Master’s thesis MSc Finance
Is the Sell in May effect still present in US markets?
A sector and industry analysis.
Matt Kleinsman S3489086
University of Groningen, Groningen, The Netherlands Supervisor: Artem Tsvetkov
Second supervisor: J.V. Tinang Nzesseu Final version date: 11-01-2021
Data from 11 US sectors and 49 US industries during a period of 1926 to 2020 is analysed to draw a conclusion about the existence of a Sell in May effect. An Ordinary Least Squares (OLS) regression is use with a seasonal dummy for May-October returns to investigate this effect. For the whole period (1926-2020) there is a highly significant Sell in May effect present with a 0.045% daily excess return on a Total level. Adjusting for the January effect and outliers results in about the same outcomes. Adjusting for highly significant (1950-1979) period returns causes the Sell in May effect to vanish from the data. Moreover, in recent year (2000-2020) is no evidence found of a Sell in May effect on a Total level. Overall we conclude that the Sell in May effect was present, but vanished over time.
1. Introduction
Every year, at the end of April, the financial news reports about selling your stocks in May and buy them back after October. By implementing this simple strategy, an investor would get higher returns compared with a simple buy and hold strategy.
The strategy to sell your stocks in May finds it origin in the old saying: Sell in May and Go Away. It implies that an investor should sell his stocks in May and Go away for some time. There are two moments when an investor should rebuy stocks. One of these moments finds his origin in England where bankers would go on vacation in May and leave the city. They returned on St. Leger's Day (12th of September) to the city and bought back stocks. This successive advice is also referred to as "remember to come back in September". The other moment to come back is at the end of October around Halloween (31st of October). Therefore, the "Sell in May and Go Away Strategy" is also called the "Halloween Strategy" or "Halloween indicator". The reasoning behind the saying is that stock returns between May and October tend to be lower than those between November and April. When selling in May, an investor would not be exposed in months which are believed to have the ‘worst’ returns of the year. These months end after October and this would be the moment to buy back the stocks. An investor would only be in the market in the ‘good’ months of the year this way and could get a higher return compared with holding stocks for the whole year (buy and hold strategy)
Bouman and Jacobson (2002) were the first researchers to give an in-depth analysis of this Sell in May and Go away strategy in-depth. They find evidence in favour of a Sell in May effect in 35 of the 37 countries studied. Many other researchers have studied this effect since with different outcomes for different countries, markets, and periods.
The main question that arises: is there still a Sell in May effect still present? Is the Sell in May strategy still a profitable strategy and how can investors profit from it? To research if the effect is still present, a dataset of 11 US sectors and 49 US industries will be studied from July 1926 till August 2020. The setup of Bouman and Jacobsen (2002) is used and will be explained in chapter 3.2. This paper will focus on the periods when the effect is present (which years and months), which US sectors and US industries show evidence of the Sell in May effect and if the effect is still present in recent years.
Because of the Halloween strategy/indicator, Sell in May and Go Away, Sell in May effect all talk about the same phenomenon, some ambiguity can arise. For the remainder of this paper, we will refer to the Sell in May effect as the effect of selling stocks at the beginning of May and buying them back at the end of October.
2. Literature review
As mentioned before, the Sell in May effect should give an investor the chance to earn a higher return with a strategy of selling stocks in May and buy them back at the end of October when compared with a buy and hold strategy. Fama (1970), Jensen (1978) argue that this would imply that the financial markets are inefficient. It would be a violation of the semi-strong form of market efficiency, which incorporates that share prices adjust quickly to new information, and no excess return can be earned. If there were something that violates the efficient market hypothesis, this would be a so-called market anomaly. Tversky and Kahneman (1986), Latif, Shanza, Mariam, and Samia (2011), pp 2,224 defined a market anomaly as: “an anomaly is a deviation from the presently accepted paradigms that are too widespread to be ignored, too systematic to be dismissed as random error and too fundamental to be accommodated by relaxing the normative system”. A study by Schwert (2003) found that if there is something like a market anomaly it would not be around for long. Investors would exploit the opportunity and the anomaly will vanish. Purely looking at the above-mentioned theories, a Sell in May effect could not exist. The only explanations for the effect are: 1) it existed, but vanished over time, or 2) financial markets are not efficient.
not yet solved. To dive deeper into the question if the Sector variable could explain the Sell in May effect, Jacobsen and Visaltanachoti (2009) took a sample containing 17 US sectors and 49 industries between 1926 and 2006. In over two/thirds of the sectors and industries, the Sell in May effect was significantly present. To exploit the effect, a sector rotation strategy was used to improve the risk-return trade-off. Jacobsen and Visaltanachoti (2009) concluded that the Sell in May effect is not weakened during the last decade of their research period.
Maberly and Pierce (2004) commented on Bouman and Jacobson (2002) with their paper Solving the "Sell in May/Buy after Halloween" puzzle. They researched the Sell in May effect with S&P500 futures returns from April 1982 till April 2003. In their beliefs, the Sell in May effect was driven by two big outliers. Namely: October 1987 (stock market crash) and August 1998 (Collapse of Long-Term Capital Management). Taking these outliers out of the data the Sell in May effect disappears for the S&P500 futures. Witte (2010) gave another insight into the methodology used by Maberly and Pierce (2004). Witte argues that the way of dealing with the outliers was unsatisfactory. The results showed that the four biggest outliers besides October 1987 and August 1998 all work against the Sell in May effect. Witte concluded that the effect of the outliers was to obscure rather than it would drive the effect. A robust regression was used which reported outcomes very similar to Bouman and Jacobson (2002) and in favour of a Sell in May effect. Similar results are found by Andrade, Chhaochharia, and Fuerst (2013). The same dataset was used with an extension of the period till 2012. They performed an out of sample test to investigate whether there is a difference in return between May-October and November-April. On average, they find that during the period of November-April stock returns are 10 percentage points higher than in May-October. These findings are in line with the Bouman and Jacobson (2002) paper and in favour of a Sell in May effect. Zhang and Jacobson (2012) extended the number of countries in their research compared with Bouman and Jacobsen (2002). A few years later, Zhang and Jacobson (2018) extended the dataset even further and 114 countries from the MSCI world index were examined up to the end of 2017. Their overall conclusion is that there still is a Sell in May effect present in the data. Furthermore, they conclude that the effect has even strengthened in recent years rather than vanished.
In contrast to Bouman and Jacobson (2002), Dichtl and Drobetz (2014) did not use the MSCI world indexes. The researchers choose to take the indices rather than the indexes. This decision was based upon the investment availability of the MSCI World indexes. Their argument was based on the fact that investors could not invest in the indexes. A research period from roughly 1980 until 2012 was used with the same methodology as Bouman and Jacobsen (2002). At first, the results were significantly in favour of a Sell in May effect. When an adjustment was made for the possibility to perform the strategy with liquid stocks, the results become insignificant and vanished during recent years.
Sell in May effect over most of the chosen time period and in recent years. An important thing to note is the difference in the dataset of previously mentioned papers. When using MSCI world indexes and or US Sector/Industries the effect seems to be present. When using country indices and account for liquidity the effect is not significantly present anymore.
To add to the currently existing theory and research available, the first adjustment is an expansion of the research period from 1926 till 2020. In contrast to Dichtl (2014), there will be 11 instead of 17 US sectors used. These 11 US sectors will be the same as used by the S&P500. Sector fabrication will be explained in section 3.3. The total period will be split up into different decenniums to investigate whether certain times in history show more evidence for a Sell in May effect than others. Besides the split, in different periods a more in-depth investigation in the different months and their returns will be given.
3. Methodology
3.1 Hypothesis development
From the beliefs of Fama (1970) and Jensen (1978), it is not possible that there is a Sell in May effect in the data. If it were present, it would vanish over time. O’Higgins and Downes (1990), Bouman and Jacobsen (2002) and Zhang and Jacobsen (2018) all find (strong) evidence of a Sell in May effect in their papers. The question which has to be answered is: are the Nov-Apr period returns significantly higher than the May-Oct period returns (Sell in May effect) over the full period from 1926 till 2020. Besides the whole period, smaller periods of time will be examined with the same question.
The following hypotheses are used to answer this question: 𝐻0: There isn't a Sell in May effect
𝐻1: There is a Sell in May effect
3.2 Research setup
One thing previously mentioned papers, that are published after 2002, have in common is the use of a somewhat same research setup. This research setup is the same as in the Bouman and Jacobson (2002) paper. For the sake of comparability, this setup will be used.
𝑟𝑡𝑖 = 𝜇 + 𝛼1𝑆𝑡+ 𝜀𝑡 (1)
Where 𝑟𝑡𝑖 is the return on time t for variable i, µ is a constant, 𝛼1is a coefficient, 𝑆𝑡 is the seasonal dummy for the returns from the May-Oct period and 𝜀𝑡 is the error term.
As mentioned before, 𝑟𝑡𝑖 is the return on time t for variable i. The variable i takes a value of one of the sectors of industries examined. When using this equation, the model can be simply modified. This way, other dummies can be tested if they give the same result. For example, the January effect can be accounted for as Bouman and Jacobsen (2002) did.
Other dummies can also be easily added which is convenient for the rest of the paper. If the dummy variable 𝑆𝑡 takes a value of zero it will drop out and the equation will be a well-known random walk.
Bouman and Jacobsen (2002) mention in their paper that the Sell in May effect could be caused by to so-called January effect. The January returns will inflate the Nov-Apr period results and could be one of the drivers of the Sell in May effect. To correct for this effect, the returns of the January months will be removed from the Nov-Apr period and added into the May-Oct period. Therefore, January returns will get a dummy value of 0 instead of 1. To see if the January returns on their own have a significant effect, the dummy variable 𝐽𝑎𝑛𝑡 is added to the equation. 𝐽𝑎𝑛𝑡 will take a value of 1 for January returns and 0 for the not January returns. The model used:
𝑟𝑡𝑖 = 𝜇 + 𝛼1𝑆𝑡+ 𝛼2𝐽𝑎𝑛𝑡+ 𝜀𝑡 (2)
Where 𝑟𝑡𝑖 are the returns on time t of variable i, µ is a constant, 𝛼1 is a coefficient, 𝑆𝑡 is the seasonal dummy for the returns from the May-October period plus the January returns that are removed from the Nov-Apr period, 𝛼2 is a coefficient, 𝐽𝑎𝑛𝑡 is the dummy for the January returns and 𝜀𝑡 is the error term. 𝑟𝑡𝑖 is the return on time t for variable i. The variable i takes a value of one of the sectors or industries examined. These sectors and industry's returns will be tested on their own to see the specific effect.
To get a better understanding of which months have a significant effect on the data, the months besides January also will be tested. First solely on their returns compared to the other months of the year. The dummy variable 𝑀𝑡 will be introduced. 𝑀𝑡 will take the value of 1 for the month tested and zero for all other months of the year. For example, if April is tested, all April returns will get a value of 1 and all other month's returns will get a value of zero.
Where 𝑟𝑡𝑖 is the return on time t for variable i, µ is a constant, 𝛼1is a coefficient, 𝑀𝑡is the month dummy and 𝜀𝑡 is the error term. The variable i takes a value of one of the sectors or industries examined. These sectors and industry's returns will be tested on their own to see the specific effect.
It could be the case, besides the January month, correcting for other months returns is appropriate. If this is the case, the same correction will be made as for the January effect. The regression equation will take the form of equation (2). For example, if it turns out that November's returns have a significant impact on the returns in the Nov-Apr period, a correction will be made. The November returns will be added to the May-Oct period and removed from the Nov-Apr period. The dummy variable for the specific months’ returns 𝐷𝑀𝑡 will account for November returns rather than January returns.
𝑟𝑡𝑖 = 𝜇 + 𝛼1𝑆𝑡+ 𝛼2𝐷𝑀𝑡+ 𝜀𝑡 (4)
Where 𝑟𝑡𝑖 are the returns on time t of variable i, µ is a constant, 𝛼1 is a coefficient, 𝑆𝑡 is the seasonal dummy for the returns from the May-October period plus the January returns, 𝛼2 is a coefficient, 𝐷𝑀𝑡 is the dummy for the corrected month returns and 𝜀𝑡 is the error term. 𝑟𝑡𝑖 is the return on time t for variable i. The variable i takes a value of one of the sectors or industries examined. These sectors and industry's returns will be tested on their own to see the specific effect.
3.3 Data
A dataset containing 49 industry portfolios composed out of the NYSE, AMEX and NASDAQ stocks will be used to examine the existence of a Sell in May effect. The data can be found in the data library of Kenneth R. French on his personal website1. These portfolios are constructed based on the Compustat SIC or CRSP SIC codes when Compustat SIC codes are not available. This construction takes place at the end of June in year t. The returns are then calculated on a daily basis starting from the 1st of July 1926 till the 31st of August 2020. An overview of these industries can be found in Appendix A, table A1.
To investigate the Sell in May effect in the bigger picture, we assign all the industries to sectors. The 49 industries are divided into the 11 sectors which are officially used by the S&P500. The sectors are: Communication Services, Consumer Discretionary, Consumer Staples, Energy, Financials, Healthcare, Industrials, Materials, Real Estate, Technology and Utilities. Sector allocation can be found in Appendix A, table A1. The daily returns per sector are computed by taking the mean of all industries which are present in that specific sector. The variable Total is added to examine the overall effect for sectors and industries together. It is computed from an average of the daily sector returns. The summary statistics for the sectors
can be found in table 1. The summary statistics of the industries can be found in Appendix B, B1.
Table 1
Summary statistics period 1926-2020 (daily return and volatility (%)) for all sectors
Sector Observations Min Max Mean Std dv
Communication Services 24,811 -16.69 15.98 .039 1.03 Consumer Discretionary 24,811 -17.53 19.18 .044 1.06 Consumer Staples 24,811 -13.58 19.71 .046 1.06 Energy 24,811 -16.92 19.05 .045 1.42 Financials 24,811 -18.06 18.23 .047 1.30 Healthcare 24,811 -18.43 36.17 .050 1.14 Industrials 24,811 -15.84 18.06 .045 1.08 Information Technology 24,811 -27.05 28.80 .053 1.54 Materials 24,811 -13.66 14.78 .046 1.17 Real Estate 24,811 -21.23 36.78 .036 2.13 Utilities 24,811 -16.61 21.91 .041 1.29 Total 24,811 -16.54 16.76 .046 1.07
The first look at the data shows some range of daily returns of -27.05% to 36.79% on a sector level. The daily return range for industries is even larger with outcomes of 150% for the MedEq industry and a loss of -53.62% for the Paper industry. In section 4.5 will be dealt with outliers in the data.
4. Results
4.1 Economic significance of the whole period (1926-2020)
Table 2
Daily mean returns and volatility per sector (1926-2020) divided into May-Oct and Nov-Apr periods.
Daily mean return Daily volatility
Sector May-Oct Nov-Apr Difference May-Oct Nov-Apr Difference
Communication Services 0.029 0.049 0.020 1.059 1.011 -0.049 Consumer Discretionary 0.025 0.064 0.039 1.112 1.016 -0.096 Consumer Staples 0.034 0.060 0.026 1.110 1.019 -0.091 Energy 0.025 0.067 0.042 1.475 1.372 -0.104 Financials 0.031 0.064 0.033 1.330 1.271 -0.059 Healthcare 0.030 0.073 0.043 1.160 1.124 -0.036 Industrials 0.020 0.060 0.040 1.418 1.300 -0.118 Information Technology 0.025 0.083 0.058 1.589 1.504 -0.085 Materials 0.020 0.074 0.055 1.218 1.136 -0.082 Real Estate -0.002 0.076 0.078 2.202 2.070 -0.132 Utilities 0.016 0.068 0.052 1.343 1.248 -0.094 Total 0.025 0.069 0.045 1.122 1.024 -0.097
As shown in table 2, the only sector with a negative return in the May-Oct month is Real Estate. This would mean that a Sell in May effect would not be optimal besides the Real Estate sector. Investors would not profit from the positive returns in the May-Oct period if they sold their stocks in May. Taking the daily risk-free rate2 Into account of 0.024 it becomes logical to Sell the stocks in May. In the May-Oct period the mean daily return is 0.025. If an investor sells his stocks in May and buys bonds (risk-free), he would get almost the same return with less risk. Overall are the Nov-Apr returns superior over the May-Oct returns (11/11 sectors show a positive difference),
In Appendix B, table B2 are the results for the industries shown. Only two of the 49 industries show a negative difference. Beer and Smoke have higher returns in the May-Oct period than the Nov-Apr period. A logical explanation could be that more people smoke and drink in the summer months because of the weather.
The second part of table 2 shows the daily volatility in the periods of May-Oct and Nov-Apr. The expected outcome is higher volatility for sectors with higher returns in the Nov-Apr month compared with May-Oct. An outcome like that would imply that there is a risk-return trade-off present, which is in line with efficient financial markets. If an investor could get higher returns with less risk, it would imply that the markets are inefficient. 11/11 sectors show a negative difference for volatility, meaning that the May-Oct volatility is higher than the Apr-Nov volatility. This contradicts Fama (1970) and Jensen (1978) on efficient financial markets
2 An average of the 10-year US treasury rate is taken from 1963 to 2020. 1963 was the first reported year in the
theories and is a remarkable result. Industries show a kind of similar result with only seven of the 49 differences in volatility positive. These industries are part of different sectors and there is no clear connection between them.
4.1.1 Statistical significance of the data (1926-2020)
In the previous section, the economic significance of the results was discussed. With the use of the dummy variable regressions (1) & (2), described in section 3.2,the following outcomes are generated. The results are shown in Table 3.
Table 3
The outcomes of regressions (1) and (2) are presented in table 3. MayOct shows the results from regression (1) where no correction van the January effect is made. MayOctCorJan and Jan show the outcomes of regression (2). For MayOctCorJan, the January returns are removed from the Nov-Apr period and added in the May-Oct period. Jan is the dummy variable for the month January. It states the results if January returns
are significantly different from other months of the year.
Sector MayOct MayOctCorJan Jan
Communication Services 0.020 0.017 0.036 (0.013) (0.014) (0.024) Consumer Discretionary 0.039*** 0.031** 0.077*** (0.014) (0.014) (0.025) Consumer Staples 0.026* 0.025* 0.029 (0.014) (0.014) (0.025) Energy 0.042** 0.043** 0.040 (0.018) (0.019) (0.034) Financials 0.033** 0.030* 0.051* (0.017) (0.017) (0.031) Healthcare 0.043*** 0.042*** 0.046* (0.015) (0.015) (0.027) Industrials 0.047*** 0.042*** 0.071*** (0.014) (0.015) (0.026) Information Technology 0.058*** 0.051** 0.090** (0.020) (0.021) (0.037) Materials 0.055*** 0.050*** 0.079*** (0.015) (0.016) (0.028) Real Estate 0.078*** 0.054* 0.193*** (0.027) (0.029) (0.051) Utilities 0.052*** 0.045*** 0.085*** (0.016) (0.017) (0.031) Total 0.045*** 0.039*** 0.071*** (0.014) (0.014) (0.025)
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
are even highly significant. When correcting for the January effect the results become less significant, but still, ten of the 11 sectors show a significant effect. Meaning that correcting for the January effect did not vanish the Sell in May. On its own, the January month shows a significant effect on eight of 11 sectors, which implies that January returns are significantly different from returns of other months. At an industry level (Appendix B Table B3), 11 of the 49 industries do not show a significant Sell in May effect. When correcting for the January effect this amount rises to 20 industries. The industries that become insignificant from the January effect correction were only significant on a significance 10% level without the correction. Industries that are insignificant are part of different sectors and do not show a clear link between them. One thing to note is that there are many industries highly significant and those results can inflate the sector's results to become significant. On a Total level are all three variables highly significant, indicating evidence of a Sell in May effect.
This way we conclude that there is strong evidence in favour of the Sell in May effect for the (1926-2020) period. Even when a correcting for the January effect. At this stage, our outcomes are in line with Bouman and Jacobsen (2002), Jacobsen and Visaltanachoti (2009), Zhang and Jacobson (2018) and in favour of the Sell in May effect.
4.2 Sell in May through the years
A question immediately arises: if there is economic significance and statistical significance, can an investor still benefit from the Sell in May effect today? Or is the effect driven by returns from a historical period and did it vanish over time. To research if the Sell in May effect is still present in today’s financial markets, the outcomes from decennium to decennium will be examined. Starting with the period 1926-1929 and after that whole decenniums. The sectors will be used to give an overview of the results. Industry outcomes will be discussed when necessary. The variables used for this analysis are the same as in the previous chapter (MayOct, MayOctCorJan and Jan). Significant decennium outcomes are shown in table 4. A complete overview per decennium is presented in Appendix B, Table B4. Figure 1 shows the difference in return between Nov-Apr and May-Oct periods.
-0,06 -0,04 -0,02 0 0,02 0,04 0,06 0,08 0,1 0,12 26-30 30-39 40-49 50-59 60-69 70-79 80-89 90-99 00-09 10-20 N O V -A PR R ET URN S MIN US MA Y-O CT R ET URN S DECENNIUM
Table 4
May-Oct regression per sector per decennium corrected for possible January effect. MayOct shows the results from regression (1) where no correction van the January effect is made. MayOctCorJan and Jan show
the outcomes of regression (2). For MayOctCorJan, the January returns are removed from the Nov-Apr period and added in the May-Oct period. Jan is the dummy variable for the month January. It states the
results if January returns are significantly different from other months of the year.
Sector MayOct MayOctCorJan Jan Sector MayOct MayOctCorJan Jan
1950-1959 1960-1969 ComSer 0.060*** 0.057*** 0.072** ComSer 0.068* 0.062* 0.097 (0.018) (0.019) (0.034) (0.035) (0.036) (0.064) ConsDis 0.047** 0.042** 0.070* ConsDis 0.093*** 0.091*** 0.100* (0.020) (0.021) (0.037) (0.030) (0.032) (0.056) ConsSup 0.052** 0.046* 0.084* ConsSup 0.077*** 0.071** 0.104** (0.024) (0.025) (0.044) (0.029) (0.030) (0.053) Energy 0.064** 0.067** 0.045 Energy 0.037 0.034 0.051 (0.031) (0.033) (0.057) (0.028) (0.030) (0.052) Financials 0.095*** 0.107*** 0.037 Financials 0.023 0.023 0.023 (0.025) (0.027) (0.046) (0.028) (0.030) (0.052) Healthcare 0.067** 0.092*** -0.052 Healthcare 0.084** 0.091** 0.052 (0.029) (0.031) (0.053) (0.035) (0.037) (0.064) Industrials 0.062*** 0.057** 0.083* Industrials 0.077*** 0.068** 0.116** (0.023) (0.025) (0.043) (0.025) (0.027) (0.047) InfoTech 0.134*** 0.170*** -0.036 InfoTech 0.096* 0.117** -0.005 (0.036) (0.038) (0.067) (0.053) (0.056) (0.099) Materials 0.045* 0.046* 0.044 Materials 0.087*** 0.075*** 0.147*** (0.026) (0.027) (0.047) (0.027) (0.029) (0.050) RealEstate 0.039 0.022 0.121 RealEstate 0.098* 0.085 0.158 (0.049) (0.052) (0.090) (0.052) (0.055) (0.096) Utilities 0.066** 0.062** 0.088* Utilities 0.093*** 0.084*** 0.136*** (0.027) (0.028) (0.049) (0.027) (0.028) (0.050) Total 0.062*** 0.063*** 0.056 Total 0.079*** 0.076*** 0.095* (0.022) (0.023) (0.041) (0.027) (0.028) (0.050) 1970-1979 1990-1999 ComSer 0.067** 0.054* 0.128** ComSer 0.013 0.019 -0.013 (0.030) (0.031) (0.056) (0.040) (0.042) (0.074) ConsDis 0.097** 0.080* 0.184** ConsDis 0.071** 0.076** 0.048 (0.040) (0.042) (0.074) (0.033) (0.035) (0.062) ConsSup 0.073** 0.059 0.143** ConsSup 0.033 0.048 -0.039 (0.034) (0.036) (0.063) (0.037) (0.039) (0.070) Energy 0.103** 0.092** 0.155* Energy 0.063 0.078* -0.014 (0.043) (0.045) (0.080) (0.041) (0.043) (0.077) Financials 0.062* 0.066* 0.042 Financials 0.082** 0.085** 0.067 (0.034) (0.036) (0.064) (0.041) (0.043) (0.076) Healthcare 0.060 0.059 0.068 Healthcare 0.037 0.043 0.006 (0.047) (0.050) (0.088) (0.038) (0.040) (0.071) Industrials 0.095*** 0.082** 0.158*** Industrials 0.089*** 0.100*** 0.032 (0.032) (0.034) (0.060) (0.027) (0.029) (0.051) InfoTech 0.175*** 0.137** 0.361*** InfoTech 0.085 0.057 0.222** (0.065) (0.068) (0.120) (0.059) (0.062) (0.110) Materials 0.122*** 0.091*** 0.271*** Materials 0.102*** 0.117*** 0.028 (0.031) (0.033) (0.058) (0.033) (0.035) (0.063) RealEstate 0.175*** 0.127** 0.405*** RealEstate 0.055 0.040 0.130 (0.055) (0.058) (0.103) (0.045) (0.048) (0.085) Utilities 0.096*** 0.084** 0.156** Utilities 0.082*** 0.075** 0.117** (0.035) (0.037) (0.065) (0.030) (0.032) (0.057) Total 0.099*** 0.083** 0.177*** Total 0.074** 0.079** 0.054 (0.036) (0.037) (0.066) (0.030) (0.032) (0.056)
Standard errors in parentheses
4.2.1 Period 1926-1949 (results in Appendix B, Table B4)
The first thing noticed for the periods 1926-1929, 1930-1939 and 1940-1949 is the fact that only one sector has a significant outcome for the variables MayOct, MayOctCorJan and Jan. Real Estate has a significant result with the MayOct variable in the period of 1926-1930. The result is significant on a 10% significance level. All the Total variables are insignificant with the 1930-1939 and 1940-1949 Total variable even negative for MayOct and MayOctCorJan. This implies that Nov-Apr returns tend to be lower than May-Oct returns, which is the opposite of the Sell in May effect. Big events that could have an impact on the data are the great depression and World War II. These events had a significant effect on the United States and maybe a reason why there is no Sell in May effect to be found in the data.
4.2.2 Period 1950-1979 (results in Table 4)
Going into the '50s, completely different results are found. The three decenniums after 1949 (1950-1959, 1960-1969 and 1970-1979) show highly significant effects in favour of the Sell in May effect. Only eight out of the possible 66 sector outcomes show an insignificant effect on the variables MayOct and MayOctCorJan. Those four sectors are Real Estate (1950-1959), Energy and Financials (1960-1969) and Healthcare (1970-1979). On a Total level are similar results found with highly significant outcomes for the MayOct variable for all three periods. Correcting for the January effect returns show the same outcome. All results for the MayOctCorJan variable are significant on a 5% significance level.
4.2.3 Period 1980-1989 (results in Appendix B, Table B4)
After three decenniums of highly significant outcomes, there is a reverse in the results. Only three sectors (Healthcare, Materials and Utilities) have significant outcomes at a 10% level. When correcting for the January effect there is not one sector that has significant outcomes. Moreover, the Total variable value indicates non-significant outcomes. It looks like the Sell in May effect found in previous decenniums has vanished over time. This would be in line with the theories of efficient financial markets by Fama (1970) and Jensen (1978).
4.2.4 Period 1990-1999 (results in Table 4)
significance level. Correcting for the January effect results in the Energy sector's addition with the sectors that already show a Sell in May effect.
4.2.5 2000-2020 (results in Appendix B, Table B4)
Three on a total of the possible 22 outcomes are significant in this period of time looking at the MayOct variable. Two of these three are Real Estate, a sector that has gained much interest in recent years.Real Estate is also the only outcome that is significant in the 2010-2020 period. The MayOctCorJan variable shows similar results. Only the Materials sector becomes significant when a correction for January returns is made. With none of the sectors highly significant and an insignificant Total variable one can conclude that in the last 20 years the Sell in May effect has vanished on a Total level.It seems to be the case that 1950-1979 return have driven the results towards a Sell in May effect which is not present anymore. Section 4.7.1 investigates this statement in further detail. After O’Higgins and Downes (1990) and Bouman and Jacobsen (2002) published their book and paper about the Sell in May effect, it seems to vanish. Schwert (2003) argues that this would happen if there is an anomaly present in the financial markets.
4.3 Which months show the biggest effect on returns?
Table 5
Regressions (3) is used to conduct a dummy variable regression with one of the months as a dummy. This way the impact of the months in examined on the results of the whole year (1926-2020)
Month Observations Min Max Mean Std dev Total
January 2,085 -5.61 5.78 0.096 0.913 0.054** (0.025) February 1,877 -4.68 6.66 0.046 0.894 -0.001 (0.026) March 2,158 -11.49 16.76 0.039 1.166 -0.008 (0.024) April 2,053 -4.94 8.09 0.065 1.007 0.020 (0.025) May 2,094 -8.09 6.55 0.024 1.006 -0.024 (0.025) June 2,090 -6.93 7.94 0.030 1.031 -0.019 (0.025) July 2,083 -12.20 10.02 0.063 0.997 0.019 (0.025) August 2,174 -7.35 6.64 0.057 0.994 0.011 (0.024) September 1,992 -7.40 11.41 -0.034 1.158 -0.088*** (0.025) October 2,160 -16.55 12.47 0.005 1.461 -0.045* (0.024) November 1,955 -9.70 7.60 0.091 1.161 0.048* (0.025) December 2,090 -9.46 8.11 0.078 0.963 0.034 (0.025)
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
From the perspective of a Sell in May strategy the outcomes seem to be logical. The expectation would be that there are months with negative or slightly positive returns in the May-Oct period. The Nov-Apr period contain the best months return wise. In total show four of the 12 months a significant effect. Two positive (January and November) and two negative (September and October). The positive months are both in the Nov-Apr period and the negative months are in the May-Oct period.
4.4 Correcting for the months September, October, and November
To correct for the September month, the same adjustment is made as for the January returns. Regression (4) is used for this adjustment. A detailed description can be found in section 3.2. In short: the September returns are removed from the May-Oct period and added in the Nov-Apr period. The same procedure is done for the October and November returns. Table 6 shows the sector regression results. Appendix B, Table B5, B6, and B7 present industry outcomes.
Table 6
Regression (4) is used to conduct a dummy variable regression for the corrections of September, October, and November returns. The MayOct variable is the standard regression (1) and is added to compare results. MayOctCorOct corrects for October returns, MayOctCorSepOct for combined September and October returns and
MayOctCorSepOctMay corrects the combined result of September, October and May returns. Sep, Oct, and May are the dummy variables for the months September, October, and May. It presents the outcomes for the regression
of the month return if they are significantly different from other months of the year. Period (1926-2020)
Sector MayOct MayOctCorSep MayOctCorSepOct MayOctCorSepOctMay Sep Oct May
ComSer3 0.020 0.016 0.014 0.012 -0.039 -0.025 -0.019 (0.013) (0.014) (0.015) (0.016) (0.025) (0.024) (0.024) ConsDis4 0.039*** 0.030** 0.024 0.016 -0.084*** -0.056** -0.046* (0.014) (0.014) (0.015) (0.016) (0.026) (0.025) (0.025) ConsSup5 0.026* 0.011 0.008 0.007 -0.104*** -0.021 -0.014 (0.014) (0.014) (0.015) (0.016) (0.026) (0.025) (0.025) Energy 0.042** 0.031 0.024 0.019 -0.104*** -0.058* -0.038 (0.018) (0.019) (0.020) (0.022) (0.034) (0.033) (0.034) Financials 0.033** 0.019 0.015 0.002 -0.110*** -0.036 -0.052* (0.017) (0.017) (0.018) (0.020) (0.031) (0.030) (0.031) Healthcare 0.043*** 0.034** 0.027* 0.025 -0.091*** -0.060** -0.035 (0.015) (0.015) (0.016) (0.018) (0.028) (0.027) (0.027) Industrials 0.047*** 0.037** 0.029* 0.025 -0.102*** -0.068*** -0.042 (0.014) (0.014) (0.015) (0.017) (0.026) (0.025) (0.026) Information Technology 0.058*** 0.046** 0.044** 0.043* -0.118*** -0.057 -0.047 (0.020) (0.021) (0.022) (0.024) (0.037) (0.036) (0.037) Materials 0.055*** 0.045*** 0.026 0.021 -0.110*** -0.118*** -0.040 (0.015) (0.016) (0.017) (0.018) (0.028) (0.027) (0.028) Real Estate 0.078*** 0.072** 0.059* 0.058* -0.110** -0.122** -0.061 (0.027) (0.028) (0.030) (0.033) (0.052) (0.050) (0.051) Utilities 0.052*** 0.037** 0.029 0.017 -0.135*** -0.068** 0.065** (0.016) (0.017) (0.018) (0.020) (0.031) (0.030) (0.031) Total 0.045*** 0.033** 0.026* 0.019 -0.104*** -0.064** -0.045* (0.014) (0.014) (0.015) (0.017) (0.026) (0.025) (0.025)
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
The variable MayOctCorSep shows the results when the returns of September are removed from the May-Oct period and added to the Nov-Apr period. Because of the highly significant negative return for this month, a decrease in findings in line with the Sell in May effect is expected. Seven of the 11 sectors still show a significant effect on a 10% significance level in favour of the Sell in May effect. This is a decline of three sectors compared to the MayOct variable where no correction is made. The Total variable is significant on a 5% significance level, indicating that even when correcting for September return, there still is a Sell in May effect present in the data. The MayOctCorSepOct variable corrects for the returns of September and October in the May-Oct period. Even with the two worst months of the May-Oct period added to the Nov-Apr period, there is still a significant outcome for the Total variable on a 10% significance level, indicating the presents of a Sell in May effect. At last, a correction is made for the results of September, October, and May altogether. With this correction the Total variable becomes insignificant. Only Information Technology and Real Estate still show a significant effect on a 10% level.
4.5 Dealing with possible outliers.
Up to this point, the data is used without making any adjustment for possible outliers. An adjustment for outliers could be appropriate because of the extreme daily returns of some of the industries. Examples of these extreme values are the Paper industry with a maximum daily return of 150% and the MedEq industry with a minimum daily return of – 53%. To deal with the outliers the 1% and 5% most extreme values are removed from the data. Data of the 1st percentile and 99th percentile are removed by the MayOct1% variable and data outside the 5th percentile and the 95th percentile is removed at the MayOct5% variable. Again, the MayOct variable presents the results without any correction. MayOct1% and MayOct5% show the results when 1% or 5% of the most extreme values are taken away from the data.
The ConsDis sector experiences the opposite effect compared with Paper industry. It was significant at a 1% level, but when the 1% outliers are removed the outcome is not significant anymore on any level. Other sectors like ConsSup and ComSer react the same as the Paper variable. A more significant outcome is found when the adjustments are made. This is especially remarkable for the ComSer sector because it shows a significant result for a Sell in May effect for the first time in this paper. A reason for this could be that all the extreme negative return values fall in the period Nov-Apr and those returns are more negative than the 1% extreme positive values.
With ComSer also showing results favouring the Sell in May effect, all sectors show a similar result at some point in this paper. The outliers prevented the Sell in May effect to happen for ComSer. Removing the outliers for other sectors does not have a significant effect on the results. All sectors which were significant with the outliers stay significant without the outliers. We conclude that the outliers are not responsible for the Sell in May effect as was argued by Witte (2010). This contradicts a belief that outliers caused the Sell in May effect, as Maberly and Pierce (2004) found. Overall, the results show a highly significant result in favour of the Sell in May effect and in line with Bouman and Jacobsen (2002).
Table 7
The outcomes of regression (1) are presented in table 7. MayOct shows the results from regression (1) where no correction van the January effect is made. MayOct1% and MayOct5% show the outcomes when values of the
1st and 99th percentile (1%) and outside the 5th and 95th percentile (5%) are removed. Period (1926-2020)
Sector MayOct MayOct1% MayOct5%
ComSer 0.020 0.039*** 0.040*** (0.013) (0.012) (0.009) ConsDis 0.039*** 0.015 0.017* (0.014) (0.011) (0.009) ConsSup 0.026* 0.033*** 0.032*** (0.014) (0.012) (0.010) Energy 0.042** 0.023* 0.022** (0.018) (0.012) (0.009) Financials 0.033** 0.039** 0.037*** (0.017) (0.016) (0.013) Healthcare 0.043*** 0.029** 0.029*** (0.015) (0.014) (0.011) Industrials 0.047*** 0.054*** 0.047*** (0.014) (0.018) (0.015) Information Technology 0.058*** 0.039*** 0.034*** (0.020) (0.013) (0.010) Materials 0.055*** 0.044*** 0.043*** (0.015) (0.012) (0.009) RealEstate 0.078*** 0.050*** 0.047*** (0.027) (0.013) (0.011) Utilities 0.052*** 0.078*** 0.071*** (0.016) (0.023) (0.018) Total 0.045*** 0.049*** 0.048*** (0.014) (0.014) (0.011)
Standard errors in parentheses
4.6.1 Sell in May effect the last 20 years (2000-2020)
Significant evidence was found in favour of the Sell in May effect for the period 1926-2020. Schwert (2003) argues that if an anomaly (like the Sell in May effect) is present, it would vanish after discovery. Therefore, following this logic, the Sell in May effect would not be around these days. To investigate if this is the case the same research setup is used as in the previous sections for the (2000-2020) period.
Table 8
Daily mean returns and volatility per sector (2000-2020) divided into May-Oct and Nov-Apr periods.
Daily returns per period
(2000-2020) Daily vol. Per period (2000-2020)
Sector May-Oct Nov-Apr Difference May-Oct Nov-Apr Difference
Communication Services 0.005 0.034 0.029 1.337 1.402 0.065 Consumer Discretionary 0.026 0.060 0.035 1.147 1.208 0.061 Consumer Staples 0.035 0.054 0.020 1.006 1.103 0.097 Energy -0.015 0.084 0.098 1.901 1.959 0.058 Financials 0.028 0.053 0.024 1.622 1.822 0.200 Healthcare 0.026 0.067 0.041 1.150 1.245 0.095 Industrials 0.026 0.066 0.040 1.153 1.208 0.056 Information Technology 0.030 0.040 0.010 1.665 1.880 0.215 Materials 0.011 0.083 0.072 1.345 1.450 0.106 Real Estate -0.028 0.113 0.140 1.850 2.056 0.206 Utilities 0.000 0.086 0.086 1.537 1.632 0.095 Total 0.019 0.068 0.050 1.227 1.314 0.087
In table 8 are the returns per sector presented (Industry results can be found in Appendix C, Table C1). Difference indicated the difference between Apr-Nov and May-Oct returns. To not violate the properties of financial efficient markets, these higher returns should come with higher volatility. This is the case for all the sectors in the (2000-2020) period. Hence, the markets are efficient on a risk-return trade-off perspective. A different result than found for the (1926-2020) period.
Table 9
May-Oct regression per sector (2000-2020) corrected for possible January effect. MayOct shows the results from regression (1) where no correction van the January effect is made. MayOctCorJan and Jan show the outcomes of regression (2). For MayOctCorJan, the January returns are removed from the Nov-Apr period
and added in the May-Oct period. Jan is the dummy variable for the month January. It states the results if January returns are significantly different from other months of the year.
Sector MayOct MayOctCorJan Jan
Communication Services 0.029 0.041 -0.033 (0.038) (0.040) (0.071) Consumer Discretionary 0.035 0.043 -0.008 (0.033) (0.035) (0.054) Consumer Staples 0.020 0.030 -0.030 (0.029) (0.031) (0.051) Energy 0.098* 0.121** -0.017 (0.054) (0.057) (0.095) Financials 0.024 0.041 -0.058 (0.048) (0.051) (0.088) Healthcare 0.041 0.041 0.042 (0.033) (0.036) (0.058) Industrials 0.040 0.055 -0.034 (0.033) (0.035) (0.056) Information Technology 0.010 0.009 0.014 (0.049) (0.052) (0.099) Materials 0.072* 0.092** -0.026 (0.039) (0.042) (0.062) Real Estate 0.140*** 0.144** 0.120 (0.054) (0.058) (0.092) Utilities 0.086* 0.101** 0.008 (0.044) (0.047) (0.074) Total 0.050 0.056 -0.014 (0.035) (0.039) (0.066)
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Four of the 11 sectors turn out to have a significant outcome on a 10% significance level, with only Real Estate highly significant. One of the reasons for this outcome could be the popularity the sector gained in recent years. On an industry level 8 of the 49 industries show a significant outcome with again only Real Estate highly significant. The Total variable is insignificant which indicates no evidence in favour of the Sell in May effect. These results are notably different from the results of the 1926-2020 period. The Total variable showed a highly significant outcome for that period, but this effect seems to have vanished over time.
none of the sectors show a significant effect for January returns. The next sections discusses the regression outcomes per month.
4.6.3 The significant effect per month for the 2000-2020 period.
Table 10
Regressions (3) is used to conduct a dummy variable regression with one of the months as a dummy. This way the impact of the months is examined on the results of the whole year.
Month Observations Min Max Mean Std dev Total
January 427 -5.37 3.87 0.009 1.102 -0.037 (0.056) February 403 -4.68 3.72 0.019 1.072 -0.026 (0.056) March 458 -11.49 10.50 0.060 1.762 0.018 (0.084) April 434 -4.94 7.50 0.142 1.209 0.108* (0.061) May 446 -4.32 5.06 0.036 1.090 -0.007 (0.055) June 448 -6.93 3.56 -0.005 1.122 -0.053 (0.056) July 444 -3.85 5.09 0.042 1.045 -0.001 (0.053) August 466 -7.35 5.12 0.026 1.191 -0.019 (0.058) September 402 -7.40 3.91 -0.044 1.249 -0.095 (0.065) October 443 -9.34 10.83 0.050 1.595 0.008 (0.078) November 409 -7.07 7.13 0.107 1.398 0.070 (0.071) December 419 -9.46 5.65 0.071 1.164 0.030 (0.060)
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
period, but there is no evidence of a Sell in May effect on a Total level. Therefore, correcting for April returns will make the result even more insignificant and is not appropriate.
4.7 Did (1950-1979) returns drive the effect for the whole period (1926-2020)
As shown in section 4.2, the period (1950-1979) has the most significant outcomes when testing for the Sell in May effect. On a Total level, all outcomes are highly significant, meaning that there is strong evidence of the Sell in May effect as seen for the 1926-2020 period. It could be the case that (1950-1979) returns inflate the effect for the whole period (1926-2020). Therefore, regression (1) and (2) are used on a period that excludes (1950-1979) returns. Table 11 presents the sector results for the (1926-1949/1980-2020) period.
Table 11
Outcomes of regression (1) and (2) are presented in this table. MayOct shows the results from regression (1) where no correction van the January effect is made. MayOctCorJan and Jan show the outcomes of regression (2). For MayOctCorJan the January returns are removed from Nov-Apr period and added in the May-Oct period. Jan is the dummy variable for the month January. It states the results if January returns are significantly different from other months of the year. Period
(1926-1949 / 1980-2020)
Sector MayOct MayOctCorJan Jan
ComSer 0.000 -0.002 0.009 (0.018) (0.018) (0.032) ConsDis 0.021 0.013 0.060** (0.018) (0.019) (0.029) ConsSup 0.007 0.010 -0.007 (0.018) (0.019) (0.029) Energy 0.031 0.033 0.021 (0.025) (0.026) (0.042) Financials 0.022 0.014 0.059 (0.023) (0.024) (0.038) Healthcare 0.031* 0.025 0.057* (0.019) (0.020) (0.030) Industrials 0.034* 0.031 0.049 (0.019) (0.020) (0.031) InfoTech 0.023 0.010 0.084* (0.025) (0.026) (0.044) Materials 0.042** 0.041* 0.046 (0.020) (0.022) (0.033) RealEstate 0.066* 0.044 0.177*** (0.037) (0.039) (0.064) Utilities 0.038* 0.032 0.067* (0.023) (0.024) (0.037) Total 0.029 0.023 0.055* (0.018) (0.019) (0.030)
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
industries (results in Appendix C, table C3) show a significant Sell in May effect on a 10% significance level. Only the Mach sector shows a highly significant outcome. Correcting for the (1950-1979) period weakens the Sell in May effect for almost every sector and industry. On the Total level becomes the Sell in May effect insignificant for the MayOct and MayOctCorJan variables. Therefore, we conclude that (1950-1979) period returns are one of the drivers of the Sell in May effect in the data.
4.7.1 The significant effect per month for the (1950-1979) period.
Without the (1950-1979) returns, there is no evidence found in our dataset in favour of the Sell in May effect. One could say that the origin of the Sell in May effect is from this period. If this is the case, the month returns should have a pattern that follows such a Sell in May effect. Table 12 shows the monthly results for the (1950-1979) period.
Table 12
Regressions (3) is used to conduct a dummy variable regression with one of the months as a dummy (1950-1979). This way the impact of the months is examined on the results of the whole year.
Month Observations Min Max Mean Std dev Total
January 650 -2.66 3.35 0.115 0.666 0.076*** (0.028) February 582 -2.34 1.85 0.046 0.600 0.001 (0.026) March 668 -2.52 2.35 0.075 0.605 0.033 (0.025) April 633 -2.18 2.63 0.040 0.628 -0.006 (0.026) May 650 -6.76 6.55 -0.013 0.849 -0.064* (0.034) June 640 -4.44 3.57 -0.020 0.774 -0.071** (0.032) July 627 -3.27 4.19 0.048 0.711 0.003 (0.030) August 662 -3.14 4.34 0.036 0.689 -0.011 (0.028) September 608 -6.10 3.24 -0.005 0.769 -0.055* (0.032) October 664 -3.86 3.90 -0.011 0.812 -0.062* (0.033) November 594 -3.72 4.62 0.138 0.820 0.101*** (0.035) December 635 -2.76 2.64 0.100 0.657 0.060** (0.027)
Standard errors in parentheses
As expected becomes the structure of the Sell in May effect quite clear. From May till October, several months have a negative mean return. This period ends at the end of October when Halloween (31st of October) is celebrated. The November returns are significantly positive which makes Halloween a good moment to buy stocks. Hence, the creation of the Halloween indicator. November, December, and January are the three months with the highest positive returns of the year. Because of the returns it could be the case that those three months inflate the Apr returns and cause the Sell in May effect to be present. The rest of the Nov-Apr period show positive mean returns but are not significant. May is the first month to show significant (negative) returns after January. As said before, in the May-Oct period, some months show significant negative returns which makes the end of April the right moment the Sell your stocks. Because Jacobsen and Bouman (2002) took a sample starting at 1970 there is a good change their results are influenced by the significance of the 1970-1979 period and this way stirred the strategy to sell in May and buy at the end of October.
4.8 Possible explanations of the Sell in May effect
4.8.1 Cyclical returns that drive the effect.
A reason why returns in the Nov-Apr period are higher than the May-Oct period could be incorporated with the product or services a company provides. Some sectors, industries and companies follow a certain cycle which is related to economic cycles. The duration of such an economic cycle depends on the characteristic of the sector, industry, and company. Morningstar founded three categories in which the sectors are categorized. The categories are: Cyclical, Defensive and Sensitive. An overview of the definitions and characteristics of the categories can be found in Appendix D, figure 1.
According to Morningstar, the sectors Basic Materials, Consumer cyclical (in this paper called Consumer discretionary), Financial Services and Real Estate are placed in Cyclical category. This category is highly sensitive to business cycle peaks and throughs.
Taking the full period (1926-2020), an outcome is expected that for those four sectors the results will be the same or show the same effect from decennium to decennium. This is not the case. The Financial Services sector does not show a significant Sell in May effect in the 1960-1969 period where the other three sectors do show an effect. The opposite is true for the 1990-1999 period. In some decennium, none of these sectors show a significant effect.
the outcomes are significant on a 5% significance level. After 1970 the results are mixed. Healthcare becomes insignificant and significant again in the (1980-1989) period where Consumer Supplies and Utilities still show a significant effect. Consumer Supplies becomes insignificant from the (1980-1989) period where Utilities still gives a significant outcome in the (2000-2010) period.
The last category is named “Sensitive” and has a moderate correlation with business cycles. Sectors in this category are: Communication Services, Energy, Industrials and Technology. Because of the moderate correlation, it would be expected that the outcomes do not show a kind of pattern. Communication Services show in none of the decenniums a significant outcome. Energy, Industrials and Technology show significant returns in several decenniums. The Energy sector shows a significant effect in the (1990-1999) where Industrials and Technology do not have significant outcomes. For the (1960-1969) period, this is vice versa, and this implies no clear pattern could be found. The same conclusion can be drawn for the other 2 categories where no clear pattern was seen. Therefore, we conclude that cyclical returns are not the reason for a Sell in May effect.
4.8.2 End of World War II
Is the Sell in May effect caused by the ending of World War II? In the period before and during World War II, the Sell in May effect was not present in our data. After World War II, the first three decenniums (1950-1979) showed a significant effect in favour of the Sell in May effect. Despite World War II ended, the United States was part of the Korean War, The cold war, and the Vietnam War during the (1950-1979) period. It seems to be the case that war is not a variable that significantly impacts the Sell in May effect. From an economic point of view, would it not make sense either. War is not a seasonal phenomenon, which means that equal returns could be expected throughout the year. Therefore, we conclude that World War II and war in general could have affected the Sell in May effect, but it is not clear in which direction.
4.8.3 Trading volume
period is mentioned. Because this paper investigates US markets and industries the trading volume of the broad S&P500 (19506 -2020) is used to come up with a conclusion.
Table 13
The mean trading volume of specific periods of history. Difference states the difference in trading volume between the Nov-Apr period and the May-Oct period. T-value is the test value if the mean of the Nov-Apr
period is significantly different from the May-Oct period
Period Mean trading volume Difference T-value
May-Oct Nov-Apr
1950-2020 1,006,695,989 1,016,668,259 9,972,271 0.68
1990-1999 391,915,269 379,319,696 -12,595,573 0.15
2000-2020 3,201,015,396 3,152,660,591 -48,354,805 0.25
1950-1979 9,285,299 9,765,196 479,897 0.02
The periods investigated are the whole period (1950-2020), the significant decenniums (1950-1979), (1990-1999) and the last 20 years (2000-2020). Difference indicates the difference between the trading volume in Nov-Apr relative to the May-Oct period. The (1950-2020) and (1950-1979) periods which both show a significant Sell in May effect, have a higher trading volume in the Nov-Apr period. The (1990-1999) and (2000-2020) periods both have a lower trading volume which is remarkable for (1990-1999) because it did show significant evidence of the Sell in May effect as 2020) and 1979). Only one period (1950-1979) shows a significantly different trading volume, and this is a different result than Jacobsen and Bouman (2002) found.
5. Conclusion
This paper aims to investigate whether there is some evidence for a Sell in May effect. For this effect to be present, the Nov-Apr returns have to be significantly higher than the May-Oct returns. Following the logic of Fama (1970) and Jensen (1978), this could not be the case. Efficient financial markets do not allow such an effect (anomaly) to be present in the data. If there was an anomaly present, it would vanish over time after discovery (Schwert, 2003). Despite the fact that efficient financial markets is a widely carried belief, some researchers find evidence of the Sell in May effect. The papers of Bouman and Jacobsen (2002), Jacobsen and Visaltanachoti (2009), Witte (2010), Andrade, Chhaochharia, and Fuerst (2013) and Zhang and Jacobsen (2018) all present results that are in line with a Sell in May effect.
When testing for the whole period (1926-2020), there is strong evidence of a Sell in May effect on a Total level with daily return difference of 0.045%. Ten of the 11 sectors and 37 of the 49 industries show a significant Sell in May effect on a 10% significance level. This is in
line with the finding of Jacobsen and Visaltanachoti (2009). Adjusting for January returns (January effect) gives similar results favouring the Sell in May effect. A correction for other significant months return (September, October, and May) weakens the effect, but still shows some evidence in favour of the Sell in May effect. Correcting for possible outliers does not change the conclusion for the (1926-2020). When the 1% and 5% extreme values are removed from the dataset the Total variable is still significant on a 1% significance level. This outcome is not in line with Maberly and Pierce (2004) but follows the logic of Witte (2010).
A more in-depth research on the Sell in May effect in different decenniums gave some remarkable results. The periods (1926-1949), (1980-1989) and (2000-2020) don’t show any evidence of a Sell in May effect on the Total level. In some of the period only one sector and a few industries had a significant effect. For the (1950-1979) and (1990-1999) periods are different results presented. These periods show significant outcomes in favour of the Sell in May effect. Especially the (1950-1979) period shows highly significant outcomes on the Total level. With outcomes of 0.062 (1950-1959), 0.079 (1960-1970) and 0.099 (1970-1979) all significant on a 1% significance level this period shows strong evidence of the Sell in May effect. Only four out of a possible 33 sectors show an insignificant effect. When the (1950-1979) results are removed from the whole period (1926-2020) there is no evidence of a Sell in May effect on a Total level.
As mentioned before, the (2000-2020) period does not show a significant effect on the Total level indicating the effect is not around anymore in recent years. Only the Real Estate sector shows a highly significant effect. This finding contradicts the beliefs of Zhang and Jacobsen (2018) who argue that the effect is still around and did not vanish. O’Higgins and Downes (1990) and Bouman and Jacobsen (2002) both published their book and paper around the (1990-1999) period in which significant evidence in favour of the Sell in May effect was found. After these publications, the effect vanished. This is in line with Schwert (2003) and with the efficient financial market theories of Fama (1970) and Jensen (1978).
Appendix A
Table A1
All 49 industries portfolio's presented with their definition. Portfolio construction and definitions are from Kenneth R. French his personal website. With the definition the industries are divided into sectors. Sector
definitions are the commonly used sectors of the S&P500.
Industrycode Definition Sectors
Telcm Communication Communication services
Smoke Tabacco Products Consumer Discretionary
Toys Recreation Consumer Discretionary
Fun Entertainment Consumer Discretionary
Books Printing and Publishing Consumer Discretionary
Hshld Consumer Goods Consumer Discretionary
Clths Apparel Consumer Discretionary
PerSv Personal Services Consumer Discretionary
Whlsl Wholesale Consumer Discretionary
Rtail Retail Consumer Discretionary
Meals Restaurant, Hotels, Motels Consumer Discretionary
Other Almost nothing Consumer Discretionary
Agric Agriculture Consumer Staples
Food Food Products Consumer Staples
Soda Candy & Soda Consumer Staples
Beer Beer & Liquor Consumer Staples
ElcEq Electrical Equipment Energy
Coal Coal Energy
Oil Petroleum and Natural Gas Energy
Banks Banking Financials
Insur Insurance Financials
Fin Trading Financials
Hlth Healthcare Healthcare
MedEq Medical Equipment Healthcare
Drugs Pharmaceutical Products Healthcare
LabEq Measuring and Control Equipment Healthcare
Autos Automobiles and Trucks Industrials
Aero Aircraft Industrials
Ships Shipbuilding, Railroad Equipment Industrials
Guns Defense Industrials
Mines Non-metallic and Industrial Metal Mining Industrials
BusSv Business Services Industrials
Paper Business Supplies Industrials
Boxes Shipping Containers Industrials
Trans Transportation Industrials
Hardw Computers Information technology
Softw Computers software Information technology
Chips Electronic Equipment Information technology
Chems Chemicals Materials
Rubbr Rubber and Plastic Products Materials
Txtls Textiles Materials
Gold Precious Metals Materials
RlEst Real Estate Real Estate
BldMt Construction Materials Utilities
Cnstr Construction Utilities
Steel Steel Works Etc Utilities
FabPr Fabricated Products Utilities
Mach Machinery Utilities
Appendix B
Table B1
Summary statistics period 1926-2020 (daily return and volatility (%)) for all sectors
Industry Observations Min Max Mean Std dv
Table B2
Daily mean return and volatility (%) per industry (1926-2020) divided into May-Oct and Nov-Apr periods
Daily mean return Daily volatility
Industry May-Oct Nov-Apr Difference May-Oct Nov-Apr Difference
Table B3
The outcomes of regressions (1) and (2) are presented in table B2. MayOct shows the results from regression (1) where no correction van the January effect is made. MayOctCorJan and Jan show the outcomes of regression (2). For MayOctCorjan the January returns are removed from Nov-Apr period and added in the May-Oct period. Jan is
the dummy variable for the month January. It states the results if January returns are significantly different from other months of the year.
Industry MayOct MayOctCorJan Jan Industry MayOct MayOctCorJan Jan
Table B4
The outcomes of regressions (1) and (2) are presented in table B3. MayOct shows the results from regression (1) where no correction van the January effect is made. MayOctCorJan and Jan show the outcomes of regression (2). For MayOctCorjan the January returns are removed from Nov-Apr period and added in the May-Oct period. Jan is the dummy variable for the month January. It states the results if January returns are
significantly different from other months of the year.
1926-1930 1930-1940
Sector MayOct MayOctCorJan Jan Sector MayOct MayOctCorJan Jan
ComSer 0.019 -0.002 0.132 ComSer -0.044 -0.053 0.000 (0.076) (0.079) (0.149) (0.048) (0.051) (0.090) ConsDis -0.013 -0.011 -0.025 ConsDis -0.030 -0.066 0.143 (0.074) (0.078) (0.146) (0.061) (0.064) (0.113) ConsSup 0.013 0.007 0.042 ConsSup -0.017 -0.025 0.026 (0.083) (0.087) (0.163) (0.067) (0.070) (0.124) Energy 0.063 0.082 -0.039 Energy -0.081 -0.114 0.078 (0.073) (0.077) (0.144) (0.082) (0.086) (0.152) Financials 0.099 0.107 0.055 Financials -0.032 -0.087 0.227 (0.112) (0.118) (0.221) (0.075) (0.079) (0.139) Healthcare -0.009 -0.014 0.019 Healthcare -0.004 -0.032 0.129 (0.074) (0.078) (0.146) (0.065) (0.068) (0.120) Industrials 0.063 0.069 0.027 Industrials -0.014 -0.047 0.141 (0.077) (0.081) (0.151) (0.072) (0.075) (0.133) InfoTech 0.073 0.073 0.074 InfoTech -0.020 -0.045 0.099 (0.127) (0.133) (0.249) (0.076) (0.080) (0.141) Materials 0.049 0.042 0.082 Materials -0.051 -0.079 0.080 (0.085) (0.089) (0.166) (0.073) (0.077) (0.135) RealEstate 0.163* 0.155 0.208 RealEstate -0.010 -0.070 0.273 (0.091) (0.096) (0.180) (0.143) (0.150) (0.264) Utilities 0.096 0.094 0.102 Utilities -0.059 -0.087 0.070 (0.071) (0.074) (0.139) (0.084) (0.089) (0.156) Total 0.044 0.046 0.031 Total -0.034 -0.066 0.120 (0.074) (0.077) (0.145) (0.064) (0.067) (0.119)
