The other January effect in U.S. sectors

(1)

The other January effect in U.S. sectors

Amsterdam Business School

Name Sven Pijlman

Student number 10650563

Program Economics & Business Specialization Finance & Organization Number of ECTS 12

Supervisor Ilko Naaborg Target completion 31 / 01 / 2017

Abstract

The ‘other January’ effect tries to predict the return in the 11-months following January. This study investigates if there are differences in the effect across different sectors in the U.S. in various time periods. This study used 17 different sectors in the U.S. These are tested for the presence of the ‘other January’ effect in the period 2007 and the sub-periods 1949-1971, 1972-1997 and 1998-2007. This is done by an OLS-regression of the returns in the remainder of the year on a dummy variable for January. The outcomes are that in consumer-oriented sectors the effect is present and in sectors that are linked to the production process the effect is absent. Also, there is found that the effect weakens over time. Further is found that a portfolio managed based on the ‘other January’ effect in different sectors does not generate significantly more money than a simple buy-and-hold strategy.

Key words: ‘other January’ effect, January barometer, seasonal anomalies, sector specific returns

(2)

2

This document is written by Sven Pijlman who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

The topic of this study is the ‘other January’ effect, also known as the January barometer. The effect is explained by the statement: ‘as January goes, so goes the year’. This means that if the market return in January is positive, the return for the remainder of the year is also positive and higher than when the market return in January is negative. The ‘original’ January effect states that the return in January has a higher mean than the return in other months, especially in small firm stocks (Bhardwaj & Brooks, 1992).

The ‘other January’ effect was first mentioned in 1974, but surprisingly there is not done much research to this topic in the first 30 years since it was first mentioned. After Cooper, McConnel and Ovtchinnikov (2006) proved the existence of such an effect for the first time with decent tests, there has been done more research on this topic. However, there is still discussion about the explanation for the presence of the ‘other January’ effect. This study tries to answer why the ‘other January’ effect is present by testing for the presence of the effect in different sectors in the U.S. and if this presence changes over time. The research question will be answered with an OLS regression.

When portfolio managers and investors know in which sector the effect is present, they can adjust their portfolios to this knowledge and may generate more money. It will be

interesting to see if the effect changes over time, because Schwert (2003) states that after an effect is discovered it will disappear.

In the next part of this study the existing literature and empirical findings on this topic are discussed. In the third section the methodology will be explained and the descriptive statistics are summarized. The fourth section discusses the test results and provides a robustness check. In the last section, the conclusion and the discussion are presented.

2. Literature review

This section will discuss the main theories in the existing literature and what theory predicts regarding the ‘other January’ effect in different U.S. sectors. Next, the empirical findings regarding the ‘other January’ effect are discussed. The last part of this section concludes and summarizes.

(5)

5 2.1 Theories regarding the ‘other January’ effect

The first one to mention the ‘other January’ effect was Yale Hirsch in the 1974 edition of the Stock trader’s almanac. He called it the January barometer and this barometer indicates that as January goes, so goes the year. So, if the return in January is positive then the return for the rest of the year will be higher than when the return in January is negative. He said that this effect was proven in 20 of the last 24 years.

In the first 20 years after Hirsch (1974) mentioned the January barometer only 2 researches are dedicated on this effect. The first one was by Fuller in 1978 and the second was by Bloch and Pupp (1983). They concluded both the same, the January barometer could not be used as a trading strategy to generate excess return in the U.S stock market. However, the tests used in these researches lacked any statistical inference. In the next decade, there was no research on this subject. Hensel and Ziemba (1995) did research on the ‘other January’ effect using data on the S&P 500. They concluded that a trading strategy based on the ‘other January’ effect is only profitable when the return in January is positive and not when it is negative. So, January has predictive power when the return in January is positive and lacks predictive power when January is negative. The conclusion that Hensel and Ziemba (1994) draw can also be doubted, because their paper give no theoretical explanation for their findings.

Brown and Luo (2006) were the next to do a research on the ‘other January’ effect. The data they used was from the NYSE, they used all the listed stocks from 1941-2003. Their overall conclusion was that a trading strategy based on the ‘other January’ effect fails when January is positive, but succeeds when January is negative and someone decides not to invest in the stock market the next 11 months.

The most well-known anomalies in the finance literature tend to disappear or become weaker after they are reported (Schwert, 2003). So, if we follow this rule than the ‘other January’ effect must have vanished away over time, starting from 1974 when Hirsch

reported it. However, Cooper et al. (2006) found that even after Hirsch (1974) reported the anomaly the effect was still present. In contrast to the findings of Cooper et al. (2006) there are the findings of Stivers, Sun and Sun (2009). They conclude that the ‘other January’ effect

(6)

6

is weaker in the period after 1974 than before and called it a ‘temporary anomaly’. More on this in the next paragraph.

The return on a stock portfolio depends on the risk of that portfolio. If the risk is higher, the return must be higher, otherwise the investor will invest in another portfolio with less risk and an equal or higher return (Berk & DeMarzo, 2014, p.327). So, it is possible that the higher excess returns for the months following a positive January is explained by a higher risk for these months. Cooper et al. (2006) looked to this possibility by comparing the volatility and the Sharpe ratio between the 11-month returns following from a positive January and the 11-month returns following from a negative January. Cooper et al. (2006) concluded that the ‘other January’ effect is not present due to risk, as measured by the volatility or the Sharpe ratio of the returns.

2.2 Empirical findings

The first empirical evidence on the existence of the ‘other January’ effect is given by Cooper et al. (2006). They did research on this effect on the stock market in the U.S. In their

research, they compared the excess returns for the rest of the year following from a positive January, with the excess returns for the rest of the year following a negative January. They used the value-weighted returns and the equally-weighted returns in their research. In both cases, they found the same results. Namely, that the January barometer described by Hirsch (1974) is present in the U.S. stock market over the time-period 1940-2003. This time-period was also cut in half to see if the effect was still present after that the January barometer was first mentioned by Hirsch. The periods tested were 1940-1972 and 1973-2003. In both periods, they found significant evidence for the January barometer and they decided to call it the ‘other January’ effect. It is possible that not only January, but every month has

predictive power for the next 11 months. Cooper et al. (2006) also tested this and concluded that January is unique in its power to predict the returns for the following 11 months.

As mentioned in the paragraph before, Stivers et al. (2009) stated that the ‘other January’ effect in the U.S. market-level returns had weakened after 1974. They did their research in 2 different subperiods. The first one from 1940-1974 and the second from 1975-2006. The subperiods are a little bit different from Cooper et al. (2006), the difference is only 2 years.

(7)

7

This may be the reason for the differences in outcomes between both researches. Just like Cooper et al. (2006) they looked at the value-weighted returns and the equally-weighted returns. In both cases the ‘other January’ effect is significant in the first period. In the case of the value-weighted return they find that the effect is still significant in the second period, but it has weakened. However, the effect on the equally-weighted returns is no longer significant in the second period. In this research, only the value-weighted returns are used, so it will be interesting to see if the ‘other January’ effect in different sectors also weakens over time.

After Cooper et al. (2006) proved the existence of the ‘other January’ effect in the U.S. stock market, Bohl and Salm (2009) wanted to test if the ‘other January’ effect also exists in other countries. Their research consisted of 19 major industrialized countries with mature stock markets. These countries were from all over the globe. In only 2 out of the 19 countries they found evidence for the existence of the ‘other January’ effect, these countries are Norway and Switzerland. Bohl and Salm (2009) conclude that the ‘other January’ effect is no international phenomenon. That is why in this research the focus will be only on the different sectors in the U.S.

The existence of the ‘other January’ effect tells not that much, it is only interesting if economic profit can be made based on the sign of the return in January. Marshall and Visaltanachoti (2010) tested if the ‘other January’ effect can be implemented to earn significant returns. They compared an investing method based on the ‘other January’ effect with a simple buy-and-hold strategy for the period 1940-2007. They found that it was not possible to earn significant returns with a strategy based on the ‘other January’ effect and concluded that this strategy underperformed compared to a buy-and-hold strategy. Marshall and Visaltanachoti (2010) give 2 reasons for their conclusion. The first one is that the ‘other January’ effect strategy misses out on the returns in January because it needs January to signal to go short or long on the market. While the returns in January tends to be higher than in other months (Bhardwaj & Brooks, 1992). The second reason is that the ‘other January’ effect gives a signal to go short following a negative January, however the average returns following a negative January is positive. An investor that based his investment on the ‘other January’ effect misses out on these positive returns.

(8)

8

The Halloween effect is another seasonal anomaly. This anomaly says that the stock market returns in the summer months tend to be significantly lower than during the winter months (Bouman & Jacobsen, 2002). The summer months are May-October and the winter months are November-April. In 2006 Jacobsen and Visaltanachoti did another research on the Halloween effect. They wanted to see if there are differences in the Halloween effect across sectors and industries in the time-period 1926-2003. In the main part of their study they look at 17 different sector portfolios. The same 17 sectors that will be compared in this research. They find that there are indeed differences in the Halloween effect across different sectors. Jacobsen and Visaltanachoti (2006) find a weak Halloween effect in the defensive consumer-oriented sectors, the most of them have a short lifespan like food, consumer and utilities. A strong Halloween effect seems to be found in sectors that are related to raw materials and the production process like construction, steel and machines. However, there are

exceptions. The clothes sector shows a strong Halloween effect and the oil sector shows no significant Halloween effect. The conclusion of Jacobsen and Visaltanachoti (2006) is that the Halloween effect is related to different sectors. It will be interesting to see if the January-effect is related to the different sectors in the same way.

2.3 Conclusion on the literature

The ‘other January’ effect was first mentioned by Hirsch in 1974. It took a long time, but in 2006 Cooper et al. were the first to provide any statistical evidence for the existence of this anomaly. Normally an anomaly like this one disappears after it was first mentioned (Schwert 2003), but this one didn’t, according to Cooper et al. (2006). However, Stivers et al (2009) disagree with this and say that the effect at least had weakened after 1974. Marshall and Visaltanachoti (2010) concluded that it was not possible to earn economically and

statistically significant returns with a strategy based on the ‘other January’ effect. The effect is not a worldwide phenomenon (Bohl & Salm, 2009). Another seasonal anomaly, the

Halloween effect, tends to have a strong effect in sectors that are related to raw materials and the production process. This effect is weaker in consumer-oriented sectors (Jacobsen & Visaltanachoti, 2009).

(9)

9 3. Methodology and Data

In this section the research method and the used data are discussed. In the first part the key paper that is used for the methodology and the added features of this research are

discussed as well as the data that is needed to do this research. After that the tests that will be used are shown. In the second part of this section the data sources will be listed and some descriptive statistics are provided.

3.1 Methodology

The key paper that is used to guide this research is the one in which Cooper et al. (2006) proved that the ‘other January’ effect existed in the U.S. over the time-period 1940-2003 as discussed in the literature review. The methods they use in their research will also be used in this research. Only on a few points this research will deviate from the one of Cooper et al. (2006) and some extra features will be added.

Cooper et al. (2006) used in their tests on the ‘other January’ effect the value-weighted (VW) returns and the equal-weighted (EW) returns. In this research, only the value-weighted returns are used, because the value-weighted returns are the appropriate data to test for an anomaly because they replicate investment performance (Bohl & Salm, 2009). Cooper et al. (2006) also tested both the raw-returns and the excess returns, in this research only the excess returns are used. The excess return will be used because advice based on the January barometer would not be especially valuable if the returns are positive, but less than the risk-free rate (Cooper et al., 2006). The excess return will be calculated by subtracting the one month treasury-bill rate from the raw return in that month for a particular sector.

The statistical significance of the ‘other January’ effect in the different sectors will be tested by comparing the average of the 11-month value-weighted excess returns following a

positive January with the average of the 11-month value-weighted excess returns following a negative January. This test is performed by estimating an ordinary least square time-series regression. The monthly excess returns for the rest of the year will be regressed on a binary variable that takes the value 1 if the excess return in January is positive and 0 otherwise. This regression is a simple means test to look at the statistical difference between the 11-month return following a positive January and the 11-month return following a negative January.

(10)

10

This regression will be done for all sectors in the different time-periods. The ‘other January’ effect in a sector is confirmed if the coefficient of the binary variable is statistically

significant. The significance of the coefficient will be tested with a t-test. The t-test is performed at different levels of significance, these are the 10%, the 5% and the 1% level.

For the main results of this research, there are no data points recognized as possible

outliers, even when there are large differences in the 11-month excess returns. Cooper et al. (2006) did research in almost the same time-period, so they must have had the same type of possible outliers. They mentioned no outliers in their paper and kept all their data in the sample, so for the main results of this research also no outliers will be recognized. In the end of this study some possible outliers in the data will be taken out, to see if this affect the results. This will be tested for the complete time-period.

The first feature that is added is that we look to the return on stocks in different sectors and Cooper et al. (2006) only tested the whole American exchange market. In this research the ‘other January’ effect will be tested in the same sectors that Jacobsen and Visaltanachoti (2009) used in their research to the Halloween effect in different U.S. sectors. These are the following 17 sectors: food, mines, oil, clothes, consumer durables, chemicals, consumer, construction, steel, fabricated products, machines, cars, transportation, utilities, retail, financial and other. In each of these sectors the existence of the ‘other January’ effect will be tested in the same way as Cooper et al. (2006) did for the ‘other January’ effect in the U.S. stock market.

The second point where this research deviates from the one of Cooper et al. (2006) is in the time-period. Cooper et al. (2006) did research on the period from 1940-2003 and in this study the time-period 1949-2007 is tested. In this research, the existence of the ‘other January’ effect is also tested for 3 sub-periods in this timespan. It can be interesting to look at different sub-periods because the way of investing changes over time. First the complete period will be examined. After that the ‘other January’ effect in the period from 1949 to 1971 is tested. This period is tested because it represents the postwar period. This period start at the point where the Marshall plan was implanted in Europe, this was in 1948 (The Marshall Plan: lessons learned for the 21st_{century, 2008). This period starts in the year after}

(11)

11

the official starting point because this research is on an anomaly that influences a complete year starting in January. So, it makes sense that each sub-period also starts in January. The second event that influenced the way of investing is the availability of new technologies. The starting point for this period will be February 1971 when NASDAQ introduced the first automatic quoting system (Santomero, 1974). Again, the period will start in the first complete year after the initial point, so in this case 1972. This sub-period will end in 1997, because in 1998 the last period will start. The last period will cover the effects of the dot-com bubble. The so-called bubble period started in 1998 (Ghosh, 2006). This period will start in 1998 because the growth of the bubble started at the beginning of this year, so it is not necessary to start this period at the beginning of the next year. This period will end in 2007 because in September 2008 the credit crisis started with the bankruptcy of Lehmann Brothers (Tietje & Lehmann, 2010). So, the last period tested is from 1998-2007.

First the data for the average-weighted returns for the different portfolio sectors are needed. Then the one month treasury-bill rates, to calculate the excess value-weighted returns, are needed. When the excess value-weighted returns are calculated, the data needs to be divided. On the one hand the value-weighted excess returns in January, to determine the value of the binary variable, and on the other hand the value-weighted excess returns for the rest of the year.

The model that is used to test the statistical significance of the ‘other January’ effect is as follows:

= + +

Where is the dependent variable and stands for the value-weighted excess return in the months from February till December. Here is the constant and is the coefficient for the variable Jan. The independent variable Jan is a binary variable that takes the value 1 if the value-weighted excess return in that year is positive and 0 otherwise. The Ɛ is the error term of the model. There is no need for control variables in this model because the research in on an anomaly. Normally there are no explanations for these anomalies, so there will be no explanatory variables.

(12)

12

The null-hypothesis will be rejected if the coefficient for the binary variable January is significant. Then the alternative hypothesis is accepted and the ‘other January’ effect is confirmed for that sector in that time-period. To decide if a coefficient is significant, the t-statistic will be used at different significance levels, these are the 10%, the 5% and the 1% level. When the p-value for the t-statistic is lower than 0.10, 0.05 or 0.01 the coefficient is significant.

After the sectors with evidence for the ‘other January’ effect are determined, it is time to test if it is possible to make significantly more money with a portfolio that is based on these sectors. This test will be done by comparing the average return of a buy-and-hold strategy with the average return of an actively managed portfolio based on the ‘other January’ effect, just like Marshall and Visaltanachoti (2010). The averages are calculated for the complete period 1949-2007. The ‘other January effect portfolio’ is managed by investing for the remainder of the year in a sector portfolio that has a positive return in January and ignore a sector portfolio for 11 months when the return in that sector is negative in January. In this comparison, only the sectors that provide evidence for the ‘other January’ effect in the period 1949-2007 will be used. The difference in the average return between the two strategies will be tested with a one-sided t-test and a significance level of 10%.

3.2 Data and descriptive statistics

The data that is used for this research is the value-weighted excess returns in the 17

different sectors. The average value-weighted returns for this sectors are obtained from Ken French’s website. On his website, French made the returns available for sector-based

portfolios. The value-weighted returns on these portfolios are used in this research. The one month treasury-bill rate, needed to calculate the excess returns, are also from French’s website.

For the complete period from 1949-2007 there are 59 observations for both the dependent and the independent variable, this is the case for all 17 sectors. In table 1 below, the

characteristics of the data for the complete time-period are summarized. In the table can be seen that the means of the returns from February till December are positive in every sector. The lowest mean belongs to the steel sector and is 4.04%, while the highest mean belongs to

(13)

13

the oil sector and equals 9.52%. There are big differences in the 11-month excess return between the years in the data. So was the minimum 11-month excess return over the complete time-period -62.53% and the maximum 11-month excess return 65.56%. The minimum excess return belongs to the clothes sector and the maximum excess return was measured in the steel sector. The transport sector had the most positive Januarys, namely 42. The chemicals sector had the most negative Januarys, it is the only sector with more negative than positive Januarys for the complete time-period. It had 29 positive Januarys and 30 negative Januarys.

Table 1: Summary statisitcs 1949-2007

Sector Mean Median Standard

deviation Minimal value Maximal value Positive January Negative January Food 7.88 6.66 14.24 -40.09 38.12 33 26 Mines Oil Clothes Consumer durables Chemicals Consumer 6.26 9.52 4.56 4.67 6.99 8.78 6.68 11.43 7.04 8.03 8.73 8.21 20.11 15.82 22.03 17.82 15.40 14.64 -32.12 -31.77 -62.53 -62.20 -34.35 -25.25 62.87 41.93 40.51 39.81 37.95 46.11 37 32 36 40 29 35 22 27 23 19 30 24 Construction 6.04 8.54 17.83 -41.70 47.52 38 21 Steel 4.04 3.89 21.90 -43.2 65.56 39 20 Fabricated products 5.86 6.58 14.03 -40.11 39.6 34 25 Machines 6.80 7.14 19.42 -43.94 54.26 37 22 Cars 5.61 5.98 22.81 -50.04 46.41 35 24 Transport 5.73 7.67 18.00 -49.65 47.87 42 17 Utilities 6.08 7.45 13.33 -30.86 33.15 38 21 Retail 6.88 8.15 18.05 -52.45 44.13 38 21 Financials 7.63 8.92 16.46 -44.01 40.88 35 24 Other 5.55 7.64 15.01 -35.94 35.23 37 22

In the first sub-period from 1949-1971 there are 23 observations for each sector. These 23 observations are for the dependent as for the independent variable. In table 2 the summary statistics for the first sub-period are listed. Again, there can be seen that the means in all sectors are positive, only they are a bit higher than in the complete period. This time the lowest mean is for the clothes sector and is 5.82% and the highest mean is for the machines sector and equals 12.87%. In this period the lowest minimum return over the months

(14)

14

February till December belongs to the transport sector and is -49.65%. The highest maximum return for the months February till December is for the steel sector and equals 57.81%. In the first sub-period the sector with the most positive Januarys is the transport sector, this sector has 18 positive Januarys. The chemicals sector is again the only sector that has more negative than positive returns in January. This time, 11 positive returns and 12 negative returns are measured in January.

Table 2: Summary statistics 1949-1971

deviation Minimal value Maximal value Positive January Negative January Food 7.91 6.66 11.60 -12.48 31.33 14 9 Mines Oil Clothes Consumer durables Chemicals Consumer 6.29 10.58 5.82 11.71 8.61 11.55 10.93 12.65 9.83 12.64 10.09 11.01 18.70 16.64 21.64 16.25 18.04 13.70 -32.12 -31.77 -37.93 -20.55 -34.35 -22.80 38.79 41.93 34 39.81 37.95 46.11 16 14 16 17 11 13 7 9 7 6 12 10 Construction 8.29 6.75 16.59 -22.09 43.89 14 9 Steel 6.28 4.10 23.99 -30.30 57.81 17 6 Fabricated products 7.74 10.01 13.96 -14.12 31.63 12 11 Machines 12.87 12.41 14.84 -15.24 46.81 13 10 Cars 12.37 17.80 22.18 -38.79 46.41 12 11 Transport 6.73 7.59 21.06 -49.65 47.87 18 5 Utilities 6.84 5.39 11.68 -22.76 27.47 17 6 Retail 10.27 10.43 15.23 -26.64 37.68 15 8 Financials 10.17 7.38 15.60 -18.04 40.88 14 9 Other 7.87 8.53 11.93 -12.55 35.23 13 10

In the second sub-period from 1972-1997 there are 26 observations per sector, the summary statistics for these observations are listed in table 3. Again, for every sector the mean is positive, but the means are lower than in the periods discussed before. The means for the 11-month returns in this period are between 0.16% and 8.73%. The lowest mean in this period is, just as in the complete period, for the steel sector. The highest mean in this sub-period is for the food sector with 8.73%. The minimum 11-month excess return in this time-period is -62.53% and belongs to the clothes sector. The highest 11-month excess return for this period is for the mine sector and equals 46.74%. There are 2 sectors that share the most

(15)

15

Januarys with a positive return in this sub-period. These are the fabricated products sector and again the transport sector. Just like in the time-periods before the oil sector has the most negative Januarys, this time they have 12.

deviation Minimal value Maximal value Positive January Negative January Food 8.73 9.24 17.41 -40.09 38.12 16 10 Mines Oil Clothes Consumer durables Chemicals Consumer 2.14 7.14 3.02 0.83 5.09 8.12 3.94 7.61 7.36 5.14 7.69 5.98 18.76 15.44 24.24 19.10 13.73 15.36 -25.92 -24.78 -62.53 -62.20 -23.80 -25.25 46.74 37.93 40.51 28.34 28.77 35.52 16 14 15 17 16 16 10 12 11 9 10 10 Construction 4.49 10.39 16.27 -41.70 26.60 18 8 Steel 0.16 2.67 12.68 -26.63 21.45 17 9 Fabricated products 3.72 6.61 13.17 -40.11 22.03 19 7 Machines 2.13 1.67 15.98 -43.94 31.38 17 9 Cars 1.45 2.99 22.80 -50.04 42.01 18 8 Transport 4.63 9.89 17.17 -35.31 34.19 19 7 Utilities 4.06 7.30 13.65 -30.86 19.78 17 9 Retail 4.47 6.11 20.57 -52.45 44.13 16 10 Financials 6.22 9.62 17.79 -44.01 32.92 17 9 Other 5.01 8.02 13.56 -35.94 23.84 18 8

Over the last sub-period there are 10 observations per sector, this period is from 1998-2007. In this last period, the means of the 11-month excess returns are no longer all positive, as can be seen in table 4. This time the lowest mean for the 11-month excess return is in the consumer durables sector and is negative, namely -1.56%. The highest mean for the 11-month excess returns in this period is higher than is seen in periods before, it is found in the mines sector and equals 16.91%. The minimum 11-month excess return in this time-period is -41.79%, while the maximum 11-month excess return equals, just like in the complete time-period, 65.56%. The minimum can be found in the machine sector, while the maximum belongs to the steel sector. This time the sector with the most Januarys with positive returns is not the transport sector, but both the retail sector and the machines sector. They have

(16)

16

both 7 Januarys with a positive return. Just like in the time periods before, the chemicals sector has the most negative Januarys, in this time-period there are 8.

deviation Minimal value Maximal value Positive January Negative January Food 5.63 7.25 11.29 -14.56 23.33 3 7 Mines Oil Clothes Consumer durables Chemicals Consumer 16.91 13.27 5.68 -1.56 8.20 4.14 13.75 20.51 0.74 -0.32 5.42 3.24 24.46 15.48 18.40 13.26 13.78 14.89 -19.90 -14.97 -18.04 -18.71 -10.13 -18.89 62.87 27.51 36.36 20.85 31.17 26.16 5 4 5 6 2 6 5 6 5 4 8 4 Construction 4.90 8.97 24.88 -33.98 47.52 6 4 Steel 8.97 13.38 33.96 -43.2 65.56 5 5 Fabricated products 7.13 5.32 16.93 -24.55 39.6 3 7 Machines 4.94 5.68 31.96 -41.79 54.26 7 3 Cars 0.92 -6.19 22.70 -21.73 44.61 5 5 Transport 6.25 4.25 13.34 -10.92 28.69 5 5 Utilities 9.59 12.22 16.34 -19.07 33.15 4 6 Retail 5.34 4.81 17.59 -22.48 38 7 3 Financials 5.48 8.42 15.68 -19.50 30.03 4 6 Other 1.63 4.34 23.70 -35.76 29.27 6 4 4. Analysis

In the first part of this section the main empirical results regarding this research are presented. In the second part the robustness check and some additional results are presented.

4.1 Empirical results

In table 5 below, the empirical results for the complete time-period from 1949-2007 are presented. In this table can be seen that there are differences in the presence of the ‘other January’ effect in the different sectors. For the food-, oil- and chemical sector there is only statistical evidence for the ‘other January’ effect at the 10% significance level. When the significance level is lowered to the 5% level there is enough evidence for the ‘other January’ effect to be present in the following sectors: consumer durables, consumer, fabricated

(17)

17

products, retail, financial and other. For these sectors, only in the consumer durables sector and the consumer sector there is enough statistical evidence for the presence of the ‘other January’ effect at the 1% significance level.

The interpretation of the results in table 5 is as follows. In the food sector the coefficient for the dummy variable January is 6.87. This means that the average return in the rest of the year following a positive January is on average 6.87% higher than following a negative January. The excess return for the 11-months following a positive January is on average 10.91%, while this for the 11-months following a negative January 4.04% is. This difference is significant at the 10% significance level. The t-statistic for the coefficient of January is 1.91 with a p-value of 0.061. In the other sectors that are significant at the 10% level the differences in the 11-month returns are a bit larger. In the oil sector the 11-month return following a positive January is 7.60% higher than the 11-month return following a negative January. In the chemicals sector this difference is 7.45% in favor of the 11-months following a positive January.

For the sectors that are significant at the 5% level the coefficient for January is higher than for the sectors that are only significant at the 10% level. This means that the differences in the 11-month return between a positive January and negative January are bigger. In the fabricated product sector the return following a positive January is on average 8.56% higher than the returns following a negative January. For the Retail sector this difference is 11.54%. While the effect in the financial sector and other sector is 9.75% and 8.50% respectively.

The highest impact of the ‘other January’ effect is measured in the Consumer durables sector. The difference in the 11-month return following a positive January and a negative January is in this sector 16.55%. The return in the 11 months following a positive January is in the consumer durables sector on average 9.99%, while the average 11-month return following a negative January is negative with -6.56%. In the other sector that is significant at the 1% level, the consumer sector, the effect is weaker. The 11-month return following a negative January is 2.32% and the 11-month return following a positive January is 13.20%. This gives a spread of 10.88%.

(18)

18 Table 5: Empirical results 1949-2007

Sector Return negative January Return positive January

Spread Robust SE t-statistic p-value

Food* 4.04 10.91 6.87 3.59 1.91 0.061 Mines 3.62 7.84 4.22 5.69 0.74 0.461 Oil* 5.40 13.00 7.60 4.15 1.83 0.072 Clothes 0.27 7.29 7.02 5.94 1.18 0.242 Consumer durables*** -6.56 9.99 16.55 5.00 3.31 0.002 Chemicals* 3.33 10.78 7.45 3.91 1.90 0.062 Consumer*** 2.32 13.20 10.88 3.67 2.97 0.004 Construction 2.51 7.99 5.48 5.18 1.06 0.294 Steel -0.86 6.55 7.41 6.28 1.18 0.243 Fabricated products** 0.93 9.49 8.56 3.76 2.28 0.027 Machines 3.37 8.84 5.47 5.37 1.02 0.313 Cars -0.13 9.56 9.69 6.16 1.57 0.121 Transport 3.50 6.63 3.13 4.99 0.63 0.533 Utilities 3.14 7.71 4.57 3.57 1.28 0.206 Retail** -0.55 10.99 11.54 4.44 2.60 0.012 Financials** -1.85 7.90 9.75 4.30 2.27 0.027 Other** 0.22 8.72 8.50 4.07 2.09 0.041

*significant at 10% level, ** significant at 5% level, ***significant at 1% level.

In table 6 the empirical results for the first sub-period from 1949-1971 are presented. There are some differences in the statistical evidence for the presence of the ‘other January’ effect compared to the complete period. The coefficient for the sectors mines, clothes, machines and cars weren’t significant in the complete period, but in the first sub-period they are. This indicates that the ‘other January’ effect won’t be significant in the following sub-periods. The coefficient for January is in the machines sector significant at the 10% level, in the mines sector at the 5% level and in the cars and the clothes sectors even at the 1% level. The other sectors where there is statistical evidence for the ‘other January’ effect are the consumer durables, the fabricated products, the retail and the financial sector. In the fabricated products and the financial sector there is evidence at the 10% significance level, in the consumer durables sector at the 5% level and in the retail sector at the 1% level.

The largest spread in this sub-period belongs to the cars sector and is 26.24%. The average 11-month return following a negative January is -1.32% and following a positive January it is 24.92%. For the clothes sector and the retail sector, the other sectors that are significant at

(19)

19

the 1% level, the spread between negative Januarys and positive Januarys is lower. The spread is 22.77% in the clothes sector and 20.94% in the retail sector. This means that the effect in these sectors is weaker than in the cars sector.

The spread in the mines sector is 20.20%. The average 11-month return following a negative January is -7.76%, while following on a positive January the 11-month return averages 12.54%. In the other sector that is significant at the 5% level, consumer durables, the

difference in the 11-month return is 15%. The average 11-month return is 0.63% following a negative January and 15.63% following a positive January.

In the sectors that are only significant at the 10% level the largest spread is in the financial sector. The average 11-month return following a negative January is 2.68%, while the

average 11-month return following a positive January equals 14.99%. This lead to a spread of 12.31%. The spread in the fabricated products sector is 9.89% and in the machines sector the spread is 10.35%.

Remarkable is that the utility sector is the only sector that has a higher 11-month return following a negative January than following a positive January in this period. The return following a negative January is 7.30% and following a positive January it is 6.69%. This leads to a negative spread of -0.61%.

In the second sub-period from 1972-1997 there is for none of the sectors evidence for the ‘other January’ effect at the 10% significance level, as can be seen in table 7. There is evidence for the ‘other January’ effect at the 1% significance level in the consumer sector and at the 5% level in the consumer durables, the fabricated products and the financial sector. In comparison with the sub-period before, the spreads are now all positive.

The consumer sector is the only one that has statistical evidence for the ‘other January’ effect at the 1% significance level. The average 11-month return following a positive January is 14.09%, while the average return following a negative January is -1.45%. This leads to a spread of 15.44%.

(20)

20 Table 6: Empirical results 1949-1971

Food 2.98 11.09 8.11 5.02 1.62 0.121 Mines** -7.76 12.54 20.20 7.75 2.61 0.016 Oil 7.07 12.84 5.77 7.83 0.74 0.469 Clothes*** -10.02 12.75 22.77 7.57 3.01 0.007 Consumer durables** 0.63 15.63 15.00 6.43 2.33 0.030 Chemicals 2.79 14.96 12.17 7.09 1.72 0.101 Consumer 7.71 14.50 6.79 5.64 1.21 0.242 Construction 4.34 10.82 6.48 7.29 0.89 0.384 Steel -6.15 10.68 16.83 10.40 1.62 0.120 Fabricated products* 2.58 12.47 9.89 5.52 1.79 0.088 Machines* 7.02 17.37 10.35 5.94 1.74 0.096 Cars*** -1.32 24.92 26.24 7.71 3.40 0.003 Transport -1.72 9.08 10.80 8.69 1.24 0.228 Utilities 7.30 6.69 -0.61 4.77 -0.13 0.899 Retail*** -3.39 17.55 20.94 5.19 4.03 0.001 Financials* 2.68 14.99 12.31 6.10 2.02 0.057 Other 3.32 11.37 8.05 4.75 1.70 0.105

Table 7: Empirical results 1972-1997

Food 4.48 11.39 6.91 6.48 1.07 0.297 Mines 0.57 3.11 2.54 7.51 0.34 0.738 Oil 2.31 11.29 8.98 6.00 1.50 0.147 Clothes -0.62 5.69 6.31 9.93 0.64 0.531 Consumer durables** -12.03 7.63 19.66 8.32 2.36 0.027 Chemicals 1.26 7.48 6.22 5.07 1.23 0.232 Consumer*** -1.45 14.09 15.54 5.48 2.84 0.009 Construction -2.00 7.37 9.37 6.74 1.39 0.178 Steel -4.55 2.66 7.21 5.59 1.29 0.210 Fabricated products** -7.91 8.00 15.91 6.17 2.58 0.016 Machines -4.24 5.50 9.74 7.47 1.30 0.204 Cars -4.30 4.00 8.30 10.68 0.78 0.445 Transport -1.52 6.90 8.42 7.75 1.09 0.288 Utilities 0.39 6.01 5.62 5.45 1.03 0.312 Retail -1.92 7.33 9.25 7.68 1.20 0.240 Financials** -4.00 11.63 15.63 7.33 2.13 0.043 Other 0.86 6.86 6.00 5.73 1.05 0.305

(21)

21

The spread is larger in the sectors that are significant at the 5% level. This is possible because the robust standard errors are larger in these sectors, what leads to a lower t-statistic and thus a higher p-value. The spread in the consumer durables sector is the largest with 19.66%. In this sector the 11-month return following a negative January is -12.03% and following a positive January 7.63%. In the fabricated products sector the spread is 15.91%, with an average 11-month return of -7.91% following a negative January and 8% following a positive January. In the financial sector the average 11-month return following a negative January is -4% and following a positive January 11.63%. That leads to a spread in this sector of 15.63%.

The last sub-period is a special case. The outcomes are different from the ones found before. The empirical results are presented in table 8. The coefficient for January is statistically significant in the clothes and the utilities sector at the 10% significance level and in the transport sector at the 5% level. However, the statistical significances of these coefficients in the clothes and the transport sector does not provide evidence for the ‘other January’ effect in these sectors, since the spread is negative. This gives statistical evidence for a contrary effect, where a negative return in January indicates a positive return in the rest of the year and vice versa. The coefficient for January is significant at the 1% level for none of the sectors.

The only sector that has evidence for the presence of the ‘other January’ effect in this time-period is the utilities sector. The 11-month return following a negative January in this sector averages 3.11% and following a positive January 19.30%. This gives a spread of 16.19%.

In the sectors where there is found statistical evidence for the contrary of the ‘other January’ effect in the last sub-period, the clothes and the transport sector, the spreads are negative. In the clothes sector the spread is -21.97%. The average 11-month return following a negative January is 16.66% and following a positive January -5.31%. In the transport sector this spread is smaller. The 11-month return averages 15.74% following a negative January and -3.25% following a positive January. The total spread equals -18.99%.

The different outcomes compared with the outcomes before may occur because of the length of this time-period compared with the others. The last sub-period is way shorter and

(22)

22

consists of only 10 observations per sector. The periods discussed before had at least 23 observations. The less amount of observations leads to higher standard errors and less precise estimations. It is also possible that the ‘other January’ effect diminishes over time like Schwert (1974) stated. It is not sure where these different outcomes come from. This is a limitation of this research.

Table 8: Empirical results 1997-2007

Food 4.78 7.60 2.82 6.54 0.43 0.677 Mines 25.63 8.19 -17.44 15.20 -1.15 0.285 Oil 9.07 19.56 10.49 8.87 1.18 0.271 Clothes* 16.66 -5.31 -21.97 9.60 -2.29 0.051 Consumer durables -5.02 0.75 5.77 9.62 0.60 0.565 Chemicals 6.71 14.14 7.43 7.35 1.01 0.342 Consumer -1.71 8.04 9.75 9.98 0.98 0.357 Construction 7.39 3.24 -4.15 18.80 -0.22 0.831 Steel 12.15 5.79 -6.36 22.67 -0.28 0.786 Fabricated products 7.18 7.00 -0.18 8.20 -0.02 0.983 Machines 13.98 1.06 -12.92 23.45 -0.55 0.597 Cars 9.15 -7.31 -16.46 14.07 -1.17 0.276 Transport** 15.74 -3.25 -18.99 5.92 -3.21 0.012 Utilities* 3.11 19.30 16.19 8.62 1.88 0,097 Retail 9.23 3.67 -5.56 11.82 -0.47 0.651 Financials 9.39 -0.38 -9.77 9.67 -1.01 0.342 Other -8.79 8.57 17.36 16.20 1.07 0.315

Now it is time to see if it is possible to earn significantly money with the information about the ‘other January’ effect based on these results. The sector portfolios considered in this test are the ones that are significant in the time-period 1949-2007. The average return of the buy-and-hold strategy is 8.17%. This is lower than the average return of the strategy based on the ‘other January effect’, which has an average return of 9.04%. The spread between the two is thus 0.87%. The t-statistic for this spread is 0.55 and has a p-value of 0.276, this means that the spread is not significantly higher than 0 at the 10% significance level. So, there is no evidence that the strategy based on the ‘other January’ effect in different sectors leads to higher profits than a simple buy-and hold strategy. There should be considered that

(23)

23

the transaction costs are ignored. This has no impact on the outcome that the strategy based on the ‘other January’ effect does not perform better than the buy-and-hold strategy

because the transaction costs in the buy-and-hold strategy are always lower than in an actively managed portfolio (Marshall & Visaltanachoti, 2010).

4.2 Robustness check

A part of the robustness check is already done by considering different sub-periods of time. The rest of the robustness check will be done by doing the same regression (OLS) as before, only this time the excess equally-weighted return will be used instead of the excess average-weighted returns. Cooper et al. (2006) also used the equal-average-weighted return in a part of their research. The model remains the same as in section 3.1, only the definitions of the variable

and the independent variable will change. will now be the excess equal-weighted return for February-December and will be a dummy variable that takes the value 1 if the excess equal-weighted return in January is positive and 0 otherwise. This check is done for the complete timespan from 1949-2007, the results are shown in the table 9 below.

The outcomes of the robustness check with equal-weighted returns are slightly different from the outcomes with value-weighted returns. The first difference is that there is no longer evidence for the presence of the ‘other January’ effect in the food sector. The second difference that draws attention is that there is evidence for the ‘other January’ effect in the clothes sector at the 5% significance level. With the value-weighted return there was no evidence for the ‘other January’ effect in this sector.

Furthermore, there are some differences in the level of significance for some sectors. When the value-weighted returns are used the consumer durables and the consumer sector are significant at the 1% level and with the equal-weighted returns only at the 10% level. Just as the fabricated products and the retail sector, before they were significant at the 5% level and now only at the 10% level. In some sectors the evidence for the ‘other January’ effect became stronger. The sectors included are the oil, the chemical and the financial sector. The oil sector goes from significant at the 10% level to significant at the 1% level, the chemical

(24)

24

sector goes from significant at the 10% level to significant at the 5% level and the financial sector goes from significant at the 5% level to significant at the 1% level.

Table 9: robustness check 1949-2007

Food 0.37 5.88 5.51 6.49 0.85 0.399 Mines -5.53 7.47 13.00 8.25 1.58 0.121 Oil*** -5.12 16.29 21.41 7.36 2.91 0.005 Clothes** -11.87 6.48 18.35 7.81 2.35 0.022 Consumer durables* -9.81 6.22 16.03 8.11 1.98 0.053 Chemicals** -1.06 10.40 11.46 4.97 2.31 0.025 Consumer* -1.03 10.58 11.61 6.17 1.88 0.065 Construction -4.03 5.82 9.85 8.26 1.19 0.238 Steel -6.39 7.03 13.42 8.03 1.67 0.100 Fabricated products* -5.02 7.77 12.79 6.81 1.88 0.066 Machines -0.82 8.37 9.19 8.10 1.13 0.261 Cars -10.11 6.53 16.64 10.27 1.62 0.111 Transport -0.41 6.29 6.70 8.14 0.82 0.414 Utilities 4.10 8.49 4.39 3.48 1.26 0.213 Retail* -7.21 7.65 14.86 8.54 1.74 0.087 Financials*** -7.24 11.51 18.75 5.88 3.19 0.002 Other** -7.60 9.23 16.83 7.00 2.41 0.019

Now some possible outliers are taken out of the dataset to see if this changes the results. The years 1973 and 1974 are taken out of the dataset for every sector, because the returns in these years, except for January, are extremely negative in every sector, possibly due to the oil crisis. Furthermore, in specific sectors some years are taken out of the dataset because the return in that year has at least one month that has a weird return. These are the years in the following sectors: 1955 in cars, 1958, 1999 and 2000 in steel, 1961 in consumer durables, 1970, 1973 and 1987 in clothes, 1973 in construction, 1987 in retail, 1998 in utilities, 2001 in consumer and 2005 in oil.

The results with outliers differ not much from the results without outliers for the complete time-period. No additional sectors became significant with outliers, only the oil sector is no longer significant. However, the level of significance changes for some sectors. Food and

(25)

25

chemicals went from significant at the 10% level to significant at the 5% level. The retails sector is now significant at the 1% level instead of the 5% level. The fabricated product and the financial sector are now only significant at the 10% level, without outliers they were significant at the 5% level. The consumer sector went from significant at the 1% level to significant at the 5% level. The results for the other sectors do not change with possible outliers taken out.

5. Conclusion and discussion

In this study the existence of the ‘other January’ effect in different U.S. sectors is examined with an OLS-regression, this is done for various time-periods. The excess returns in the months February-December are regressed on a binary variable for January that takes the value 1 if the excess return in January is positive and 0 otherwise. Over the complete time-period there is found evidence for the ‘other January’ effect in 9 of the 17 sectors tested in this study. These are the following sectors: food, oil, consumer durables, chemicals,

consumer, fabricated products, retail, financial and other. Thereafter the complete time-period was split into 3 different sub-time-periods. In the first sub-time-period from 1949-1971 there is found evidence for the ‘other January’ effect in the following 8 sectors: mines, clothes, consumer durables, fabricated products, machines, cars, retail and financial. In the second sub-period from 1972-1997 there are less sectors that provide evidence for the ‘other January’ effect, namely 4. These are the consumer durables, the consumer, the fabricated products and the financial sector. In the last period from 1998-2007 there is only in the utilities sector evidence for the presence of the ‘other January’ effect. The effect is more likely to be opposite in this time-period. The robustness check provides more or less the same results as the initial method. A strategy based on the findings for the 1949-2007 period does not give a probability to earn more money compared to a simple buy-and-hold

strategy.

The sectors that contain evidence for the ‘other January’ effect are the consumer-oriented sectors like food, consumer durables and consumer. While the effect seems to be absent in sectors that are related to the production process like construction, steel and transport. This is in contrast with the findings of Jacobsen and Visaltanachoti (2009) in their study on the Halloween effect in U.S. sectors. The findings were expected to be equal because the return

(26)

26

of a sector portfolio is expected to be based on sector-specific characteristics. That the ‘other January’ effect weakens over time is line with the theory of Schwert (2003) and the findings of Stiver et al. (2009). They predicted that the ‘other January’ effect will weaken or disappear over time. In line with the findings of Marshall and Visaltanchoti (2010) it is not possible to earn extra profit by managing a portfolio based on the existence of the ‘other January’ effect in different sectors.

This study contains some limitations. The first one is the last sub-period contains only 10 observations. This makes that the estimations are not very precise and maybe biased. The results in this period should be interpreted with caution. The second limitation is that this study tested the ‘other January effect’ in 17 different sectors, due to data availability, but there is data available for more sectors in more recent time-periods.

Investors and portfolio managers could take these results into account in their decision-making process. They can decide to add or leave out an investment opportunity in their portfolio based on the return of a stock in January. However, this needs to be done with caution because the results show that the effect changes over time. Besides that, it may be better to invest in a simple buy-and-hold strategy because a portfolio based on the ‘other January’ effect in different sectors does not generate significantly more money.

For future research, it may be interesting to test the ‘other January’ effect for more than 17 sectors in a more recent time-period, this could lead to a better understanding of the ‘other January’ effect. The question why this effect is present is still unanswered. Future research should try to answer this. It will be also interesting to see if the ‘other January’ effect still exists and in which sectors in the time-period after the credit crisis. Furthermore, it will be interesting to test if it is possible to make significantly more profit with a strategy based on the ‘other January effect’ in different sectors with 12-month returns instead of 11-month returns. In this way, the mostly higher January returns are captured in the strategy based on the ‘other January’ effect in different sectors.

(27)

27 Reference list:

Berk, J., DeMarzo, P., (2014). Corporate Finance. United Kingdom: Pearson Education Limited.

Bhardwaj, R.K., & Brooks L.D. (1992). The January Anomaly: Effects of Low Share Price, Transaction Costs, and Bid-Ask Bias. Journal of Finance, 47(2), 553-575.

Bloch, H., & Pupp, R., (1983). The January Barometer revisited and rejected. The Journal of Portfolio Management, 9(2), 48-50.

Bohl, M. T., & Salm, C.A. (2010). The other January effect: International evidence. The European Journal of Finance, 16(2), 173-182.

Bouman, S., & Jacobsen, B. (2002). The Halloween Indicator, Sell in May an Go Away: Another Puzzle. American Economic Review, 92(5), 1618-1635.

Brown, l., & Luo, L. (2006). The January barometer: further evidence.(trading analysis). Journal of investing, (15)1, 25-31.

Cooper, M.J., McConnel, J.J., & Ovtchinnikov, A.V. (2006). The other January effect. Journal of Financial Economics, 82(2), 315-341.

Fuller, R.J. (1978). The January Barometer: What’s its batting average? The Journal of Portfolio Management, 4(2), 5-7.

Ghosh, A. (2006). The IPO phenomenon in the 1990s. The Social Science Journal, 43(3), 487 -95.

Hensel, C.R., & Ziemba, W.T. (1995). The January Barometer. Journal of investing, 4(2), 67-70.

Hirsch, Y., 1974. Stock Trader’s Almanac. The Hirsch Organization, Nyack, NY.

Jacobsen, B., & Visaltanachoti, N. (2009). The Halloween effect in U.S. sectors. Financial Review, 44(3), 437-459.

Lehmann, M., & Tietje, C. (2010). The Role and Prospects of International Law in Financial Regulation and Supervision. Journal of International Economic Law, 13(3), 663-682. Marshall, B.R., Visaltanachoti, N. (2010). The Other January Effect: Evidence against market

efficiency?. Journal of Banking and Finance, 34(10), 2413-2424.

Organisation For Economic Co-Operation And Development., (2008). The Marshall plan lessons learned for the 21st_{century. Paris: OECD Publishing.}

(28)

28

Santomero, A.M. (1974). The Economic Effects of NASDAQ: Some Preliminary Results. Journal of Financial and Quantitative Analyses, 9(1), 13-24.

Schwert, G.W., 2003. Anomalies and market efficiency. In: Constantinides, G., Harris, M., Stulz, R. (Eds.), Handbook of the Economics of Finance. North-Holland, Amsterdam, pp. 937–972.

Stivers, C., Sun, L., & Sun., Y (2009). The other January effect: International, style and subperiod evidence. Journal of Financial Markets, 12(3), 521-546.

The other January effect in U.S. sectors

The other January effect in U.S. sectors

Table of contents