ssrn id2154873

Transcript

1 , “Sell in May and go i ndicator The Halloween Away”: an even bigger puzzle * Ben Jacobsen University of Edinburgh Business School [email protected] Cherry Y. Zhang , China Nottingham University Business School N ottingham.edu.cn [email protected] Z i. Y - C herry not only defies stock market Our simple new test for the Sell in May effect shows it h efficiency but also challenges the existence of a positive risk return trade off. W e we n examine the effect using all historical data for all s tock ma rket indices worldwide, we only of a significa - - sk return’ nt positive ‘ri trade October) - during summer (May off find evidence in Mauritius . P ooling all country data we find excess returns during summer are . Over the full year we find ( - 1.2% based on 33,348 monthly returns ) significant ly negative - premium of 3.7% annually value 7.65) . (t a positive estimate for the e quity * Place, Edinburgh, Scotland Corresponding Author: University of Edinburgh Business School, 29 Buccleuch EH8 9JS, United Kingdom, Tel: +44 131 651 5978 http Electronic 2154873 = abstract / ssrn.com / / : : at available copy

2 1. Introduction Since 2002 when Bouman and Jacobsen published their study on the Halloween Indicator , in the American Economic Review Sell in May and go away ’ effect , also known as the ‘ popular press. the ir study has stirred a fierce debate both in the academic literature and the that Bouman and Jacobsen (2002 ) find returns during winter (Nov ember through April) are - October) in 36 out of the 37 countries in higher than during summer (May significantly their study. As it was a new market efficiency anomaly they called it: ‘another puzzle’ One purpose of this paper is to rigorously re - examine the Halloween or Sell in May puzzle tatistical and address issues raised in t h e d ebate on data mining, sample selection bias, s 1 e importantly, w e Mor economic significance. add a new simple also problems, outliers and dom. We add this new test for two reasons. test for this market wis Firstly, one could argue a that the test in Bouman and Jacobsen (2002) is not proper test of the Sell in May effect. Bouman and Jacobsen test whether winter returns are higher than summer returns. However, the market wisdom suggests all not invest in stock markets during the is that one should , of the adage would be whether summer months. So a better test summer returns are significantly higher than short term interest rates. If excess returns are not significantly it , invest in negative to for risk averse investors makes no sense different from zero , or even 2 The second reason for the stock market during summer. This is the new test we perform. ot . N the Sell in May effect aspect of , mostly ignored , it reveals another this new test is that ket wisdom vary predictably s defy market efficiency because return would the mar only the challenge with the seasons. It would also existence of a positive risk return trade off 3 . a violation during a substantial part of the year and predictably so suggest would of This as be the most fundamental relations in finance. For t hat reason we want thorough to one of consider all stock markets worldwide using the full history of stock market as we can and 1 Maberly & Pierce, 2003; Maberly & Pierce, 2004; Lucey & Zhao, 2007; Zhang & Jacobsen, See for instance, 2012; Powell, Shi, Smith, & Whaley, 2009. 2 In the Bouman and Jacobsen te st, summer returns may be lower than winter returns but if summer returns are higher than the short term interest rates it might still pay to stay in the stock market. 3 This test is also interesting as we still lack a proper explanation on what causes the effect (see for instance, cast doubt on explanations that rely only on behavioral changes Jacobsen & Marquering, 2008) and this tests in risk aversion to explain the effect. Investors have to become systematically risk seeking to explain zero or negative equity premia in the long run. 2 / 2154873 = abstract / ssrn.com / Electronic : http : at available copy

3 4 indices available for each market. study to date which has done We are not aware of any his seems probably the best safeguard against data mining and sample selection bias. so but t Or, as an author on the Seeking Alpha website described our approach: “it is the lethal 5 weapon against skepticism.” stock markets with stock market indices in the world for which Our data consists of all 10 9 exist The sample starts with the UK stock market in 1693 and ends with the price indices . 6 arket of Syrian Arab Republic which starts in 2010. stock m For our tests addition of the total return data and short term interest rates historical equity premia we rely on for the 7 available for 65 stock markets . all For each individual market we use which are jointly additional historical data available for that market. An this approach is , that advantage of cross country estimate s of the equity we get what might be one of the most accurate . An estimate based on all historical total return data and short term interest data premium world wide . On average we find an historical estimate for th e equity premium available for these 65 countries % annually (significant with of 3.7 based on the 33,348 observations - value of 7.65) . While lower than 4.5% estimated in Dimson, Marsh and Staunton (2011) , a t inte he good news of our study is that t more extended his rnational evidence also suggests t there is an equity premium. Results are less comforting w we consider whether excess returns in summer are hen . n none of the 65 countries for which we have total returns significantly higher than zero I – i us - can we reject a and short term interest rates available with the exception of Maurit based on our new test . For no other stock market in the world do we find Sell in May effect evidence of sign i tive excess returns during summer , o r, i n other words , a ficantly posi 4 Another reason why we use all data in all countries is that Zhang and Jacobsen (2012) show even with an extremely large sample for just one country (the same UK data set we use here) it is hard to determine anomalies exist. The problem is the same as put forward by Lakonishok and Schmidt (1988): whether monthly To detect monthly anomalies one needs samples of at least ninety years, or longer, to get any reliable estimates. Looking at all historical data across all countries seems the best remedy. It seems fair to say that at least this makes the ‘Sell in May’ effect the most extensively tested anomaly in the world. 5 - seasonal - patterns - in - stock - markets http://seekingalpha.com/article/1183461 319 - years - of - evidence . - 6 Initially, we find 143 countries with active stock exchanges. B ut many newly established markets only trade a limited number of stocks and do not maintain a market index. We exclude Cambodia, Laos, Fiji and Zimbabwe as they have fewer than a year of observations. 7 While we have the data for Brazil as well we exclude them because of long periods of hyperinflation. 3

4 positive risk return trade off . Figure 1 summarises our main result. It plots the risk premia for 65 countries during the summer months . Please insert Figure 1 here not only not significantly positive, the y are in most cases Unfortunatel y, these results are even marginally positive. I not n 46 countries the excess returns during summer have been negative, and in 9 significantly so. Only Mauritius shows a significant positive relat ion between risk and return in summer and only at the 10% level. Overall based on 33,348 observations we find that average stock market returns (including dividends) during May to or 0.20% per month) lower than the short term inter est rate and October have been 1.17% ( these negative excess returns are significantly different from zero (t - - 3.36). This value of absence of evidence of an equity premium during summer motivated the part of ‘an even bigger puzzle’ in our title. nly in the winter months do we fin d evidence of a positive risk O return relation. Average excess returns from November to April are 4.89% or (0.41% per - value of 14.52. Of course, risk would be an month) and these are significant with a t obvious (partial) d deviations are higher during explanation but if anything standar 8 summer. T he evidence on negativ e risk premia we report here suggests that the Halloween effect differs from other seasonalities like for instance the same month seasonal reported by recently Heston and Sadka (200 ) or ‘Day - of - the - w eek ’ - effect. Both seasonals are , 2010 8 considered by Keloharju, Linnainmaa, and Nyberg (2013) and they find these seasonals may be risk related if risk factor loadings may not accrue evenly through the year. Apart from this new violation of the risk return trade off, there are more reasons why the Sel l in May effect seems to be the anomalous anomaly and remains interesting to study . 8 In Appendix 3 we test this possibility in more detail using GARCH(1,1) models where we can assess risk onjunction with differences in mean returns between su mmer and winter. In 23 out of the 57 differences in c countries (and also for the world market index) for which we have enough da ta to test for risk differences, we find that risk is significantly higher in summer than winter. Winter shows significantly higher risk o nly in 13 countries. This suggests that not only stock market returns may be lower during summer. If anything, after correcting for Sell in May mean effects and volatility clustering effects, volatility may b e higher too, further increasing the puzzle on the risk return trade off. 4

5 The adage has been ‘publicly available information’ for a very long time even before the 9 ) sample. Nevertheless, it seems to defy economic gravity. It Bouman and Jacobsen (2002 does not disappear or reverse itself, as theory dictates it should (Campbell, 2000 and Schwert, 2002), or seems to happen to many other anomalies (Dimson and Marsh, 1999 and McLean and Pontiff 2014). In fact, a number of papers have appeared recently that find results similar to ours with respect to the Bouman and Jacobsen (2002) out of sample some 10 evidence. The fact that trading on this strategy is particularly simple makes its continued ence even more surprising. exist Apart from our new test for a Sell in May effect , our comprehensive dataset allows us to revisit the old test in Bouman and Ja cobsen (2002) . Moreover, we deal with the important issues raised in the debate which followed their p ublication . In short, we find that - based on all avai l able data - none of the criticism survives closer scrutiny . Here are our main findings. Overall, the 56,679 monthly observations over 319 years show a strong Halloween effect suggested in Bouman and Jacobsen (2002). Winter returns when measured the way as – November through April - are 4.5% (t - value 11.42 ) higher than summer returns. The Halloween effect is prevailing around the world to the extent that the mean returns are 9 - April than for May - higher for the period of No 2 out of 10 vember countries. October in 8 T he difference is statistically significant in 35 countries, compared to only 2 countries having significantly higher May - October returns. Our evidence reveals that the size of the Halloween eff ect does vary cross - nation. It is stronger in developed and emerging markets than in frontier and rarely studied markets. Geographically, the Halloween effect is more prevalent in countries located in Europe, North America and Asia than in other areas. As we show, however, this may also be due to the small sample sizes yet available for many of these newly emerged markets. Th e effect is even more robust in our total return and risk premium estimates. Out of the 65 markets, 58 total market returns (and 56 ri sk premium 9 As we show here the market wisdom was already reported in 1935 and at that time already well known, at least in the United Kingdom. 10 See for instance, Andrade, Chhaochharia, & F uerst, 2012; Grimbacher, Swinkels, & van Vliet, 2010; & Visaltanachoti, 2009. Jacobsen 5

6 series) show positive point estimates for a and for 34 (and 32) markets Halloween effect, are statistically significant. these results Using time series subsample period analysis by pooling all market indices together, we - ye ar sub - periods 24 have November - show over 31 ten - April returns higher than the May October returns. The difference becomes statistically significant in the last 50 years starting from the 1960s. The difference in these two 6 month period returns is very persistent and - econ omically large ranging from 5.08% to 8.91% for the most recent five 10 - year sub - periods. The world index from Global Financial Data reveals a similar trend. Subsample period analysis of 28 individual countries with data available for over 60 years also con firms this strengthening trend in the Halloween effect. More specifically, measured over 28 all these countries t he Halloween effect emerges around the 1960s, with 27 out of the se countries revealing positive coefficient estimates in the 10 year sub 1970. - - period of 1961 Both the magnitude and statistical significance of the Halloween effect keeps increasing - period 1991 to 2000 showing the strongest Halloween effect among over time, with the sub countries. Consistent with country by country whole sample peri od results, the Halloween effect is stronger in Western European countries. We show the economic significance of the Halloween effect by investigating the out - of - sample performance of the trading strategy in the 37 countries used in Bouman and Jacobsen ( 2002). The Halloween effect is present in all 37 countries for the out - of - sample period September 1998 to sample gains from the Halloween strategy July 2011. The out - of - are still higher than the buy and hold strategy in 31 of the 37 countries; after taking risk into account, the Halloween strategy outperforms the buy and hold strategy in 36 of the 37 countries. In addition, given that the United Kingdom is the home of this old market wisdom (and has shown a Halloween effect throughout its history) we examin e the performance consistency of the trading strategy using long time series of over 300 years of UK data. The result shows that investors with a longer horizon would have had remarkable estment horizons over odds beating the market using this trading strategy: Over 80% for inv 5 years; and over 90% for horizons over 10 years, with returns on average around 3 times higher than the market. 6

7 We also address a number of methodological issues concerning the sample size, impact of er time varying volatility, outli s and problems with statistical inference using UK long time series data of over 300 year. In particular, extending the evidence in Zhang and Jacobsen (2012), we revisit the UK evidence and provide rolling regressions for the Halloween effect with a large sample size of 100 - year time intervals. The results show that the Halloween most often si gnifica nt if measured this way. Although even within this long sample effect is there are subsamples where the effect is not always significant. oint estimates are always P , b king GARCH effects into account estimates ta positive based on traditional regressions ut or outlier robust regressions occasionally show negative point estimates halfway through the previous century. This dataset also allows us to test an argume nt put forward by Powell et al. (2009). They question the accuracy of the statistical inference drawn from standard OLS estimation with Newey and West (1987) standard errors when the regressor is persistent, or has a highly d the dependent variable is positively autocorrelated. autocorrelated dummy variable an They suggest that this may affect the statistical significance of the Halloween effect. This argument has been echoed in Ferson (2007). With the benefit of long time series data, we address this concer n by regressions using 6 monthly, rather than monthly, returns. The bias if any seems marginal e find almost similar standard errors regardless of whether at best. W we use the 6 - month intervals, or the monthly data, to estimate the effect. We feel our pa per adds to the literature in a number of ways. Firstly, we provide the lethal weapon to answer the skeptics when it comes to the Sell in May effect by looking at all available data. Based on all historical returns of 109 countries the Halloween effect see ms a bigger puzzle than we may have realised before. Secondly, we introduce a simple new tests that not only shows that the Halloween effect is interesting from a market efficiency point of view but highlights how the empirical evidence systematically se ems to violate the positive long run relation we would expect to see between risk and return. In this sense we reveal what may be the most puzzling aspect 7

8 of this phenomenon: in no country – apart from Mauritius – do we find evidence of a itive risk premium during the summer months. One could argue this seems significantly pos entional asset pricing theory. to pose a major challenge for conv n interesting by - Thirdly, a and one might call this another contribution is that we product provide a new estimate for the equity premium ( 3.7 %) using probably the largest cross county data set over the most historically long period available. Fourthly , we show how none of the arguments a ffect gainst the existence of the Halloween e es closer scrutiny. The effect holds out put forward to date surviv of - sample and cannot be - explained by outliers, or the frequency used (monthly or six monthly) to measure it. The effect is economically large an d seems to be increasing in the last fifty years . E ven when in doubt of the statistical evidence, it seems that investors may want to give this effect the benefit of the doubt, as trading strategies suggest a high chance of outperforming the market for inv estors with a horizon of five years or more. Of course, just as with in - sample - results, past out - of - sample data do not guarantee future out sample results. of - In short the results we provide here suggest that , based on all country evidence , there is a Hall oween or most often Sell in May effect. While it may not be present in all countries, all the time, it is. Last but not least our results help to contribute on answering what may cause the effect, it , seems that given all the statistical issues it might be difficult to rely on cross sectional evidence to find a definite answer. What we can say is that any explanation should allow for time variation in the effect and should be able to explain why the effect has increased so strongly in the last fifty years . If we assume human behaviour does not change over time this seems to rule out just explanations and suggest changes in socie ty play a behavioural role. Additionally , and maybe more importantly from a theoretical perspective , this explanation should also be able to account for the negative excess returns during the May - October period in stock markets around the world. While it seems unlikely that we will ever find a smoking gun, the circumstantial evidence we report confirm s more recent Kaustia and Rantapuska, 2012 Zhang, 2014) that vacations are the and empirical evidence ( 8

9 most likely explanation. At least, the vacation explanation is consistent with all empirical evidence to date. 2 A short b ackground on the Sell in May or Halloween effect a c obsen (2002) test for the existence of a seasonal effect based on the old Bouman and J ‘Sell in May and go away’ so named because investors should sell their market wisdom stocks in May markets tend to go down during summer. While many people in the because US are unfa miliar w ith this saying there is a similar indicator known as the Halloween indicator, which suggest s leav ing the market in May and com ing back after Halloween (31 (May through October) October). Bouman and Jacobsen (2002) find that summer returns tantially low are subs 36 of the 37 winter returns (November through April) in er than countries o 0 through to August 1998. They find no ver the period from January 197 evidence that the effect can be explained by factors like risk, cross correlation between markets, or – except for the US - the January effect. Jacobsen, Mamun and Visaltanachoti (2005) show that the Halloween effect is a market wide phenomenon, which is not related to the common anomalies such as size, Book to Market ratios and dividend yield. Jacobsen and Visaltanachoti (2009) investigate the Halloween effect among US stock market sectors. They find the effects is stro ngest in production related sectors. The Halloween effect is also studied in Arabic stock markets by Zarour (2007) and in Asian stock markets by Lean (2011). Zarour (2007) finds that the Halloween effect is present in 7 of the 9 Arabic markets in the sam ple period from 1991 to 2004. Lean (2011) investigates 6 Asian countries for the period 1991 to 2008, and shows that the Halloween effect is only significant in Malaysia and Singapore if modelled with OLS, but that 3 additional countries (China, India and Japan) become statistically significant when time varying volatility is modelled explicitly using GARCH models. While Bouman and Jacobsen (2002) cannot trace the origin of this market wisdom, they are back to 1964 before the start of their able to find a quote from the Financial Times dating sample. This makes the anomaly particularly interesting. Contrary to, for instance, the iven inference, but based January effect (Wachtel, 1942), the Halloween effect is not data dr 9

10 on an old rs c ould have been aware of. This reduces the market wisdom that investo 11 Bouman and Jacobsen investigate several possible explanations, likelihood of data mining. but find none, although they cannot r eject that the Halloween effect might be caused by summer vacations, which would a lso explain why the effect is predominantly European. - term history of UK data is especially interesting, as the United Kingdom is the Our long home of the market wisdom “Sell in May and go away”. Popular wisdom suggests that the e English upper class spending winter months in London, but effect originated from th spending summer away from the stock market on their estates in the country: An extended version of summer vacations as we know them today. Jacobsen and Bouman (2002) report e Financial Times as the oldest reference they co uld find at the time. a quote from 1964 in th ith more and more information becoming online we can now report a written W accessible in the Financial mention of the market wisdom “Sell in May” Times of Friday 10 of May 1935. It states: “A shrewd North Country correspondent who likes stock exchange flutter now and again writes me that he and his friends are at present drawing in their horns on the strength of the old adage ‘ Sell in M ay and go aw ay.’ ” The suggestion is that , at th at time , it is already an old market saying. This is confirmed by a more recent article in the Telegraph 12 in 2005 In the article “Should you ‘ S ell in May and buy another day?’” the journalist . ouglas Eaton, who in that year was 88 and was still working as George Trefgarne refers to D a broker at Walker, Cripps, Weddle & Beck. “He says he remembers old brokers using the Button, or messenger, in adage when he first worked on the floor of the exchange as a Blue It 1934. ‘ was a lways sell in May,’ he says. ‘ I think it came about because that is when so many of those who originate the business in the market start to take their holidays, go to Lord’s, [Lord’s cricket ground] and all that sort of thing. ’ ” Thus, if the Sell - in - May an omaly should be significantly present in one country over a long period, one would expect it to be the United Kingdom. Many of the early newspaper articles link the adage to vacation behaviour. 11 F or instance, an implication is that Bouman and Jacobsen (2002) need not consider all possible com binations of six month periods. 12 another - day.html http://www.telegraph.co.uk/finance/2914779/Should - you - sell - in - May - and - buy - 10

11 Gerlach (2007) attributes the significantly higher 3 - month r eturns from October through December in the US market to higher macroeconomic news announcements during the period. Gugten (2010) finds, however, that macroeconomic news announcements have no effect on the Halloween anomaly. Bouman and Jacobsen (2002) fin d that only summer vacations as a possible explanation survive closer scrutiny or liquidity his might either be caused by changing risk aversion . T , the size of the effect is significantly related to both length and that constraints . They report timing of vacations and also to the impact of vacations on trading activity in different countries. Hong and Yu ( show that trading activity is lower during the three summer 2009) . i dence in the se paper s The ev support s the popular holiday months in many countries dom the most convincing evidence b ut probably wis to date comes from recent studies by , Zhang (2014) and ) . Zhang looks at vacation data in 34 Kaustia and Rantapuska (2012 countries and finds strong support for vacation behaviour as an explanation for the lower summer return effect, especially among European countries. Kaustia and Rantapuska (2012) ding decisions of Finnish consider actual tra to be consistent investors and find these trades They also report evidence which is with the vacation hypothesis. istent with the incons Seasonal Affective Disorder hypothesis put forward by Kamstra, Kramer and Levi (SAD) (2003). in stock returns , bu t Kamstra, Kramer and Levi (2003) document a similar pattern attribute it to mood changes of investors caused by a Seasonal Affective Disorder . N ot only , however, does the new evidence i n Kaustia and Rantapuska (2012 ) not sup p ort the SAD hypothesis , the Kamstra, Kramer and Levy (2003) study itself has been critisiced in a but Meschke for its methodological flaws ( number of papers instance, Kelly & f , 2010 ; Keef or & Khaled , 2011 ; Jacobsen & Marquering , 2008, 2009 ). B y itself this does n ot mean , play a role in financial markets however, the SAD effect could not that . B ut our evidence of the absence of such an effect in some periods, coupled with a strong increase in the prevalence of this effect in the last fifty years seems hard to reconcile with a SAD effect . If it was a mood effect o relatively constant over time . Moreover, ne would expect it to be increased risk aversion c aused by SAD might explain lower returns but still would not explain persistent negative excess returns or negative risk premia as we report here . The uggested s as same argument also applies for a mood effect caused by temperature changes , 11

12 by Cao and Wei ( 2005 who find a high correlation with temperature and stock market ) , returns. The long time series data we use here allows us to address a number of methodological that have emerged regarding test ing for the Halloween effect . In particular, there has issues on the robustness of the Halloween effect under alternative model been a debate specifications. For example, Maberly and Pierce (2004) re - examine the Halloween effect in that the Halloween effect in the US is the US market for the period to 1998 and argue caused by two extreme negative returns in October 1987 and August 1998. Using a similar metho ) claim that the Halloween effect is only present in dology, Maberly and Pierce (2003 the Japanese market before 1986. Haggard and Witte (2010) show , however, t hat the identification of the two extreme outliers lack s an objective basis. Using a robust regression een effect is robust technique tha t limits the influence of outli ers, they find that the Hallow 08. 1954 to 20 from outli ers and significant for the period of U - year sub - period analysis over the period of sing 20 Lucey and Zhao (2007) 1926 to 2002, reconfirm the finding of Bouman and Jacobsen (2002) that the Halloween effect in the US may be related to the January effect . Haggard and Witte (2010) show , howe ver, that the to i be attributed nsignificant Halloween effect may the small sample size used, which reduces the power of the test. Wit h long time series data of 17 countries for over 90 years, we are abl ers , as well as increase the sample size in e to reduce the impact of outli examining the out of sample robustness and the persistence of the Hallow een effect in these countries . As we noted earlier, Powell et al. (2009) question the accuracy of the statistical inference drawn from standard OLS estimati on with Newey and West (1987) standard errors when the regressor is persistent, or has a highly autocorrelated dummy variable , and the dependent variabl e is positively autocorrelated. This argument by itself may seem strange as a regression with a dummy va riable is no thing else than a difference in mean test. to explicitly address the issue. Still, it may be worthwhile 12

13 3 . and Methodology Data 13 (GFD) Datastream , and , We collect monthly price index data from Global Financial Data s individual for all the countries in the world that have stock market indices stock exchange I nitially , we find a total of 143 countries with active stock exchanges, but many available. established markets only trade a limited number of stocks and do not newly maintain a market ind ex. We also require the countries to have at least one year of data to be included 14 analysis in . As a result our the sample size reduces to 109 countries , consisting of all 24 developed markets, 21 emerging markets, 30 frontier markets classified by the MSCI market classificatio 4 countries that are not included in the n framework and an additional 3 15 Our sample has e denote them as rarely studied markets . MSCI market classification . W countries of course a considerable geographical coverage : we have 1 6 African countries, 19 39 countries from in Asia, ntries cou Europe, 1 3 countries located in the Middle East, 11 from North America and 9 from South America, as well as 2 countries in Ocean ia. W e also 16 obtain total return indice data for 65 countries in order to address the s and risk free rate possible impact of dividend payments and reveal the pattern of market risk premium s . T his smaller stock markets for which we can find total market return sample covers all the indices . We use Treasury bills or the nearest comparable short term instrument as the proxy for risk free rates. Appendix 1 presents the source s and sample period s of the price index, bas total return index and the proxy of the risk free rate for each country grouped on the is of their . For many of the countries, the MSCI market classification and geographic region the entire trading history of their stock ma rket. time series In particular, we almost cover have over 310 year monthly market index prices for the U nited Kingdom, more than s of 13 When data is available from both GFD and Datastream , we choose the one with longer sample periods. 14 e due to insufficient observations. Cambodia, Laos, Fiji and Zimbabwe are excluded from our sampl 15 Our market classification is based on “MSCI Global Investable Market Indices Methodology” published in August 2011. MSCI classifies markets based on economic development, size and liquidity, as well as market accessib ility. In addition to the developed market and emerging markets, MSCI launched frontier market indices in 2007; they define the frontier markets as “all equity markets not included in the MSCI Emerging Market Index that (1) demonstrate a relative openness and accessibility for foreign investors, (2) are generally not considered as part of the developed market universe, (3) do not belong to countries undergoing a period of extreme economic or political instability, (4) a minimum of two companies with securit ies eligible for the Standard Index” (p.58). The countries classified as rarely studied markets in our sample are not necessarily the countries that are less developed than the frontier markets; they can be countries that are considered part of the develop ed markets’ universe with relatively small size; for example, Luxembourg and Iceland; which are excluded from the developed market category by MSCI. 16 tes, We excluded Brazil from the sample even we do have the date of total returns and short term interest ra because of the extremely high observations due to the hyper inflation from 1980s to 1994. 13

14 210 years for the United States and over 100 years data for another 7 countries. The world index is the GFD world price index and GFD world return index that goes back to 1919 and 17 1926 respectively s provided in the first row. For the price , the information for the index i here are 28 countries in total h aving data available for over 60 years. These long indices, t ta allows us to examine the evolution of the Halloween effect by conducting time series da sub - period analysis. Although the countries with long time series data in our sample are primarily developed European and North American countries, we do have over 100 years data for Australia, South Africa and Japan, and over 90 years data for India. We also have countries with data countries with very small sample size for example, there are 1 3 for less ; than 10 years. All price indices are quoted at local currency, except Georgia where the only index data available is in USD. on whether excess returns in summer are significantly positive we Apart from our new test also investigate the statistical significance of the Halloween effect using the Halloween dummy regression model the traditional way : (1) is the Halloween is the continuously compounded monthly index returns and where dummy, which equals one if the month falls in the period of November through April and is zero otherwise. If a Halloween effect is pr esent we expect the coefficient estimate to be significantly positive, as it represents the difference between the mean returns for the two October. 6 - month periods of November - April and May - 17 The index is capitalisation weighted starting from 1970 and using the same countries that are included in the MSCI indices. Prior to 1970, the index consists o f North America 44% (USA 41%, Canada 3%), Europe 44% (United Kingdom 12%, Germany 8%, France 8%, Italy 4%, Switzerland 2.5%, the Netherlands 2.5%, Belgium 2%, Spain 2%, Denmark 1%, Norway 1% and Sweden 1%), Asia and the Far East 12% (Japan 6%, ustralia 2%, South Africa Gold 1%, South Africa Industrials 1%) , weighted in January 1919. The India 2%, A country weights were assumed unchanged until 1970. The local index values were converted into a dollar index by dividing the local index by the exchange rate. 14

15 4. Price Returns, Risk Premiums and Dividend Yields 4.1. Overall results first calculate continuously compounded monthly returns for both price i ndices and We . We also estimate the risk premiums for the countries by subtracting total return indices s risk free rate from the total return series. Table 1 p resent summary monthly statistics of the price returns, total returns and risk premium s . Please insert Table 1 around here The top section of the table show s the annualised mean returns and standard deviations for the world index and pooled countries. The statistics for the price returns are calculated from 56,679 sample observations over 109 countries from year 1693 to 2011, and the results for ium are computed based on 33,348 observations from 65 the total return and risk prem countries for % 2 the period 1694 to 2011. The average price returns and total returns are 9. and 10.8 over the entire sample , if we only consider the 65 countries that have total return % data available, the mean capital gain is about 7% per annum, which lead to an estimation of average dividend yield of 3.8%. the This result coincides with a similar dividend historical yield of 3.6% inferred from the world total return and price return indices over the p eriod 1926 - 2011. Figure 2 plots 30 - year moving averages of total returns, price returns, risk premiums and dividend yield from pooled 65 countries over the period 1694 to 2011. In Figure 3 we zoom in on the more recent period period results are based on a larger number of as for that countries . Figure 2 makes clear that d ividend yield weights a large portion of total returns in the first the total returns centuries , in fact, dividend is almost the sole contributor to two up since 1910s. We up to around 1850 s. The weight of the price returns starts catch ing observe a continuous trend of declining dividend yields accompanied with increased price 15

16 returns over the recent 50 years beginning from 1960s . For example, the dividend yield 18 s for 30% of the total return in the latest 30 - year observation. t only weigh Please insert Figure 2 and 3 around here we observe lower mean returns with relatively smaller standard For individual countries, her markets, and the emerging deviations for countries in developed markets than the ot ns with the largest volatility. For examp le, market tends to have the highest average retur price returns for all develop ed markets in our sample is 6.5 % , which the average annualised just over about one - third of the average return of the emerging ma rkets (1 6.8 %) and is only half the si ze of the frontier markets (11. %) and the rarely studied markets (1 0. 8 %). 4 , with an Meanwhile, the volatility for the emerging markets is among the highest ised standard deviation of 35 . annual % comparing to 2 1 % for the developed markets, and 2 29. 3 % and 3 3 . 5 % for the frontier and rarely studied markets. Despite of a sma ller sample are size s reveal a similar pattern, the mean returns (standard deviations) , total return 9.5% (20.9%), 16.4% (33.5% ) , 12.7% (29.4%) and 5.4% (38.8%) for developed, emerging, The highest increase in monthly index frontier and rarely studied markets, respectively. returns is and the largest plunge in index prices in a % 213.1 2007 % in Uganda in October single month is 465.7 in July 2008 (Note that because we use log returns, drops % in Egypt . T he unequal sample size among the countries does, of more than 100% are possible) however, make direct comparison across nations difficult. We address this by applying sub - period an of the paper . alysis in the later sections Table 1 also reveals some interesting observations about the risk premium. The pooled 65 countries’ result over 3 18 - year s history suggests an average and significant risk premium of 3.7% . This is a bit lower than 4.5% estimated in Dimson, Marsh and Staunton (2011) using data 19 countries bu t it s confirms their argument that a 6% over the period 1900 to 2011, risk premium commonly used in finance text books is too high. The green line of Figure 2 depicts a 30 year moving average of the risk premiums of the pooled countries. The risk - 18 It seems this offsetting trend between dividend yield and price returns are driven by three major markets: UK, US and Australia, the level of dividend yields tend to be quite fixed over time for other countries. In - year moving averages for 11 countries that have data available for over 60 years. Appendix 2 we plot the 30 16

17 premium s in the first 230 years. It grows up to 10% in the late 1940s, then rarely excess 4% T his confirms the widely held gradually declines to about 3% in the latest observation. eve that the high risk premium in the recent past may be due to the exceptional growth beli the economies around the world . in 4. 2 Total returns and risk premiums in summer and winter The total return data and short term interest rates allow us to investigate the behaviour of risk premiums in summer and winter. As we discussed before “Sell in May and go away” suggests leaving the stock market altogether. Even summer returns are significantly lower than winter returns , investors might still be better off to remain in the market if these returns are greater than the risk free rate . Hence, one could argue that a better test of the Sell in May effect is whether excess returns a re positive during summer. If summer returns rom (or even significantly lower than) interest rates the are not significantly different f correlate positively market wisdom seems to holds. The results of this test will, of course , wit While the Bouman and Jacobsen (2002) reveals h the Bouman and Jacobsen (2002) test. an interesting pat tern, t he advantage s of our new test are two - fold. Firstly, this test is more illustrate in line with the actual market wisdom , and , additionally , this new test s much more clearly what makes the anomaly interesting beyond a market efficiency point of view . It not only violates the notion that returns should be difficult to predict, but also that there is no risk return trade off during long predi we plot the risk ctable time periods. In Figure 4 emia in summer (as in Figure 1) pr but add the winter risk prem ia for comparison. Please insert Figure 4 around here. month periods for 65 compares the total return and risk premium bet ween two 6 - Table 2 markets. For comparison we also include the Halloween dummy based on the old test. around here Please insert Table 2 17

18 We observe the presence of negative summer risk premiu m in 45 out of 65 countries. I n 8 A verage excess summer returns countries these risk premia are significantly below zero. are lower than winter returns for most of the countries except f or 8 markets. Summer returns tend to be insignificant even befor This is in striking e deducting the risk free rates. contrast with winter (excess) returns which are often significantly greater than zero, When w especially in developed and emerging markets. e pool the data we find that over the entire 33348 monthly observations, during 6 - month summer the average risk premium period is - 1.17% (t - winter with 4.89% (t - value 14.52 ) during the value 3.36) compared summer is worrying from a risk return negative excess return during months period. This perspective. Why would risk averse investors invest during summer if all historical data tell them that if past returns offer any indications for future returns, these re turns are likely to be negative? Note that this finding also indicates that explanations solely based on changes in risk aversion of investors might not fully explain the effect. The coefficient estimates of the Halloween dummy is statistically significa nt in 34 (and 32) of the 65 countries’ total return indices (and risk premium indices), which is even more pronounced than the res ults 19 price return indices as we wi ll show below . Substantial risk differences might for our explain a huge difference in returns between summer and wint er. However, simple standard deviations do not indicate a difference. If anything risk is higher during summer. We address in more detail later in Appendix 3 5 . The Halloween indicator revisited As noted before the existence of a Halloween effect has bee n debated. It may be good to consider some of the argum ents put forward in the debate. We do this based on the old test which allows comparison with previous results in the literature. We also use price indices as this allows us to test an even bigger samp le of countries (and as we have shown above dividends hardly seem to affect results). Moreover, we include some additional tests that may help shed further light on what or what may not cause this effect. 19 This also reinforces the finding of Zhang and Jacobsen (2013) that there is no strong seasonal effect in dividend payments. 8 1

19 5 .1 Out of sample performance first sure that the Halloween effect still exists beyond the ori ginal To be relevant we in must study. The Bouman and Jacobsen (2002) analysis ends in August 1998. Campbell (2000) ir suggest that if an anomaly is truly anomalous , it should be quickly and Schwert (2002) aged away by rational investors. ( Note that this ar g u ment also should have applied to arbitr (2002) tudy itself , as the market wisdom the Bouman and Jacobsen was known before s their sample period. ) . Many anomalies indeed seem to follow t he theoretical prediction . McLean and Pontiff (2014) investigates the performance of 95 published stock return predictors out of sample and post publication, they show that predictor’s return declines 31% average after taking es into account. on statistical bias investigate whether the Halloween effect has weakened, we start with an out of sample To test of the Halloween effect in the 37 countries examined in Bouman and Jacobsen (2002) . 20 Table 3 compares in - sample performance for the period 1970 to August 1998 with out - of - sample sample performance for the period of September 1998 to November 2011. The in - s test using a different dataset present with Bouman and Jacobsen (2002), similar results to stock market returns from November through April being higher than from May through tober in 34 of the 37 countries, and the difference of statistically significant in 20 Oc being the countries. Although a small sample size may reduce the power of the test, the out of sample performa nce is still very impressive. All 37 countries show positi ve point estimates of the Halloween effect . For 15 countries the effect is statistically significant out of sample . The Halloween effect seems have weakened in the recent years. Moreover, the point not to estimates in the out - of - sample test of 18 countries are even higher than for the in - sample the test. The a verage coefficient estimate in - out of - sample test ing is 8.9 % , compar ed to 8.2 % in the in - Column s 4 and 7 show the percentage of years that November - April sample test. - October returns in the sample for each countr y. Most of the countries returns beats May have a val ue greater than 50% , suggest ing that the positive Hallow een effect is not due to 20 In their study, they have 18 countri es’ data starting from January 1970, 1 country starting in 1973 and 18 - sample test begins from 1970 for those countries with data available in countries starting from 1988. Our in (refer to Table 1 for the starting our sample prior to 1970. We use the earliest data available in our dataset data of each country) for the 7 countries for which data starts later than 1970. 19

20 outli ers. r 10 years since Bouman and Jacobsen (2002) published their study, the It is ove Halloween effect still remain significant making it an even more puzzling anomaly. around here Please insert Table 3 5 .2 Overall results deal with Using all historical data for all countries available seems the most logical way to lection bias and data mining issues. sample se A ll 56,679 monthly observations for all 109 countries over 319 years combined (reported in t he first ro w of Table 4 give a general ) impression of how strong the Halloween effect is. The a verage 6 - month w inter return compar ( 6.9 % , ) is ed to the summer return ( May through October) November through April of 2.4 % . This difference between w inter and summer returns is 4.5 % , highly significant with a t value of 11.42 . Despite - the possibility that the statistical significance might be overstated due to cross correlations between markets , t hese results do provide a n overall feeling f the strength of the Halloween effect . To co ntrol for these cross correlations we o he Halloween effect ese the world index returns in the second row . Th consider t using is 4.5 reveal average 6 - month winter return a similar result. % (t - value 3.64 ) higher The than the 6 - month summer return . Please around here insert Table 4 5 .3 Country by country analysis Many explanations suggest cross - country variations of the strength of the Halloween effect . This section conducts - nation Halloween effect analysis on the most comprehensive cross that the countries with stock market indices available . The evidence shows all 109 Halloween effect is preva lent around the world to the extent that the mean returns are 20

21 higher for the period of - April than for May - October in 82 out of 109 countries November t to only 2 the difference is statistically significant in 35 countries, compar ed and hat ly higher May countries having significant October returns . - 5 .3.1 Market development status, geographical location and the Halloween effect Figure 5 (A - D) plots the November - April and the May - October price return s for all 109 countries in four charts grouped by market classification , each chart is ordered by descending summer returns. An overall picture is that the Halloween effect is more pronounced in dev udied eloped and emerging markets than in the frontier and rarely st - A compare s markets. Figure 5 - month period returns for the 24 developed markets ; the two 6 with Finland being the only exception, 23 countries exhibit higher average November - April return s than May - October return s. T he differences are quite large for many countries primarily due to the low returns during May October, with 12 countries even hav ing - negative average retur ns for the period May - October. The char t for emerging mark ets ; (Figure 5 - B) shows a similar pattern 19 of the 21 countries have November - April return s that ed the May - October return s , and 7 countries have negative mean return s for May - exce October. As we move to the frontier and rarely studied markets, this pattern becomes less d s 5 - C and 5 - D reveal that 21 out of 30 (70 %) countrie s in the frontier istinctive. Figure %) countries in the rarely studied markets have November - markets and 19 out of 34 (56 s s greater than their May April return October return . - Please in sert Figure 5 around here Table 4 provides statistical support for the Halloween effect across countries. The table s reports average value s for the two 6 - month period returns , the and standard deviation coefficient estimate s and t - statistics for the Halloween regression E quation (1) , as well as May the the November - April return s beat the percentage of years that - October return s for each country. The countries are grouped based on market classifications and geographical region s. For the developed markets, a statistically significant Halloween effect is prevalent countries located in Asia and the among the European countries, but also not only among 21

22 North America. In fact, the strongest Halloween effect in our sample is , which has in Japan of 8.3% with a t a difference in returns . The Halloween effect is - statistic of 3.37 dle East and The statistically significant in 17 out of 24 (71%) developed markets. Mid none of the countries exhibit a Oceania are the only two continents where significant his difference in the two 6 - month returns cannot be justified by risk T Halloween effect. since we observe similar or even lower standard , measured with standard deviations in the November April returns deviations . The number of countries with a statistically - significant Halloween effect reduces as we move to less developed markets. Among 21 countries have November April returns significantly higher than emerging countries, 10 - - October returns. The Halloween effect is more prevalent in Asian and European their May in other regions. Brazil is the only country in North and South countries than America where we find a significant effect. lthough over 70% (21/30 ) of For the frontier markets, a - May during the countries show higher average returns during November - April than October, only countries have significant t - statistics. For the rarely studied markets, the 4 countries with a significant Hallo ween effect drops to 4 out of 34 . At this stage we are still anomaly, not able to identify the root of this seasonal nonetheless, over the total 109 countries, we only observe 2 countries (Bangladesh and Nepal from the frontier and rarely groups fect ; the studied markets ) to have a statistically significant negative Halloween ef , so far at least , suggests that the Halloween effect is a puzzling anomaly that overall picture s prevail around the world. Another interesting observation that might be noted from the table is that , among the countries with a significant Halloween effect, the difference between two 6 - mont h period returns is much larger for the countries in the emerging, frontier and rarely studied markets than for the countries in the developed markets. Th e average difference in 6 - returns among countries with significant Halloween effect in month oped markets is % in the emerging markets, 2 %, comparing to 1 3.5 the devel 0.6 % in the 5.7 s and 14 % in the rarely studied markets. However, w e need to be careful frontier market before making any judgement on the finding since the sample size tend s to be smaller in em tier and rarely studied markets . In addition, the observations in those newly erging, fron emerged markets tend to be more recent. If the overall strength of the Halloween effect is stronger in recent samples than in earlier samples, we may observe higher p oint estima tes ing conduct for the countries with shorter sample periods. We will address this issue by 22

23 cross sectional comparison within the - period analysis in same time interval using sub ection 5 .4 S . 5 the Halloween effect over time .4 The evolution of .4 .1 Pooled sub - sample period regression analysis 5 of We provide an overview how the Halloween effect has evolved over time using time series analysis by pooling all countries in our sample together. This gives us a long time We divide the entire sample into series data from 1693 to 2011. - one 10 - year sub - thirty 21 . and compare the two 6 - month period returns in Table 5 periods These sub - period estimates allow us to detect whether there is any trend over time in general . The second - column reports the number of countries in each sub period. There is only one country in the increasing to 6 countries by the end of 1900. sample during the entire eighteenth century, The number of countries expands rapidly in the late twentie century and reaches 108 in th the most recent subsample period. Columns 4 to 7 report the mean returns and standard deviations for the two 6 - month periods. The average 6 - month return over the entire sample duri - April is 6.9 % , compar ed ng November to only 2. 4 % for the period of May - October. 10 Figure 6 graphically plots the 6 - month return differences of 31 year sub - - periods ; 24 of the 31 10 - year sub - periods have November - April return s higher than their May - October s ence between the volatilities in the two 6 - month return . In addition, there is not much differ ; if anything, the standard deviation in November - April tends to be even lower than periods - entire sample is - month standard deviation over the in May October. For example, the 6 - % for November - 17.3 19.9 % for May October , indicating that the higher return is April and not due to higher risk , at least measured by the second moment. Columns 8 and 9 of Table 5 sh E quation (1) and the corresponding t - statistics ow the Halloween coefficients of - the est standard errors. Although corrected with Newey November - April returns are W the May - October returns, the t - statistics are not consistently frequently higher than significant until the 1960s. For the most recent 50 years, the Halloween effect is very persistent and econom - April return s are over 5% higher than the ically large. The November 21 To be precise, the first sub - period is 8 years from 1693 - 1710 and the last sub - period is about 11 years from July 2011. 2001 to 23

24 May - in all of the sub - periods, and this difference is strongly significant at October return s 22 April return We report the percentage of times that November - 1% level. s beat May - the return s in the last column. This non - parametric test provides consistent evidence October ; 24 of the 31 sub - periods have greater returns for the with the parametric regression test October for over 50% of of April than for May - - the years. November period i nsert Table 5 and Figure 6 around here Please The standard errors estimated from pooled OLS regressions may be biased due to cross - sectional correlations between countries. Thus, we also reveal the trend of the Halloween world index returns from 1919 to 2011. Figure 7 plots effect in the Global Financial Data’s the Halloween effects using 10 - - year and 50 - year rolling window regressions. The year, 30 and dark solid line shows the coefficient estimates of the effect, we also indicate the upper and lower 95% confi dence intervels for the estimates with lighter dotted lines. The plots reveal that the Halloween effect is quite prevelant over the previous century. For example, - always significantly positive. with a 50 year rolling window, the Halloween effect is almost - year rolling window, which is a considerably small sample size, the Even with a 10 around the World War II period. In coefficient estimates only appears negative in the 1940s addition, all of the plots exhibit an increasing trend of the Halloween effect st arting from around the 1950s and 1960s. The point estimates have become quite stable since the 1960s. Figure 7 around here Please insert 5 .4.2 Country by country subsample period analysis Understanding how persistent the Halloween effect is and when it emerged and beca me prevalent among countries is important since it may help to validate some explanations , 22 knowledge that there are many problems with this simple pooled OLS regression technique. Our We ac intention here is, however, only to provide the reader with a general indication on the trend of the Halloween andom effects model also gives a similar conclusions. effect over time. The panel data analysis using a r 24

25 while rul ing out others. To be specific, if the Halloween effect is related to some fundamental factors that do not change over time, one would expect a very persistent Halloween effect in the markets . If the Halloween effect is triggered by some fundamental of institutional factors in the economy, we would expect to observe the H alloween changes effect emerg ing around the same period . Alternatively, if the Halloween effect is simply a a fluke or a market mistake, we would expect arbitragers to take the riskless profit with way, a weakening Halloween effect following its discovery. Longer t ime series data is essential for the subsample period analysis. In this section, we divide countries with over 60 years’ data into several 10 - year subsample periods to test whether or not there is any persistence of the Hallowe en effect in the market. ite small sample size may reduce the power of Desp e choose 10 - the test, w for the purpose to reveal the trend of the Halloween year subsamples effect. Table 6 presents the sub - period results for 28 countries that meet the sample size criterion grouped according to market classification and regions. It consists of 20 countries , from the developed markets, 6 from the emerging markets and 2 from the rarely studied markets. Geographically, we have 14 countries in Europe, 2 countries in Oceania, 2 co untri es in Asia, 1 African country, 3 North American countries, and 5 countries from Sout h America . The table reports coefficient estimates and t - statistics of the Halloween effect regression f or the whole sample period and 11 - sub sample periods. The sub - p eriod analysis not only enables us to investigate the persistence of the effect for each individual country, it also allows a direct comparison of the size of the anomaly between countries but pheno menon that emerges within the same time frame. The Halloween effect seems to be a the 1960s and ha s become stronger over time , from among the European especially countries . T he coefficient esti mates become positive in 27 of the 28 countries, in which 4 are statistically significant during the 10 year period from 1961 to 1970. The number of countries with statistically significant Halloween effect keeps growing with time . Sub - 1991 - period shows the strongest Halloween effect especially for the Western 2000 European countries. Of 27 countries, 25 have lower average May - October returns than the rest of the year, in which 14 countries are statistically significant compris es of , this group all the Western European countries except Denmark. In addition, the sizes of the Halloween effects are much stronger in European cou ntries than in other areas. Although the most - weaker Halloween effect, the higher November a April recent 10 year period reveal s 25

26 return s are present in all the markets except Chile. For the five 10 - year sub - periods since stently 1960, the point estimates are persi in Japan, Canada, the United States, positive Australia, New Zealand, South Africa and almost all western European countries except Denmark, Finland and Portugal. Countries like Austria, Finland, Portugal and South Africa en effect over the whole sample also exhibit a significant that do not have a Hallowe Halloween effect in the recent sub - periods. The sizes of the Halloween effect in recent subsample periods are also considerably larger compar ed to the earlier sub - periods and whole sample periods. Since the data for most of the emerging/frontier/rarely studied markets that have a Halloween effect starts within the past 30 years , i f we focus our comparison to the most recent 30 year sub - periods, the difference in size of the Halloween effect between the developed markets and less developed markets noted in the previous section in T able 4 is reduced substantially: The average size of the coefficient estimates for the countries with significant Halloween effe ct in developed markets is 12.7 % for the for peri od of 2000 - 2011, 15 % The Halloween effect 1991 - 2000 and 16.5 % for 1981 - 1990. does not appear in Israel, S outh American area. India, and all the countries located in around here Please insert Table 6 6 . Economic significance - .1 Out - of 6 sample performance in 37 countries examined in Bouman and Jacobsen (2002) Bouman and Jacobsen (2002 ) develop a simple trading strategy based on the Halloween indicator and the Sell - in - May effect, which invests in a market portfolio at the end of October for six months and sells the portfolio at the beginning of May, using the proceeds to purchase risk free short term Treasury b ill s and hold these from the beginning of May to the end of October. They find that the Halloween strategy outperforms a buy and hold sample stra tegy even after taking transaction costs into account. We investigate the out - of - performance of this trading strategy in this section . 26

27 Please insert Table 7 around here approach is to see how investors might profit from the Halloween effect if Our they follow November 1998 to April 2011. able 7 shows the from the Halloween trading strategies T of out sample performance of the Halloween trading strategy relative to the Buy and Hold - - . We use 3 - strategy of the 37 countries originally tested in Bouman and Jacobsen (2002) s in the local currency of each country as the risk free rate. The month Treasury Bill Yield s ed average returns reported in the second and the fifth columns reveal that the annuali y Halloween strateg . The Halloween strategy frequently beats a buy and hold strategy bu s . The standard returns are higher than the y and hold strategy in 31 of the 37 market deviations of the Halloween strategy are always lower than the buy and hold strategy, this s the S harpe ratios of the Halloween strategy to be higher than the buy and hold lead The finding indicates that after the publication of strategy in all 37 markets except Chile. cobsen (2002), investors using a the Halloween strategy Bouman and J are still able to mak e higher risk adjusted returns using the buy and hold strategy . than 6 .2 Long term performance of the Halloween strategy in the UK data for UK stock market returns, w e are able to With the availability of long time series data 300 years. Investigating the long examine the performance of this Halloween strategy over term performance of the strategy in the UK market is especially interesting, since the United Kingdom is the origin of the market adage “Sell in May and go away”. Thi s has been referred as an old market saying as ear ly as 1935 , indicating that UK investors are to aware of the trading strategy over a long time period . Table 8 presents the performance of the Halloween strategy relative to the buy and hold strategy over different subsample periods. Please insert Table 8 around here 27

28 The average annual returns reported in the second and the fifth columns reveal that the ample period, Halloween strategy consistently beats a buy and hold strategy over the whole s in all 100 - year and 50 - year subsamples. It only underperforms the buy and hold and 30 - year subsamples (1941 - strategy in one out of ten of the 1970). The magnitude with which the Halloween strategy outperforms the market is also considerable. For example, the r eturns of the Halloween strategy are almost three times as large as the market returns over the whole sample. In addition, the risk of the Halloween strategy, as measured by the standard deviation of the annual returns is, in general, smaller than for the buy and hold strategy. This is evident in all of the sample periods we examine. Sharpe ratios for each strategy are shown in the fourth and seventh columns. Sharpe ratios for the Halloween strategy are unanimously higher than those for th e buy and hold str ategy. Table 8 also reveals the persistence of the outperformance of the Halloween strategy within each of the subsample periods by indicating the percentage of years that the Halloween strategy beats , the Halloween strategy the buy and hold strategy. Over the whole sample period outperforms the buy and hold strategy 63.09% (200/317) of the years. All of the year - 100 and - year subsample periods have a winning rate higher than 50%. Only one of the 30 - 50 year subsamples has a winning rate below 50% (1941 - 1970, 43.33%). Most investors will, however, have shorter investment horizons than the subsample periods used above. Using this large sample of observations allows us a realistic indication of the erm investment horizons. Table strategy over different short t 9 contains our results. It compares the descriptive statistics of both strategies over incremental investment horizons, ranging from one year to twenty years. Returns, standard deviations, and maximum and minimum values are annuali s ed to make the statistic s of different holding periods comparable. The upper panel shows the results calculated from overlapping samples and the lower panel contains the results for non - overlapping samples. around here. Please insert Table 9 28

29 The two sampling methods produce similar results. For every horizon, average returns are significantly higher for the Halloween strategy: Roughly three times as high as for the buy and hold strategy. For shorter horizons the standard deviation is lower for the Halloween strategy than for the buy and hold strategy. For longer investment horizons, however, the standard deviation is higher. This seems to be the result of positive skewness, indicating een strategy than for the buy that we observe more extreme positive returns for the Hallow Figure 8 confirm this. The graphs and hold strategy. The frequency distribution plots in reveal that the returns of the Halloween strategy produce less extreme negative values, and more extreme positive values, than the buy an d hold strategy. around here. Please insert figure 8 This is also confirmed if we consider the maximum and minimum returns of the strategies shown in Table 9 . Except for the one - year holding horizon, the maximum returns for the Halloween strategy of di fferent investment horizons are always higher than for the buy and hold strategy, whereas the minimum returns are always lower for the buy and hold strategy. The last column of Tabl e 9 presents the percentage of times that the Halloween strategy outperform s the buy and hold strategy. The results calculated from the overlapping sample - indicate that, for example, when investing in the Halloween strategy for any two year of beat horizon over the 317 years, an investor would have a 70.57% chance the market. ing The percentage of winnings computed from the non - overlapping sample, shown in the lower panel, yield similar results. Once we expand the holding period for the Halloween trading strategy, the possibility of beating the market increases dramatically. If an investor uses a Halloween strategy with an investment horizon of five years, the chances of beating the market rises to 82.11%. As the horizon expands to ten years this probability increases to a striking 91.56%. As a last indication of the persistency of the Halloween strategy in the UK market over time, in Figure 9 we compare the cumulative annual return over the three centuries. The ote that n buy and hold strategy hardly shows any increase in wealth until 1950 ( this is a 29

30 price index and the series do not include dividends). The cumulative wealth of the Halloween strategy increases gradually over time and at an even faster rate since 1950. around here Please insert figure 9 7 . Methodological issues .1 Sample S ize and the Halloween effect 7 From Table 4 , we observe that the Halloween effect is stronger in the developed markets than in the other markets. The sample size for the developed market tends, however, to be considerably larger than the sample size fo r the emerging, frontier, or rarely studied, markets. For example, the country with the smallest sample size among developed markets is Norway, which has 40 years data starting from 1970, while the sample starting date for many less developed countries is around the 1990s, or even after 2000. The difference in the strength of the Halloween effect between developed markets with large sized samples and other markets with small sized samples may not have any meaningful implication, as it may just be caused by noise. The importance of a large sample size to cope with noisy data is emphasized in Lakonishok and Smidt (1988), in that: Monthly data provides a good illustration of Black's (1986) point about the “ difficulty of testing hypotheses with noisy data. It is quite possible that some month is indeed unique, but even with 90 years of data the standard deviation of the mean monthly r eturn is very high (around 0.5 percent). Therefore, unless the unique month outperforms other months by more than 1 percent, it would not be identified as a special month.” We examine whether there is a possible linkage between the Halloween effect and th e sample size among countries. Figure 10 plots each country’s number of observations indicate 5% against its Halloween regression t - statistics. Two solid lines at 30

31 el. significance level, and two dotted lines at indicate a 10% significance lev The graph reveals that a small sample size seems to have some adverse effects on detecting a significant Halloween effect. In particular, a large proportion of countries with an insignificant Halloween effect is concentrated in the area of below 500 (a round 40 years) observations, with most of the negative coefficient estimates from those countries with less than 360 (30 years) observations. As the sample size increases, the proportion of countries with a significant Halloween effect increases as well. 10 around here Please insert Figure If we follow the advice of Lakonishok and Schmidt (1988) to the letter and only consider countries for which we have stock market data for more than ninety years, we find strong evidence of a Halloween effect. It is s ignificantly present in 13 out of these 17 countries and the world market index. Three countries (Australia, India and South Africa have positive coefficients that are not significant and only for Finland we find a negative but not significant Halloween ef fect.) The long time series of over 300 years UK monthly stock market index returns allow s us to addres s this issue in another way using rolling windows larger than 90 years. Figure 11 of the extend s the evidence in Zhang and Jacobsen (2012) and shows the Halloween effect The dark solid line indicates the UK market over 100 - year ro lling window regressions . estimates of the Halloween effect, and the light dotted lines show the 95% confidence - West standard errors. Th e Halloween effect seems to interval calculated based on Newey be persistently present in the UK market for a long time period . Point estimates for the and the size of the effect is quite stable in effect are always positive , the eighteenth and . nineteenth centuries the effect is not always E ven with this large sam ple size, however, statistically significant. The first half of the twentieth century shows a weakening Halloween effect . Consistent with the results of the world index in Figure 7 and the sub - sample period analysis in Table 5 and 6 , the Halloween effect keeps increasing in strength starting from the second half of the twentieth century. 31

32 Please insert figure 11 around here. 2 T ime varying volatility and outliers 7. the impact of volatility clustering and outliers in the monthly index return w e To verify also show the rolling window estimates controlling for a conditional heteroscedasticity using 12 ) and outliers using OLS robust regressi ons (Figure GARCH model (Figure 1 3 ). W e use a GARCH (1, 1) model , since this simple parsimonious representation generally captures volatility clustering well in monthly data with a window of years or more (Jacobsen & 50 The model is given by: Dannenburg, 2003). ) ( ) 2 ( For the robust regression, we use the M estimation introduced by Huber (1973), which is - considered appropriate when the dependent variable may contain outliers. Please insert figu re 12 and figure 1 3 around here The result s from the GARCH rolling window are consistent with the OLS regressions. The estimates of the Halloween effect are always positive over the three centuries, and the century strength of the effect reduces during the first half of the twentieth , while it increases in the second half of the century. from the robust regressions Although the result reveals a similar trend, the point estimates become negative during the 1940s and 1950s. 7 .3 M easuring the effect with a six month dummy et al. (2009) question the accuracy of the statistical inference drawn from standard Powell OLS estimation with Newey and West (1987) standard errors when the regressor is 32

33 persistent, or has a highly autocorrelated dummy variable and the dependent variable is positively autocorrelated. They suggest that this may affect the statistical significance of the Halloween effect. This argument has been echoed i n Ferson (2007). H owever, it is easy to show that this is not a concern here. We fi nd that statistical significance is not affected if - month summer and we examine the statistical significance of the Halloween effect using 6 - yearly Halloween dummy is negatively winter returns. By construction, this half et al. autocorrelated. Powell (2009 ) show that the confidence intervals actually narrow relative to conventional confidence intervals when the regressor’s autocorrelation is - This causes the standard t - statistics to under negative. reject, rather than over - reject, the null hypothesis of no e ffect. Thus, as a robustness check, it seems safe to test the statistics adjusted with Newey and West (1987) standard Halloween effect using standard t - - annual re errors from semi - the coefficient estimates and t turn data . Table 10 presents statistics. Please insert Table 10 around here. The results drawn from semi - annual data do not change our earlier conclusion based on monthly returns. If anything, these results show an even stronger Halloween effect. The periods with significant Halloween effects in our earlier tests remain statistically significant, - with t values based on semi - annual data. The first hundred years (1693 - 1800) period was not statistically significant using the monthly data, but now becomes significant at the 10% level. As a final test , we use a simple equality in means test. In this case, we also reject the hypothesis that summer and winter returns are different, with almost the same, highly value (4.20). significant, t - 33

34 8 . Conclusion tes the Halloween effect countries market price returns and 66 This study investiga for 109 over all the period s for which data is available. market total returns and risk premium Based on 33,348 monthly returns, we find an overall historical market risk premium of 3.7%, however, this premium is sol ely contributed from the returns generated from - April, overall, summer returns (May November October) is significantly lower than the - risk free rate by 1.17%, 46 out of 66 market show negative average risk premium during summer time. This finding does not only challenge the notion of market efficiency but also defies traditional economic theory in an even more fundamental way. The Halloween effect is prevailing around the world to the extent that mean returns price of November are higher for the period April than for May - October in 81 out of 108 - countries, and the difference is statistically significant in 35 countries compar ed to only 2 October returns. countries having significant higher May - ly The results are even stronger if we consider total return s and risk premiums: 58 out of 66 (56 out of 66) countries show positive point estimates on the Halloween effect in the total return (risk premium) series, in which the effect is statistically significant in 34 (32) countries. that Our evidence reveals the - nation. It is stronger in developed and the size of Halloween effect does vary cross emerging markets than in frontier and rarely studied markets. Geographically, the Halloween effect is more prevalent in countries located in Europe, North America and Asia than in other areas. that the strongest Halloween effect Subsample period analysis shows among countries are observed in the past 50 years since 1960 and concentrated in developed Western European countries. The Halloween effect is still present out - of - sample in the 37 countries used in Bouman and the Jacobsen (2002) . The out - of - sample risk adjusted payoff from Halloween trading buy and hold strategy strategy still higher than for the is in 36 of the 37 countries . When considering trading strateg ies assuming different investment horizons, the UK evidence reveals that investors with a long horizon would have remarkable odds of beating the ; with market the , for example, an investment horizon of 5 years, the chances that 34

35 buy and hold strategy is 80%, with the probability of the Halloween strategy outperforms beating the market increas ing to 90% if we expand the investment horizon to 10 years. maly that Overall, our evidence suggest s that the Halloween effect is a strong market ano has strengthened rather th an weakened in the recent year the s . P lausible explanations o f Halloween effect should be able to allow for time variation in the effect and explain why the effect has strengthened in the last 50 years. 35

36 Reference s Andrade, S. C., Chhaochharia, V., & Fuerst, M. (2012). "Sell in May and Go Away" Just Working Paper Won't Go Away. , University of Miami. Jacobsen, B. (2002). The Halloween indicator,"Sell in May and go away": Bouman, S., & , 1618 - 1635. Another puzzle. American Economic Review, 92 . Y. ( 2000 ). Asset pricing at the millennium . Journal of Finance , 55, 1515 - 1567. Campbell, J Wei, J. (2005). Stock market returns: A note on temperature anomaly. Journal Cao, M., & 1573. , 1559 - of Banking & Finance, 29 David, M. R., & Pontiff, J. (2004). Does academic research destroy stock return predictability. .com/abstract=2156623 . working Paper, Available at SSRN: http://ssrn Journal of Portfolio Dimson, E., & Marsh, P. (1999). Murphy's law and market anomalies. (2), 53 - 69. Management, 25 Working Dimson, E., Marsh, P., & Staunton, M. (2011). Equity Premia Around the World. Paper . Ferson, W. E. (2007). Mark et efficiency and forecasting : UK: Elsevier . ISBN# 978 - 0 - 7506 - 8321 0. - Gerlach, J. R. (2007). Macroeconomic news and stock market calendar and wealther Journal of Financial Research, XXX (2), 283 - 300. anomalies. Vliet, P. V. (2010). An Anatomy of Calendar Grimbacher, S. B., Swinkels, L. A. P., & Effects. Robeco Investment rasmus Univeristy Rotterdam, Working Paper , E Soluation & Research, Asset management. Robeco Gugten, T. v. d. (2010). Stock market calendar anomalies and macroeconomic news announcement Working Paper , Erasmus University Rotterdam. s. Haggard, K. S., & Witte, H. D. (2010). The Halloween effect: Trick or treat? International Review of Financial Analysis, 19 (5), 379 - 387. Heston, S. L. and R. Sadka (2008). Seasonality in the cross - section of stock returns. Journal of Financial Economics 87(2), 418 – 445. Heston, S. L. and R. Sadka (2010). Seasonality in the cross - section of stock returns: The interna tional evidence. Journal of Financial and Quantitative Analysis 45(5), 1133 – 1160. Hong, H., & Yu , J. (2009). Gone fishin': Seasonality in trading activity and asset prices. Journal of Financial Markets, 12 , 672 702. - Journal of Banking & Finance, Jacobsen, B., & Marquering, W. (2008). Is it the weather? , 526 - 540. 32 Jacobsen, B., & Marquering, W. (2009). Is it the weather? Response. Journal of Banking & Finance, 33 , 583 - 587. Jacobsen, B., & Visaltanachoti, N. (2009). The Halloween Effect in US Sectors. Financial Review, 44 (3), 437 - 459. Jacobsen, B., Mamun, A., & Visaltanachoti, N. ( 2005). Seasonal, Size and Value Anomalies. Working Paper , Massey Univeristy, University of Saskatchewan. Kamstra, M. J., Kramer, L. A., & Levi, M. (2003). Winter Blues: A SAD Stock Market Cycle. American Economic Review, 93 (1), 324 - 343. Working Paper, apuska, E. (2012). Does mood affect trading behavior? Kaustia, M., & Rant Aalto University School of Economics. 36

37 Keef, S. P., & Khaled, M. S. (2011). A review of the seasonal affective disorder hypothesis. Journal of Socio , 959 - 967. - Economics, 40 Meschke, F. (2010). Sentiment and stock returns: T he SAD anomaly Kelly, P. J., & Journal of Banking & Finance, 34 - 1326. revisited. , 1308 M., and Linnainmaa, J. T. & Nyberg, P. M., Common Factors in Stock Ma rket Keloharju, 2013). Fama Miller Working Paper; Chicago Booth Research Paper - Seasonalities ( No. 13 - 15. Available at SSRN: http://ssrn.com/abstract=2224246 - year perspective. Lakonishok, J., & Smidt, S. (1988). Are seasonal anomalies real? A ninety (4), 403 - Review of Fianncial Studies, 1 425. Lean, H. H. (2011). The Halloween puzzle in selected Asian stock markets. Int. Journal of Economics and Management 5 (1), 216 - 225. Lucey, B M, & Zhao, S. ( 2007 ). Halloween or J anuary? Yet another puzzle . Intern ational . Review of Financial Analysis , 17 , 1055 - 1069. Maberly, E. D., & Pierce, R. M. (2003). The Halloween Effect and Japanese Equity Prices: Myth or Exploitable Anomaly. Asia - Pacific Financial Markets, 10 , 319 - 334. Stock market efficiency withstands another Maberly, E . D., & Pierce, R. P. ( 2004 ). alloween" puzzle Econ Journal . c h allenge: Solving the "sell in M ay/ buy after H Watch , 1, 29 - 46. Moller, N., & Zilca, S. (2008). The evolution of the January effect. Journal of Banking & Finance, 32 , 447 - 457. Newey, W. K., & West, K. D. (1987). A simple, positive semi - definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica, 55 , 7003 - 7708. Whaley, Powell, J G., Shi, J., Smith, T., & . R. E. ( 2009 ). Political regimes, business c ycles, - seasonalities, and returns . Journal of Banking & Finance , 33, 1112 1128. Schwert, W. G. (200 2 ). Anomalies and market efficiency. In G. M. Constantinides, M. and , Han d book of E conomic s Harris, & R. M. Stulz (Eds.) F inance. Amsterdam, Netherlands: North - Holland. Sullivan, R., Timmermann, A., & White, H. (2001). Dangers of data mining: The case of calendar effects in stock returns. , 249 - 286. Journal of Econometrics, 105 Wachtel, S. B. (1942). Certain Observations on Seasonal Movements in Stock Prices . Journal of Business, 15 (2), 184 - 193. Zarour, B. A. (2007). The Halloween effect anomaly: Evidence from some Arab countries equity markets. Studies in Business and Economics, 13 (1), 68 - 76. Zhang, C. Y., & Jacobsen, B. (2013). Are monthly seasonals real? A three century 1785. - perspective. Review of Finance, 17 (5), 1743 Zhang, C.Y. (2014) Vacation behaviours and seasonal patterns of stock market returns, Working Paper FMA 2014. 37

38 Tab le 1. Summary statistics for market price returns, total returns and risk premiums number of observations, as well as some basic descriptive statistics s starting date, ending date and market price indices, 65 market total return indices, and the , for 109 The table present world index . The statistics for pooled price returns are calculated based on 109 stock market price indices, while for pooled total return s and risk premiums are calculated based on 65 stock market total return indices. Risk premium is the difference between monthly total market returns and risk free rates. Mean and standard deviation expressed as percentage are ication and geographical annualised by multiplying by 12 and MSCI market classif , t - value shows if the mean is significantly different from zero. Countries are grouped based on the √ regions. *** denotes significance at 1% level; **denotes significance at 5% level; *den otes significance at 10% level. Price Return Risk Premium Country Region S tatus Total Return Obs S t Dev S tart End Obs Mean S t Dev Mean S t Dev t-value t-value t-value S tart End Mean - 13.23 07-2011 1027 8.38 5.29 *** 14.67 - *** - World 3.03 4.17 1110 07-2011 02-1919 01-1926 26.39 Pooled 109 countries 02-1693 07-2011 56679 9.24 24.05 - - - *** - - - - - - 7.65 24.99 3.72 09-1694 *** 25.10 Pooled 65 countries 07-2011 33348 10.75 22.69 *** - - - - - - * 32.42 07-2011 499 15.79 3.01 *** 33.81 9.85 1.87 01-1970 33.95 ** 11.52 564 07-2011 08-1964 Hong Kong Asia Develop ed 2.44 01-1921 07-2011 1077 10.91 5.10 *** 20.27 5.82 2.86 *** 18.91 Jap an 08-1914 07-2011 1154 6.30 2.84 *** 21.77 08-1973 07-2011 6.69 1.56 26.52 2.39 0.55 26.57 ** 2.05 7.04 552 07-2011 08-1965 Singap ore 23.32 456 19.71 27.52 499 7.43 2.44 ** 19.67 1.43 0.47 07-2011 *** 3.02 9.04 1018 07-2011 02-1922 Austria Europ e 01-1970 01-1951 07-2011 727 8.85 4.54 *** 15.19 2.81 1.44 15.21 Belgium 02-1897 07-2011 1302 3.91 2.27 ** 17.90 01-1970 1.37 499 11.51 4.28 *** 17.35 3.72 12.87 16.66 Denmark 01-1921 07-2011 1086 4.31 3.18 *** 07-2011 20.89 11-1912 13.14 6.25 *** 20.84 6.36 3.02 *** 1179 20.51 07-2011 Finland 11-1912 07-2011 1179 8.30 4.01 *** 01-1898 07-2011 1348 10.08 5.61 *** 19.02 4.92 2.66 France 19.19 01-1898 07-2011 1348 6.67 3.76 *** 18.82 *** 01-1870 1692 5.71 1.88 * 36.07 0.60 0.20 07-2011 36.10 25.03 1.21 2.55 1692 07-2011 01-1870 Germany 26.33 01-1977 07-2011 415 14.12 2.57 ** 32.37 2.47 0.45 32.29 Greece 01-1954 07-2011 690 9.51 2.74 *** 16.29 *** 463 10.67 2.84 *** 23.35 2.81 0.75 23.31 3.06 5.67 930 07-2011 02-1934 Ireland 01-1973 07-2011 01-1925 07-2011 10.30 3.78 *** 25.34 3.86 1.42 25.36 ** 2.33 5.44 1264 07-2011 10-1905 Italy 23.95 1038 17.11 16.97 727 10.31 4.71 *** 17.05 6.00 2.73 *** 07-2011 ** 2.05 3.65 1086 07-2011 02-1919 Netherlands Europ e 01-1951 02-1980 07-2011 378 11.71 2.54 ** 25.86 3.59 0.78 25.95 Norway 01-1970 07-2011 499 10.81 2.86 *** 24.37 02-1988 30.93 282 3.98 0.98 19.68 -2.18 -0.53 19.81 Portugal 01-1934 07-2011 897 6.09 1.70 * 07-2011 04-1940 ** 856 11.35 5.32 *** 18.01 4.70 2.20 17.31 18.08 Sp ain 01-1915 07-2011 1116 5.35 2.98 *** 07-2011 ** 16.86 1111 9.65 5.42 *** 17.13 4.39 2.46 07-2011 17.15 *** 3.35 5.50 1265 07-2011 01-1906 Sweden 01-1919 15.24 02-1966 07-2011 546 6.92 2.85 *** 16.41 3.76 1.54 16.45 Switzerland 01-1914 07-2011 1155 3.19 2.05 ** * 1.86 07-2011 3798 6.52 9.20 *** 12.61 2.10 2.96 *** 12.61 1.44 3817 07-2011 02/1693 United 13.86 09-1694 Kingdom *** 01-1993 05-2011 221 9.90 1.89 * 22.49 1.93 0.37 8.08 22.52 M id East Israel 02-1949 05-2011 748 23.66 23.12 15.14 16.12 929 9.41 5.50 *** 15.05 4.85 2.82 *** 07-2011 *** 3.02 5.03 1124 07-2011 12-1917 Canada North 03-1934 America 07-2011 2538 8.10 7.77 *** 15.17 4.07 3.90 *** 15.21 United States 09/1791 07-2011 2639 2.81 2.77 *** 15.06 02-1800 07-2011 07-1928 10.87 6.33 *** 15.65 5.66 3.29 *** 15.69 997 13.51 *** 4.31 4.99 1638 07-2011 02/1875 Australia Oceania 07-2011 301 5.00 1.34 18.67 -3.01 -0.80 18.91 New Zealand 01-1931 07-2011 967 4.33 2.73 *** 14.22 07-1986 38

39 Table 1. (continued) Price Return S tatus Region Country Total Return Risk Premium Obs End S tart End Obs Mean S t Dev Mean S t Dev S tart t-value t-value Mean t-value S t Dev -7.37 07-2011 10-1996 07-2011 177 13.91 1.71 * 31.22 4.93 0.61 31.25 01-1993 Egy p t Africa Emerging -0.28 222 112.88 2.33 14.93 07-2011 208 13.78 3.65 *** 15.71 8.77 04-1994 ** 15.69 4.36 13.49 279 07-2011 01-1988 M orocco *** *** 16.76 02-1960 07-2011 618 15.19 5.03 *** 21.68 South Africa 1.99 ** 21.75 02-1910 07-2011 1218 7.67 4.61 6.05 14.83 48.14 01-1993 07-2011 223 0.01 0.00 36.12 -4.86 -0.58 36.23 Asia China 01-1991 07-2011 247 1.40 4.09 *** 01-1988 07-2011 283 17.48 2.70 *** 31.47 19.26 0.62 28.51 5.88 1080 07-2011 08-1920 India 2.89 2.25 ** 31.02 01-1988 07-2011 283 19.50 2.40 ** 39.45 Indonesia 0.66 39.77 04-1983 07-2011 340 13.13 5.40 13.47 ** 39.03 02-1962 07-2011 592 21.53 3.83 *** 39.52 9.23 1.64 39.53 Korea 02-1962 07-2011 592 2.42 1.64 451 07-2011 451 9.69 2.10 ** 28.24 5.40 1.17 28.30 07-2011 01-1974 M alay sia 27.19 7.29 01-1974 0.49 0.76 07-2011 355 15.23 2.80 *** 29.57 2.65 01-1982 29.65 2.87 703 07-2011 01-1953 Philip p ines 28.93 2.04 ** 33.21 01-1988 07-2011 283 8.02 1.11 34.95 Taiwan 0.58 34.99 02-1967 07-2011 534 10.16 4.18 6.70 29.14 05-1975 07-2011 435 11.60 2.27 ** 30.84 6.58 1.15 30.66 Thailand 05-1975 07-2011 435 1.38 27.48 07-2011 0.99 30.06 12-1993 07-2011 212 10.46 1.60 7.07 5.30 0.81 27.61 10-1993 Europ e Czech 214 Rep ublic 16.01 2.10 ** 30.99 01-1995 07-2011 199 16.21 2.13 ** 30.92 3.72 0.49 30.80 Hungary 01-1995 07-2011 199 5.28 0.66 07-2011 207 8.43 1.09 32.12 -3.60 -0.46 32.39 Poland 05-1994 07-2011 33.44 207 05-1994 *** -0.63 01-1995 06-2011 198 16.05 1.13 57.44 -8.42 3.42 51.58 Russia 10-1993 07-2011 213 41.72 51.37 -5.58 4.07 02-1986 07-2011 306 47.51 4.48 *** 53.53 53.65 -0.53 53.49 43.29 306 07-2011 02-1986 Turkey *** 978 16.21 5.70 *** 25.66 01-1988 07-2011 283 26.33 4.64 *** North 6.82 1.23 26.88 27.54 M exico 02-1930 07-2011 America 67.65 *** 56.46 - - - - - - - - - South Brazil 01-1990 07-2011 258 5.56 America *** 01-1983 07-2011 546 18.94 6.05 *** 21.13 9.28 8.52 *** 20.82 3.01 Chile 01-1927 07-2011 1015 27.36 29.53 -0.20 19.94 07-2011 283 16.92 1.44 57.05 -2.33 01-1988 57.50 4.49 9.74 1014 07-2011 02-1927 Colombia *** *** 39.15 01-1993 07-2011 223 20.66 2.73 *** 32.62 10.61 1.39 32.84 Peru 01-1933 07-2011 943 31.15 7.05 19.29 266 14.70 - - - - - - - - - 07-2011 06-1989 Botswana Africa Frontier 6.18 *** 18.49 - - - - - - ** - - Ghana 01-1996 07-2011 187 11.62 2.48 - - - - - - - - - - 1.38 7.11 258 07-2011 02-1990 Keny a 23.94 *** 16.42 08-1989 07-2011 264 18.09 5.20 *** 16.33 9.34 2.67 *** 16.45 M auritius 08-1989 07-2011 264 13.16 3.76 - *** - - - - - - - - 21.61 4.62 20.69 280 07-2011 01-1988 Nigeria 16.62 - - - - - - - - - Tunisia 01-1996 07-2011 187 3.44 0.82 39

40 Table 1. (continued) S tatus Country Price Return Risk Premium Total Return Region Obs Mean S t Dev S tart End Obs Mean S t Dev Mean S t Dev t-value t-value t-value S tart End 258 02-1990 33.37 - - - - - - - - - Bangladesh 11.39 Asia 07-2011 Frontier 1.58 - ** - - - - - - - - 2.13 Kazakhstan 08-2000 07-2011 132 24.53 38.13 *** 23.34 01-1988 07-2011 280 15.08 1.97 ** 36.98 5.26 0.68 37.14 Pakistan 08-1960 07-2011 608 9.61 2.93 *** 3.18 06-1987 07-2011 290 16.91 2.94 *** 28.25 4.06 0.70 28.41 Sri Lanka 01-1985 07-2011 319 25.81 15.90 - 41.63 - - - - - - - - 0.52 127 07-2011 01-2001 Viet Nam 6.66 81 -8.45 Europ e 32.26 - - - - - - - - - Bosnia And -0.68 11-2004 07-2011 Herzegowina 12.34 35.83 11-2000 07-2011 129 22.03 1.97 ** 36.71 18.25 1.63 36.82 11-2000 Bulgaria 07-2011 129 1.13 0.58 174 - - - - - - - - 07-2011 02-1997 Croatia 32.44 4.91 - 9.98 1.36 07-2011 181 13.10 1.36 37.47 07-1996 1.03 37.55 13.10 181 07-2011 07-1996 Estonia 37.48 0.64 28.57 01-1996 07-2011 187 9.34 1.10 Lithuania 3.01 0.35 33.84 01-1996 07-2011 187 4.65 33.62 12.44 38.79 10-1997 07-2011 166 13.94 1.19 43.44 -18.11 -1.54 43.82 Romania 10-1997 07-2011 166 1.19 - -0.54 - - - - - - 60.86 - - 36 07-2011 08-2008 Serbia -18.94 6.66 1.04 25.32 01-1999 07-2011 151 4.73 0.90 18.57 -1.88 -0.36 18.33 Slovenia 01-1996 07-2011 187 19.19 1.59 - - - - - - - - - 07-2011 02-1998 Ukraine 162 44.43 15.55 3.48 13.57 01-2004 07-2011 91 2.98 0.53 253 0.35 0.06 15.52 07-2011 07-1990 Bahrain M id East 1.18 6.46 1.64 22.76 07-2006 07-2011 61 0.97 0.10 21.90 -3.64 -0.37 21.89 Jordan 02-1978 07-2011 402 10.96 2.29 19.53 01-2004 07-2011 91 7.09 0.83 23.43 4.37 0.51 23.45 07-2011 01-1995 Kuwait 199 ** 0.34 - - - - - - - 2.45 - - Lebanon 02-1996 07-2011 186 28.23 22.27 1.79 20.56 10-2005 07-2011 70 8.11 0.88 8.54 6.02 0.65 22.27 224 07-2011 12-1992 Oman * 15.41 1.76 * 30.03 01-2004 07-2011 91 14.34 1.05 37.59 11.15 0.82 37.65 Qatar 10-1999 07-2011 142 236 26.74 2.87 *** 19.65 01-2004 09-2008 56 30.12 1.82 * 35.83 09-2008 1.61 35.92 United Arab 01-1988 12.73 Emirates - *** 25.60 - - - - - - 4.08 - - 07-1969 01-2011 North Jamaica 499 16.21 America 3.47 *** 14.40 - - - - - - Trinidad And - - - 01-1996 07-2011 187 12.67 Tobago South 535 *** 62.03 08-1993 07-2011 216 13.32 1.80 * 31.47 3.82 0.50 32.18 63.70 6.86 Argentina 01-1967 07-2011 America - 17.38 - - - - - - 0.65 - - 07-2011 07-1997 2.99 Rarely Cote D`Ivoire Africa 169 Studied 38.02 * - - - - - - - 1.83 22.63 114 01-2011 04-2001 M alawi - - - - - - - - - - 24.88 Namibia 03-1993 07-2011 218 11.59 1.99 ** - 15.18 - - - - - - - - 0.43 2.39 88 04-2007 01-2000 Swaziland - - - - - - - - - - Tanzania 12-2006 07-2011 56 5.11 1.44 7.66 - - - - - - - - - 148.36 0.04 3.14 54 07-2011 02-2007 Uganda 25.27 - - - - - - - - - Zambia 02-1997 07-2011 174 25.52 3.85 *** 40

41 Table 1. (continued) S tatus Total Return Region Country Price Return Risk Premium S t Dev S tart End t-value Mean S t Dev Mean S t Dev t-value S tart End Obs Mean t-value Obs 0.53 42.52 - - - - - - - - - Asia Ky rgy zstan 01-2000 Rarely 137 6.68 05-2011 Studied 48.16 - - - - - - - - 2.42 29.33 189 05-2011 09-1995 M ongolia ** - Nep al - - - - - - - - 01-1996 07-2011 186 3.56 0.61 23.03 - 01-1993 07-2011 223 6.14 0.66 39.90 1.75 0.19 39.91 Europ e Cy p rus 01-1984 07-2011 331 2.98 0.46 34.04 - - - - - - - - Georgia 11-2008 07-2011 33 32.74 0.79 68.50 - 50.65 07-2002 -6.19 -0.37 50.17 -15.60 -0.93 109 36.53 0.29 2.47 223 07-2011 01-1993 Iceland 07-2011 05-1996 07-2011 183 10.74 1.18 35.57 5.87 0.65 35.51 Latvia 02-1996 07-2011 186 9.89 1.11 35.18 16.79 01-1985 07-2011 319 10.10 2.59 *** 20.13 4.78 1.22 20.16 Luxembourg 01-1954 07-2011 691 8.17 3.69 *** - 1.03 - - - - - - 12.50 117 07-2011 11-2001 M acedonia - 37.87 - 17.32 02-2000 1.40 0.28 17.24 -1.92 -0.38 138 18.89 1.57 7.51 187 07-2011 01-1996 M alta 07-2011 44.42 - - - - - - - - M ontenegro 04-2003 07-2011 100 29.25 1.90 * - 0.59 - - - - - - - - - Slovak 10-1993 07-2011 214 4.54 32.33 Rep ublic - 18.77 - - - - - - 25.90 - - 255 06-2011 04-1990 Iran 6.36 M id East *** 59.11 - - - - - - - - - Iraq 11-2004 07-2011 79 10.88 0.47 40.51 1.05 - - - - - - - 11.48 166 07-2011 08-1997 Palestine - - - - - - - - - - 1.21 6.59 222 07-2011 01-1993 Saudi Arabia 23.43 - - 28.18 - - - - - - - 0.12 2.70 19 07-2011 01-2010 Sy rian Arab - Rep ublic 7.87 - - - - - - - - - Bahamas 12-2002 07-2011 North 98 5.67 2.06 ** America - - - - - - - Barbados 04-1989 02-2011 263 - 4.24 1.42 13.99 - - - - - - - - - - 20.48 0.33 1.78 170 10-2010 09-1996 Bermuda Costa Rica - - - - - - - 10-1997 02-2011 161 13.90 2.37 ** 21.48 - - - - - - - - - - - El Salvador 01-2004 07-2011 91 7.41 2.53 ** 8.07 - 11.18 - - - - - - *** 5.43 14.08 223 07-2011 01-1993 Panama - - 23.17 - - - - - - - - - 0.32 02-1994 South 1.80 210 07-2011 Ecuador America - - - - - - 11-1993 09-2008 176 11.15 4.06 *** 10.52 - - Paraguay - - - - - - - - Uruguay 02-1925 12-1995 848 13.10 2.65 *** 41.57 - - *** 84 0.70 38.19 -11.72 -0.80 38.83 12-2003 12-1996 23.59 10.16 4.94 13.51 891 07-2011 01-1937 Venezuela 41

42 Table 2 Halloween effect in market total returns and risk premiums This table provides two 6 - month period (November - April a nd May - October) mean returns, standard deviations ( at percentage ) , and t - values of the zero mean test of the two periods, as wel l as the coefficient estimates and t - statistics for the Halloween effect regression represents for 65 markets and world index’s total returns and risk premiums. the 6 - month mean returns difference between November - April and May - Octo ber. T - values are adjusted using Newey - West standard errors. The 6 - month mean returns (standard deviations) are calculated by multiplying monthly returns (standard deviations) by 6 ( ). √ *** denotes significance at 1% level; **denotes significance at 5% level; *denotes significance at 10% level. Countries are g rouped based on the MSCI market classification and geographical regions. Total Return Risk Premium Nov-Apr S tatus Region Country May-Oct Halloween Nov-Apr May-Oct Halloween Mean Mean S t Dev Mean S t Dev Mean t-value S t Dev t-value S t Dev t-value t-value t-value t-value β β Hal Hal - - - - - - - - World 6.59 9.61 6.35 *** 1.78 11.00 1.50 4.81 3.45 *** 4.89 17.39 *** -1.17 18.01 -3.36 *** 6.06 12.28 *** 12.76 6.18 *** 6.71 17.95 2.29 *** 25.82 17.29 8.46 Pooled 65 countries *** 14.52 3.49 23.81 3.21 24.20 0.86 3.41 0.64 6.62 1.80 0.66 * 1.64 24.08 6.15 *** 2.62 23.76 9.63 Hong Kong Asia Develop ed Jap an *** -0.74 12.39 -0.56 7.29 3.75 *** 9.74 15.29 6.03 *** 1.18 13.10 0.86 8.56 3.61 *** 6.55 14.13 4.31 17.07 2.08 ** -3.37 20.26 -1.03 9.13 1.96 * Singap ore 7.89 17.06 2.85 *** -1.20 20.17 -0.37 9.09 1.96 * 5.76 Europ e 5.87 *** -4.46 14.89 -1.93 * 10.33 3.37 *** *** 3.00 Austria 8.86 12.58 4.55 *** -1.45 14.85 -0.63 10.31 3.37 12.61 10.09 4.14 *** -2.57 11.15 -1.80 * 7.95 4.17 *** Belgium 8.40 10.13 6.47 *** 0.44 11.10 0.31 7.97 4.18 *** 5.38 11.68 11.29 0.52 12.25 0.26 2.68 1.17 3.20 Denmark 7.45 * 4.12 *** 4.05 12.82 2.04 ** 3.39 1.51 1.74 6.39 14.89 4.25 *** -0.03 14.53 -0.02 6.42 2.96 *** Finland 9.81 14.89 6.53 *** 3.33 14.48 2.28 ** 6.47 2.98 *** France 13.72 -1.12 13.25 -0.87 7.14 3.39 *** 6.03 4.56 *** 8.50 13.66 6.60 *** 1.57 13.09 1.27 6.94 3.42 *** 21.46 2.74 *** -4.34 28.88 -1.79 * 9.30 3.03 Germany 7.51 21.50 4.14 *** -1.78 28.85 -0.73 9.29 3.03 *** 4.96 *** 22.27 -2.60 23.32 -0.66 7.66 1.30 5.06 1.27 1.34 0.83 23.34 3.30 *** 2.84 22.38 10.80 Greece 7.49 8.23 15.33 3.34 *** -5.45 17.15 -1.97 ** 13.68 3.95 *** Ireland 12.28 15.39 4.96 *** -1.64 17.13 -0.59 13.92 4.01 *** 5.36 ** *** -1.49 18.20 -0.76 6.85 2.50 ** 2.33 6.60 0.94 18.13 1.84 *** 4.46 17.61 8.44 Italy 17.56 2.84 11.05 *** -1.45 12.84 -0.88 8.88 4.23 *** 4.24 8.85 0.44 12.79 0.73 *** 6.78 11.01 5.24 9.58 Netherlands 7.43 *** * 1.79 19.16 -0.58 7.56 1.68 * 17.44 * 5.57 -1.99 1.68 7.56 0.61 19.10 2.08 *** 3.12 17.35 9.64 Norway Portugal -5.24 14.60 -1.74 * 8.30 1.90 * 6.13 13.07 2.27 ** -2.15 14.56 -0.72 8.28 1.90 * 3.06 13.23 1.12 12.69 3.94 *** -1.19 12.76 -0.79 7.12 3.58 *** Sp ain 9.19 12.64 6.13 *** 2.18 12.69 1.45 7.01 3.54 *** 5.92 12.30 12.32 *** -1.01 11.83 -0.82 6.40 3.64 *** Sweden 8.01 6.26 *** 1.64 11.78 1.34 6.37 3.65 *** 5.39 4.22 *** 10.39 -1.15 12.69 -0.61 6.07 2.67 *** 4.92 3.19 *** 2.61 5.92 0.27 12.65 0.50 *** 4.19 10.34 6.42 Switzerland United Kingdom 8.79 4.68 *** -0.21 9.04 -0.42 2.52 3.65 *** 4.54 8.81 9.17 *** 1.98 8.99 3.92 *** 2.56 3.71 *** 2.31 ** 5.11 16.07 1.38 -3.30 15.68 -0.90 8.41 2.10 ** M id East Israel 9.06 16.04 2.44 ** 0.73 15.65 0.20 8.33 2.08 *** 3.59 4.82 *** -0.51 11.44 -0.39 5.88 3.58 *** 5.87 1.38 11.35 1.78 *** 6.89 9.75 7.64 Canada North America 5.37 9.80 United States 2.89 *** 1.18 11.41 1.51 1.70 1.65 * * 10.04 4.18 4.91 10.03 7.11 *** 3.20 11.38 4.09 *** 1.71 1.65 3.70 9.72 3.47 *** 1.97 12.32 1.45 1.73 Oceania Australia 6.31 9.72 5.91 *** 4.56 12.26 3.39 *** 1.74 1.04 1.03 -0.87 14.60 -0.32 -1.18 -0.33 12.10 -2.10 -0.33 -0.92 1.06 14.52 3.07 0.82 11.77 1.93 New Zealand -1.14 2.07 ** 8.96 23.64 1.46 -4.11 20.21 -0.78 13.07 2.09 ** Emerging Africa Egy p t 13.38 23.64 2.18 ** 0.46 20.16 0.09 12.92 *** 10.46 *** -1.57 9.85 -0.67 12.04 3.21 *** 11.78 3.68 3.20 12.00 0.40 9.92 0.95 *** 4.57 11.74 12.95 M orocco * 6.20 14.48 3.07 *** -0.15 16.17 -0.07 6.35 1.72 * South Africa 10.77 14.46 5.34 *** 4.42 16.07 1.98 ** 6.34 1.73 42

43 Risk Premium Total Return Region Country May-Oct S tatus Nov-Apr May-Oct Halloween Nov-Apr Halloween t-value t-value Mean S t Dev Mean S t Dev t-value t-value Mean S t Dev Mean S t Dev t-value t-value β β Hal Hal 1.63 25.21 -0.69 -0.81 26.11 -0.13 -3.23 -0.39 -1.60 China Asia Emerging 25.12 -3.24 -0.39 -4.04 26.05 -0.28 0.27 3.09 1.78 0.99 19.79 0.22 2.10 0.32 * 1.74 21.83 7.85 ** 2.06 22.74 9.63 India 20.58 0.27 0.65 ** 25.67 -5.71 30.10 -0.92 16.76 2.06 ** 11.05 2.09 ** 2.07 16.78 0.22 29.52 1.33 *** 3.41 25.84 18.11 Indonesia Korea 2.63 *** -1.63 26.58 -0.43 12.53 1.91 * 17.08 29.07 4.12 *** 4.49 26.58 1.19 12.59 1.92 * 10.90 29.12 7.42 18.74 2.43 ** -2.04 21.04 -0.59 9.46 1.99 M alay sia 9.55 18.75 3.13 *** 0.12 20.99 0.04 9.43 1.98 ** ** 3.13 0.93 -0.47 23.39 -0.11 3.60 0.71 Philip p ines 9.70 18.30 2.89 *** 5.51 23.26 1.29 4.19 0.83 18.22 12.91 *** 2.64 *** -8.81 24.94 -1.71 * 21.72 2.91 *** Taiwan 14.83 23.84 3.02 *** -6.88 24.86 -1.34 21.71 23.81 2.91 0.87 18.79 1.27 24.18 0.28 4.12 0.75 5.38 4.69 1.53 24.16 3.47 ** 2.56 19.14 8.16 Thailand 0.87 1.73 * 7.00 21.41 1.38 -1.77 17.29 Europ e 8.77 1.72 * Czech Rep ublic 9.56 21.35 1.89 * 0.82 17.18 0.20 8.75 -0.43 2.12 7.98 20.53 1.59 -4.31 22.76 -0.77 12.29 2.11 ** Hungary 14.23 20.71 2.80 *** 1.92 22.80 0.34 12.30 ** *** 16.56 1.41 -10.09 25.25 -1.67 * 16.80 2.76 *** -0.66 24.86 -3.95 *** 2.63 19.81 12.61 Poland 6.72 2.73 19.72 23.21 36.35 36.67 -1.01 10.71 0.75 1.00 0.11 1.54 -9.71 -0.33 45.44 -3.70 ** 2.29 34.85 19.51 Russia Turkey 39.22 0.35 -8.32 36.37 -1.15 11.01 1.10 29.61 39.76 3.76 *** 17.89 35.82 2.52 ** 11.72 1.16 2.70 8.19 1.76 * 6.35 18.57 1.66 * 0.47 19.42 0.12 North America 1.20 M exico 17.25 19.24 4.36 *** 9.06 19.63 2.24 ** 5.88 *** 1.02 6.43 14.55 2.99 *** 2.83 14.87 1.28 3.60 0.93 South America Chile 11.44 14.92 5.18 *** 7.48 14.94 3.37 3.97 -1.51 -0.06 -0.13 -0.82 18.91 -0.21 -0.69 -0.04 Colombia 8.24 54.02 0.74 8.69 18.42 2.29 ** -0.45 54.40 1.37 0.66 1.56 2.34 23.64 0.43 5.90 0.69 5.69 22.85 23.49 7.47 ** 2.51 22.69 13.16 Peru 8.24 3.70 *** 0.09 0.03 4.66 11.78 1.85 * 4.68 11.49 1.91 * -0.02 -0.01 Frontier Africa M auritius 9.09 11.72 3.64 *** 9.00 11.42 11.00 0.39 8.14 27.04 1.45 -2.88 25.33 -0.55 11.01 1.82 * 25.26 2.04 ** 2.34 26.91 13.04 Pakistan Asia 1.83 * -0.14 -4.28 4.16 20.09 1.02 -4.30 -0.65 *** 2.62 19.91 10.58 1.54 20.07 6.30 Sri Lanka 20.09 -0.65 -0.03 Europ e 10.70 7.47 27.75 0.87 3.23 0.23 0.22 24.49 1.45 Bulgaria 12.56 24.41 1.71 * 9.39 27.67 1.10 3.17 15.95 25.94 2.38 ** -5.85 26.53 -0.86 21.80 2.58 Estonia 17.48 25.91 2.61 *** -4.26 26.46 -0.63 21.74 2.58 ** ** 25.13 -3.08 22.63 -0.53 9.17 1.02 6.09 1.04 0.95 -0.01 22.35 -0.06 1.48 25.08 9.35 Lithuania 9.42 31.33 -0.48 -14.20 30.67 -1.71 * Romania 0.94 12.56 31.03 1.51 1.24 30.40 0.15 11.32 1.07 -4.03 10.18 -3.18 -0.84 1.27 12.69 0.35 -4.45 -1.10 Slovenia 0.79 13.57 0.21 3.96 12.73 1.10 -3.16 -0.78 13.25 10.39 1.21 -0.89 11.64 -0.21 2.09 0.35 0.34 M id East 0.32 Bahrain 2.51 10.37 0.67 0.45 11.71 0.10 2.06 13.23 -0.14 -2.80 17.61 -0.36 2.00 0.41 Jordan 1.45 13.24 0.25 -0.45 17.60 -0.06 1.90 0.39 -0.80 -0.66 17.98 4.04 15.21 0.73 -3.68 -0.66 0.37 0.06 -3.61 0.97 15.18 5.37 0.27 17.95 Kuwait 1.76 3.16 12.61 0.61 2.86 18.69 0.36 0.30 0.03 Oman 4.19 12.55 0.82 3.91 18.75 0.50 0.28 0.03 8.35 0.44 2.74 20.62 0.36 5.61 0.45 5.46 0.59 20.61 4.41 0.87 31.54 9.87 Qatar 31.62 0.73 0.52 0.88 1.55 8.34 25.33 0.72 10.43 0.53 25.31 10.11 * 1.68 25.66 20.38 United Arab 18.77 10.26 25.77 Emirates 20.14 24.64 1.25 -3.46 20.55 -0.71 10.74 1.38 0.26 10.84 1.24 1.36 ** 2.13 24.07 12.08 Argentina South America 7.27 -0.55 -1.64 25.06 -0.28 3.43 31.13 0.47 -5.07 -0.54 Europ e Cy p rus 0.53 25.09 0.09 5.63 31.13 0.78 Rarely Studied -5.11 -4.04 -0.52 -11.50 45.10 -0.77 7.46 0.67 0.69 23.12 -0.47 44.80 -6.93 0.11 22.67 0.82 Iceland 7.75 5.81 23.27 0.97 0.15 26.87 0.02 5.67 0.58 Latvia 8.21 23.24 1.37 2.62 26.95 0.38 5.59 0.56 3.36 Luxembourg 12.20 3.97 *** -4.64 15.58 -1.53 14.01 3.34 *** 12.05 12.17 5.11 *** -1.99 15.56 -0.66 14.03 9.37 *** 3.31 13.82 -2.64 10.53 -0.85 3.37 0.81 0.73 0.18 -0.31 10.44 -0.95 0.58 13.76 2.36 M alta 0.80 -0.81 -11.96 20.97 -1.51 0.24 32.77 0.02 -12.19 -0.75 South America Venezuela -1.27 20.84 -0.16 11.43 32.07 0.94 -12.70 43

44 In - sample and o ut - of - sample comparison of the Halloween effect Table 3. statistics for the regression The table shows the coefficient estimates and t - , as well as the percentage of times that November October returns for the in - sample April returns beat May - - - period and out of sample period of 37 countries. The in sample period refers to the sample period examined in Bouman and Jacobsen (2002) and runs from January 1970 (or the earliest date in our sample depending on data availability) to Au gust 1998. The out - of - sample period is from September 1998 to July month return difference between November - April and May - 2011. The coefficient β represents the 6 - - values are adjusted using Newey West standard errors. *** denotes significance a t 1% level; October. T - **denotes significance at 5% level; *denotes significance at 10% level. IN SAMPLE OUT OF SAMPLE Country β - value %+ β t - value %+ t Argentina 3.64 0.28 0.66 15.26 1.51 0.57 2.91 1.49 0.59 5.39 0.89 0.50 Australia Austria 8.79 2.72 *** 0.69 2.84 *** 0.71 14.11 12.44 5.21 *** 0.90 Belgium 6.96 1.48 0.71 Brazil 37.43 1.72 * 0.67 9.58 1.29 0.50 Canada 7.72 2.57 ** 0.69 5.98 1.54 0.50 - 7.44 - 0.7 0.45 1.43 0.37 0.57 Chile Denmark 3.82 1.55 0.66 4.89 1.19 0.71 Finland 9.28 3.01 *** 0.76 12.42 1.74 * 0.64 14.22 3.99 *** 0.79 9.59 2.32 ** 0.64 France Germany 8.34 2.91 *** 0.69 11.61 2.35 ** 0.79 10.96 1.94 * 0.62 3.99 0.55 0.50 Greece Hong Kong 5.18 0.75 0.66 0.11 0.01 0.43 12.60 1.5 0.56 Indonesia 14.60 1.89 * 0.57 Ireland 8.42 2.17 ** 0.62 13.77 2.70 *** 0.79 Italy 14.98 3.59 *** 0.76 14.18 2.85 *** 0.71 Japan 7.76 0.64 2.41 ** 0.76 11.83 2.14 ** Jordan 4.52 1.08 0.52 3.06 0.72 0.43 Korea 0.43 0.55 12.82 1.67 1.70 * 0.71 12.86 1.9 * 0.68 Malaysia 5.83 1.04 0.57 Mexico 5.06 0.82 0.59 1.36 0.50 8.15 Netherlands 11.86 4.1 *** 0.86 10.38 1.93 * 0.64 3.12 0.83 0.52 4.31 1.41 0.64 New Zealand Norway 6.34 1.38 0.52 10.36 1.69 * 0.57 13.01 1.96 * 0.62 2.56 0.36 0.43 Philippines 3.59 0.34 0.67 8.37 1.67 * 0.79 Portugal Russia - 6.37 - 0.15 0.50 26.62 2.41 ** 0.79 Singapore 7.78 1.52 0.62 4.74 0.78 0.50 6.21 1.18 0.59 1.98 0.35 0.50 South Africa Spain 11.91 3.31 *** 0.76 6.09 1.26 0.71 Sweden 11.70 3.44 *** 0.76 13.80 2.95 *** 0.79 Switzerland 6.29 2.2 ** 0.72 5.03 1.30 0.71 Taiwan 20.11 3.44 *** 0.72 15.00 1.69 * 0.79 0.42 - 0.29 - 0.04 Thailand 5.64 0.66 0.50 Turkey 0.73 0.05 0.46 18.75 1.48 0.50 0.59 Kingdom 12.37 2.89 *** United 6.56 1.85 * 0.64 United States 5.82 2.45 ** 0.72 4.90 1.57 0.57 44

45 – 4 ross country analysis . C market price returns Table month (November - - October) mean returns and standard deviations at percentage, the - This table provides two 6 April and May - coefficient estimates and t , as well as percentage of times that November - April statistics for the regression October return for 109 countries’ market index and the world index. represents the 6 - return beats May - month mean returns April and May October. T - values are adjusted using Newey - West standard errors. The 6 - month mean difference between November - - ). retur ns (standard deviations) are calculated by multiplying monthly returns (standard deviations) by 6 ( √ rouped *** denotes significance at 1% level; **denotes significance at 5% level; *denotes significance at 10% level. Countries are g arket classification and geographical regions. based on the MSCI m Halloween Effect Oct - May Apr - Nov End Status Country Start Date Region Mean St Dev Mean St Dev β t - value %+ Date 17.34 02/1693 07 - 2011 - 6.88 Pooled 109 2.35 19.86 4.53 11.42 *** 58% Countries 02 - 1919 - 2011 - 4.35 8.75 - 0.18 9.84 4.53 3.64 *** 67% World 07 Developed Asia 11 - 1964 07 - 2011 Hong Kong 7.08 22.48 4.44 23.39 2.64 0.55 58% 1914 07 - 2011 Japan 7.31 16.05 - 1.00 14.52 8.31 3.37 *** 66% 11 - 1.87 11 1965 07 - Singapore 6.91 15.79 0.13 17.08 6.78 - * 60% 2011 Europe 02 - 1922 07 - 2011 Austria 5.35 17.31 3.69 21.41 1.66 0.41 56% 0.10 07 - 2011 Belgium 3.99 12.03 - 13.22 4.09 2.61 *** 62% 02/1897 01 - 1921 07 - 2011 Denmark 3.74 9.15 0.56 9.01 3.18 2.59 *** 64% - 1912 07 - 2011 Finland 4.08 14.14 11 4.22 14.87 - 0.14 - 0.06 50% 01/1898 07 - 2011 France 7.05 13.50 - 0.39 12.95 7.45 3.69 *** 66% 07 - 2011 Germany 4.09 14.36 - 1.53 20.44 5.63 2.42 ** 59% 01/1870 01 1954 07 - 2011 Greece 8.65 18.50 0.84 18.63 7.81 2.10 ** 55% - - 1934 07 - 2011 Ireland 6.14 10.85 - 0.48 12.01 6.62 3.63 *** 69% 02 11 - 1905 07 - 2011 Italy 6.11 16.89 - 0.69 16.88 6.80 2.75 *** 60% 02 - 1919 07 - 2011 Netherlands 5.62 10.90 - 1.97 12.83 7.59 4.28 *** 67% - 1970 07 - 2011 Norway 9.19 16.18 01 1.60 18.13 7.58 2.06 ** 55% 01 - 1934 07 - 2011 Portugal 4.87 26.91 1.21 15.20 3.66 0.95 62% - 1915 07 - 2011 Spain 6.26 12.47 - 0.91 11.83 7.16 4.18 *** 69% 01 - 1906 07 - 2011 Sweden 5.52 12.32 01 - 0.03 11.41 5.56 3.34 *** 63% 01 - 1914 07 - 2011 Switzerland 3.91 9.41 - 0.73 11.92 4.64 3.09 *** 66% 07 - 2011 United Kingdom 2.40 9.34 02/1693 - 0.96 10.19 3.37 4.34 *** 59% 02 - 1949 05 - 2011 Israel 13.56 16.74 10.09 15.93 Mid East 3.46 1.43 62% 61% North *** 12 - 1917 07 - 2011 Canada 5.29 9.94 - 0.28 12.61 5.57 3.59 America 57% 11.27 11/1791 07 - 2011 United States 2.24 9.98 0.57 1.67 1.67 * 1.88 Oceania 07 - 2011 Australia 3.11 8.59 02/1875 10.43 1.22 1.12 53% 01 - 1931 07 - New Zealand 2.69 9.71 1.63 10.39 1.06 0.65 51% 2011 Emerging Africa 01 - 1993 07 - 2011 Egypt 14.89 22.01 - 58% 110.45 37.15 1.31 22.26 - 1988 07 - 2011 Morocco 12.40 01 1.05 9.67 11.35 3.68 *** 71% 10.92 02 - 1910 07 - 2011 South Africa 4.78 11.59 2.89 12.10 1.88 0.90 53% Asia 01 - 1991 07 - 2011 China 12.75 26.86 2.04 39.99 10.72 1.00 67% - 1920 07 - 2011 India 3.52 13.63 2.35 13.61 1.17 0.57 45% 11 - - 1983 07 - 2011 Indonesia 13.40 21.29 0.18 22.27 13.58 2.30 ** 55% 04 - 1962 07 - 2011 Korea 12.25 28.77 02 1.26 26.24 11.00 1.71 * 62% 01 - 1974 07 - 2011 Malaysia 8.86 18.56 - 1.59 19.69 10.46 2.17 ** 63% - 1953 07 - 2011 Philippines 6.23 19.59 01 - 3.37 21.13 9.60 2.31 ** 58% 02 - 1967 07 - 2011 Taiwan 13.74 21.48 - 3.58 24.87 17.31 3.76 *** 76% 46% 11 - 1975 07 - 2011 Thailand 4.29 17.99 2.42 22.93 1.87 0.35 45

46 Table 4 . (continued) May - Apr Nov - Oct Halloween Effect Region Country Status End Date Start Date value St Dev β t - %+ St Dev Mean Mean 11 - 1993 07 - 2011 Czech Republic 9.00 22.27 - 2.03 20.01 11.03 Emerging 2.10 ** 68% Europe 01 1995 07 - 2011 Hungary 14.69 21.23 - 22.35 13.42 2.42 ** 71% 1.26 - 1994 07 - 2011 Poland 11.27 21.29 - 5.75 25.35 17.02 2.56 ** 72% 11 - 1993 07 - 2011 Russian 29.49 29.42 11.99 42.11 17.50 1.17 68% 11 0.94 02 1986 07 - Turkey 26.51 39.78 16.78 36.02 9.73 - 46% 2011 North 17.74 02 - 1930 07 - 2011 Mexico 9.76 6.45 18.53 3.30 1.35 56% America 23.72 Brazil - 1990 07 - 2011 43.92 39.80 01 39.77 20.20 2.02 ** 59% South America - 1927 07 - 2011 Chile 11.70 17.01 15.66 24.13 - 3.97 - 0.98 52% 01 - 1927 07 - 2011 Colombia 6.29 14.43 02 3.45 13.76 2.85 1.32 56% 01 - 1933 07 - 2011 Peru 13.72 23.77 17.43 31.13 - 3.72 - 0.81 49% Africa 11 - 1989 07 - 2011 Botswana 6.90 9.16 12.35 11.41 - 5.45 - 1.51 48% Frontier 14.12 - 1996 07 - 2011 Ghana 8.46 01 11.91 5.33 1.08 63% 3.13 02 - 1990 07 - 2011 Kenya 5.65 20.36 1.46 12.63 4.19 0.97 59% - 1989 07 - 2011 Mauritius 6.32 11.80 6.84 11.46 - 0.52 - 0.16 57% 11 01 1988 07 - 2011 Nigeria 11.18 13.88 9.48 16.65 1.69 0.38 58% - - 1996 07 - 2011 Tunisia 3.89 12.58 - 0.47 10.84 4.35 1.19 81% 01 Asia 02 - 1990 07 - 2011 Bangladesh - 5.45 24.43 16.84 21.89 - 22.29 - 2.15 ** 23% 23.30 - 2000 07 - 2011 Kazakhstan 26.90 1.23 26.47 22.07 1.49 67% 11 - 1960 07 - 2011 Pakistan 8.56 16.61 11 1.04 16.28 7.52 2.62 *** 62% 01 - 1985 07 - 2011 Sri Lanka 6.22 18.72 9.69 17.81 - 3.46 - 0.62 52% - 2001 07 - 2011 Viet Nam 11.88 29.98 - 5.36 28.67 17.23 1.12 64% 01 Europe 11 - 2004 07 - 2011 Bosnia a nd 50% - 0.84 26.83 - 7.87 17.73 7.03 0.53 Herzegowina 11 - 2000 07 - 2011 Bulgaria 1.91 23.63 10.64 27.07 - 8.73 - 0.90 33% - 1997 07 - 2011 Croatia 9.33 20.74 02 - 4.42 24.74 13.76 2.03 ** 60% 11 - 1996 07 - 2011 Estonia 17.59 25.93 - 4.38 26.45 21.97 2.62 *** 81% - 01 - 1996 07 - 2011 Lithuania 5.92 17.94 1.31 22.26 7.22 0.97 56% 47% - 11 - 1997 07 2011 Romania 9.56 27.50 2.81 27.46 6.75 0.64 - 2008 07 - 2011 Serbia - 3.70 37.88 - 15.23 48.65 11.53 0.36 75% 11 - 1996 07 - 2011 Slovenia 1.79 19.62 4.88 16.08 - 3.09 - 0.63 31% 01 - 1998 07 - 2011 Ukraine 29.22 29.26 02 - 10.03 31.63 39.25 3.46 *** 79% 9.05 11 - 1990 07 - 2011 Bahrain - 0.79 41% 4.25 10.05 - 5.04 - 1.57 Mid East 02 - 1978 07 - 2011 Jordan 5.21 15.66 1.25 16.51 3.96 1.31 50% 01 - 1995 07 - 2011 Kuwait 4.31 13.80 6.67 13.88 - 2.36 - 0.48 41% - 02 1996 07 - 2011 Lebanon - 3.57 19.44 6.02 20.39 - 9.60 - 1.27 63% - 1992 07 - 2011 Oman 5.16 13.89 3.36 15.22 1.80 0.39 45% 12 - 1999 07 - 2011 Qatar 8.13 23.11 7.27 19.28 0.86 0.10 46% 11 13.34 - 1988 09 - 2008 United Arab 48% 6.51 01 6.22 14.48 0.29 0.06 Emirates 11 - 1969 01 - 2011 Jamaica 11.48 56% 18.34 4.74 17.79 6.74 1.56 North America 01 - 1996 07 - 2011 Trinidad a nd 63% 8.73 10.65 3.91 9.65 4.82 1.36 Tobago 64% South 0.91 01 - 1967 07 - 2011 Argentina 35.90 38.66 27.78 48.55 8.12 America 46

47 Table 4 . (continued) May - Apr Nov - Oct Halloween Effect Region Country Status End Date Start Date value St Dev β t - %+ St Dev Mean Mean Africa 11 - 1997 07 Rarely 2011 Cote D`Ivoire 3.66 11.87 - 0.65 12.69 4.31 1.08 80% - Studied - 2001 01 - 2011 Malawi 11.87 26.66 10.82 27.31 1.05 0.09 18% 04 - 1993 07 - 2011 Namibia 10.93 15.14 0.66 19.60 10.26 1.84 * 68% 03 01 2000 04 - 2007 Swaziland 2.15 14.14 0.15 4.96 2.00 0.38 13% - 3.91 12 2006 07 - 2011 Tanzania 1.30 2.95 7.22 - 2.62 - 0.86 17% - 02 - 2007 07 - 2011 Uganda 14.46 21.78 - 11.32 147.99 25.78 1.33 80% 1997 07 - 2011 Zambia 7.34 15.70 18.18 19.64 - 10.84 - 1.62 47% 02 - 01 - 2000 05 - 2011 Kyrgyzstan 13.05 32.15 - 6.80 27.34 19.84 2.12 ** 75% Asia - 1995 05 - 2011 Mongolia 13.33 31.09 11 37.03 - 2.71 - 0.25 41% 16.04 01 - 1996 07 - 2011 Nepal - 4.54 16.90 8.11 15.30 - 12.65 - 2.18 ** 31% Europe 01 - 1984 07 - 2011 Cyprus 1.07 22.59 1.91 25.53 - 0.84 - 0.13 61% - 2008 07 - 2011 Georgia 2.50 59.57 11 33.02 31.03 - 30.52 - 0.87 50% 01 - 1993 07 - 2011 Iceland 4.52 17.91 - 2.08 31.93 6.60 1.10 58% - 1996 07 - 2011 Latvia 8.32 23.17 1.56 26.53 6.76 0.70 69% 02 01 1954 07 - 2011 Luxembourg 8.72 10.63 - 0.56 12.74 9.28 3.78 *** 71% - 8.21 2001 07 - 2011 Macedonia 4.39 27.27 - 26.47 - 3.82 - 0.30 55% 11 01 - 1996 07 - 2011 Malta 6.39 1.09 11.33 5.30 1.17 69% 15.09 04 - 2003 07 - 2011 Montenegro 13.08 29.86 16.11 33.11 - 3.02 - 0.19 56% - 1993 07 - 2011 Slovak Republic 6.74 28.41 11 - 2.29 15.19 9.03 1.16 68% Mid East 04 - 1990 06 - 2011 Iran 11.43 10.97 14.46 15.24 - 3.03 - 0.71 55% - 2004 07 - 2011 Iraq 15.88 40.08 11 - 6.41 43.71 22.29 0.65 50% 11 - 1997 07 - 2011 Palestine 10.42 35.87 1.06 18.90 9.36 1.18 73% 16.52 - 1993 07 - 2011 Saudi Arabia 3.87 01 2.72 16.68 1.15 0.24 53% 01 - 2010 07 - 2011 Syrian Arab 0% - 7.26 21.16 10.92 18.89 - 18.18 - 1.05 Republic 1.96 North 40% 12 - 2002 07 - 2011 Bahamas 3.67 6.25 4.78 1.71 0.79 America 04 - 1989 02 - 2011 Barbados 0.37 8.52 3.85 11.08 - 3.48 - 1.15 43% 60% 11 - 1996 10 - 2010 Bermuda 1.23 15.28 0.55 13.75 0.68 0.10 17.57 - 1997 02 - 2011 Costa Rica 7.42 11 6.46 12.36 0.96 0.17 47% 01 - 2004 07 - 2011 El Salvador 2.82 7.17 4.61 3.70 - 1.78 - 0.59 13% 8.15 01 - 1993 07 - 2011 Panama 7.09 6.99 7.68 0.10 0.03 53% 15.05 02 - 1994 07 - 2011 Ecuador - 1.95 3.74 17.61 - 5.69 - 1.08 56% - 1993 09 - 2008 Paraguay 3.40 7.24 11 7.85 7.58 - 4.45 - 1.48 19% 02 - 1925 12 - 1995 Uruguay 14.86 34.28 - 1.80 23.03 16.66 3.73 *** 62% 53% 01 - 1937 07 - 2011 Venezuela 6.70 16.52 6.81 16.85 - 0.10 - 0.05 47

48 year sub - period analysis . Pooled 10 Table 5 - - April and May October), the - This table provides mean 6 month returns and standard deviations for two periods (November - statistics for the regression , as well as the percentage of times that the - coefficient estimates and t - April return beats the May - year subsample periods. represents 6 - month mean returns November - October return for 31 ten - October. T - values are adjusted using Newey - West standard errors. The 6 - - differences between November April and May deviations) are calculated by multiplying monthly returns (standard deviations) by 6 ( ). month mean returns (standard √ *** denotes significance at 1% level; **denotes significance at 5% level; *denotes significance at 10% level. - Nov - Apr lloween Effect May Ha Oct Sample % of No of Period value - t Mean St Dev Mean St Dev β Size Countries Positive 58% 6.88 *** 2.35 19.86 4.53 11.42 17.34 - 2011 1693 56679 109 3.70 1710 1693 0.83 - 0.07 14.13 - - 15.40 3.63 215 61% 1 - 1720 - 1.14 8.72 12.38 1711 2.01 32.95 10.73 60% 1 120 - 1730 0.40 - 1.63 7.90 1721 - 0.63 8.58 - 1.00 - 120 50% 1 - 1740 ** 2.04 0.64 - 2.93 1731 2.59 4.96 3.24 120 1 80% 3.68 - 1741 * - 0.65 4.72 2.10 - 2.75 - 1.75 1750 120 20% 1 - 1760 1.43 - 0.75 3.12 1751 - 2.13 2.94 1.39 1 80% 120 1.36 - 1.52 2.65 1761 5.41 - 1770 6.10 4.00 120 1 70% 1771 - 1780 0.14 - - 1.16 5.60 - 0.75 3.77 - 0.41 120 1 60% 1.10 - * 1.93 3.32 1781 - 1790 5.19 4.41 5.52 120 1 70% 1791 - 1800 0.85 - - 0.76 7.34 0.97 7.06 - 1.72 2 50% 232 1801 - 1810 0.26 0.43 4.64 0.03 5.36 0.40 2 30% 240 1811 - 1820 0.62 - 2.15 4.30 2.77 1.95 * 3.88 2 240 70% - 1.51 1821 - 1830 0.84 3.91 2.40 17.00 6.50 240 2 70% 0.82 - 1831 0.03 - 0.75 7.64 - 7.06 0.07 1840 240 55% 2 - 1850 0.46 1.17 8.69 1841 0.16 7.09 1.32 - 2 60% 240 3.48 - 1.04 1.39 1851 10.13 - 1860 10.16 4.86 240 2 75% 1861 - 1870 0.39 3.60 7.52 2.50 9.30 1.10 252 3 52% 0.02 - 0.45 1.06 1871 - 1880 9.24 1.08 8.96 431 4 53% 1881 - 1890 1.54 - - 0.40 5.61 1.89 5.91 - 2.29 4 43% 480 1891 - 1900 1.12 2.24 6.97 0.10 7.34 2.15 6 62% 563 1901 - 1910 0.95 1.83 6.16 0.51 6.72 1.33 854 9 51% 0.90 1911 - 1920 0.16 - - 0.29 11.71 - 0.61 10.88 - 1383 16 55% - 1930 1.34 2.54 13.54 - 0.36 18.76 2.90 1921 2313 63% 22 - 1940 0.80 1931 1.85 13.60 0.22 14.85 1.63 54% 27 2977 - 1950 1941 0.02 3.12 14.85 3.09 15.87 0.03 45% 28 3182 10.11 - 4.05 10.01 1951 4.91 1960 - 0.86 - 0.77 32 3628 46% 1961 1970 *** 4.63 4.80 13.56 - 0.76 13.49 5.56 - 39 64% 4211 18.44 - 9.09 1971 4.00 1980 5.08 2.88 *** 20.05 42 4831 60% 1981 *** 2.79 14.90 22.98 8.79 1990 26.48 6.12 - 5558 57 64% 1991 - 2000 *** 5.50 11.56 21.12 2.65 21.42 8.91 9151 96 63% 3.40 2001 - 2011 5.64 25.61 7.09 18.58 1.45 *** 12908 108 57% 48

49 - periods analysis Table 6. Country by country sub - statistics for the regression , for 28 countries that have data This table provide the coefficient estimates and t available over 60 years and the world market over the whole sample period and several 10 - year sub - periods. The coefficient estimate October. T represents 6 month mean returns differences between November - April and M ay - - - values are adjusted using Newey - West standard errors. *** denotes significance at 1% level; **denotes significance at 5% level; *denotes significance at 10% level . Whole Sample 1911-1920 1921-1930 Prior to 1911 1931-1940 1941-1950 End Start Status Region Country t-value t-value t-value t-value t-value t-value Date β β β β β β Date Hal Hal Hal Hal Hal Hal - 1.77 * 24.64 1.77 * -3.26 *** -0.37 6.27 1.52 9.67 Developed Asia Japan 08/1914 07/2011 8.31 3.60 - 1.09 4.71 Mid East Israel 02/1949 05/2011 3.46 0.84 - - - - - - - - -1.09 -0.27 North America Canada 12/1917 07/2011 5.57 3.34 *** - - -3.47 -0.86 4.58 1.01 3.81 0.50 0.70 -0.68 -1.08 -10.19 1.31 6.70 -0.15 -0.68 -3.31 * 1.66 1.67 07/2011 09/1791 UnitedStates 0.85 -0.72 -2.75 -0.98 Oceania Australia 02/1875 07/2011 1.22 1.07 -1.29 -0.92 6.64 2.28 ** -1.17 -0.31 -2.67 - -0.54 01/1931 07/2011 1.06 0.66 - New Zealand - - - - -1.62 -0.47 -1.09 -0.44 Western 02/1922 07/2011 1.66 0.44 - - - - -29.99 -1.26 9.31 1.09 -9.11 Austria Europe -0.56 1.88 -0.42 -3.18 -0.21 -1.27 0.11 0.43 ** 2.47 4.09 07/2011 02/1897 Belgium 0.23 -2.93 1.08 07/2011 Denmark 0.20 0.53 -0.49 -1.58 0.27 01/1921 - - - - ** 2.20 3.18 Finland -0.14 -0.06 - - -19.35 -2.00 11/1912 -0.77 -0.16 -6.42 -1.62 -18.20 -1.93 * 07/2011 ** France 01/1898 07/2011 7.45 3.87 *** 2.62 1.35 4.34 0.82 2.95 0.54 16.90 2.47 ** -8.86 -0.85 12.31 Germany 01/1870 07/2011 5.63 2.44 ** -0.65 -0.41 -3.07 -0.39 22.54 1.05 11.54 1.98 * 0.82 - 1.05 1.72 4.66 - - - - * - *** 3.35 6.62 07/2011 02/1934 Ireland 1.84 6.77 0.40 Italy 10/1905 07/2011 6.80 2.67 *** 6.77 2.19 ** 3.96 0.63 3.77 0.58 -4.06 -0.73 1.37 7.62 -0.30 -2.04 1.18 6.31 -1.19 -13.92 - - *** 4.05 7.59 07/2011 02/1919 Netherlands - Portugal 1.18 0.96 5.52 - - 0.26 - - - 0.94 3.66 07/2011 01/1934 Spain 01/1915 3.75 *** - - 5.80 1.51 8.58 2.06 ** 10.85 1.18 0.39 07/2011 0.07 7.16 0.09 1.27 -0.56 -4.74 1.52 6.81 1.23 5.11 0.45 0.47 *** 3.14 5.56 07/2011 01/1906 Sweden 07/2011 4.64 2.94 *** - - 9.03 1.61 0.67 0.19 4.19 0.66 -2.92 -1.10 Switzerland 01/1914 -0.70 -0.20 *** 02/1693 07/2011 3.37 4.06 *** 2.54 2.75 United -1.39 -0.62 1.68 0.66 1.22 0.21 Kingdom -1.87 Emerging Africa South Africa 02/1910 07/2011 1.88 0.97 4.29 0.80 -5.07 -1.57 -2.62 -0.97 5.57 0.97 -0.48 - -3.28 -2.33 0.46 1.64 - - - -0.54 0.52 1.17 07/2011 08/1920 India Asia -0.71 * Chile 01/1927 07/2011 -3.97 -0.94 - - - - 6.80 0.80 4.39 0.53 Central/South -5.85 -1.69 America &the - 07/2011 2.85 1.20 - - - 02/1927 -3.52 -0.79 -2.66 -0.47 -1.21 -5.31 Colombia Caribbean 02/1930 0.64 -4.37 3.30 -0.90 0.58 1.13 0.18 07/2011 - - Mexico - - 6.37 - -0.33 -0.61 - - - - -1.25 - -0.68 -3.72 07/2011 01/1933 Peru -2.09 Central/ South - 25.42 Least Uruguay 02/1925 12/1995 16.66 3.52 *** 1.31 9.85 0.40 4.92 1.44 - - - America & the Developed 01/1937 07/2011 0.62 Venezuela 1.54 0.33 1.97 -0.10 -0.04 - - - - - - Caribbean *** -2.58 0.10 0.50 ** 2.25 6.60 -1.47 -7.89 - - -0.81 3.31 4.53 07/2011 02/1919 World 49

50 Table 6 . Continued Start 1951-1960 2001-2011 1991-2000 End 1961-1970 1971-1980 1981-1990 Region Status Country Date t-value t-value t-value t-value t-value t-value Date β β β β β β Hal Hal Hal Hal Hal Hal 08/1914 11.27 1.53 Developed Asia -0.72 07/2011 8.66 1.53 10.74 1.99 ** 10.53 1.91 * 6.06 Japan 0.99 -4.32 * 7.85 0.85 6.41 0.40 3.90 -0.25 -2.07 1.20 5.30 -0.10 -0.78 05/2011 02/1949 Israel Mid East 1.30 North America Canada 1.19 5.21 1.53 8.82 * 1.66 9.27 6.20 *** 2.98 9.61 1.50 6.56 07/2011 12/1917 1.20 09/1791 07/2011 5.02 1.40 5.54 1.47 6.66 1.50 6.62 1.42 4.20 1.38 5.65 1.17 UnitedStates 5.52 0.96 1.87 1.63 7.02 0.85 6.11 0.80 0.40 4.03 -0.97 -3.35 07/2011 02/1875 Australia Oceania 01/1931 New Zealand 07/2011 -6.51 -2.17 ** 3.25 1.16 8.41 1.69 * 0.79 0.10 2.26 0.44 2.87 0.73 13.40 2.25 Western 14.88 1.96 ** Austria 02/1922 07/2011 -10.52 -2.11 ** 6.17 1.15 4.16 1.67 * 10.91 1.56 Europe 12.01 *** 8.10 1.27 2.73 Belgium 02/1897 07/2011 -3.22 -1.09 7.50 2.54 ** 10.92 *** 12.85 2.30 ** 2.95 6.05 6.41 0.94 5.44 -0.43 -1.85 *** 3.07 8.96 * 1.77 3.45 07/2011 01/1921 Denmark 0.99 1.24 8.38 11/1912 0.58 5.21 ** 2.52 21.11 1.56 Finland 1.50 7.88 -0.39 -1.28 -0.49 -2.43 07/2011 1.40 01/1898 07/2011 1.30 0.26 11.78 2.53 ** France 1.03 20.45 3.47 *** 16.77 3.65 *** 8.54 7.12 Germany 1.45 * 01/1870 07/2011 -5.19 -0.97 5.17 1.10 9.80 2.04 ** 5.31 0.93 13.88 2.67 *** 9.94 13.08 Ireland 02/1934 07/2011 -0.88 -0.31 3.68 1.17 4.56 0.64 8.81 1.27 16.27 2.83 *** 1.77 1.02 *** * 3.67 23.97 ** 2.54 22.48 0.12 1.93 1.02 5.49 -1.58 -7.44 07/2011 10/1905 Italy 11.71 2.67 *** 9.28 1.26 Netherlands 02/1919 07/2011 3.19 0.75 7.50 1.58 16.04 3.07 *** 11.72 2.54 ** 12.39 8.11 ** 1.98 14.01 -0.12 -1.63 -0.09 -2.90 0.74 2.22 0.56 1.39 07/2011 01/1934 Portugal 1.21 4.87 *** 2.86 16.95 1.19 9.88 * 1.76 0.77 10.36 0.47 1.65 0.80 3.20 07/2011 01/1915 Spain *** 1.65 11.12 ** 2.37 16.76 1.26 8.79 Sweden 3.61 14.37 0.68 2.85 -1.36 -4.33 07/2011 01/1906 01/1914 07/2011 3.39 0.78 7.74 1.40 8.08 1.49 3.54 0.79 9.74 2.20 ** 4.86 Switzerland 0.89 ** 6.30 1.24 UnitedKingdom 02/1693 07/2011 -2.19 -0.49 7.09 1.54 17.13 1.71 * 14.93 2.90 *** 7.34 1.99 2.10 * 2.69 0.40 Emerging Africa South Africa 02/1910 07/2011 -6.08 -1.66 9.37 1.22 2.25 0.25 0.27 0.03 14.12 ** 0.70 0.02 0.94 11.67 -0.63 -4.52 1.59 6.78 0.16 1.96 -0.46 -1.42 07/2011 08/1920 India Asia -1.55 Chile 01/1927 07/2011 -11.77 -1.32 2.87 0.33 -40.24 -1.68 * 13.29 1.74 * 2.79 0.36 Central/South -0.33 Colombia 02/1927 07/2011 1.73 0.87 3.13 1.40 7.31 1.46 -3.35 -0.37 12.83 1.14 America & the 10.83 1.25 2.50 1.39 0.86 Caribbean 7.86 -1.00 -14.49 ** 9.19 21.87 1.28 2.40 0.93 2.35 07/2011 02/1930 Mexico 0.24 0.23 -8.22 -0.92 -29.37 -0.91 -0.83 -0.06 07/2011 13.63 1.29 Peru -2.50 01/1933 -1.29 0.51 - - - 2.95 55.39 0.88 9.26 0.04 - 0.28 1.56 Central/South 12/1995 02/1925 Uruguay Least *** -0.50 1.99 0.97 -3.85 -0.82 1.75 0.18 -1.30 -0.11 America & the 0.03 0.00 Venezuela 01/1937 07/2011 Developed -1.97 Caribbean 1.98 * 1.84 5.77 1.18 ** 2.16 10.66 1.58 7.27 ** 6.49 5.77 0.89 2.34 07/2011 02/1919 World 50

51 - of - sample Performance of Buy & Hold strategy versus Halloween strategy Table 7 . Out The table presents the annualized average returns, standard deviations in percentages, and Sharpe ratios of the buy n strategy and hold strategy and the Halloween strategy, as well as the percentage of years that the Hallowee outperforms the Buy & Hold strategy for the sample period from October 1998 to April 2011. Halloween Strategy Buy & Hold Strategy Percentage Country of Winning St Dev Sharpe Return St Dev Sharpe Return 18.67 32.19 0.58 21.53 24.15 0.89 38% Argentina 13.29 0.37 6.42 8.56 0.75 46% Australia 4.92 6.68 20.59 0.32 11.43 12.15 0.94 46% Austria Belgium 17.78 0.03 4.50 12.09 0.37 38% 0.46 26.54 0.65 21.52 19.37 1.11 54% Brazil 17.25 6.47 16.03 0.40 7.96 10.61 0.75 31% Canada 14.34 1.06 10.66 15.23 10.89 0.98 38% Chile Denmark 6.78 18.58 0.36 6.47 12.71 0.51 23% Finland 30.05 0.14 9.14 23.26 0.39 38% 4.14 19.05 0.12 6.85 12.86 0.53 38% France 2.29 1.78 22.20 0.08 7.66 Germany 15.16 0.51 46% Greece - 3.28 28.81 - 0.11 1.81 19.10 0.09 54% Hong Kong 6.79 23.59 0.29 5.74 16.42 0.35 38% 19.03 Indonesia 27.92 0.73 20.33 18.34 1.04 23% Ireland - 2.87 22.17 - 0.13 6.74 13.85 0.49 46% Italy - 0.51 20.54 - 0.02 7.30 15.09 0.48 46% Japan - 2.56 20.73 - 0.12 4.74 13.58 0.35 62% 20.47 Jordan 8.96 46% 0.44 7.70 14.86 0.52 Korea 28.44 0.48 15.90 20.99 0.76 46% 13.54 16.14 10.65 0.51 10.94 20.92 0.68 23% Malaysia 17.64 22.10 0.80 18.60 Mexico 1.16 38% 16.09 0.95 20.91 - 0.05 - 13.36 0.42 46% Netherlands 5.59 1.60 13.13 0.12 New Zealand 5.78 8.61 0.67 62% Norway 10.71 22.97 0.47 12.50 14.69 0.85 38% 23.57 0.31 9.59 Philippines 16.05 0.60 38% 7.21 2.47 19.46 - 0.13 Portugal 3.83 13.44 0.29 46% - Russia 33.89 38.71 0.88 36.05 28.23 1.28 38% 0.30 Singapore 6.94 22.86 7.67 14.37 0.53 31% South Africa 19.31 0.74 13.11 13.36 0.98 31% 14.35 19.69 0.15 5.57 13.64 0.41 38% Spain 2.90 5.90 21.57 0.27 Sweden 15.46 0.69 38% 10.74 Switzerland 0.86 14.53 0.06 10.25 0.29 54% 3.02 Taiwan 1.83 26.92 0.07 9.75 18.53 0.53 54% 10.80 Thailand 27.84 0.34 9.55 18.53 0.58 54% Turkey 27.61 45.88 0.60 38.98 38.52 1.01 46% United Kingdom 1.85 15.15 0.12 6.23 9.79 0.64 46% 46% United States 1.73 16.28 0.11 5.02 11.32 0.44 51

52 Table 8 Buy & Hold strategy versus Halloween strategy of the UK market . Annual performance of The table presents the average annual returns, standard deviations in percentages, and Sharpe ratios of the buy and hold strategy number of years, and the percentage of times that the Halloween strategy outperforms and the Halloween strategy, as well as the - the Buy & Hold strategy for the whole sample period from 1693 2009 of the UK market index returns, three subsamples of - year subsamples. around 100 years, six 50 year subsamples, and ten 30 - Halloween Strategy Buy & Hold Strategy of Number Percentage Sample Period Obs. s Sharp e Sharp e of Winning Winning Return Std. Dev. ratio Return ratio Std. Dev. - 2009 1.38 14.58 0.09 4.52 10.71 0.42 316 200 63.29% 1693 - interval year 100 - 1800 - 0.52 11.54 - 0.05 2.95 8.92 0.33 107 70 65.42% 1693 0.06 - 1900 0.68 11.90 100 3.86 8.20 0.47 1801 69 69.00% 1901 - 3.91 18.71 0.21 6.69 13.68 0.49 109 61 55.96% 2009 year interval 50 - 1693 - 1750 - 0.49 13.16 - 0.04 3.19 10.82 0.29 57 32 56.14% 1751 - 0.56 9.45 - 0.06 2.66 1800 6.14 0.43 50 38 76.00% - - 1850 - 0.21 14.81 - 0.01 4.62 10.46 0.44 50 38 76.00% 1801 1851 - 1900 1.58 8.07 0.20 3.10 5.01 0.62 50 31 62.00% 0.20 1901 - 1950 56.00% 11.07 0.02 1.59 6.00 0.26 50 28 1950 - 7.05 22.95 0.31 11.01 16.64 0.66 59 33 55.93% 2009 year interval 30 - 1693 - 1730 - 0.62 15.52 - 0.04 3.83 13.16 0.29 37 22 59.46% 1731 1760 - 1.12 6.60 - 0.17 1.71 3.50 0.49 30 20 66.67% - 1761 1790 0.28 9.77 0.03 4.00 - 6.60 0.61 30 22 73.33% 1791 - 1820 - 0.22 11.48 - 0.02 21 3.04 5.75 0.53 30 70.00% 1821 - - 0.39 16.82 - 0.02 4.69 12.93 0.36 30 23 76.67% 1850 5.57 - 1.45 9.03 0.16 3.45 1880 0.62 30 18 60.00% 1851 1881 - 1910 0.84 6.73 0.13 2.31 3.59 0.64 30 20 66.67% 1911 1940 - 1.19 11.86 - 0.10 1.12 7.01 0.16 30 17 56.67% - 1941 1970 5.84 14.89 0.39 5.21 - 9.30 0.56 30 13 43.33% 61.54% 1971 - 2009 7.61 25.75 0.30 13.36 18.68 0.72 39 24 52

53 Table 9 . Strategy performance over different trading horizons of the UK market standard deviations, skewness, and the maximum and minimum values of the buy and The table shows average returns, hold strategy and the Halloween strategy for different holding . The average returns and the standard deviations are annuali year to years of the UN market index returns from 1693 - 2009 s ed by dividing the total returns horizons from twenty one (standard dev ). The No. of Winning and the % of Winning are the number of times and the percentage of times that the Halloween strategy be iations) by n ( ats the Buy & Hold √ the overlapping s ample, and the lower panel are the resul ts from the non - . The upper panel presents the results calculated using , respectively strategy overlapping sample. Overlapping Sample Holding Buy & Hold Strategy Halloween Strategy No. of Horizon % Win Obs. Win St. Dev. Skew ness Max imum Min imum Return Return St. Dev. Skew ness Max imum Min imum - Year 1.38 14.58 0.12 86.01 - 80.60 4.52 10.71 2.06 83.59 - 30.96 317 200 63.09% 1 - 1.42 14.50 - 0.39 41.03 - 59.11 4.56 11.16 1.60 59.91 - 28.78 316 223 70.57% Year 2 - Year 1.50 14.00 0.10 38.85 - 35.39 4.61 11.09 1.75 46.05 - 11.12 315 236 74.92% 3 - 1.55 13.50 0.31 29.79 - 25.50 4.63 Year 11.40 1.58 35.02 4 7.86 314 250 79.62% - 5 - Year 1.59 13.12 0.58 24.68 - 16.06 4.64 11.92 1.59 33.33 - 6.28 313 257 82.11% 6 - 1.60 12.96 0.77 24.56 - 15.91 4.65 12.34 1.66 29.53 - 3.66 312 258 82.69% Year 7 1.60 12.75 1.01 22.05 - 12.75 4.65 Year 12.76 1.76 29.35 - 4.07 311 267 85.85% - 8 - Year 1.59 12.67 1.27 21.79 - 10.89 4.66 13.21 1.81 27.33 - 2.46 310 271 87.42% 9 1.59 12.78 1.35 21.67 - 7.98 4.66 Year 13.73 1.87 27.15 - 2.83 309 281 90.94% - - Year 1.61 13.00 1.43 21.82 - 8.16 4.67 14.23 1.91 27.06 - 2.89 308 282 91.56% 10 15 - Year 1.63 13.98 1.56 19.27 - 6.52 4.67 16.27 2.04 24.81 - 0.20 303 282 93.07% - 20 - Year 1.61 14.75 1.72 15.62 94.30% 3.56 4.64 17.82 2.04 20.57 0.18 298 281 Overlapping Sample Non - Buy & Hold Strategy Halloween Strategy No. of % Win Obs. Win St. Dev. Skew ness Max imum Min imum Return St. Dev. Skew ness Max imum Return imum Min - - - - - - - Year - - - - - - - 1 - Year 1.33 16.35 - 0.59 41.03 59.11 4.53 12.50 1.66 59.91 - 28.78 158 2 110 69.62% - 3 Year 1.46 16.12 0.15 38.85 - 35.39 4.55 12.51 2.22 46.05 - 11.12 105 80 76.19% - 4.53 4 1.33 15.87 - 0.14 21.70 - 25.50 Year 11.63 1.01 23.35 - 7.86 79 60 75.95% - 5 - Year 1.46 13.36 - 0.01 16.46 - 16.06 4.55 11.49 1.01 22.53 - 6.28 63 51 80.95% 16.41 6 - Year 1.37 0.72 24.56 - 15.91 4.52 14.23 2.23 29.53 - 3.01 52 42 80.77% 7 - 1.46 13.39 0.79 18.44 - 8.76 4.55 13.55 1.15 20.27 - 4.07 45 41 91.11% Year 8 1.37 11.73 1.13 14.43 - 6.98 4.52 Year 12.58 1.64 20.17 - 1.70 39 36 92.31% - 9 - Year 1.46 13.15 0.99 15.75 - 7.98 4.55 14.06 1.85 21.66 - 2.40 35 32 91.43% 10 Year 1.30 11.82 1.19 12.72 - 5.45 4.51 13.80 1.73 18.57 - 1.51 31 29 93.55% - 15 Year 1.46 15.36 0.88 12.33 - 4.08 - 4.55 16.47 1.77 17.75 0.38 21 20 95.24% 14 93.33% 20 - Year 1.24 15.36 1.53 9.16 - 2.51 4.36 18.77 2.39 17.34 0.18 15 53

54 . Halloween effect semi - Table 10 annual data versus monthly data The table compares the regression results of the Halloween effect using - semi annual data and monthly data. Coefficient estimates are in statistics are calculated based on Newey - West percentage terms. T - divided into three sub - periods of - standard errors. The sample is sub - year intervals and six sub - periods of 50 approximately 100 year intervals. - *** ** denotes significance at 5% level; denotes significance at the 1% level; * denotes significance at 10% level Monthly data - Semi annual data Sample s Period β t - value β t - value *** *** 3.36 4.39 1693 2009 0.56 4.26 - - 100 year Interval * - 1800 2.03 1.71 1693 0.34 1.6 *** *** 1801 - 1900 3.14 3.03 2.71 0.52 *** *** 4.87 3.04 1901 2009 0.80 3.03 - year Interval - 50 1693 - 1750 2.83 1.47 0.48 1.29 1751 - 1800 1.10 0.88 0.93 0.18 *** ** 5.06 2.88 1801 1850 0.84 2.29 - 1851 - 1900 1.22 1.33 0.20 1.46 1901 1950 0.67 0.4 - 0.08 0.31 *** *** 1951 - 2009 8.43 3.59 3.33 1.40 54

55 - October) risk premiums for (May 65 countries Figure 1. Summer 10.00 5.00 0.00 -5.00 Return (%) -10.00 -15.00 US NZ UK HK Italy Peru India UAE Chile Israel Qatar Japan Spain Malta China Egypt Oman Korea Latvia Jordan Russia France Poland Greece Ireland Turkey Cyprus Iceland Austria Kuwait Taiwan Estonia Canada Mexico Finland Sweden Bahrain Norway Portugal Pakistan Bulgaria Belgium Slovenia Hungary Thailand Romania Morocco Malaysia Australia Denmark Germany Lithuania Mauritius Indonesia Sri Lanka Colombia Argentina Singapore Venezuela Philippines Switzerland Netherlands South Africa Luxembourg Czech Republic Country 55

56 30 Figure 2 . - year moving average of pooled 65 countries’ price returns, total returns, risk premiums and dividend yield for the period 1694 to 2011. 20.0 Price Return Total Return Risk Premium Dividend Yield 15.0 10.0 Return (%) 5.0 0.0 -5.0 1694 1702 1710 1718 1726 1734 1742 1750 1758 1766 1774 1782 1790 1798 1806 1814 1822 1830 1838 1846 1854 1862 1870 1878 1886 1894 1902 1910 1918 1926 1934 1942 1950 1958 1966 1974 1982 1990 1998 2006 Year 56

57 30 Figure 3 . - year moving average of pooled 65 countries’ price returns, total returns, risk premiums and dividend yield for the period 1951 to 2011. 18.0 Price Return Total Return 16.0 Dividend Yield Risk Premium 14.0 12.0 10.0 Return (%) 8.0 6.0 4.0 2.0 0.0 1951 1953 1955 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 Year 57

58 April) - October) and Winter (November - (May risk premiums for 65 cou ntries Summer Figure 4. 20.00 15.00 10.00 5.00 Return (%) 0.00 -5.00 -10.00 -15.00 UAE HK US Peru Chile Oman Qatar Kuwait Cyprus Bulgaria India UK Mauritius Sri Lanka Latvia NZ Slovenia Australia Mexico Thailand Japan Finland China Denmark Canada Italy Spain Venezuela France Bahrain Sweden Korea Colombia Philippines South Africa Malta Morocco Norway Greece Israel Jordan Belgium Malaysia Switzerland Egypt Netherlands Pakistan Lithuania Austria Hungary Ireland Singapore Argentina Germany Portugal Estonia Turkey Russia Taiwan Czech Republic Poland Indonesia Iceland Luxembourg Romania Country 58

59 - month sub - period (November . Two 6 April and October - May) returns comparison for the developed markets, Figure 5 - rarely studied emerging markets, frontier markets and markets (A) Developed Markets 14.00 12.00 10.00 8.00 6.00 4.00 Return (%) 2.00 0.00 -2.00 Israel Finland Austria Greece Norway Australia Portugal Italy Hong Kong Sweden France Canada Denmark Ireland Spain Belgium Singapore Japan New Zealand United States Germany Switzerland Country Netherlands United Kingdom (B) Emerging Markets 50.00 40.00 30.00 20.00 10.00 0.00 Return (%) -10.00 -20.00 -30.00 Peru Brazil Chile Turkey India Russian Mexico China Korea Thailand Colombia Hungary Morocco South Africa Egypt Malaysia Indonesia Taiwan Poland Philippines Country Czech Republic 59

60 Figure 5 . continued (C) Frontier Markets 40.00 30.00 20.00 10.00 0.00 Return (%) -10.00 -20.00 Qatar Nigeria Bulgaria Kuwait Argentina Botswana Oman Sri Lanka Ghana Jamaica Lebanon Mauritius Bahrain Bangladesh Slovenia Kenya Jordan Romania Tunisia Pakistan Estonia Croatia Serbia Lithuania Ukraine Kazakhstan Viet Nam Zimbabwe United Arab Emirates Trinidad And Tobago Country Bosnia And Herzegowina (D) Rarely Studied Markets 35.00 30.00 25.00 20.00 15.00 10.00 5.00 Return (%) 0.00 -5.00 -10.00 Iran Nepal Zambia Georgia Malawi Panama Mongolia Malta Latvia Paraguay Cyprus Iraq Ecuador Tanzania Montenegro Macedonia Barbados Venezuela Costa Rica Namibia Palestine Iceland El Salvador Bermuda Uruguay Swaziland Saudi Arabia Luxembourg Kyrgyzstan Cote D`Ivoire Syrian Arab Republic Slovak Republic Country 60

61 October) for 31 periods from 109 Figure 6 . Size of the Halloween effect (difference between 6 - month returns November - April and May - pooled countries over the ten - year sub - period 1693 - 2011 Halloween Effect 12.00 10.00 8.00 6.00 4.00 2.00 Returns (%) 0.00 -2.00 -4.00 1693-1710 1711-1720 1721-1730 1731-1740 1741-1750 1751-1760 1761-1770 1771-1780 1781-1790 1791-1800 1801-1810 1811-1820 1821-1830 1831-1840 1841-1850 1851-1860 1861-1870 1871-1880 1881-1890 1891-1900 1901-1910 1911-1920 1921-1930 1931-1940 1941-1950 1951-1960 1961-1970 periods - 10 Year Sub 1971-1980 1981-1990 1991-2000 2001-2011 61

62 Figure 7 . Rolling window regressions of the Halloween effect in the GFD world index returns (1919 - 2011) - year rolling window, a 30 - The figure plots Halloween effects in the GFD world index returns from 1919 to 2011 using a 10 y ear rolling window and a 50 - year rolling window. The dark solid line indicates the coefficient estimates of the effect, the light dotted lines indicates the upper and lower 95% confidence interval based on Newey - West standard errors 62

63 Figure 8 Return frequency distribution of Buy & Hold strategy and Halloween strategy . n years, ten years, fifteen years and twenty years. Th e The figure shows the return frequencies of the Buy & Hold strategy and the Halloween strategy for the holding periods of seve returns are annualised and expressed in percentages. 90 120 80 100 70 80 60 50 60 Buy & Hold Buy & Hold 40 Frequency Frequency Halloween Halloween 40 30 20 20 10 0 0 -8 -12 8 12 16 0 24 28 -4 4 -6 -4 -2 0 2 4 6 8 10 12 14 16 18 20 22 24 26 20 year Return (%) 7 - year Return (%) - 15 140 100 90 120 80 100 70 60 80 50 Buy & Hold Buy & Hold 60 40 Frequency Frequency Halloween Halloween 30 40 20 20 10 0 0 8 10 12 14 16 18 20 22 24 26 28 -8 -6 22 20 18 16 -2 0 2 4 -4 6 14 12 10 8 -4 -2 0 2 4 6 year Return (%) year Return (%) - 10 - 20 63

64 Figure 9 . End of period wealth for the buy and hold strategy and the Halloween strategy for the period 1693 to 2009 16 Buy & Hold 14 Halloween 12 10 8 6 %Returns 4 2 0 -2 1801 1693 1702 1711 1720 1729 1738 1747 1756 1765 1774 1783 1792 2008 1810 1819 1828 1837 1846 1855 1864 1873 1882 1891 1900 1909 1918 1927 1936 1945 1954 1963 1972 1981 1990 1999 Year 64

65 Figure 10. Halloween effect & sample size 5.00 4.00 3.00 2.00 1.00 statistic - t 0.00 1500 4000 3500 0 500 1000 3000 2000 2500 -1.00 -2.00 -3.00 Sample Size 65

66 . UK Halloween effect 100 - year rolling window OLS regressions Figure 11 plots 100 The figure - year rolling window estimates of the Halloween effect for the UK monthly stock market index returns over the period 1693 to 2010. The dark - West standard solid line indicates the coefficient estimates of the effect, the light dotted lines show the upper and lower 95% bounds calc ulated based on Newey errors. 1.60 Hal U95B L95B 1.40 1.20 1.00 0.80 0.60 0.40 Calendar Month Effect (%) 0.20 0.00 -0.20 -0.40 1838 1793 1797 1802 1806 1811 1815 1820 1824 1829 1833 2009 1842 1847 1851 1856 1860 1865 1869 1874 1878 1883 1887 1892 1896 1901 1905 1910 1914 1919 1923 1928 1932 1937 1941 1946 1950 1955 1959 1964 1968 1973 1977 1982 1986 1991 1995 2000 2004 Year 66

67 re 12 . UK Halloween effect 100 - year rolling window regressions estimated with GARCH (1,1) Figu - The figure plots 100 year rolling window estimates of the Halloween effect based on time varying volatility GARCH (1,1) model for the UK monthly s tock market index r eturns over the period 1693 to 2010. The dark solid line indicates the coefficient estimates of the effect and the light dott ed lines show the upper and lower 95% bounds. 1.00 Hal U95B L95B 0.80 0.60 0.40 0.20 Calendar Month Effect (%) 0.00 -0.20 1793 1797 1802 1806 1811 1815 1820 1824 1829 1833 1838 1842 1847 1851 1856 1860 1865 1869 1874 1878 1883 1887 1892 1896 1901 1905 1910 1914 1919 1923 1928 1932 1937 1941 1946 1950 1955 1959 1964 1968 1973 1977 1982 1986 1991 1995 2000 2004 2009 Year 67

68 th Robust Regressions Figure 13 . UK Halloween effect 100 - year rolling window regressions estimated wi - year rolling window estimates of the Halloween effect from robust regressions based on M estimation introduced in Huber (1973) for the UK monthly The figure plots 100 - stock market index returns over the period 1693 to 2010. The dark solid line indicates the coefficient estimates of the effect and the light dotted lines show the upper and lower 95% bounds. 1.00 Hal U95B L95B 0.80 0.60 0.40 0.20 0.00 Calendar Month Effect (%) -0.20 -0.40 -0.60 1846 2009 2005 1793 1797 1801 1805 1809 1813 1817 1821 1825 1829 1833 1837 1842 2001 1850 1854 1858 1862 1866 1870 1874 1878 1882 1886 1891 1895 1899 1903 1907 1911 1915 1919 1923 1927 1931 1935 1940 1944 1948 1952 1956 1960 1964 1968 1972 1976 1980 1984 1989 1993 1997 Year 8 6

69 Appendix 1. Data sources Sample Period Sample Period Status Market Total Return Indices Proxy for the Risk Free Rate Region Country Market Price Index Name End Start End Start World GFD World Price Index 02 - 1919 07 - 2011 GFD World Return Index 01 - 1926 12 - 2012 - Hang Seng Composite Return Asia Hong Hong Kong Hang Seng Composite 08 - 1964 07 - 2011 Developed Hong Kong 3 - month HIBOR 01 - 1970 07 - 2011 Index Index Kong Japan Topix Total Return Index Japan Nikkei 225 Stock Average 2011 08 - 1914 07 - 2011 Japan Overnight LIBOR , Japan 3 - - 07 01 - 1921 month Treasury Bill Yield from (w/GFD extension) Jan 1960 Singapore Singapore FTSE All - Share Index 08 - 1965 07 - 2011 Singapore SE Return Index Singapore 3 - month SIBOR 08 - 1973 07 - 2011 Europe Austria Austria Wiener Boersekammer 2011 02 - 1922 07 - 2011 Vienna SE ATX Total Return - 07 Austria 3 - month Treasury Bill Rate 1970 - 01 mth - from Jan 1970, Europe 3 Index Share Index (WBKI) EURIBOR from Dec 1990 2011 Belgium Brussels All - Share Price Index - 07 02 - 1897 07 - 2011 Brussels All - Share Return Index 1951 Belgium 3 - month Treasury Bill - 01 Yield (GFD extension) (w/GFD extension) 2011 - Denmark OMX Copenhagen All - Share 07 01 - 1921 07 - 2011 OMX Copenhagen All - Share 1970 - Denmark National Bank Discount 01 - Rate, Denmark 3 Gross Index Price Index month Treasury Bill Yield from Jan 1976 OMX Helsinki All Finland OMX Helsinki All - Share Price 2011 11 - 1912 - 2011 - Share Gross - 07 Finland Central Bank Discount 1912 - 11 07 Rate, Bank of Finland Repo Rate Index Index from Dec 1977 2011 France CAC All France - Tradable Index - 07 01 - 1898 07 - 2011 France CAC All - Tradable Total 1898 Bank of France Discount Rate, - 01 month Treasury Bill - France 3 Return Index (w/GFD extension) Yield from Jan 1931 - 2011 Germany - 01 - 1870 07 - 2011 Germany CDAX Total Return Germany Bundesbank Discount 07 1870 01 Germany CDAX Composite Index - (w/GFD extension) Index (w/GFD extension) month Treasury Rate, Germany 3 Yield from Jan 1953 Bill ASE Total Return General Index Greece Athens SE General Index (w/GFD 2011 01 - 1954 07 - 2011 Bank of Greece Discount Rate, - 07 1977 - 01 month Treasury Bill - Greece 3 extension) Yield from Jan 1980 2011 - Ireland Ireland ISEQ Overall Price Index 07 02 - 1934 07 - 2011 Datastream Global Equity 1973 - Ireland 3 - month Treasury Bill 01 Indices (w/GFD extension) Yield - 2011 Italy Banca Commerciale Italiana Index 1925 10 - 07 07 - 2011 Italy BCI Global Return Index - 01 Bank of Italy Discount Rate, Italy 1905 3 - month Treasury Bill Yield from (w/GFD extension) (w/GFD extension) Jan 1940 69

70 Appendix 1. (continued) Sample Period Sample Period Region Country Market Total Return Indices Rate Market Price Index Name Status Proxy for the Risk Free End Start End Start Europe Netherlands Netherlands All - Share Price Share Return 02 - 1919 07 - 2011 Netherlands All - Developed 2011 - Netherlands 3 - month Treasury 07 01 - 1951 Index (w/GFD extensio Bill Yield Index (w/GFD extension Norway Oslo SE All - Share Index 01 - 1970 07 - 2011 Datastream Global Equity Norway 3 - month OIBOR 02 - 1980 07 - 2011 Indices Portugal - Oporto PSI - 20 Index 01 - 1934 07 - 2011 Lisbon BVL General Return 2011 Portugal 3 month Treasury Bill - 07 02 - 1988 Index Yield 2011 - 07 Spain Madrid SE General Index 1940 - 01 - 1915 07 - 2011 Barcelona SE - 30 Return Index 04 Bank of Spain Discount Rate, Spain 3 month MIBOR from Jun (w/GFD extension) (w/GFD extension) - 1973 , Spain 3 - month T - Bill Yield from Jul 1982 Sweden Sweden OMX Aff?rsv?rldens 2011 01 - 1906 07 - 2011 OMX Stockholm Benchmark - 07 Sweden Riksbank Reference Rate, 1919 - 01 month Treasury Bill - Sweden 3 Gross Index (GFD extension General Index Yield from Jan 1955 Switzerland Switzerland Price Index (w/GFD 2011 - 01 - 1914 07 - 2011 Swiss Performance Index Switzerland Overnight LIBOR, 07 1966 - 02 month Secondary extension) - Switzerland 3 Bill Yield from Jan - Market T 1980 UK 3 month Treasury United 2011 - UK FTSE All - Share Index 07 02/1693 07 - 2011 UK FTSE All - Share Return 1694 - 09 - Bill Yield, Bank of England Base Lending (w/GFD extension) Kingdom Index (w/GFD extension) Rate from Jan 1900 Mid East Israel Tel Aviv All - Share Index 02 - 1949 05 - 2011 Datastream Global Equity 2011 Israel 3 - month Treasury Bill 01 - 1993 05 - Indices Yield 300 Total North Canada Canada S&P/TSX 300 Composite 12 - 1917 07 - 2011 Canada S&P/TSX - 2011 - 07 Canada 3 - month Treasury Bill 1934 03 - Yield Return Index America (w/GFD extension) United - S&P 500 Composite Price Index 09/1791 07 2011 S&P 500 Total Return Index GFD Central Bank Discount Rate 2011 - 07 1800 - 02 States (w/GFD extension) (w/GFD extension) Index at annual frequency (interest rates are treated as same every month within a year), USA - day T - Government 90 Bills Secondary Market from Jan 1920 Oceania Australia Australia ASX All - Ordinaries 2011 02/1875 07 - 2011 Australia ASX Accumulation - 07 1928 Australia 3 - month Treasury Bill - 07 All Ordinaries - Index (w/GFD extension) Yield from Jul 1928 New 2011 - New Zealand SE All - Share 07 01 - 07 - 2011 New Zealand SE Gross All - 1986 - New Zealand 3 - month Treasury 07 1931 Bill Yield Share Index Capital Index Zealand 70

71 Appendix 1. (continued) Sample Period Sample Period Region Market Total Return Indices Proxy for the Risk Free Rate Country Market Price Index Name Status End Start End Start Africa Egypt Emerging Cairo SE EFG General Index 01 - 1993 07 - 2011 Datastream Global Equity Egypt 3 - month Treasury Bill 2011 - 10 - 1996 07 Yields Indices 07 Casablanca Financial Group 25 2011 01 - 1988 - 2011 Datastream Global Equity - 07 Morocco Interbank Offer Rate, 1994 - 04 Morocco month Treasury Bill - Morocco 3 Indices Share Index Yield from Jan 2008 2011 South 2011 FTSE/JSE All - Share Index - 07 02 1910 07 - Johannesburg SE Return Index South Africa 3 - month Treasury 1960 02 - - Bill Yield (w/GFD extension) Africa Asia China Shanghai SE Composite 01 - 1991 07 - 2011 China Stock Return Index China Central Bank Discount Rate, 2011 - 01 - 1993 07 China 3 Month Repo on Treasury Bills from Mar 1990 India Bombay SE Sensitive Index 2011 - 08 - 1920 07 2011 India Stocks Total Return Index India 3 - month Treasury Bill Yield 01 - 1988 07 - extension) (w/GFD Indonesia Jakarta SE Composite Index 04 - 1983 07 - 2011 Indonesia Stock Return Index Indonesia Overnight Interbank 2011 - 07 01 - 1988 month JIBOR - Rate, Indonesia 3 from Dec 1993 Korea Korea SE Stock Price Index 2011 - 02 - 1962 - 2011 Korea Stocks Total Return Index Bank of Korea Discount Rate, 07 1962 02 - 07 Korea Overnight Interbank Rate (KOSPI) Aug 1976 bill Discount Malaysia Malaysia KLSE Composite 01 - 1974 07 - 2011 Kuala Lumpur SE Return Index Malaysia 3 - month T - 2011 - 01 - 1974 07 Rate Philippines Return Stock Index Philippines Manila SE Composite Index 01 - 1953 07 - 2011 Philippines 3 - month Treasury Bill 2011 01 - 1982 07 - Yield Taiwan Taiwan SE Capitalization 2011 02 - 1967 07 - 2011 Taiwan FTSE/TSE - 50 Return - 07 Taiwan 3 - month T - bill Yield 01 - 1988 Weighted Index Index Thailand Thailand SET General Index 05 - 1975 07 - 2011 Bangkok SE Return Index Bank of Thailand 1 - day 2011 - 07 05 - 1975 - 3 Repurchase Rate, Thailand month Treasury Bill Yield Jan 1977 Czech Europe Czech Republic 3 2011 Prague SE PX Index 10 - 1993 07 - 2011 Datastream Global Equity - 07 - month Treasury 1993 12 - Bill Yield Indices Republic Budapest Stock Exchange Index Hungary Vienna OETEB Hungary Traded 2011 01 - 1995 07 2011 - 07 Hungary 3 - month Treasury Bill 1995 01 - - Yield (BUX) Index (Forint) Poland Warsaw SE 20 - Share Composite 05 - 1994 07 - 2011 Warsaw SE General Index 2011 Poland 3 - month WIBOR 05 - 1994 07 - (WIG) Russia Russia AK&M Composite (50 2011 - 10 - 1993 07 - Russian Depository Total Return 06 Russia 3 - month Treasury Bill 1995 - 01 2011 Yield Index shares) - 2011 Turkey Istanbul SE IMKB - 100 Price 1986 02 - 1986 07 - 2011 Turkey ISE - 100 Total Return - 02 Turkey 3 - 6 month Treasury Bill 07 Yield Index Index 71

72 Appendix 1. (continued) Sample Period Sample Period Market Total Return Indices Region Market Price Index Name Country Status Proxy for the Risk Free Rate End Start End Start North 1988 Emerging Mexico Mexico SE Indice de Precios y 2011 02 - 1930 07 - 2011 Mexico SE Return Index Mexico 3 - month Cetes Yield 01 - 07 - America Cotizaciones (IPC) - Brazil MSCI Brazil 01 - 1990 07 2011 - - - - South America Chile Santiago SE Indice General de - 01 - 1927 07 - 2011 Santiago SE Return Index Chile Central Bank Mimimum 2011 01 - 1983 07 Precios de Acciones Interest Rate, Chile Repo 7 Day from Aug 1994 Colombia Stock Return Index Colombia Colombia IGBC General Index 2011 02 - 1927 07 - 2011 - Colombia Bank of the Republic 07 1988 - 01 (w/GFD extension) Discount Rate, Colombia TBS Interbank Rate from Jan 1989, month Treasury Bill - Colombia 3 Yield from Jan 1998 Peru Lima SE General Index (w/GFD 2011 01 - 1933 07 - 2011 Peru Stock Return Index Central Bank of Peru Discount - 07 01 - 1993 Rate, Peru Interbank Offer Rate extension) Sep 1995 Frontier Africa Botswana Botswana SE Domestic 06 1989 07 - 2011 - - - Companies Index Ghana Standard and Poor's Ghana Broad 01 - 1996 07 - 2011 - - Market Index Kenya Kenya Nairobi Stock Exchange 02 - 1990 07 - 2011 - - Mauritius Securities Exchange of Mauritius 2011 08 - 1989 07 - 2011 Mauritius Semdex Total Return - Mauritius Interbank Overnight 07 1989 - 08 - Treasury month Index (SEMDEX) Rate, Mauritius 3 Index Rupees Bill Yield from Dec 1996 Nigeria Nigeria SE Index 01 - 1988 07 - 2011 - - Tunisia Standard and Poor's Tunisia Broad 01 - 1996 07 - 2011 - - Market Index - - Asia Bangladesh - Bangladesh Stock Exchange All 02 - 1990 07 2011 - - Share Price 2011 Kazakhstan Kazakhstan SE KASE Index 08 - 2000 07 - - - - Pakistan Pakistan Karachi SE - 100 Index 08 - 1960 07 - 2011 Pakistan Stock Return Index Pakistan Overnight Repo Rate, 2011 01 - 1988 07 month Treasury Bill Pakistan 3 - Rate from Mar 1991 Sri Lanka Colombo SE All - Share Index 01 - 1985 07 - Datastream Global Equity 2011 - Sri Lanka 3 - month Treasury Bill 07 06 - 1987 2011 Yield Indices Viet Nam Viet Nam Stock Exchange Index 01 - 2001 07 - 2011 - - 72

73 Appendix 1. (continued) Sample Period Sample Period Market Price Index Name Market Total Return Indices Proxy for the Risk Free Rate Country Region Status Start End End Start Bosnia And 2011 Frontier Sarajevo SE Bosnian Investment 11 - 2004 07 - Europe - - Herzegowina Funds Index Bulgaria SE SOFIX Index 11 - 2000 07 - 2011 Datastream Global Equity Bulgaria Bulgaria 1 - Mth Sofibor 11 - 2000 07 - 2011 Indices Croatia Croatia Bourse Index (CROBEX) 02 - 1997 07 - 2011 - - Estonia OMX Tallinn (Omxt) 07 - 1996 07 - 2011 OMX Talinn SE Total Return Europe 3 - mth EURIBOR 07 - 1996 07 - 2011 Index Lithuania Standard and Poor's Lithuania 2011 - 01 - 1996 07 - 2011 OMX Vilnius VILSE Total 07 Lithuania 3 - month Treasury Bill 1996 - 01 Yield Return Index Broad Market Index 2011 Romania Bucharest SE Index in Lei 10 - 1997 07 - 2011 Datastream Global Equity Romania National Bank - 07 10 - 1997 Indices Refinancing Rate - Serbia MSCI Serbia 08 - 07 2011 - - 2008 Slovenia HSBC Slovenia Euro 01 - 1996 07 - 2011 Datastream Global Equity 2011 Slovenia 3 - month T - bill Yield 01 - 1999 07 - Indices Ukraine Ukraine PFTS OTC Index 02 - 1998 07 - 2011 - - Mid East Bahrain Bahrain BSE 2011 Composite Index - 07 - 1990 07 - 2011 Datastream Global Equity 07 Bahrain 3 - month Treasury Bill 2004 - 01 Yield Indices Jordan 6 Jordan Jordan AFM General Index 02 - 1978 07 - 2011 Datastream Global Equity - 12 - month Treasury Bill 2011 - 07 - 2006 07 Yield Indices Kuwait Kuwait SE Index 01 - 1995 07 - 2011 Datastream Global Equity 2011 Kuwait 3 - month Interbank Offer - 2004 01 - 07 Indices Rate 2011 Lebanon Beirut Stock Exchange Index 02 - 1996 07 - - - Oman Muscat Stock Market General 2011 12 - 1992 07 - 2011 Datastream Global Equity - 07 Oman 3 - month Interbank Rate 10 - 2005 Indices Index - - Qatar Qatar SE Index 10 - 1999 07 - 2011 Datastream Global Equity 07 Qatar 3 2011 month Interbank Rate 01 - 2004 Indices 2004 United Arab 2008 United Arab Emirates SE Index 01 - 1988 09 - 2008 Datastream Global Equity - 09 United Arab Emirate 3 - month 01 - Indices Emirates Interbank Rate - North Jamaica Jamaica Stock Exchange All - - 07 1969 01 - 2011 - America Share Composite Index Trinidad - Standard and Poor's Trinidad and - 01 - 1996 07 - 2011 Tobago And Tobago Broad Market Index 2011 - South 1993 Argentina Buenos Aires SE General Index - 08 01 07 1967 07 - 2011 Datastream Global Equity - Argentina Interbank up to 15 day - Indices term (IVBNG) America 73

74 Appendix 1. (continued) Sample Period Sample Period Market Total Return Indices Proxy for the Risk Free Rate Market Price Index Name Country Status Region End Start End Start Rarely Africa Cote Cote d'Ivoire Stock Market Index 07 - 1997 07 - 2011 - - D`Ivoire Studied Malawi Malawi SE Index 04 - 2001 01 - 2011 - - - Namibia Namibia Stock Exchange Overall 03 - 1993 07 2011 - - Index Swaziland Swaziland Stock Market Index 01 - 2000 04 - 2007 - - Tanzania Dar - Es - Saleem SE Index 12 - 2006 07 - 2011 - - Uganda USE All Share Index 02 - 2007 07 - 2011 - - - Zambia Zambia Lusaka All Share (n/a) 02 - 1997 07 2011 - - Asia Kyrgyzstan Kyrgyz Stock Exchange Index 01 - 2000 05 - 2011 - - Mongolia Mongolia SE Top - 20 Index 09 - 1995 05 - 2011 - - Nepal Nepal NEPSE Stock Index 01 - 1996 07 - 2011 - - 07 Cyprus Cyprus CSE All Share Composite 01 - 1984 07 - 2011 Datastream Global Equity 2011 Cyprus 3 - month Treasury Bill - Europe 1993 - 01 mth EURIBOR - Yield, Europe 3 Indices from Nov 1982 07 Georgia Standard and Poor's/IFCG - 2008 - 2011 - - 11 Extended Front 150 Georgia Dollar Iceland OMX Iceland All - Share Price 2011 01 - 1993 07 - 2011 OMX Iceland All - Share Gross - 07 Iceland 3 - month Treasury Bill 2002 07 - Index Index Yield Latvia Nomura Latvia 02 - 1996 07 - 2011 OMX Riga SE Total Return 2011 Latvia 3 - month Treasury Bill - 07 05 - 1996 Yield Index Luxembourg Luxembourg SE LUXX Index 2011 01 - 1954 07 - 2011 Luxembourg SE Total Return - 07 Europe 3 - mth EURIBOR 01 - 1985 Index (w/GFD extension) 07 Macedonia Macedonia MBI - 10 Index 11 - - 2011 2001 Malta Malta SE Index 01 - 1996 07 - 2011 Datastream Global Equity 2011 Malta 3 - month T - bill Yield 02 - 2000 07 - Indices Montenegro Montenegro NEX - 20 Index 04 - 2003 07 - 2011 - - Slovak Bratislava SE SAX Index 10 - 1993 07 - 2011 - - Republic 74

75 Appendix 1. (continued) Sample Period Sample Period Country Market Price Index Name Market Total Return Indices Status Proxy for the Risk Free Rate Region Start End Start End ely Mid East Iran Tehran SE Price Index (TEPIX) Rar 04 - 1990 06 - 2011 - - Studied Iraq Iraq SE ISX Index 11 - 2004 07 2011 - - - Palestine Palestine Al - Quds Index 08 - 1997 07 - 2011 - - 07 Saudi Saudi Arabia Tadawul SE Index 01 - 1993 - 2011 - - Arabia Syrian Damascus Securities Exchange 01 - 2010 07 - 2011 - - Arab Weighted Index Republic North Bahamas BISX All Share Index 12 - 2002 07 - 2011 - - America Barbados Barbados SE Local Stock Index 04 - 1989 02 - 2011 - - - Bermuda Bermuda Royal Gazette BSX 09 - 1996 10 2010 - - Composite Index Costa Rica BCT Corp. Costa Rica Stock 02 10 - 1997 - 2011 - - Market Index El El Salvador Stock Market Index 01 - 2004 07 - 2011 - - Salvador Panama Panama Stock Exchange Index 01 - 1993 07 - 2011 - - (BVPSI) 07 Ecuador Ecuador Bolsa de Valores de 02 - 1994 South - 2011 - - Guayaquil (Dollars) America Paraguay Asuncion SE PDV General Index 11 - 1993 09 - 2008 - - 12 Uruguay Uruguay Stock Exchange Index 02 - 1925 - 1995 - - 2003 - Venezuela Caracas SE General Index (w/GFD 1996 01 - 1937 07 - 2011 Datastream Global Equity - 12 Venezuela 3 - month Treasury Bill 12 Yields Indices extension) 75

76 Appendix 2 . 30 - year moving average of price returns, total returns, risk premiums and dividend yield for individual countries that have over 60 years data available, the charts are arranged by descending order of sample siz e United Kingdom 1.6 Price Return Total Return 1.4 Risk Premium Dividend Yield 1.2 1 0.8 0.6 Return (%) 0.4 0.2 0 0.2 - 0.4 - 1790 1706 1712 1718 1724 1730 1736 1742 1748 1754 1760 1766 1772 1778 1784 1994 1796 1802 1808 1814 1820 1826 1832 1838 1844 1850 1856 1862 1868 1874 1880 1886 1892 1898 1904 1910 1916 1922 1928 1934 1940 1946 1952 1958 1964 1970 1976 1982 1988 1694 2000 2006 1700 Year United States 1.2 Total Return Price Return 1 Dividend Yield Risk Premium 0.8 0.6 0.4 Return (%) 0.2 0 0.2 - 0.4 - 1800 1804 1808 1812 1816 1820 1824 1828 1832 1836 1840 1844 1848 1852 1856 1860 1864 1868 1872 1876 1880 1884 1888 1892 1896 1900 1904 1908 1912 1916 1920 1924 1928 1932 1936 1940 1944 1948 1952 1956 1960 1964 1968 1972 1976 1980 1984 1988 1992 1996 2000 2004 2008 Year 76

77 Appendix 2 . Continued Germany 1.5 Price Return Total Return Risk Premium Dividend Yield 1 0.5 Return (%) 0 0.5 - 1 - 1924 1879 1882 1885 1888 1891 1894 1897 1900 1903 1906 1909 1912 1915 1918 1921 2005 1927 1930 1933 1936 1939 1942 1945 1948 1951 1954 1957 1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 1870 2008 2011 1873 1876 Year Australia 1.4 Total Return Price Return Dividend Yield Risk Premium 1.2 1 0.8 Return (%) 0.6 0.4 0.2 0 1930 1932 1934 1936 1938 1940 1942 1944 1946 1948 1950 1952 1954 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 1928 Year 77

78 Appendix 2 . Continued France 1.6 Price Return Total Return 1.4 Risk Premium Dividend Yield 1.2 1 0.8 0.6 Return (%) 0.4 0.2 0 - 0.2 1904 1907 1910 1913 1916 1919 1922 1925 1928 1931 1934 1937 1940 1943 1946 1949 1952 1955 1958 1961 1964 1967 1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003 2006 2009 1898 1901 Year Finland 2 Total Return Price Return 1.8 Risk Premium Dividend Yield 1.6 1.4 1.2 1 Return (%) 0.8 0.6 0.4 0.2 0 1914 1916 1918 1920 1922 1924 1926 1928 1930 1932 1934 1936 1938 1940 1942 1944 1946 1948 1950 1952 1954 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 1912 Year 78

79 Appendix 2 . Continued Sweden 2 Total Return Price Return Risk Premium Dividend Yield 1.5 1 0.5 Return (%) 0 0.5 - 1 - 1925 1927 1929 1931 1933 1935 1937 1939 1941 1943 1945 1947 1949 1951 1953 1955 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 1919 1921 1923 Year Japan 2 Total Return Price Return Risk Premium Dividend Yield 1.5 1 0.5 Return (%) 0 0.5 - 1 - 1923 1925 1927 1929 1931 1933 1935 1937 1939 1941 1943 1945 1947 1949 1951 1953 1955 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 1921 Year 79

80 Continued . Appendix 2 Italy 2 Total Return Price Return Dividend Yield Risk Premium 1.5 1 Return (%) 0.5 0 0.5 - 1961 1933 1935 1937 1939 1941 1943 1945 1947 1949 1951 1953 1955 1957 1959 2011 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 1925 1927 1929 1931 Year Canada 1.2 Total Return Price Return Dividend Yield Risk Premium 1 0.8 0.6 Return 0.4 0.2 0 1936 1938 1940 1942 1944 1946 1948 1950 1952 1954 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 1934 Year 80

81 Appendix 2 . Continued Spain 1.4 Total Return Price Return Dividend Yield Risk Premium 1.2 1 0.8 Return (%) 0.6 0.4 0.2 0 1966 1948 1950 1952 1954 1956 1958 1960 1962 1964 2008 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 1940 2010 1942 1944 1946 Year Netherlands 1.4 Total Return Price Return Risk Premium Dividend Yield 1.2 1 0.8 Returns 0.6 0.4 0.2 0 1953 1955 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 1951 Year 81

82 Continued Appendix 2 . Belgium 1.2 Total Return Price Return Dividend Yield Risk Premium 1 0.8 0.6 Returns (%) 0.4 0.2 0 0.2 - 1953 1955 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 1951 Year 82

83 Appendix 3. Halloween effect in variance both mean and (E quation 2) t o model a Halloween We use GARCH (1,1) with a Halloween dummy in the variance equation seasonal in the mean and variance of returns. , ( ) (2) We include 57 countries that have price indices world index and for over 20 years . Table 11 data available reports the Halloween effects in r eturns and in variance for the world and individual markets . In t heory, if there is a significantly hi we would expect the variance to be higher in winter than in summer. gher winter return, How e ver 35 of the 57 countries actually have a smaller variance in winter than in summer , in which 23 are , suggests risk difference can even significant. This evidence not explain the existence of the Halloween effect, if anything, the risk in summer months is strikingly higher than winter months. 83

84 - M model GARCH (1,1) - Table 11. Halloween effect This table provides the results for the Halloween effect estimated with GARCH (1,1) in mean model: , ) ( for 57 countries that have data for over 20 years and the world one if the month falls in the period of November through April T values are is the Halloween dummy that equals . . - index adjusted using Newey - West standard errors. *** denotes significance at 1% level; **denotes significance at 5% level; *denotes significance at 1 0% level. Countries are grouped based on the MSCI market classification and geographical regions. Variance Halloween Effect S tatus Region S tart Date End Date Country t-value β βv t-value H 1.88 Asia 07-2011 Hong Kong Develop ed 11-1964 * -11.52 -3.62 *** 1.11 11-1914 07-2011 Jap an 0.89 2.88 *** -2.70 -6.17 *** 11-1965 07-2011 0.81 1.68 * -4.36 -2.88 *** Singap ore 02-1922 Austria 1.09 3.53 *** 3.82 10.13 *** Europ e 07-2011 07-2011 Belgium 0.24 1.22 -0.79 -1.42 02/1897 07-2011 01-1921 0.30 1.85 * -0.32 -1.11 Denmark 11-1912 Finland -0.28 -1.15 -2.52 -5.31 *** 07-2011 01/1898 07-2011 France 0.60 2.92 *** -1.84 -7.59 *** 01/1870 07-2011 0.42 1.03 -23.76 -63.32 *** Germany 07-2011 ** 0.74 2.11 01-1954 0.39 0.39 Greece 02-1934 Ireland 0.08 0.59 -0.19 -1.57 07-2011 11-1905 07-2011 Italy 0.44 1.32 -2.90 -2.79 *** 02-1919 Netherlands 1.03 3.35 *** -3.76 -5.80 *** 07-2011 *** 07-2011 Norway 0.99 1.41 -7.93 -3.50 01-1970 3.72 01-1934 07-2011 Portugal 0.83 1.45 *** 0.59 01-1915 07-2011 0.57 2.48 ** 0.18 0.35 Sp ain 07-2011 Sweden 1.06 -0.98 -2.08 ** 01-1906 0.23 07-2011 0.41 1.66 * -3.38 -9.41 *** 01-1914 Switzerland 07-2011 UnitedKingdom 0.25 3.46 *** 02/1693 -3.78 *** -0.24 M id East 05-2011 Israel 0.57 1.51 1.77 1.44 02-1949 12-1917 07-2011 Canada 0.55 2.37 ** -2.24 -4.36 *** North America 11/1791 07-2011 UnitedStates 0.10 1.00 -0.24 -2.06 ** Oceania 02/1875 Australia 0.08 0.66 -1.12 -6.07 *** 07-2011 07-2011 New Zealand 0.02 0.11 -0.40 -1.88 * 01-1931 Emerging Africa 01-1988 07-2011 M orocco 0.91 1.89 * 4.48 3.33 *** 02-1910 South Africa -0.06 -0.30 0.32 0.94 07-2011 Asia 01-1991 07-2011 China 0.62 0.56 -3.42 -0.42 0.08 11-1920 07-2011 India 0.44 *** 0.68 6.25 04-1983 07-2011 1.35 2.25 ** -5.19 -2.57 *** Indonesia 07-2011 Korea -0.34 2.12 0.89 02-1962 -0.19 07-2011 1.32 2.35 ** -1.58 -0.83 01-1974 M alay sia 07-2011 Philip p ines 0.67 1.16 01-1953 -1.55 -0.69 02-1967 Taiwan 1.63 2.93 *** 3.94 2.52 ** 07-2011 07-2011 Thailand 0.06 0.12 -7.20 -3.72 *** 11-1975 Europ e 02-1986 07-2011 Turkey -0.08 -0.05 33.68 2.51 ** North America 02-1930 M exico 0.19 1.53 0.04 0.33 07-2011 South America 07-2011 Brazil 1.02 0.91 -12.47 -2.35 ** 01-1990 01-1927 07-2011 Chile -0.29 -0.86 -5.06 -8.56 *** 02-1927 Colombia 0.11 0.58 0.93 4.18 *** 07-2011 *** 07-2011 Peru -0.15 -1.19 1.63 22.69 01-1933 84

85 Table 11. continued Variance Halloween Effect Country End Date S tatus Region S tart Date t-value β βv t-value H Africa 11-1989 07-2011 Botswana 0.58 2.05 ** 1.18 2.04 ** Frontier 02-1990 07-2011 0.68 1.12 5.57 2.71 *** Keny a 11-1989 M auritius 0.29 0.54 -1.29 -1.01 07-2011 01-1988 07-2011 Nigeria -0.04 -0.11 -3.16 -2.33 ** Asia 02-1990 Bangladesh -1.60 -1.54 -5.69 -1.14 07-2011 07-2011 Pakistan 0.64 2.00 ** -0.29 -0.39 11-1960 01-1985 07-2011 Sri Lanka -0.34 -0.39 -1.06 -0.21 M id East 11-1990 07-2011 Bahrain -0.44 -0.97 1.95 0.96 0.42 02-1978 07-2011 Jordan 0.47 1.01 0.45 North America 11-1969 Jamaica 0.01 0.02 -3.23 -1.19 01-2011 South America 07-2011 Argentina 0.74 0.94 -8.56 -1.90 * 01-1967 Rarely Studied Europ e 01-1984 07-2011 Cy p rus -0.07 -0.14 2.07 2.92 *** 01-1954 07-2011 Luxembourg 0.60 1.97 ** -2.20 -3.00 *** -0.08 M id East 04-1990 06-2011 Iran -0.11 -3.71 -1.14 North America 04-1989 Barbados -0.09 -0.27 3.13 6.39 *** 02-2011 12-1995 Uruguay 0.93 1.54 86.25 22.31 *** 02-1925 01-1937 07-2011 Venezuela 0.10 0.51 0.76 4.05 *** World 02-1919 07-2011 - 0.51 2.38 ** -1.21 -3.21 *** 85

Related documents

Microsoft Word   Remuneration7 4 04 1.doc

Microsoft Word Remuneration7 4 04 1.doc

Remuneration: Where we’ve been, how we got to here, what are the problems, and how to fix them Michael C. Jensen Finance Working Paper N°. 44/2004 Harvard Business School, Monitor Group, July 2004 Cam...

More info »
Corporate Surveillance in Everyday Life

Corporate Surveillance in Everyday Life

Wolfie Christl CORPORATE SURVEILLANCE IN EVERYDAY LIFE How Companies Collect, Combine, Analyze, Trade, and Use Personal Data on Billions EPORT BY CRACKED LABS A R Vienna, June 2017 Author: Wolfie Chri...

More info »
Christl Networks  K .indd

Christl Networks K .indd

Wolfie Christl, Sarah Spiekermann Networks of Control

More info »
Access to Capital and Market Liquidity

Access to Capital and Market Liquidity

Report to Congress Access to Capital and Market Liquidity As Directed by the Explanatory Statement to the Consolidated Appropriations Act, 2016 (P.L. 114-113) This is a report by the Staff of the Divi...

More info »
Who Is On the Other Side

Who Is On the Other Side

BLUEMOUNTAIN INVESTMENT RESEARCH W H O I S O N T H E O T H E R S I D E ? optimism or pessimism, leaving mispriced Being successful as an active investment securities in our wake. And then there are ma...

More info »
AI Now 2017 Report

AI Now 2017 Report

AI ​ ​ Now ​ ​ 2017 ​ ​ Report Authors ​ Campolo, ​ ​ New ​ ​ York ​ ​ University Alex ​ ​ ​ ​ New ​ Sanfilippo, York ​ ​ University ​ ​ Madelyn ​ Whittaker, ​ ​ Google ​ ​ Open ​ Meredith ​ Research,...

More info »
Gilson&Kraakman Book final

Gilson&Kraakman Book final

G ILSON RAAKMAN _B OOK (D O N OT D ELETE ) 3/19/2014 9:12 AM &K MARKET EFFICIENCY AFTER THE FINANCIAL CRISIS: IT’S STILL A MATTER OF INFORMATION COSTS ** * and Reinier Kraakman Ronald J. Gilson I NTRO...

More info »
kd0419345enn

kd0419345enn

Competition policy for the digital era A report by Jacques Crémer Yves-Alexandre de Montjoye Heike Schweitzer Competition

More info »
Automation, Big Data, and Politics: A Research Review

Automation, Big Data, and Politics: A Research Review

Internat (20 16 ), 5032 – 5055 1932 – 8036/2016 000 5 10 ional Journal of Communication Automation, Big Data, and Politics: A Research Review 1 SAMANTHA SHOREY University of Washington, USA PHILIP HOW...

More info »
12 Risks with infinite impact

12 Risks with infinite impact

Global System Major Asteroid Global Artificial Future Bad Extreme Collapse Impact Pandemic Intelligence Global Governance Climate Change Global System Major Asteroid Global Artificial Future Bad Extre...

More info »
v 1.0 2012 02 16

v 1.0 2012 02 16

Written Testimony of John W. Van Alst Attorney, National Consumer Law Center and Director of NCLC’s Working Cars for Working Families Project Before the Committee on Financial Services Subcommittee on...

More info »
Media Manipulation and Disinformation Online

Media Manipulation and Disinformation Online

Media Manipulation and Disinformation Online Alice Marwick and Rebecca Lewis

More info »
MGI big data full report

MGI big data full report

McKinsey Global Institute June 2011 Big data: The next frontier for innovation, competition, and productivity

More info »
Microsoft Word   Buccafusco Online

Microsoft Word Buccafusco Online

C OPYRIGHT 2016, V IRGINIA L AW R EVIEW A SSOCIATION © A THEORY OF COPYRIGHT AUTHORSHIP * Christopher Buccafusco I ... NTRODUCTION ... 1229 L AW AND I. C RITINGS OF A UTHORS ... 1234 OPYRIGHT W Feist ...

More info »
10749

10749

Cash transfers: what does the evidence say? A rigorous review of programme impact and of the role of design and implementation features Francesca Bastagli, Jessica Hagen-Zanker, Luke Harman, Valentina...

More info »
Entering the Law Teaching Market

Entering the Law Teaching Market

Entering the Law Teaching Market Entering the Law Teaching Market • 2015-2016 YALE LAW SCHOOL • CAREER DEVELOPMENT OFFICE • NEW HAVEN, CT 2018-2019 DEVELOPMENT SCHOOL • YALE LAW OFFICE CAREER www.law....

More info »
A Financial System That Creates Economic Opportunities: Capital Markets

A Financial System That Creates Economic Opportunities: Capital Markets

U.S. DEPARTMENT OF THE TREASURY A Financial System That Creates Economic Opportunities Capital Markets OCTOBER 2017

More info »
Establishing cryptocurrency equilibria through game theory

Establishing cryptocurrency equilibria through game theory

AIMS Mathematics, 4(3): 420–436. DOI:10.3934 / math.2019.3.420 Received: 02 March 2019 Accepted: 25 March 2019 Published: 07 May 2019 http: journal / Math // / www.aimspress.com Research article Estab...

More info »
What Do We Know About Base Erosion and Profit Shifting? A Review of the Empirical Literature

What Do We Know About Base Erosion and Profit Shifting? A Review of the Empirical Literature

Univ ersit hic ago L aw Sc hool y of C ago U d un Chic nbo Coase-Sandor Working Paper Series in Law and Coase-Sandor Institute for Law and Economics Economics 2014 What Do We Kno w About B ase Erosion...

More info »
WorkingPaper 2

WorkingPaper 2

Has the Fed Been a Failure? George Selgin Department of Economics Terry College of Business University of Georgia William D. Lastrapes Department of Economics Terry College of Business University of G...

More info »