STOCK MARKET SEASONALITY: DAY OF THE WEEK EFFECT AND JANUARY EFFECT

Lukas Mazal
Supervised by Javier

Gardeazabal

2008-2009

Abstract
This master’s thesis uses a dummy variable approach and an extended dummy variable approach to test for the existence of calendar effects in the rates of return of common stocks. It applies the extended dummy variable approach based on a factor model to returns of 30 stocks traded at the German Stock Exchange and the dummy variable approach to returns of 28 world indices. Furthermore, it investigates time persistence and evolution of these calendar effects. Finally, it simulates two portfolio strategies based on the Monday effect and the September effect. By estimating a rolling dummy variable regression this thesis provides evidence confirming that the day of the week effect started disappearing in the second half of 1990s. The simulated portfolios are able to outperform the buy and hold strategy in all the eight indices considered. This means existence of unexploited profit opportunities, which seriously undermines the efficient market hypothesis.

Stock Market Seasonality: Day of the Week Effect and January Effect

Acknowledgments
I would like to thank my personal tutor Mr. Javier Gardeazabal for his support and helpful comments that were inspirational. He was always accommodating and willing to discuss economic and econometric issues. I also would like to thank to my friend José Antonio Acosta Ramirez for his financial and psychical support and kind help over the entire academic year.

2

Stock Market Seasonality: Day of the Week Effect and January Effect

Table of contents
Stock Market Seasonality: Day of the Week Effect and January Effect ................................................................. 1 Abstract ................................................................................................................................................................... 1 Acknowledgments ................................................................................................................................................... 2 Table of contents ..................................................................................................................................................... 3 1. Introduction ......................................................................................................................................................... 4 2. Literature Review ................................................................................................................................................ 5 2.1. Efficient market hypothesis .......................................................................................................................... 5 2.2. Random walk hypothesis.............................................................................................................................. 6 2.3. Calendar effects ............................................................................................................................................ 6 2.3.1. The day of the week effect .................................................................................................................... 7 2.3.2. The January effect ................................................................................................................................. 7 3. Methodology ....................................................................................................................................................... 9 3.1. Dummy variable approach............................................................................................................................ 9 3.2. Extended dummy variable approach .......................................................................................................... 10 3.3. Heteroskedasticity ...................................................................................................................................... 11 3.4. Data set ....................................................................................................................................................... 11 4. Results ............................................................................................................................................................... 13 4.1. Extended dummy variable approach: DAX Index...................................................................................... 13 4.2. International evidence of the day of the week effect and the January effect .............................................. 16 4.3. Time persistence of the day of the week effect .......................................................................................... 20 4.4. Simulated portfolio strategies ..................................................................................................................... 24 5. Conclusion......................................................................................................................................................... 31 References ............................................................................................................................................................. 32

3

This study undertakes a further investigation of the day of the week effect and the January effect. Introduction In recent years. they cast doubt on the widely accepted efficient market hypothesis. The second part considers a methodology used in this thesis. These calendar anomalies are empirical results that are inconsistent with maintained theories of asset-pricing behavior. The research question that should be answered is: “Is there still seasonality at world stock exchanges?” The research objectives of this thesis are: • to reinvestigate the day of the week effect by using an extended dummy variable approach for stocks of the German Stock Exchange. 4 . The purpose of this thesis is to investigate these effects on major world stock markets by using index data as well as returns of individual stocks. They indicate either market inefficiency or inadequacies in the underlying asset-pricing models. • to investigate time persistence of the day of the week effect in relation to the efficient market hypothesis. • to examine a market timing strategy based on the seasonality. The third part of this thesis presents the empirical results and the last section includes conclusion that summarizes results and also comprises possible theoretical consequences of the calendar effects. • to explore international presence of the day of the week effect and the January effect by using a simple dummy variable approach. These anomalies include the January effect and the day of the week effect. Particularly.Stock Market Seasonality: Day of the Week Effect and January Effect 1. The first section of this thesis presents a brief literature review describing the efficient market hypothesis and calendar effects. The most notable effects are a negative return for Monday and a positive return for January. there has been a proliferation of empirical studies indicating that the distribution of stock returns varies by day of the week and month of the year.

Strategies based on This expression states that the expected price of j-th security in time t+1 equals the expected rate of future return conditional on the set of information available at time t multiplied by the price of j-th security at time t. Efficient market hypothesis The father of the efficient market hypothesis Eugene Fama (1970) first defined the term efficient market in his groundbreaking study as “a market in which prices always fully reflect available information” (p. Table 1.” (Sharpe et al. Sharpe et al. 384).t the price at t. summarizes the efficient market concept stating that the efficient market is a market in which “the security’s price will be a good estimate of its investment value. p jt is a price of security j at time t. 1970.1. Φ t is a general symbol for whatever set of information is assumed to be "fully reflected" in Consequently Ε(x j . p j .. 384).. p. 93). 1999.t +1 Φ t = 1 + Ε r j .t +1 Φ t ) . (1999) distinguishes three forms of efficiency based on a set of information that is available to investors.Stock Market Seasonality: Day of the Week Effect and January Effect 2. This means that an efficient market is one where all unexploited profit opportunities are eliminated by arbitrage. p.t +1 Φ t p jt where E is expected value operator. where the investment value is the present value of the security’s future prospects. 5 .“ (Fama.t +1 = p j . 93) Set of Information Reflected in Security Prices Previous prices of securities All publicly available information All information.t . p. Literature Review 2. 1999. as estimated by well-informed and skilful analysts who use the information that is currently at hand” (Sharpe et al. Sharpe et al. this means that any expected undervaluation or overvaluation is equal to zero. in average share prices are valued correctly.. p. In other words. Furthermore. 1999.t +1 − p j . t +1 is its price at t+1 (with reinvestment of any intermediate cash income form the security). Fama (1970) suggests to measure undervaluation or overvaluation of security as x j .t +1 Φ t ) = 0 . Table 1 provides these three forms of efficiency. The logical consequence of this definition means that “Market is efficient with respect to a particular set of information if it is impossible to make abnormal profits (other than by chance) by using this set of information to formulate buying and selling decisions.t +1 is one period-period percentage return p j . (1999). both public and private Fama (1970) provides exact formulation of the good estimate of the investment value: “ Ε p j . The efficient market hypothesis predicts that security prices follow a random walk and it should be impossible to predict future returns based on publicly available information. r j . Form of efficiency Weak Semistrong Strong (Sharpe et al. ( ) [ ( )] ( ) p j .t +1 − Ε( p j . 93).

6 .3. The only non-zero parameter should be the constant term.Stock Market Seasonality: Day of the Week Effect and January Effect the information set Φ t cannot achieve higher expected returns than are equilibrium expected returns. because the stock prices are completely random. However. 2. all kinds of calendar anomalies in stock market returns have been documented extensively in the finance literature. The term ‘random walk’ was further popularized by the 1973 book. because the stock markets are efficient. If the random walk hypothesis holds. the time path of stock prices is more appropriately specified by a 2 random walk plus drift model.. there is no seasonality in stock prices. The calendar effects should be short lasting.2.. Calendar effects Since the efficient market hypothesis was introduced. Random walk hypothesis The random walk hypothesis is closely connected with the efficient market hypothesis. Hence.. Therefore. 2. Whereas white noise is a sequence of random variables {ε t } such that E (ε t ) = E (ε t −1 ) = . which is the drift. This hypothesis states that stocks move randomly. where {ε t } is heteroskedastic E (ε t ) = σ t2 . The most common calendar anomalies are the January effect and the day of the week effect. Since then. Thus. trading based on exploiting a calendar pattern of returns should yield extraordinary profits at least for a short time. a great deal of research was devoted to investigating the efficiency of capital markets. where pt = ln Pt . it t is generally accepted that stock market returns do not have a zero mean and are heteroskedastic. Walter Enders (2004) defines random walk as a cumulative sum of a white noise process. 1973). and all excess returns achieved by investors are only by chance. Yet. Showing that market returns follow a seasonal pattern violates the assumption of weak market efficiency in that by observing the past development of returns market participants can lead to extraordinary profits. Let us have a model treating any kind of seasonality by using dummy variables Rit = α i + δ 1i D1t + δ 2i D2t + . the random walk hypothesis is a direct consequence of the efficient market hypothesis. any such model must have all the parameters referring to the seasonality equal to zero. consequently the random walk is defined as pt = ∑i =1 ε i . if a calendar effect exists. = 0 . + δ ki Dkt + ε it . such trading strategies affect the market in that further profits should not be possible: the calendar effect should disappear. = σ 2 and E (ε t ε t − s ) = E (ε t − j ε t − j − s ) = 0 for all j 2 and s.. There is no known strategy. t Under the random walk hypothesis.. which could continually outperform stock market averages.. as market participants can learn from past experience. A Random Walk Down Wall Street (Malkiel. The random walk hypothesis was introduced by Kendall (1953) and it was later confirmed by Fama (1965). E (ε t ) = E (ε 2 t −1 ) = . This model can be defined as pt = a ⋅ t + ∑i =1 ε i or after taking first differences ∆pt = a + ε t .

Rozeff and Kinney (1976) first observed that the average return of an equal-weighted index of the New York Stock Exchange in January is statistically significantly higher than the average return for the other months in the period 1904-1974. The literature on monthly effects. they point out that the January effect is stronger in case of small firms than in case of well-established companies with high capitalization. Gibbson and Hess (1981) provide further evidence of negative return for Monday. Tokyo New Stock Exchange Index and the Straits Times Index (Singapore) in the period January 1969 to December 1984. 1990. Dubois and Louvry (1995) provide probably the most comprehensive study confirming the day-of-the-week effect. “negative mean weekend returns do appear to be the norm rather than the exception in a range of capital markets around the world” (p. (1987) used seven indices from seven countries. This effect refers to a phenomenon that the average return on Monday is significantly lower than the average return for the other days. Similarly. Wang. Condoyanni et al. Lakonishok and Smidt (1988) explore 90 years of the Dow Jones Industrial Average Index in the period 1897-1986. Then. French (1980) first observes negative average daily share returns for Monday in the period 1953-1977 on the New York Stock Exchange. Haugen and Jorion (1996) provide evidence confirming the persistent existence of the January effect.Stock Market Seasonality: Day of the Week Effect and January Effect 2. 1995. They conclude. 19). 1041). which are compensated by abnormal positive returns on Wednesday. Most recently. Toronto Composite Index.3. p. there is a large body of literature on the day of the week effect of stock returns. Their data set ranges from 1998 to 2000. Most studies dealing with this anomaly use a simple dummy variable approach based on a linear regression with 5 dummy variables referring to the days of the week. generally. Li and Erickson (1997) confirm the day of the week effect in the period 1962-1993 by using three indices from the New York Stock Exchange. The day of the week effect Despite the efficient market hypothesis and its consequences. 1993). Gardeazabal and Regulez (2004) reach a similar conclusion. They observe the Monday effect in the 1962-1993 periods. They conclude that the January effect still exists despite the fact that it was well known for reasonably long time and therefore should have disappeared. The January effect refers to the significantly higher average share market returns in January when compared with other months. All-Share (United Kingdom). He concludes that “low Tuesday and high Wednesday returns are observed for the close-to-close returns. Paris CAC Industrial. Kato (1990) provides evidence of the day of the week effect from the Tokyo Stock Exchange from the period 1974-1987.” (Kato. Australian Stock Exchange’s All Ordinaries Share Price Index.3. They suggest using the so called “Extended Dummy Variable Approach” (EDVA) derived from the Fama-French three-factor model (Fama and French. Furthermore. Gardeazabal and Regulez (2004) used the data from the Spanish stock market. Similarly. They observed.1. Dow Jones Industrial Average Index. “We find negative returns on Monday.” (Dubios and Louvry. They examined eleven indexes form nine countries in the period 1969-1992. p. They conclude that by using the EDVA approach they had found even stronger seasonal effects than by using the simple DVA.2.T. Keim and Stambaugh (1983) identify the Monday effect in S&P Composite Index returns form 1928 to 1982.14%. The January effect One of the most puzzling empirical findings is that the sample distributions of monthly stock returns vary by the month of the year. They 7 . They observe persistent seasonality in Dow Jones Industrial Average Index returns confirming the-day-of-the-week effect with Monday’s average returns -0. 2. confirmed higher returns in January. 174). F. American Stock Exchange and NASDAQ. Similarly.

8 . International evidence of the January effect is provided by Kato and Schallheim (1985). “the January effect is still going strong 17 years after its discovery” (Haugen and Jorion. 27).Stock Market Seasonality: Day of the Week Effect and January Effect conclude. 1996. p.

etc. 226) state that. By substituting this result into ~ ~ ~ ~ the regression equation we will get Rit = α i + δ1i D1t + δ 2i D2t + δ 3i D3t + δ 5i D5t + vit . where each individual dummy variable accounts for the excess return for the particular day. which means E ( D1t ) = E ( D2t ) = E ( D3t ) = E ( D4t ) = E ( D5t ) = 1 / 5. Nonetheless. “This restriction says that the coefficients of the dummy variables add up to zero and can be interpreted as a normalizing restriction. However. Dummy variable approach The so called “Dummy Variable Approach” to the stock market seasonality is based on estimating a simple regression model. This implies that δ1i + δ 2i + δ 3i + δ 4i + δ 5i = 0 .” By imposing this restriction we can solve both the exact multicolinearity and identification problems. where Rit is the return of index (or security) i in period t. They state that this way we can solve both the exact multicolinearity problem and the problem of identification of the parameters. p. all the dummy variables are expressed as deviations with respect to the omitted dummy D4t and each slope coefficient δ ki measures the deviation with respect to the overall mean α i . there is a problem of a perfect mulitcolinearity. Methodology 3.1. 2003). D1t is a dummy variable for Monday (i. The statistical inference can be conducted by using the t-test on each parameter. as Gardeazabal and Regulez (2003) state. Gardeazabal and Regulez (2003. Gardeazabal ∑ T t =1 T t =1 T t =1 T t =1 T t =1 and Regulez (2003) further assume that the proportion of the returns for the individual days equal 1/5. The standard error of the ˆ 1 where 1 refers to a 4 × 1 vector of ones and Σ ˆ is a coefficient can be obtained as 1' Σ heteroskedasticity-consistent estimator of the covariance matrix provided by White (1980). D2t is a dummy variable for Tuesday. If we solve for the dummy variable referring to Thursday δ 4i . In this case the basic regression equation can be defined as Rit = α i + δ1i D1t + δ 2i D2t + δ 3i D3t + δ 4i D4t + δ 5i D5t + δ 6i D6t + δ 7 i D7t + δ 8i D8t + δ 9i D9t + + δ10i D10t + δ11i D11t + δ12i D12t + vit 9 . The model is analogous in case of the January effect. we get δ 4i = −(δ1i + δ 2i + δ 3i + δ 5i ) . because it contains all the dummy variables and the constant term.. proceeding in this way solves the perfect multicolinearity. Now. if α i is the mean return. Gardeazabal and Regulez (2003) suggest imposing a restriction on the parameters of the model to tackle this problem. According to Gardeazabal and Regulez (2003). D1t = 1 if observation t falls on a Monday and 0 otherwise). A consistent estimate of δ 4i can be recovered from the estimates of the other coefficients in the following way δ 4i = −(δ1i + δ 2i + δ 3i + δ 5i ) . vast majority of authors uses elimination of the constant term or one of the dummy variables from this regression equation. (Gardeazabal and Regulez. This model can be written in the following way: Rit = α i + δ 1i D1t + δ 2i D2t + δ 3i D3t + δ 4i D4t + δ 5i D5t + vit . it is not possible to estimate this equation in the above form. vit is a zero-mean disturbance. In order to cope with this problem.Stock Market Seasonality: Day of the Week Effect and January Effect 3. Therefore. where ~ Dkt = Dkt − D4t . but leaves the parameters of the model unidentified.e. it must hold that δ1i 1 T 1 T 1 T 1 T 1 T D1t + δ 2i ∑ D2t + δ 3i ∑ D3t + δ 4i ∑ D4t + δ 5i ∑ D5t = 0 .

h ˆ . This approach is derived from the Fama-French three-factor model (Fama and French.. We assume that E ( D1t ) = E ( D2t ) = E ( D3t ) = E ( D4t ) = E ( D5t ) = E ( D6t ) = E ( D7 t ) = E ( D8t ) = E ( D9t ) = E ( D10t ) = = E ( D11t ) = E ( D12t ) = 1 / 12 and we restrict the parameters such that δ 1i + δ 2i + δ 3i + δ 4i + δ 5i + δ 6i + δ 7i + δ 8i + δ 9i + δ10i + δ11i + δ12i = 0 . 3. where hjt is the j-th factor is j =1 J orthogonal on the daily dummies. Similarly as for the day ˆ 1 where 1 of the week effect. the dummy variable approach to seasonality leaves too much variability of stock returns unexplained. They propose the so called ‘Extended Dummy Variable Approach’ which leaves a lower fraction of stock returns variability unexplained.2. Extended dummy variable approach According to Gardeazabal and Regulez (2004). and the return on a portfolio of high book-tomarket firms minus the return on a portfolio of low book-to-market firms. 1993). REMRt.h ˆ where h EMRt SMBt HMLt are defined as 10 . the standard error of the coefficient can be obtained as 1' Σ ˆ is a White’s heteroskedasticity-consistent estimator of refers to a 12 × 1 vector of ones and Σ the covariance matrix (White. They state that the portion of unexplained variability increases with sample frequency and inference on daily seasonality usually leads to weak or null evidence of seasonality. it i 1i 1t 2i 2t 3i 3t 4i 4t 5i 5t EMRi EMRt SMBi SMBt HMLi HMLt it ˆ . In general.e. D2t is a dummy variable for February. By substituting this result into the regression equation we will get ~ ~ ~ ~ ~ ~ ~ ~ Rit = α i + δ1i D1t + δ 2i D2t + δ 3i D3t + δ 4i D4t + δ 5i D5t + δ 6i D6t + δ 7 i D7 t + δ 9i D9t + ~ ~ ~ + δ10i D10t + δ11i D11t + δ12i D12t + vit ~ where Dkt = Dkt − D8t . D1t = 1 if observation t falls on a January and 0 otherwise). Taking into account only these three factors. RSMBt. RHMLt. The three factors that Fama and French (1993) used are: the excess return on the market portfolio. If we solve for the dummy variable referring to August δ 8i . we get δ 8i = −(δ1i + δ 2i + δ 3i + δ 4i + δ 5i + δ 6i + δ 7i + δ 9i + δ10i + δ11i + δ12i ) . A consistent estimate of the δ 8i can be computed from the estimates of the other coefficients by using the following expression δ 8i = −(δ1i + δ 2i + δ 3i + δ 4i + δ 5i + δ 6i + δ 7i + δ 9i + δ10i + δ11i + δ12i ) . the return on a portfolio of small firms minus the return on a portfolio of large firms. 1980). vit is a zero-mean disturbance. the above mentioned equation can be written as: ˆ ˆ ˆ R =α +δ D +δ D +δ D +δ D +δ D + β h +β h +β h +u . their model can be defined in the following way: Rit = α i + δ1i D1t + δ 2i D2t + δ 3i D3t + δ 4i D4t + δ 5i D5t + ∑ β ji h jt + uit . D1t is a dummy variable for January (i. etc.Stock Market Seasonality: Day of the Week Effect and January Effect where Rit is the return of index (or security) i in period t.

Kuala Lumpur Stock Exchange Index (KLSE). Mid-cap Deutscher Aktien Index. SMB it i 1i 1t 2i 2t 3i 3t 4i 4t 5i 5t SMBt ˆ R HML it = α i + δ1i D1t + δ 2i D2t + δ 3i D3t + δ 4i D4t + δ 5i D5t + h HMLt . 1980). because “the omitted factors are orthogonal to the daily seasonals” (p. Euronext BEL 20 Index (BEL 20). Small-cap Deutscher Aktien Index.12. because it holds that σ vi = E (vit ) and σ ui = E (uit ). NASDAQ Cmposite Idex. − σ ui = ∑ j =1 β ji σ j > 0 . Therefore. the shortest data range is for BOVESPA index and it is from 1. Gardeazabal and Regulez (2004) point out that the unexplained part of stock returns volatility is lower in case of the extended dummy variable 2 2 2 2 2 2 2 2 approach. it is necessary to use some heteroskedasticity robust method.2. Furthermore. 226). Small-cap Deutscher Aktien Index (SDAX) and Dividend Deutscher Aktien Index index (DivDAX).199). Standard & Poor's Mid Cap Index. Austrian Trade Index (ATX).1997 to 8. Standard & Poor's Small Cap Index. Taiwan Weighted Index (TSEC).3. Hang Seng Index (HSI).Stock Market Seasonality: Day of the Week Effect and January Effect ˆ R EMR it = α i + δ1i D1t + δ 2i D2t + δ 3i D3t + δ 4i D4t + δ 5i D5t + h EMRt . the statistical inference is based on this White heteroskedasticityconsistent estimator (White.2009. Heteroskedasticity It is generally accepted that variance of stock market returns is not constant.1.1955 to 8. For more details. 11 . Australia All Ordinaries Index (AORD). 3. Russel 3000 Index. Bovespa Index. Shanghai Composite Index (SSE). Jakarta Composite Index (JKSE). White (1980) provides a heteroskedasticity-consistent estimator of the covariance matrix.5. Amsterdam Exchange Index (AEX). CAC 40 Index (CAC). However. Standard & Poor's 500 Index. please see the reference list. Gardeazabal and Regulez (2004) state that the model based on the dummy variable approach is “misspecified in the sense that it leaves the risk factors out of the regression equation” (p. where e refers to Σ t t t t t t t Σ n n nΣ n t =1 t =1 residuals from an OLS regression (Greene. The longest data span is for the Standard & Poor's 500 Index and it ranges from 1. In case of all the regressions estimated in this thesis.2008. IPC Mexico Index. the OLS estimator of the parameters is consistent. p. The data range depends on availability of the data for a certain index. Deutscher Aktien Index. Belgium Exchange Index (BEL). Bombay Stock Exchange 30 Index (BSE 30). Data set The data set used for the extended dummy variable approach estimations includes daily market returns of 30 German companies traded at the German Stock Exchange and daily returns of the Deutscher Aktien Index (DAX). Soul Composite Index (KOSPI). Technology Deutscher Aktien Index (TecDAX). 226).4. Tel Aviv TA-100 Iindex (TA 100). This estimator is defined in the following way: T T ˆ = 1 ( 1 X ′X ) −1 ( 1 e 2 x x ′ )( 1 X ′X ) −1 = ( X ′X ) −1 ( e 2 x x ′ )( X ′X ) −1 . All the market data was downloaded from the Yahoo Deutschland Finanzen (2008) and from the Yahoo Finance (2009).2006 to 31. Streits Times Index (STI).2009. Swiss Market Index (SMI). NIKKIE 225.1. The data used for the dummy variable approach includes daily returns of the following 28 indexes: Financial Times Index 100. where σ vi J 3.5. The data set ranges from 1. 2002. ˆ R =α +δ D +δ D +δ D +δ D +δ D + h .

It is computed as “a weighted average of all overnight unsecured lending transactions undertaken in the interbank market. The Eonia is the effective overnight reference rate for the Euro. The time series of the Eonia was downloaded from the Euribor FBE (2000). For more details. 2000).” (Euribor FBE. initiated within the Euro zone by the contributing banks.Stock Market Seasonality: Day of the Week Effect and January Effect The Euro OverNight Index Average rate was used as a risk free rate of interest. 12 . please see the reference list.

the first factor is the excess return on the market portfolio. Fig. we can use the DivDAX index. it is significantly different from 0. the combination of t-statistic in absolute value and coefficient estimate is represented by a blue cross. the only difference are factors deployed. However.Stock Market Seasonality: Day of the Week Effect and January Effect 4. high book to market ratio and high dividend yield. Monday 13 . Extended dummy variable approach: DAX Index A model extended by three factors was estimated. In general. Similarly as in Gardeazabal and Regulez (2003). This index consists of technology stocks. It is generally known that firms with high book to market ratio have also high dividend yield. which is an index of companies paying high dividends. Results 4. the second factor refers to the difference between returns of small-cap SDAX and large-cap DAX index. with low expected growth of earnings low price to earnings ratio. If the cross falls above the 95% confidence interval line. On the other hand the TecDax index represents a portfolio of “growth” companies. For each stock. which means fast growing companies with very high expected growth of earnings and consequently very high price to book ratio and low book to market ratio. 1. Each figure contains the results for one explanatory variable. Figures 1 to 5 report the results. The third factor is the difference between returns of the DivDAX index and TecDAX index. the companies of the DivDAX index can be regarded as “value” companies. Therefore. The horizontal line represents the 95% confidence interval.1. This model is based on the same logic as the model estimated by Gardeazabal and Regulez (2003).

2.Stock Market Seasonality: Day of the Week Effect and January Effect Fig. 3. Tuesday Fig. Wednesday 14 .

Thursday Fig.Stock Market Seasonality: Day of the Week Effect and January Effect Fig. 5. 4. Friday 15 .

0276 -0.0758 0.1619 0.8.0121 0. Therefore.7474 2.0100 0.4452 1.0400 0.0688 0.4208 0.6292 1. The model estimated in this thesis joins both the day of the week effect and the January effect by using dummy variables for months and for days.0066 0.1390 0.8823 0.1.0702 0.0563 -0.3224 1.0668 0. 12 refer to monthly seasonal dummies.1269 0.3457 0.3752 1.0545 S&P 500 2.0283 -0.0074 0. W.7635 1.1985 0.0739 0.9605 S&P small cap 15.0956 -0.0059 0.6456 0. 5.0354 -0.0810 0.0069 -0.2886 0.0473 -0.1971 0. 5% means 2 crosses.2026 1.0352 0.1134 1. t-statistic in absolute value 2.1190 t-statistic in absolute value 2.5020 0.2007 1. H.3701 1.6334 0.3588 1. The model has the following form: Rit = α i + δ Mi DMt + δ Ti DTt + δWi DWt + δ Hi DHt + δ Fi DFt + δ1i D1t + δ 2i D2t + δ 3i D3t + δ 4i D4t + + δ 5i D5t + δ 6i D6t + δ 7i D7 t + δ 8i D8t + δ 9i D9t + δ10i D10t + δ11i D11t + δ12i D12t + uit Where Dk.0132 0.7947 0.0873 0. 4.0561 0. 6.2723 0.4732 0.0524 -0. k = 1.0563 0. Indeed.0541 0.2157 1.0959 -0. k = M.8890 1.0254 0.0228 -0.2632 0. 10.1488 2.1257 0. from the practical point of view it is much more exciting to investigate the seasonal patterns of stock market indices rather than seasonal patterns of individual stocks.9602 t-statistic in absolute value 2.7806 0. 1999).0459 0. This can be easily conducted even by an individual investor by trading futures contracts fixed on various indices.0509 -0.0284 0.9333 0. there is never more than 1 cross above the line..0491 0.1475 1.9602 S&P mid cap 20. Since we have 30 stocks.0402 -0.4039 1.0409 0.0238 0. 4. the investor will get a sufficient financial leverage at a negligible cost and will completely eliminate the specific risk of individual stocks.1074 0. T.0489 0.2167 1.6284 0.2206 2.5739 0.9857 0.0310 0.1696 -0.0595 0.0248 -0.8874 0.1251 0.0689 -0.1170 t-statistic in absolute value 1.0968 -0.6609 0.0270 0.5620 1. This can be done by using the simple dummy variable approach.0127 -0.7776 0.0696 -0.0411 0.2635 1.Stock Market Seasonality: Day of the Week Effect and January Effect Under the assumption of no seasonality we expect not more than 5% of crosses to be above the line.0318 -0.0164 -0.3549 0. 3. it is much better to exploit the profits of the calendar effects by trading the market portfolios represented by the whole index.1359 -0.1436 1. 11.1991 0.8234 1.2.0380 1.0194 -0.2. Table 2. International evidence of the day of the week effect and the January effect In order to cope with the specific risk of individual stocks.7909 0.4156 0.8735 2.6225 2. Tables 1 to 7 provide the results of this regression model estimated on the data of 28 world stock exchange indices.1123 1.8778 1.0435 -0.0689 0.0543 -0.3619 1. 2.0354 -0.4705 1.8782 3.3194 0. 7.3960 1.1259 0.0180 0. This supports evidence of no seasonality in the returns of the 30 companies of the DAX index.0134 0.4694 0.8.1493 1.8220 0.0772 -0.9607 Index Data range constant Monday Tuesday Wednesday Thursday Friday January February March April May June July August September October November December critical value NASDAQ 5.3306 2.0216 0. F refer to daily dummies and Dkt.8568 0.0134 -0.3907 1.1995 0.1734 1.7732 6. 9.5655 0.0049 -0. This way.0315 -0.0564 -0.0060 -0.0292 0. 8.0255 0.0441 0. in the sense of the security characteristic line (Sharpe et al.0511 -0.0398 0.1975 0.0445 16 .

0930 0.0614 -0.0129 -0.1.0582 -0.0340 -0.0138 -0.0016 0.6585 0.0185 0.7505 0.0040 0.8830 0.1532 0.3413 0.9751 1.0259 -0.11.0534 -0.3142 0.0525 0.8155 1.3186 2.2114 0.2032 0.2003 0. t-statistic in absolute value 0.0244 0.3619 1.5511 1.0417 SMI 9.0125 0.0327 0.0991 0.1984 0.9489 0.0642 0.0310 0.8217 0.0300 0.2906 0.0256 -0.0933 2.0624 0.9189 0.4811 0.0596 1.0153 -0.9490 0.0838 -0.0889 0.0110 0.0543 1.4387 0.1027 0.0720 -0.7448 2.5322 1.1005 0.2000 0.0099 0.0846 t-statistic in absolute value 1.0244 -0.0413 0.0586 0.0433 -0.6966 1.0334 0.7053 0.0567 0.0026 -0.4176 0.4197 1.6780 1.0553 0.0652 -0.0075 0.0314 -0.1554 2.1.2234 0.0086 0.0489 0.1028 MDAX 3.7830 0.9.0150 -0.0052 -0.9605 8.4063 0.4701 0.1069 0.0247 0.0530 -0.0178 -0.2847 1.0346 -0.4056 0.7268 0.0148 0.0785 -0.0027 -0.9610 t-statistic in absolute value 1.7215 1.1857 0.1805 1.2005 1.0423 -0.1347 0.9615 t-statistic in t-statistic in absolute absolute value value IPC Mexico 2.1645 0.0837 1.0006 0.1581 0.6113 0.0398 -0.0151 0.0521 0.3982 0.0523 -0.1987 0.2775 0.0140 0.5475 1.0320 -0.0022 -0.7331 0.0817 0.3259 3.5670 0.0059 0.2985 0.0144 0.0026 -0.0600 17 .2529 2.0564 Table 4.0435 0.0566 -0.0651 -0.0265 0.0945 -0.7913 0.0096 1.0772 -0.0143 -0.4580 0.8821 0.0217 -0.3084 0.0001 -0.1765 0.0286 0.0329 -0.7757 0.1558 1.2870 2.1990 0.0792 0.0287 -0.1000 -0.0375 0.2642 0.7855 1.2300 2.0432 -0.9605 t-statistic in absolute value 0.0578 0.0028 -0.7610 0.0156 -0.0317 0.1900 0.2428 0.0520 -0.7056 1.2279 -0.5957 0.2559 0.0086 -0.0332 0.4.0623 -0.0360 -0.0607 0.3318 1.0545 -0.7614 1.1250 -0.0287 -0.0789 -0.0762 0.8270 0.0014 -0.3823 0.0499 0.8590 0.1781 2.9102 0.0292 0.5597 1.0534 -0.0513 0.3443 0.3676 0.1936 0.2000 0.2390 0.1457 -0.6278 3.7131 0.8584 0.3776 1.0317 1. Russell 3000 10.0205 -0.5105 1.8681 0.9805 2.5717 0.4854 1.0318 0.0703 0.1270 0.1498 0.2395 1.1560 0.0169 -0.7657 0.0432 0.0644 0.3465 0.1299 0.8248 0.0054 0.6462 1.0405 -0.8848 0.6171 1.5070 0.0457 -0.2304 0.2272 0.0453 0.2947 2.7051 0.7153 0.0340 -0.1182 0.0127 -0.1934 1.9784 0.2166 1.1253 0.1885 1.4280 0.1991 0.1140 0.3.0340 TecDAX 24.1120 0.0076 -0.7405 0.4543 0.0364 -0.0038 -0.9605 Index Data range constant Monday Tuesday Wednesday Thursday Friday January February March April May June July August September October November December critical value SDAX 3.Stock Market Seasonality: Day of the Week Effect and January Effect Table 3.1389 1.8440 1.11.2349 -0.0091 -0.0074 0.0677 0.1489 0.9610 Index Data range constant Monday Tuesday Wednesday Thursday Friday January February March April May June July August September October November December critical value FTSE 100 2.9603 t-statistic in absolute value 1.1023 -0.9604 t-statistic in absolute value 2.9740 1.2465 0.0264 0.0323 0.0384 0.9799 0.0197 0.0792 0.4142 0.11.0025 0.0409 0.0078 -0.8400 0.0244 -0.1336 0.1005 -0.0387 0.2531 0.0353 0.5658 0.0199 -0.0735 0.0612 0.1990 0.4027 0.8957 0.0282 -0.2125 0.1570 3.0932 DAX 26.0547 -0.9585 1.0165 0.0319 -0.

0774 -0.8447 0.2348 1.1403 -0.2197 2.5721 1.1178 0.4553 1.0204 0.0354 -0.0487 -0.0848 0.0726 0.0989 0.1030 0.0709 -0.0101 0.4866 0.0516 -0.7783 0.7520 1.0488 0.0762 0.8003 1.1950 0.9437 0.1670 0.3731 2.0019 0.0266 0.1085 0.12.0248 -0.1938 0.0847 1.1245 -0.3.0763 4.2739 1.3965 1.0376 -0.0326 -0.1589 -0.2186 0.4759 1.5345 0.7724 1.0295 0.1187 -0.0315 0.0551 0.8191 0.0837 -0.0424 1.2686 1.1243 0.9790 1.2797 CAC 40 1.4518 1.0913 -0.7212 2.0415 0.0191 0.1914 1.1465 0.0061 -0.1572 -0.1579 0.4388 1.0028 -0.0136 0.5907 0.9606 Jakarta Comosite 1.0188 0.4289 0.3995 0.6433 0.1992 HSI 2.4695 1.2220 0.1.9603 t-statistic in absolute value 1.4802 1.7.0612 -0.0335 -0.0024 0.1160 -0.2963 1.4494 1.1165 -0.6546 0.1742 0.0732 0.0047 0.0526 0.0182 -0.0208 1.0751 -0.1355 -0.1566 0.6453 1.5054 0.4138 1.9607 t-statistic in absolute value 1.4210 0.5313 0.8770 0.4.9668 0.4336 0.7.0318 -0.1535 -0.1646 1.1050 -0.0875 0.0055 -0.0836 1.3229 0.9785 2.1047 -0.2770 1.4760 0.1993 0.0331 -0.4301 0.0535 -0.0620 -0.1115 0.5922 0.0591 Table 6.0126 -0.0517 -0.1990 0.3645 0.7077 0.0171 0.1023 -0. t-statistic in absolute value 1.1650 0.2080 1.9608 Index Data range constant Monday Tuesday Wednesday Thursday Friday January February March April May June July August September October November December critical value Bovespa 27.0146 -0.6811 0.8341 0.0797 -0.1313 1.2759 0.1316 0.0342 -0.0963 -0.1987 0.1396 0.7869 1.0220 -0.2539 0.7044 2.9608 t-statistic in absolute value 2.0614 -0.0599 0.9370 1.2392 t-statistic in absolute value 1.0190 0.0469 -0.0265 -0.1121 0.0693 -0.0382 0.0949 0.0651 -0.1036 0.0155 -0.0622 t-statistic in absolute value 2. Australia All Ordinaries 3.0272 0.0771 -0.3322 0.7193 0.6691 1.4728 0.1504 0.9483 2.4640 1.6377 0.1259 0.0467 -0.0502 0.5311 1.0361 -0.2860 1.3061 1.4005 1.9412 0.9610 Index Data range constant Monday Tuesday Wednesday Thursday Friday January February March April May June July August September October November December critical value BSE 30 1.0014 0.2414 2.2979 0.8141 0.1996 2.1463 0.0858 18 .0131 0.0964 0.2730 1.0131 -0.3148 0.6726 1.3080 0.0102 0.1096 0.0290 0.0738 1.5641 1.0064 0.0692 -0.0059 -0.8.9605 Kuala Lumpur 3.0308 2.9604 Shanghai Composite 4.9698 0.0449 0.0746 -0.5509 0.0369 0.4962 -0.0627 0.0901 1.3461 0.0635 -0.0567 0.6434 0.0135 0.2048 0.1910 0.1984 0.9870 0.0685 0.4594 1.1246 1.4619 0.4899 2.0396 0.1246 0.0483 0.0289 0.4984 0.1997 0.0272 -0.0419 0.0557 1.4961 2.0025 0.0255 0.0948 t-statistic in absolute value 1.2477 0.0157 1.0614 -0.4334 0.0055 -0.7262 0.0319 1.1595 0.5742 2.1908 0.4675 0.0232 0.7866 0.2428 0.0416 0.6450 1.0863 -0.0006 -0.2312 0.2074 t-statistic in absolute value 0.3371 0.8480 0.0811 -0.0431 1.0329 0.0628 0.1537 0.0518 -0.0209 0.8315 0.3110 1.0280 0.2000 0.4884 1.0461 0.1.3472 0.0470 0.1375 -0.1882 -0.1793 0.3152 0.0342 0.0247 0.6393 1.4544 0.0485 0.0122 0.1997 0.4229 1.0094 0.6649 0.1596 -0.0465 0.0395 -0.Stock Market Seasonality: Day of the Week Effect and January Effect Table 5.0331 0.1993 0.0693 0.

2774 0.0729 0.Stock Market Seasonality: Day of the Week Effect and January Effect Table 7.0751 0.0455 -0.0437 -0.0013 1.2875 1.3897 1.1406 -0. which exactly confirms existing literature.1782 -0.9605 t-statistic in absolute value 1.2766 1.9608 t-statistic in absolute value 1.0016 -0.0204 -0.7397 0.9605 t-statistic in absolute value 0.5279 1.0486 0.0255 0.1991 0.0268 -0.3286 2.0716 -0.2058 -0.0055 -0.0057 0.3063 1.1225 -0.0323 Table 9 summarizes the results by providing the number of cases out of 28 in which the parameters were statistically significant.5470 0.1162 0.2361 0.6559 1.7334 0.1049 -0.1582 -0.4931 0.5429 0.0849 -0.1324 0.3255 0.7816 0.4843 0.0931 BEL 20 9.0310 0.1084 -0.1297 -0.0315 -0.0540 1.3224 1.1821 -0.0851 0.10.1984 0.9121 0.6189 0.1538 0.7089 1.1292 0.0650 0.7999 0.3100 -0. It is apparent from this table that the most significant day of the week effect was Monday.3757 2.0625 0.0078 0.9719 0.7.1930 1.1.0794 0.0245 0.0033 0.0277 0.1997 0.1997 0.0439 0.0572 -0.5411 0.0010 -0.0128 -0.0505 0.0548 1.0003 0. t-statistic in absolute value 1.5188 2.1451 0.0355 0.0970 0.3616 0.3053 1.9609 Index Data range constant Monday Tuesday Wednesday Thursday Friday January February March April May June July August September October November December critical value AEX 12.9604 t-statistic in absolute value 1.7871 1.1484 0.0355 -0.2371 0.0450 1.1635 0. The January 19 .1270 1.0268 0.7824 1.7048 0.6417 0.7.7.4078 0.6372 0.2304 1.1.3749 0.0060 -0.2810 0.0003 0.4968 1.0226 0.2113 0.1139 -0.1212 1.0736 0.3082 0.6860 1.3172 0.0102 0.0156 0.0067 -0.0050 0.0209 0.0035 0.7316 0.3847 0.0084 0.0429 -0.4308 1.0171 0.1272 0.0530 0.11.2019 0.1981 0.8824 0.0133 0.0765 0.0535 0.0595 -0.6604 0.8082 0.7266 0.0285 -0.2123 0.1269 Table 8.1162 TA 100 1.0214 0.2101 2.0277 0.0019 -0.0234 -0.6500 0.6853 0.2300 2.0307 -0.4616 0.0572 0.0144 -0.0239 0.1194 0.0232 0.0060 0.7823 1.0843 2.6821 0.1549 -0.2821 0.0579 -0. Straits Times 4.2485 0.0142 0.0371 0.0868 0.2851 1.0776 -0.0579 0.0037 1.2632 2.0195 -0.9603 Index Data range constant Monday Tuesday Wednesday Thursday Friday January February March April May June July August September October November December Critical value ATX 11.5630 0.5413 1.0577 0.0381 -0.4019 1.0315 -0.0818 Nikkei 225 4.0024 0.0768 0.0977 1.1992 0.0355 -0.2477 1.0385 0.1438 0.0232 0.0333 -0.2281 0.0184 0.5156 1.0034 -0.6239 0.0233 0.0113 0.4032 1.0649 1.2034 0.0557 0. only 1 index out of 28 achieved statistically significant excess returns in January.0832 0.3809 0.9361 0.1992 0.0557 0.0422 0.1259 0.0272 -0.1057 0.2164 -0.6363 0.4676 0.1152 -0.4998 1.6766 0.9273 0.0616 0.7017 1.0357 -0.0862 0.1086 t-statistic in absolute value 1.0004 0.3981 0.7605 0.0333 0.0772 0.0054 -0.1757 -0.0111 -0.1087 2.0358 -0.4021 1.1997 0.6699 0.1221 0.1988 0.7186 1.4366 0.4975 2.0289 -0.2155 0.4749 2.5074 1.0089 -0.0419 2.1418 -0.0327 0.6013 1.2108 0.1108 0.6411 0.2134 1.3988 0.0064 0.0285 -0.9795 1.0730 0.0838 -0.0000 -0.1062 0.1161 Kospi 1.9605 t-statistic in absolute value 1.5654 0.2341 0.0548 0.0037 1.0262 -0.4339 1.1282 TSEC 2. However.0036 -0.0571 1.3452 1.4.4184 1.0131 -0.6525 0.6109 0.0248 -0.7777 0.9671 0.0357 0.1082 -0.0221 -0.0094 0.0359 0.0390 0.0382 -0.0079 -0.6977 0.6474 0.4954 2.0342 0.0177 -0.3574 1.1112 -0.1653 2.5875 0.3008 2.9608 t-statistic in absolute value 0.0695 -0.

20 . once discovered this effect should disappear in a very short time. The time path of the Monday effect is computed by estimating a rolling regression. and then this regression equation is reestimated by moving towards the end of the time series by 5 trading days. The reason why the Monday has been chosen is that this effect was the most significant. The other day’s effects were also considered. NASDAQ Composite Index.Stock Market Seasonality: Day of the Week Effect and January Effect effect was overtaken by April. Time persistence of the day of the week effect This thesis also examines time persistence of the Monday effect of 7 out of 28 indices. The 7 figures bellow provide the time path of the t-statistics of the Monday effect for Financial Times Index 100. The regression equation is estimated on the first 1250 trading days. December. their evolution through time is very similar to the Monday effect. Therefore. Australia All Ordinaries Index. Number of Statistically Significant Cases Monday Tuesday Wednesday Thursday Friday January February March April May June July August September October November December 9 0 2 2 8 1 0 0 10 0 0 0 3 7 0 0 8 4. According to the efficient market hypothesis. It should be pointed out that parameters for April and December were positive and parameters for August and September were negative. Nikkei 225. Hang Seng Index. Straits Times Index. September and August. their time paths are not reported.3. Table 9. However. and Standard & Poor's 500 Index. which approximately represents a 5 years period.

FTSE 100 Fig.Stock Market Seasonality: Day of the Week Effect and January Effect Fig. 7. Australia All Ordinaries 21 . 6.

Straits Times Index 22 . Hang Seng Index Fig. 9.Stock Market Seasonality: Day of the Week Effect and January Effect Fig. 8.

11. 10. NASDAQ Composite Index 23 . Nikkei 225 Fig.Stock Market Seasonality: Day of the Week Effect and January Effect Fig.

Probably the longest time persistence of this effect is in case of the Straits Times Index and Nikkei 225.4. It occurred in the 1960s. Nikkei 225. The same 7 indices out of 28 were considered. and it was very strong and stable till the second half of the 1990s. The blue line represents the buy and hold strategy. NASDAQ Composite Index. This thesis aims to consider this issue by focusing on the negative September effect and Monday effect only. we can observe that this effect disappeared first from the S&P 500 index. 24 . Figures 13 to 19 graphically compare performance of 3 market timing strategies by plotting their time path and a relative performance of a strategy based on Monday effect and a strategy based on September effect with respect to the buy and hold strategy. Hang Seng Index.Stock Market Seasonality: Day of the Week Effect and January Effect Fig. Straits Times Index. Standard & Poor’s 500 Index It is apparent from the 7 figures above that the day of the week effect has very similar patter for all the indices examined. and Standard & Poor's 500 Index. Australia All Ordinaries Index. Furthermore. In other words. which can be seen on the last figure. The green line corresponds to a market timing strategy based on selling the index in August at a closing price and buying again in September at a closing price. 4. where it stayed significant till 1998. it has to be assessed if it was possible to achieve profits that would be able to outperform the buy and hold strategy. Simulated portfolio strategies The calendar effects should also be considered in terms of their exploitability. The red line marks a market timing strategy based on selling the index on Friday at a closing price and buying again on Monday at a closing price. Financial Times Index 100. 12.

14. FTSE 100 Fig.Stock Market Seasonality: Day of the Week Effect and January Effect Fig. 13. Australia all Ordinaries 25 .

15. Straits Times Index 26 .Stock Market Seasonality: Day of the Week Effect and January Effect Fig. 16. Hang Seng Index Fig.

17. 18. Nikkei 225 Fig. NASDAQ Composite Index 27 .Stock Market Seasonality: Day of the Week Effect and January Effect Fig.

Nikkei 225. The strategies based on calendar effects did not perform very well on the Australian stock market. This implies that a risk averse investor would prefer this strategy. The best strategy in terms of mean return was the strategy based on the Monday effect. 19. In this case. This is in fact the reward-to-variability ratio (Sharpe. If we had allowed short selling. we would have outperformed the buy and hold strategy by almost 400%. It should be pointed out that this strategy was the best in all the 7 cases out of 7 in terms of the standard 28 .Stock Market Seasonality: Day of the Week Effect and January Effect Fig. The most pronounced is the day of the week effect in case of the S&P 500 and NASDAQ Composite. NASDAQ and S&P 500. The best performing strategy in terms of the standard deviation of returns was also the strategy based on the Monday effect. It is apparent from this table that the buy and hold strategy preformed always worst. the additional profits achieved by these strategies would not be probably high enough to cover transaction costs. Standard & Poor’s 500 Index In case of the FTSE 100. 1994). Moreover. The buy and hold strategy is not the best strategy in any of the considered indices. which implies higher returns at a lower risk. its standard deviation. 1966) or the Sharpe ratio with a constant risk free rate (Sharpe. and a ratio of these two variables. Table 10 further summarizes the results by providing the mean daily return. the standard deviation of returns would have been lower than in case of the buy and hold strategy. Straits Times. which means twice as big performance as is achieved by the buy and hold. we would have moderated the crash that followed the DotCom bubble. the buy and hold strategy was the worst performing strategy. which means 5 indices out of 8. It might be appropriate to mention that if we had used the strategy based on the Monday effect for the Nikkei 225. The September effect would have led to a remarkable outperformance of 100%. we would have not suffered a loss connected with the so called “Lost decade” on the Japan Stock Exchange. If we had used the strategy based on short selling of the portfolio every September in case of the FTSE 100.

0325 0.7293 1.1104 1.0435 0.0240 0.2756 0.0203 0.0283 0.0322 0.0341 0.0327 0. If we assume a constant rate of substitution between risk and return. it can be concluded that this strategy is superior both in terms of risk and return.9841 0. It can provide a higher return with a lower risk than the buy and hold strategy.0629 1.0266 0. n1 − 1. the equality of variances must be tested by a F-test.3284 1.0345 0. At first. we can conclude that the best performing strategy was the strategy based on the ration between the mean return and standard deviation of returns.9833 1.0343 Index Australia All Ordinaries Strategy Buy and Hold Long Positions Monday Long Positions September Mean Return % 0.0565 0.8447 0.0343 0.0430 0.9378 FTSE 100 Buy and Hold Long Positions Monday Long Positions September Hang Seng Buy and Hold Long Positions Monday Long Positions September NASDAQ Composite Buy and Hold Long Positions Monday Long Positions September Nikkei 225 Buy and Hold Long Positions Monday Long Positions September Straits Times Buy and Hold Long Positions Monday Long Positions September SP 500 Buy and Hold Long Positions Monday Long Positions September However. Ratio of the mean return and its standard deviation 0.9537 1.0375 0.0418 0.0252 0. n2 − 1) .0386 0.0322 Standard Deviation of Returns 0. On base of this ex post observation.0426 0. Then if s2 variances are unequal.0319 0.0635 0.1144 1.0283 0.0127 0. it is necessary.4102 1.0288 0.0100 0. which was again the strategy based on the Monday effect.1232 0.4709 1.0340 0.0335 0.0506 0.0339 0. which is s2 2 defined as F = 12 .2597 1.7901 1.2099 1. Table 10.9898 0.0502 0.0504 0.8789 0.0186 0.0627 0. to test if the differences in returns and standard deviations are statistically significant.0290 0.0179 0. This probably means that Monday returns are very volatile.0271 0.Stock Market Seasonality: Day of the Week Effect and January Effect deviation of returns.0269 0.4607 1.0068 0.0284 0.0298 0. H0 states σ 12 = σ 2 and is rejected if F > F (α / 2.0309 0. equality of means can be tested by the following t-statistic 29 .2944 1.

9602 1.9602 1.0422 1. in terms of standard deviation of returns. The strategy based on avoiding September is able to beat the buy and hold strategy in 6 cases out of 7. df ) . This implies that the two strategies considered are not able to achieve statistically significantly higher mean returns.0285 t-statistic in absolute value 0.0424 1.0879 1. However.3050 0.2797 0.9601 1.5019 1.0461 1.3573 1.3267 critical value of t-distribution 1. where df = 2   s12 s 2   + n  n 2   1 2 2 2  s12   s2     n  n   1   + 2 n1 − 1 n 2 − 1 2 .0288 0. Particularly the strategy based on avoiding Mondays is very powerful in this sense.2774 0.0424 1.9601 1.2419 0.2815 1. mean returns are never statistically significantly different.4157 0.9601 30 .0426 1.0426 1.0341 1. Table 11 provides values of t-statistics and critical values of a relevant distribution.0771 1.9602 1.5642 0.2913 1.Stock Market Seasonality: Day of the Week Effect and January Effect t= ( x1 − x 2 ) 2  (n1 − 1) s + (n 2 − 1) s 2   n1 + n 2 − 2  2 1  1 1  +  n n 2  1 .0341 1.0452 1.9602 1.1167 1.0120 1. It has statistically significantly lower standard deviations of returns than the buy and hold strategy in all the 7 cases considered.2884 0.9602 1. Table 11.0285 1.0452 1. Nonetheless.0840 1. they can significantly lower risk in terms of standard deviations.9602 1.9602 1. It is apparent from this table that variances are always statistically significantly different from the buy and hold strategy.0845 1.6781 0.2870 1.0716 1.9602 1.9602 1.3061 0.4743 0. Index Australia All Ordinaries Strategy Long Positions Monday Long Positions September FTSE 100 Long Positions Monday Long Positions September Hang Seng Long Positions Monday Long Positions September NASDAQ Composite Long Positions Monday Long Positions September Nikkei 225 Long Positions Monday Long Positions September Straits Times Long Positions Monday Long Positions September SP 500 Long Positions Monday Long Positions September F-statistic 1.0422 1.4209 1.2683 1.3048 1.1012 critical value of F-distribution 1. H0 stating x1 − x 2 is rejected if t > t (1 − α / 2.0461 1.9601 1.9602 1.

The day of the week effect was found in case of 12 indices out of 28 considered and the month of the year effect in case of 17 out of 28 by using a simple dummy variable approach. significantly overvalued for far too long time for the market to be at least weakly efficient. it took approximately 1820 weeks or 9100 trading days to the investors to exploit this effect and to wipe it out. However. it still implies an unexploited profit opportunity that should have been eliminated by rational investors once it was discovered.Stock Market Seasonality: Day of the Week Effect and January Effect 5. this outperformance was not statistically significant. in terms of standard deviations. They were also able to outperform the buy and hold strategy in terms of the mean return. 31 . The international evidence in support of the calendar effects is mixed. Furthermore. In other words. which the efficient market hypothesis implicitly recommends. If we recall the formula from Fama (1970) expressing undervaluation and overvaluation of security price Ε (x j . This was concluded by using time series of 7 indices with the longest data range. It might be appropriate to mention that the day of the week effect was discovered in 1960 and it disappeared around 1995. It has been also found that the day of the week effect became insignificant during the second half of 1990s.t +1 Φ t ) = 0 . Nonetheless. four market timing strategies were simulated on the data sets of these 7 indices. it appears that in terms of the day of the week effect the indices were persistently. Conclusion This thesis has examined 30 individual stocks of the German Stock Exchange for the presence of the day of the week effect and 28 world indices for presence of the day of the week effect and the month of the year effect. This seems to violate the efficient market hypothesis. No evidence of the day of the week effect has been found for stocks of the German Stock Exchange where an extended dummy variable approach was applied. These strategies were able to statistically significantly outperform the buy and hold strategy.

:pp. F.com>. [21] Lakonishok.ebsco. and Schallheim. 21 (1): pp. [23] Rozeff. G. 3. Issue 1: pp. no. pp.jstor.org/fred2/> [Accessed 9th May 2009].com>. (1991) ‘Efficient Capital Markets: II‘ Journal of finance. 4. 2186 [online]. Available at: <http://www. [online]. Issue 4:pp. E. Available at:<http://de. B.sciencedirect.org>. 40. K.. S. 39 (1): pp. Available at: <http://www. Gordon.com>. [2] Ajay.sciencedirect. Journal of the Royal Statistical Society. and Jorion. Available at: <http://www. [19] Keim. H. 243-260 [online]. Inc. [6] Fama. W. 3-56 [online].). 1.jstor. W. pp. 20. 159-174 [online]. vol. (1980) ‘Stock returns and the weekend effect’ Journal of financial economics. and Louvet. and Kishore. Available at: <http://www. vol. 21 (5): pp. [9] Fama. F.yahoo. F. (1990) ‘Weekly Patterns in Japanese Stock Returns‘ Management science. Perry. (1994) ‘Anomalies or illusions? Evidence from stock markets in eighteen countries‘ Journal of International Money and Finance. Available at: <http://www. Available at: <http://www.com>. and Kinney Jr.W. M.com>. vol. J.com> [11] © 2009 Yahoo! Inc.org>. vol.stlouisfed. [15] Greene. No.. F. 55–59. M.jstor.ebscohost.jstor. R.org/html/content/eonia_data. M. Available at: <http://www. vol. [3] Condoyann. 8.blackwell-synergy. [22] Malkiel. pp. F. [17] Kato. J.behaviouralfinance. E. Available at: <http://www. 39. A. O’Hanilon. Available at: <http://ejournals. 5: pp. W.ebsco. 119-138 [online]. Issue 2: pp. W. [13] French. A. Bailey. R. Vol. (1996) ‘The January Effect: Still there after all these years’ Financial analyst journal. E. 9. Inc. (1987) ‘Day of the Week Effects on Stock Returns: International Evidence‘Journal of Business Finance & Accounting.com/ >. Upper Saddler River. R. Available at: < http://ejournals. R. vol. 819-835 [online].org>.jstor. G.sciencedirect.org>. J. 55-69 [online]. A Random Walk Down Wall Street (6th ed. No.html/> [Accessed 9th May 2009].euribor. (1983) ‘A Further Investigation of the Weekend Effect in Stock Returns’ The Journal Of finance. 44. 36. NJ: Prentice Hall. 1575-1617 [online]. And French. R. 579-596 [online]. 52. J.jstor.sciencedirect.. ‘The Analysis of Economic Time-Series-Part I: Prices’. 39. T. J. 83-106 [online]. and Hess P.com>. and Stambaugh. Issue 1. 27-31 [online]. 224-236. Available at:<http://research. (Yahoo Finanzen – Kurslisten) [online]. 49–58. [24] Sharpe. (1995) ‘The day-of-the-week effect: The international evidence‘ Journal of Bankiny and Finance.. (1999) Investments. 4. Available at: <http://www.org>. M. P. No.com/ >.net/weekend-effect/>. [18] Kato. pp. A. D.sciencedirect.Stock Market Seasonality: Day of the Week Effect and January Effect References [1] Agrawal. A. vol. vol. [20] Kendall. [7] Fama. K.ebsco. (1973). New Persey: Upper Saddle River [16] Haugen.. Available at: <http://www. (1994) ‘The Sharpe Ratio’ Journal of Portfolio Management. vol. [12] © 2008 Yahoo! Deutschland GmbH. vol. 379-402 [online]. 13. [online].3. vol. 383-417 [online]. No.New York: W.25. K. 54. (1953). 53-62 [online]. (1988) ‘Are Seasonal Anomalies Real? A Ninety-Year Perspective’ The Review of financial studies. (1976) ‘Capital market seasonality: the case of stock retursn’ Journal of financial economics.org>. (1981) ‘Day of the Week Effects and Asset Returns‘ The Journal of Business.. W. (1965) ‘"Random Walks In Stock Market Prices". V. Mehdian S. (2002) Econometric analysis. 5th ed. [8] Fama. [25] Sharpe. (2004) ‘A factor model of seasonality in stock returns’ The Quarterly Review of Economics and Finance. (1970) ‘Efficient capital markets: a review of theory and empirical‘ Journal of finance. Financial Analysts Journal. and Smidt. [10] Gardezabal.com>. pp. F. S. (1985) ‘Seasonal and Size Anomalies in the Japanese Stock Market’ The Journal of Financial and Quantitative Analysis. 20. vol. 32 . vol. Available at: <http://www. E. S. And Regulez. J. Ward C. No. [4] Dubouis. F. Available at: < http://ejournals... F.jstor.html> [Accessed 9th May 2009]. L. Available at: <http:// http://www. 2. P. (Yahoo Finance – Market Stats) [online]. (1993) ‘Common risk factors in the returns on stocks and bonds’ Journal of Financial economics. vol. 1031-1043 [online].com/de/overview. B. M. (2004) ‘The Day-of-the-Week Effect in Stock Returns : Further Evidence from Eastern European Emerging Markets’ Emerging Markets Finance and Trade. Issue 4: pp. A (General) 116 (1): 11–34. W. F. [26] Sharpe. Available at: <http://www. Available at: <http://www. (1966) ‘Mutual Fund Performance’ Journal of Business. 403-425 [online]. R.org>. Available at:<http://www. Available at: < http://calendareffects. vol. vol. [5] © 2000 Euribor FBE (Eonia Historical data) [online]. pp. [14] Gibbson M. Issue 1: pp. pp. Available at: <http://web. Norton & Company.46. Issue 9: pp..sciencedirect. R. Available at: <http://www. 1463-1484 [online].com/ >. vol. 14 Issue 2: pp. J.org>. vol.biz.jstor.

Vol.. 52. 1980). vol. 2171-2186 [online].. (2006) Introductory econometrics : a modern approach. J. [28] White.jstor.jstor. and Erickson. K. 817-838 [online]. 3rd ed. Y. J. 5. Available at: <http://www. pp. [29] Wooldridge. (1980) ‘A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity’ Econometrica.Stock Market Seasonality: Day of the Week Effect and January Effect [27] Wang. (1997) ‘A New Look at the Monday Effect‘ The Journal of Finance.org>.org/pss/1912934>. pp. Available at: <http://www. Mason: Thomson/South-Western 33 . 48. 4 (May. No. No. H. Li.

Master your semester with Scribd & The New York Times

Special offer for students: Only $4.99/month.

Master your semester with Scribd & The New York Times

Cancel anytime.