

19
8. Appendix
Table 1: Summary statistics period June 1963 – December 2014
The statistics cover the period 1-6-1963 until 31-12-2014. The dependent variable is return (%). The first 2 years of observations in the
COMPUSTAT database are removed. The book value is measured on an annual basis. It is calculated according to the method used by Fama
and French (1993). I first took the total parent stockholders’ equity (SEQ). Then I added the Deferred Taxes and Investment Tax Credit
(TXDITC) where available, after which I subtracted the preferred stock values. The preferred stock value consists of one of three variables.
Wherever it was available the Preferred Stock Redemption Value (PSTKRV) was used. If the PSTKRV was missing the Preferred Stock
Liquidating Value (PSTKL) was used instead. As a last option I used the Preferred/Preference Stock (Capital) – Total value (PSTK) variable,
otherwise a missing value is noted. The size is calculated by multiplying the shares outstanding times the price. The size and return are
measured on a monthly basis and are downloaded from the CRSP database. Book values are downloaded from COMPUSTAT. All financial
companies with a GIC code of 40 are removed. Negative equity values are dropped. Only stocks with ordinary common equity which are
traded on the NYSE, AMEX or NASDAQ are included. Book-to-market ratios higher than 10 and lower than 0,1 are dropped. The book-to-
market and size sub-samples are the upper and lower quintiles of these variables.

Variable Obs. Mean Std. Dev. Min. Max.


return (%) 933266 1,252857 19,18824 -98,1295 1349,505
BE/ME 231609 0,985337 0,8903096 0,1000006 9,979913
size/106 939594 1958,79 12001,12 0,0105625 602432,9

Table 2: Summary statistics period June 1963 – December 1989


The statistics cover the period 1-6-1963 until 31-12-1989. The dependent variable is return (%). The first 2 years of observations in the
COMPUSTAT database are removed. The book value is measured on an annual basis. It is calculated according to the method used by Fama
and French (1993). I first took the total parent stockholders’ equity (SEQ). Then I added the Deferred Taxes and Investment Tax Credit
(TXDITC) where available, after which I subtracted the preferred stock values. The preferred stock value consists of one of three variables.
Wherever it was available the Preferred Stock Redemption Value (PSTKRV) was used. If the PSTKRV was missing the Preferred Stock
Liquidating Value (PSTKL) was used instead. As a last option I used the Preferred/Preference Stock (Capital) – Total value (PSTK) variable,
otherwise a missing value is noted. The size is calculated by multiplying the shares outstanding times the price. The size and return are
measured on a monthly basis and are downloaded from the CRSP database. Book values are downloaded from COMPUSTAT. All financial
companies with a GIC code of 40 are removed. Negative equity values are dropped. Only stocks with ordinary common equity which are
traded on the NYSE, AMEX or NASDAQ are included. Book-to-market ratios higher than 10 and lower than 0,1 are dropped. The book-to-
market and size sub-samples are the upper and lower quintiles of these variables.

Variable Obs. Mean Std. Dev. Min. Max.


return (%) 352060 1,24339 14,86636 -85 556,25
BE/ME 124026 1,108676 0,8369842 0,100012 9,979913
size/106 354280 485,366 2178,389 0,0385 102022,3

Table 3: Summary statistics period January 1990 - December 2014


The statistics cover the period horizon 1-1-1990 until 31-12-2014. The dependent variable is return (%). The first 2 years of observations in
the COMPUSTAT database are removed. The book value is measured on an annual basis. It is calculated according to the method used by
Fama and French (1993). I first took the total parent stockholders’ equity (SEQ). Then I added the Deferred Taxes and Investment Tax Credit
(TXDITC) where available, after which I subtracted the preferred stock values. The preferred stock value consists of one of three variables.
Wherever it was available the Preferred Stock Redemption Value (PSTKRV) was used. If the PSTKRV was missing the Preferred Stock
Liquidating Value (PSTKL) was used instead. As a last option I used the Preferred/Preference Stock (Capital) – Total value (PSTK) variable,
otherwise a missing value is noted. The size is calculated by multiplying the shares outstanding times the price. The size and return are
measured on a monthly basis and are downloaded from the CRSP database. Book values are downloaded from COMPUSTAT. All financial
companies with a GIC code of 40 are removed. Negative equity values are dropped. Only stocks with ordinary common equity which are
traded on the NYSE, AMEX or NASDAQ are included. Book-to-market ratios higher than 10 and lower than 0,1 are dropped. The book-to-
market and size sub-samples are the upper and lower quintiles of these variables.

Variable Obs. Mean Std. Dev. Min. Max.


return (%) 581206 1,258591 21,38555 -98,1295 1349,505
BE/ME 107583 0,843146 0,927949 0,100001 9,969596
size/106 585314 2850,628 15040,69 0,0105625 602432,9

20
Table 4: Benchmark results
The regression covers the time horizon 1-6-1963 until 31-12-2014 and the two sub-periods 1-6-1963 – 31-12-1989 and 01-01-1990 – 31-12-
2014. The dependent variable is return (%). The average expresses the sample average. The WINTER dummy is 1 for the months November
until April. The WINTER coefficient is the coefficient in the regression Ri,t= β0 + β1WINTERt + εit, The robustness check for the WINTER
coefficient is the WINTER coefficient in the regression Ri,t= β0 + β1JANt + β2WINTERt + εit. The first 2 years of observations in the
COMPUSTAT database are removed. The book value is measured on an annual basis. It is calculated according to the method used by Fama
and French (1993). I first took the total parent stockholders’ equity (SEQ). Then I added the Deferred Taxes and Investment Tax Credit
(TXDITC) where available, after which I subtracted the preferred stock values. The preferred stock value consists of one of three variables.
Wherever it was available the Preferred Stock Redemption Value (PSTKRV) was used. If the PSTKRV was missing the Preferred Stock
Liquidating Value (PSTKL) was used instead. As a last option I used the Preferred/Preference Stock (Capital) – Total value (PSTK) variable,
otherwise a missing value is noted. The size is calculated by multiplying the shares outstanding times the price. The size and return are
measured on a monthly basis and are downloaded from the CRSP database. Book values are downloaded from COMPUSTAT. All financial
companies with a GIC code of 40 are dropped. Negative equity values are dropped. Only stocks with ordinary common equity which are
traded on the NYSE, AMEX or NASDAQ are included. Book-to-market ratios higher than 10 and lower than 0,1 are dropped.

07/1963-12/2014 07/1963-12/1989 01/1990-12/2014

JANUARY 6.006*** 7.253*** 5.266***


(0,0854) (0,111) (0,119)
FEBRUARY 1.323*** 1.805*** 1.034***
(0,0692) (0,0856) (0,0979)
MARCH 1.421*** 1.824*** 1.180***
(0,0655) (0,0864) (0,0909)
APRIL 1.465*** 1.363*** 1.527***
(0,0682) (0,0813) (0,0976)
MAY 1.460*** 0.436*** 2.073***
(0,0671) (0,0851) (0,0943)
JUNE 0.644*** 0.772*** 0.567***
(0,0628) (0,0806) (0,088)
JULY -0.276*** 0.370*** -0.671***
(0,0606) (0,0782) (0,0851)
AUGUST 0.163*** 1.068*** -0.390***
(0,0622) (0,0788) (0,0879)
SEPTEMBER 0.00688 -0.11 0.0785
(0,0625) (0,0797) (0,0883)
OCTOBER -0.335*** -1.691*** 0.499***
(0,0691) (0,0934) (0,0955)
NOVEMBER 1.426*** 0.877*** 1.764***
(0,0773) (0,0863) (0,113)
DECEMBER 1.792*** 1.161*** 2.182***
(0,0691) (0,0815) (0,0999)
P-value KW-test <0,0001 <0,0001 <0,0001
R-squared 0.011 0.025 0.008
Average 1.253*** 1.243*** 1.259***
(0,0199) (0,0251) (0,0281)
Panel B: Ri,t= β0 + β1WINTERt + εit
WINTER 1.961*** 2.225*** 1.81***
(0.03967) (0.0499) (0.056)
Panel C: Ri,t= β0 + β1JANt + β2WINTERt + εit
Robustness check 1.212*** 1.265*** 1.179***
WINTER (0.0408) (0.0507) (0.0579)

21
Observations 933,266 352,060 581,206
Securities 6,421 2,548 5,448
Error Robust Robust Robust

Table 5: Robustness check period June 1963 – December 2014


The regression covers the time horizon 1-6-1963 until 31-12-2014. The dependent variable is return (%). The average expresses the sample
average. The WINTER dummy is 1 for the months November until April. The WINTER coefficient is the coefficient in the regression Ri,t= β0
+ β1WINTERt + εit, The robustness check for the WINTER coefficient is the WINTER coefficient in the regression Ri,t= β0 + β1JANt +
β2WINTERt + εit. The first 2 years of observations in the COMPUSTAT database are removed. The book value is measured on an annual
basis. It is calculated according to the method used by Fama and French (1993). I first took the total parent stockholders’ equity (SEQ). Then
I added the Deferred Taxes and Investment Tax Credit (TXDITC) where available, after which I subtracted the preferred stock values. The
preferred stock value consists of one of three variables. Wherever it was available the Preferred Stock Redemption Value (PSTKRV) was
used. If the PSTKRV was missing the Preferred Stock Liquidating Value (PSTKL) was used instead. As a last option I used the
Preferred/Preference Stock (Capital) – Total value (PSTK) variable, otherwise a missing value is noted. The size is calculated by multiplying
the shares outstanding times the price. The size and return are measured on a monthly basis and are downloaded from the CRSP database.
Book values are downloaded from COMPUSTAT. All financial companies with a GIC code of 40 are dropped. Negative equity values are
dropped. Only stocks with ordinary common equity which are traded on the NYSE, AMEX or NASDAQ are included. Book-to-market ratios
higher than 10 and lower than 0,1 are dropped.

07/1963-12/2014 Low BE/ME High BE/ME Small stocks Big Stocks

JANUARY 5.066*** 8.727*** 11.12*** 2.170***


(0.301) (0.388) (0.263) (0.109)
FEBRUARY 2.534*** -0.899*** 1.074*** 2.439***
(0.300) (0.268) (0.193) (0.119)
MARCH 2.172*** -0.0404 0.238 2.194***
(0.298) (0.263) (0.201) (0.0919)
APRIL 1.682*** -0.463* -0.639*** 2.612***
(0.307) (0.246) (0.183) (0.116)
MAY 1.226*** -1.546*** -1.527*** 1.631***
(0.306) (0.223) (0.174) (0.102)
JUNE 2.682*** -1.134*** -0.339* 1.738***
(0.306) (0.235) (0.188) (0.0855)
JULY 0.392 -1.881*** -1.512*** 0.833***
(0.302) (0.219) (0.165) (0.0904)
AUGUST 1.055*** -2.032*** -2.696*** 1.446***
(0.303) (0.248) (0.174) (0.0907)
SEPTEMBER 2.291*** -4.021*** -2.377*** 0.769***
(0.305) (0.231) (0.181) (0.0943)
OCTOBER 1.243*** -3.287*** -4.019*** 2.429***
(0.307) (0.288) (0.187) (0.108)
NOVEMBER 2.631*** -1.932*** -2.575*** 2.788***
(0.306) (0.305) (0.203) (0.102)
DECEMBER 2.574*** -2.544*** -3.208*** 3.656***
(0.304) (0.255) (0.203) (0.104)
P-value KW-test <0,0001 <0,0001 <0,0001 <0,0001
R-squared 0.016 0.036 0.024 0.029
Average 2.135 -1.012 -0.5755 2.062
0.0878 0.0782 0.0568 0.0294
Panel B: Ri,t= β0 + β1WINTERt + εit

22
WINTER 1.306*** 2.691*** 3.042*** 1.176***
(0.175) (0.157) (0.113) (0.0587)
Robustness check 0.843*** 1.116*** 1.040*** 1.271***
WINTER (0.184) (0.156) (0.114) (0.0618)
Panel C: Ri,t= β0 + β1JANt + β2WINTERt + εit
Observations 46,159 46,288 186,382 187,448
Securities 2,246 1,580 3.425 2,034
Error Normal Robust Robust Robust

Table 6: Robustness check period June 1963 – December 1989


The regression covers the time horizon 1-6-1963 until 31-12-1989. The dependent variable is return (%). The average expresses the sample
average. The WINTER dummy is 1 for the months November until April. The WINTER coefficient is the coefficient in the regression Ri,t= β0
+ β1WINTERt + εit, The robustness check for the WINTER coefficient is the WINTER coefficient in the regression Ri,t= β0 + β1JANt +
β2WINTERt + εit. The first 2 years of observations in the COMPUSTAT database are removed. The book value is measured on an annual
basis. It is calculated according to the method used by Fama and French (1993). I first took the total parent stockholders’ equity (SEQ). Then
I added the Deferred Taxes and Investment Tax Credit (TXDITC) where available, after which I subtracted the preferred stock values. The
preferred stock value consists of one of three variables. Wherever it was available the Preferred Stock Redemption Value (PSTKRV) was
used. If the PSTKRV was missing the Preferred Stock Liquidating Value (PSTKL) was used instead. As a last option I used the
Preferred/Preference Stock (Capital) – Total value (PSTK) variable, otherwise a missing value is noted. The size is calculated by multiplying
the shares outstanding times the price. The size and return are measured on a monthly basis and are downloaded from the CRSP database.
Book values are downloaded from COMPUSTAT. All financial companies with a GIC code of 40 are dropped. Negative equity values are
dropped. Only stocks with ordinary common equity which are traded on the NYSE, AMEX or NASDAQ are included. Book-to-market ratios
higher than 10 and lower than 0,1 are dropped.

07/1963-12/1989 Low BE/ME High BE/ME Small stocks Big Stocks

JANUARY 4.494*** 11.98*** 12.02*** 3.285***


(0.324) (0.472) (0.371) (0.136)
FEBRUARY 2.338*** 0.00250 2.076*** 1.363***
(0.298) (0.299) (0.282) (0.110)
MARCH 2.196*** 0.647** 0.851*** 1.937***
(0.274) (0.294) (0.288) (0.108)
APRIL 2.026*** -0.681*** -0.502** 1.846***
(0.281) (0.250) (0.251) (0.107)
MAY 1.718*** -0.580** -0.999*** 2.017***
(0.304) (0.244) (0.237) (0.107)
JUNE 1.136*** -1.794*** -0.872*** 1.206***
(0.301) (0.246) (0.288) (0.103)
JULY 0.455 -0.946*** -0.448* 0.687***
(0.290) (0.244) (0.236) (0.114)
AUGUST 2.833*** -1.637*** -2.151*** 2.667***
(0.289) (0.253) (0.232) (0.111)
SEPTEMBER 1.708*** -2.876*** -2.198*** 0.236**
(0.291) (0.235) (0.253) (0.103)
OCTOBER 0.132 -3.292*** -4.434*** 0.320**
(0.350) (0.292) (0.254) (0.149)
NOVEMBER 2.229*** -2.170*** -3.204*** 2.448***
(0.288) (0.287) (0.238) (0.111)
DECEMBER 2.723*** -2.179*** -2.304*** 2.019***
(0.283) (0.275) (0.253) (0.102)

23
P-value KW-test <0,0001 <0,0001 <0,0001 <0,0001
R-squared 0.028 0.076 0.037 0.045
Average 2.024 -0.474 -0.2776 1.666
0.086 0.0853 0.0784 0.0331
Panel B: Ri,t= β0 + β1WINTERt + εit
WINTER 1.343*** 2.878*** 3.221*** 0.9601***
(0.173) (0.171) (0.156) (0.0661)
Panel C: Ri,t= β0 + β1JANt + β2WINTERt + εit
Robustness check 0.972*** 0.901*** 1.177*** 0.738***
WINTER (0.178) (0.163) (0.156) (0.0675)
Observations 24,745 24,788 70,126 70,786
Securities 967 745 1,352 654
Error Robust Robust Robust Robust

Table 7: Robustness check period January 1990 – December 2014


The regression covers the time horizon 1-1-1990 until 31-12-2014. The dependent variable is return (%). The average expresses the sample
average. The WINTER dummy is 1 for the months November until April. The WINTER coefficient is the coefficient in the regression Ri,t= β0
+ β1WINTERt + εit, The robustness check for the WINTER coefficient is the WINTER coefficient in the regression Ri,t= β0 + β1JANt +
β2WINTERt + εit. The first 2 years of observations in the COMPUSTAT database are removed. The book value is measured on an annual
basis. It is calculated according to the method used by Fama and French (1993). I first took the total parent stockholders’ equity (SEQ). Then
I added the Deferred Taxes and Investment Tax Credit (TXDITC) where available, after which I subtracted the preferred stock values. The
preferred stock value consists of one of three variables. Wherever it was available the Preferred Stock Redemption Value (PSTKRV) was
used. If the PSTKRV was missing the Preferred Stock Liquidating Value (PSTKL) was used instead. As a last option I used the
Preferred/Preference Stock (Capital) – Total value (PSTK) variable, otherwise a missing value is noted. The size is calculated by multiplying
the shares outstanding times the price. The size and return are measured on a monthly basis and are downloaded from the CRSP database.
Book values are downloaded from COMPUSTAT. All financial companies with a GIC code of 40 are dropped. Negative equity values are
dropped. Only stocks with ordinary common equity which are traded on the NYSE, AMEX or NASDAQ are included. Book-to-market ratios
higher than 10 and lower than 0,1 are dropped.

01/1990-12/2014 Low BE/ME High BE/ME Small stocks Big Stocks

JANUARY 6.207*** 4.239*** 10.27*** 1.825***


(0.516) (0.632) (0.389) (0.141)
FEBRUARY 2.915*** -1.839*** -0.0508 2.690***
(0.511) (0.473) (0.269) (0.159)
MARCH 2.356*** -1.369*** -0.959*** 2.297***
(0.513) (0.439) (0.286) (0.117)
APRIL 1.280** -0.445 -0.961*** 2.772***
(0.517) (0.482) (0.278) (0.135)
MAY 1.647*** -3.126*** -2.301*** 1.283***
(0.512) (0.411) (0.260) (0.130)
JUNE 3.463*** 0.624 0.668** 1.786***
(0.515) (0.482) (0.272) (0.106)
JULY 0.428 -3.073*** -2.649*** 0.814***
(0.515) (0.407) (0.245) (0.114)
AUGUST 0.402 -3.485*** -4.036*** 1.108***
(0.518) (0.472) (0.254) (0.115)
SEPTEMBER 3.407*** -6.250*** -3.352*** 0.826***
(0.516) (0.439) (0.260) (0.122)
OCTOBER 1.731*** -2.918*** -3.237*** 2.903***

24
(0.523) (0.547) (0.277) (0.136)
NOVEMBER 3.088*** -1.987*** -2.028*** 2.863***
(0.524) (0.611) (0.333) (0.128)
DECEMBER 3.009*** -3.135*** -4.594*** 3.854***
(0.526) (0.501) (0.305) (0.140)
P-value KW-test <0,0001 <0,0001 <0,0001 <0,0001
R-squared 0.018 0.022 0.019 0.031
Average 2.494 -1.924 -1.106 2.09
0.1496 0.1445 0.084 0.0374
Panel B: Ri,t= β0 + β1WINTERt + εit
WINTER 1.295*** 2.295*** 2.776*** 1.270***
(0.299) (0.289) (0.168) (0.0748)
Panel C: Ri,t= β0 + β1JANt + β2WINTERt + εit
Robustness check 0.677** 1.308*** 0.779*** 1.447***
WINTER (0.314) (0.296) (0.170) (0.0786)
Observations 21,420 21,501 116,372 116,803
Securities 1,485 1,146 2,935 1,505
Error normal robust robust robust

Table 8: Descriptive statistics daily regression 01-01-2010 – 31-12-2014


The statistics cover the period 1-1-2010 until 31-12-2014. The dependent variable is return (%). The first 2 years of observations in the
COMPUSTAT database are removed. The book value is measured on an annual basis. It is calculated according to the method used by Fama
and French (1993). I first took the total parent stockholders’ equity (SEQ). Then I added the Deferred Taxes and Investment Tax Credit
(TXDITC) where available, after which I subtracted the preferred stock values. The preferred stock value consists of one of three variables.
Wherever it was available the Preferred Stock Redemption Value (PSTKRV) was used. If the PSTKRV was missing the Preferred Stock
Liquidating Value (PSTKL) was used instead. As a last option I used the Preferred/Preference Stock (Capital) – Total value (PSTK) variable,
otherwise a missing value is noted. The size is calculated by multiplying the shares outstanding times the price. The size and return are
measured on a daily basis and are downloaded from the CRSP database. Book values are downloaded from COMPUSTAT. All financial
companies with a GIC code of 40 are removed. Negative equity values are dropped. Only stocks with ordinary common equity which are
traded on the NYSE, AMEX or NASDAQ are included. Book-to-market ratios higher than 10 and lower than 0,1 are dropped. The book-to-
market and size sub-samples are the upper and lower quintiles of these variables.

Variable Obs. Mean Std. Dev. Min. Max.


return (%) 1952705 0.077831 3.640.521 -853.511 581.672
BE/ME 225976 0.812296 0.891661 0.100003 9.990.519
size/106 1953204 5.346.647 21109.17 0.458158 448246.8

Table 9: Regression daily returns 01-01-2010 – 31-12-2014


The regression covers the period 1-1-2010 until 31-12-2014. Dependent variable is return (%). The average expresses the average of the
whole sample. The first 2 years of observations in the COMPUSTAT database are being removed The book value is measured on an annual
basis. It is calculated according to the method used by Fama and French (1993). I first took the total parent stockholders’ equity (SEQ). Then
I added the Deferred Taxes and Investment Tax Credit (TXDITC) where available, after which I subtracted the preferred stock values. The
preferred stock value consists of one of three variables. Wherever it was available the Preferred Stock Redemption Value (PSTKRV) was
used. If the PSTKRV was missing the Preferred Stock Liquidating Value (PSTKL) was used instead. As a last option I used the
Preferred/Preference Stock (Capital) – Total value (PSTK) variable, otherwise a missing value is noted. The size is calculated by multiplying
the shares outstanding times the price. The size and return are measured on a daily basis and downloaded from the CRSP database. Book
values are downloaded from COMPUSTAT. All financial companies with a GIC code of 40 are removed. Negative equity values are
dropped. Only stocks with ordinary common equity which are traded on the NYSE, AMEX or NASDAQ are included. Book-to-market ratios
higher than 10 and lower than 0,1 are removed. The book-to-market and size sub-samples are the upper and lower two deciles of these
variables.

VARIABLES Complete Sample Low BE/ME High BE/ME Small stocks Big Stocks

MONDAY -0.0119** 0.158*** -0.294*** -0.136*** -0.00147

25
(0.00601) (0.0433) (0.0510) (0.0197) (0.00744)
TUESDAY 0.145*** 0.189*** 0.0133 -0.0452** 0.178***
(0.00575) (0.0414) (0.0487) (0.0189) (0.00712)
WEDNESDAY 0.0413*** 0.114*** 0.0560 0.0265 0.0640***
(0.00575) (0.0414) (0.0486) (0.0188) (0.00713)
THURSDAY 0.101*** 0.135*** -0.107** 0.00222 0.139***
(0.00581) (0.0418) (0.0492) (0.0190) (0.00720)
FRIDAY 0.107*** 0.194*** -0.0125 0.0799*** 0.123***
(0.00581) (0.0419) (0.0491) (0.0190) (0.00719)
Average 0.0778*** 0.1577*** -0.0648*** -0.0128* 0.1019***
(0.0026) (0.0188) 0,022 0,0085 0,0032
Observations 1,952,705 45,179 45,193 390,556 390,620
Securities 2,096 205 159 744 520
P-value for F-test <0.0000 <0.0000 <0.0000 <0.0000 <0.0000
R-squared 0.001 0.002 0.001 0.000 0.003
error standard standard standard standard standard

26

You might also like