The existence and historical development of the holiday effect on the Swedish stock market

ABSTRACT

This paper examines the holiday effect on the Swedish stock market over a 40-year period. We use a regression-based approach on daily price data to ascertain if the holiday effect is present on the Swedish stock market, analyse its historical development using 10-year subsamples, and assess whether its effects vary for different holidays. We find evidence for a positive post-holiday effect using the full sample period. When looking at the subsamples, however, we only find evidence for its existence in the 1990s and 2000s. We do not find evidence for the existence of a pre-holiday effect for any period. No holiday, considered by itself, shows evidence of a pre-holiday effect over the full sample period. For the holidays included, we only find evidence of a post-holiday effect after New Year’s.

KEYWORDS:

• Swedish stock market
• holiday effect
• stock market anomaly
• market efficiency

JEL CLASSIFICATION:

• G10
• G12

I. Introduction

A growing body of research presents evidence that the price movements of stock markets are subject to abnormalities that defy the predictions of the efficient market hypothesis (see, e.g. Rossi 2015 for a comprehensive literature review). A particular type of market anomaly is the holiday effect that describes the tendency for stock markets to showcase abnormal returns on trading days adjacent to holidays. Evidence for the holiday effect was first documented on the US markets (see, e.g. Lakonishok and Smidt 1988) and later found in, among others, China, Hong Kong, Japan, Malaysia, and South Korea (Yuan and Gupta 2014), and Central and Eastern Europe (Dodd and Gakhovich 2011). However, Cadsby and Ratner (1992) found no evidence for a pre-holiday effect in the UK, Switzerland, West Germany, and France in the 1980s, suggesting that developed European countries might not experience a pre-holiday effect. Various explanations have been proposed for its existence, including investor exuberance and overconfidence (Frieder and Subrahmanyam 2004) or habits of preferred trading times (Ariel 1990).

The existence and strength of a holiday effect is an indicator of market efficiency, which reflects the extent of the incorporation of information into prices (Fama 1998). As market anomalies defy the tenets of various asset pricing models and lead to increased risk of suboptimal asset allocation, further research into anomalies contributes to the testing and development of pricing models and to a judgment of market efficiency. Hitherto no study has examined the holiday effect on the Swedish stock market. Studying the Swedish market complements the existing literature by providing an understanding of the existence and development of the holiday effect in Sweden. Thus, it contributes to a judgment of the overall efficiency of the Swedish stock market, provides additional evidence to test explanatory hypotheses about the holiday effect, adds to the understanding of the holiday effect in developed European countries, and elucidates possible investment opportunities. Moreover, Sweden as a country with generous annual vacation policies makes an analysis of such policies’ impact on the holiday effect possible. By studying the historical development, we also provide insights into whether the information liberalization and increased market attention concomitant with the burgeoning of the internet and information technology in the 1990s affected the holiday effect. The aim of this study is thus to ascertain whether the holiday effect exists on the Swedish stock market, assess how it has developed over time, and whether differences in its impact during various holidays exist.

II. Materials and methods

As the basis for our analysis, we collected daily price data for the AFGX stock market index from 1980–01-01 to 2019–12-31. The AFGX is a value-weighted index that covers the entire Stockholm Stock Exchange (Affärsvärlden n.d.). To calculate daily returns, we let $P_t$ be the price of the AFGX at time t and define returns as

(1) $r_t=\frac{P_t-P_{t-1}}{p_{t-1}}\times 100$(1)

where $r_t$ denotes the returns in percent.

We adopt a regression-based methodology to study the holiday effect on the Swedish stock market and use a standard GARCH (1,1) model with the conditional error term assumed to follow a student’s t-distribution with 7 degrees of freedom to account for the observed heteroskedasticity and fat-tailed distribution of the data. Our baseline regression model is defined as

(2) $r_t={\alpha }_0+a_1{D}_{t}pre+a_2{D}_{t}post+{\epsilon }_t$(2)

where $D_{t}pre$ is the dummy variable for pre-holiday trading days and $D_{t}post$ is the dummy variable for post-holiday trading days. To study the historical development of the effect, we split our dataset into subsamples of 10-year periods on which we run separate regressions utilizing model (2).

The holidays we include in our study are Epiphany, Easter, Walpurgis Night, Ascension Day, Midsummer, Christmas, and New Year. To capture the pre-holiday effect for each holiday, we define our regression model as

(3) $r_t=a_0+a_1{D}_{t}Pr{e}_{Easter}+a_2{D}_{t}Pr{e}_{WalpurgisNight}+a_3{D}_{t}Pr{e}_{AscensionDay}+a_4{D}_{t}Pr{e}_{Epiphany}+a_5{D}_{t}Pr{e}_{Midsummer}+a_6{D}_{t}Pr{e}_{Christmas}+a_7{D}_{t}Pr{e}_{NewYear}+{\epsilon }_t$(3)

where $D_{t}Pr{e}_{Easter}$ to $D_{t}Pr{e}_{NewYear}$ are pre-holiday dummy variables for each respective holiday.

To capture post-holiday returns for each holiday, we define our model as¹

(4) $r_t=a_0+a_1{D}_{t}Pos{t}_{Easter}+a_2{D}_{t}Pos{t}_{WalpurgisNight}+a_3{D}_{t}Pos{t}_{Ephiphany}+a_4{D}_{t}Pos{t}_{Ascension}+a_5{D}_{t}Pos{t}_{Christmas}+a_6{D}_{t}Pos{t}_{NewYear}+{\epsilon }_t$(4)

where $D_{t}Pos{t}_{Easter}$ to $D_{t}Pos{t}_{NewYear}$ are post-holiday dummy variables for each holiday. We do not study the historical development of each holiday’s effect on returns, as there are not enough observations to provide accurate results in 10-year subsamples.

III. Results

Descriptive statistics

Table 1 reports descriptive statistics for our dependent variable AFGX index returns. For the entire sample period, the mean daily return was 0.055%, while the mean unadjusted pre-holiday return was 0.149% and the mean post-holiday return 0.279%.

Table 1. Descriptive statistics for AFGX returns (%) during 1980–2019.

Variable(s)	Obs.	Mean	Std.Dev.	Min	Max
Full Sample Return	10,019	0.055	1.236	−8.738	10.309
Pre-Holiday Return	231	0.149	0.948	−3.832	3.246
Post-Holiday Return	190	0.279	1.408	−4.468	4.903

Pre- and post-holiday returns

Table 2 presents the results for the pre- and post-holiday effects using the full sample period. Only the post-holiday returns, with a mean of 0.181%, achieve statistical significance (1% level). Thus, we conclude that a post-holiday effect existed on the Swedish stock market when considering the full sample period, which cannot be concluded regarding the pre-holiday effect.

Table 2. Pre-and post-holiday effect for AFGX (%) during 1980–2019.

Variable(s)	Coefficient	Obs.	(St. Error)	p-value
Pre-Holiday	0.057	231	(0.061)	0.349
Post-Holiday	0.181***	190	(0.062)	0.003
Constant	0.090***		(0.009)	0.000
L1.Arch	0.111***		(0.008)	0.000
L1.Garch	0.884***		(0.007)	0.000

Historical Development

Table 3–6present the results for the historical development of the pre- and post-holiday effects. Considering the development of the pre-holiday effect, we find no evidence of its existence for any period included in the sample. On the other hand, the post-holiday effect is positive and significant in the 1990s (1% level) and 2000s (5% level) respectively but falls short of statistical significance in the 1980s and the 2010s. Interestingly, the appearance of the post-holiday effect in the 1990s is concomitant with the start of the rapid growth of the internet. A common occurrence in Sweden is people staying on vacation after regular holidays. The development of the internet in the 1990s allowed for increased market attention when investors are on vacation, which could be a possible explanation for the appearance of the post-holiday effect in the 1990s. The decreased strength of the effect in the 2000s and its disappearance in the 2010s suggests that the market eventually corrected for the anomaly.

Table 3. Pre-and post-holiday effect for AFGX (%) during 1980–1989.

Variable(s)	Coefficient	Obs.	(St. Error)	p-value
Pre-Holiday	0.089	57	(0.117)	0.447
Post-Holiday	0.035	44	(0.116)	0.761
Constant	0.109***		(0.015)	0.000
L1.Arch	0.133***		(0.017)	0.000
L1.Garch	0.853***		(0.016)	0.000

Table 4. Pre-and post-holiday effect for AFGX (%) during 1990–1999.

Variable(s)	Coefficient	Obs.	(St. Error)	p-value
Pre-Holiday	0.126	61	(0.142)	0.375
Post-Holiday	0.417***	47	(0.130)	0.001
Constant	0.083***		(0.019)	0.000
L1.Arch	0.120***		(0.018)	0.000
L1.Garch	0.851***		(0.020)	0.000

Table 5. Pre-and post-holiday effect for AFGX (%) during 2000–2009.

Variable(s)	Coefficient	Obs.	(St. Error)	p-value
Pre-Holiday	0.004	55	(0.144)	0.980
Post-Holiday	0.326**	49	(0.162)	0.044
Constant	0.086***		(0.022)	0.000
L1.Arch	0.073***		(0.011)	0.000
L1.Garch	0.926***		(0.010)	0.000

Table 6. Pre-and post-holiday effect for AFGX (%) during 2010–2019.

Variable(s)	Coefficient	Obs.	(St. Error)	p-value
Pre-Holiday	−0.041	58	(0.104)	0.689
Post-Holiday	0.073	50	(0.111)	0.507
Constant	0.079***		(0.017)	0.000
L1.Arch	0.111***		(0.016)	0.000
L1.Garch	0.871***		(0.017)	0.000

Differentiating between holidays

Tables 7 and 8 present the results of individual holiday’s pre- and post-holiday effects using the full sample period. The pre-holiday effect lacks statistical significance for all holidays included. We find evidence at the 10% level for a positive post-holiday effect after New Year’s. No other holiday produces a significant post-holiday effect.

Table 7. Pre-holiday effect individual holidays for AFGX (%) during 1980–2019.

Variable(s)	Coefficient	Obs.	(St. Error)	p-value
Pre-Easter	0.163	40	(0.182)	0.370
Pre-NewYe	0.034	29	(0.152)	0.821
Pre-Epiphany	−0.274	26	(0.175)	0.117
Pre-Midsum	−0.064	40	(0.121)	0.597
Pre-Walpurg	−0.083	28	(0.182)	0.648
Pre-Christm	0.253	29	(0.215)	0.240
Pre-Ascens	−0.005	40	(0.168)	0.975
Constant	0.095***		(0.009)	0.000
L1.Arch	0.111***		(0.007)	0.000
L1.Garch	0.884***		(0.007)	0.000

Table 8. Post-holiday effect individual holidays for AFGX (%) during 1980–2019.

Variable(s)	Coefficient	Obs.	(St. Error)	p-value
Post-Easter	0.134	40	(0.132)	0.311
Post-NewYe	0.278*	28	(0.145)	0.056
Post-Epiph	0.151	28	(0.161)	0.348
Post-Walpurg	0.177	27	(0.147)	0.229
Post-Christm	0.211	28	(0.181)	0.244
Post-Ascens	0.184	40	(0.136)	0.175
Constant	0.092***		(0.009)	0.000
L1.Arch	0.111***		(0.007)	0.000
L1.Garch	0.884***		(0.007)	0.000

IV. Discussion

For the full sample period, we find evidence of a positive post-holiday effect on the Swedish stock market. Upon closer examination, however, the 10-year subsamples only provide evidence for the existence of a post-holiday effect during the 1990s and 2000s. A possible explanation for the time-varying significance of the post-holiday effect is the rapid growth of the internet in the 1990s, allowing for greater market attention when investors are on vacation, thus causing the effect, which was then corrected for by the market in the 2010s. Our results are in line with Cadsby and Ratner (1992) as we find no evidence for the existence of a pre-holiday effect on the Swedish stock market. One possible reason for the absence of the pre-holiday effect on the Swedish stock market can be Sweden’s generous vacation policies lowering investor pre-holiday excitement. When analysing individual holiday’s impacts on the Swedish stock market for the full sample period, we only find evidence of a holiday effect after New Year’s. The observed holiday effect after New Year’s could possibly be explained by tax customs. Overall, our findings suggest that the Swedish stock market is largely efficient. Future research could consider exploring the causes behind the observed post-holiday effect after New Year’s or differentiating between small and large-cap stocks.

Acknowledgments

We thank Jarkko Peltomäki for his valuable feedback.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Thomson Reuters Eikon is a terminal for checking and downloading financial data and there is no particular link that leads to the data used in our paper. However, we are able to provide the data upon request.

Additional information

Funding

This research did not receive funding.

Notes

¹ Midsummer is excluded in this regression as it is celebrated on a Saturday