How Does Market Celebrate Holiday? : a Comprehensive Analysis of Holiday Effect on Equity and Bond Market

(1)

How Does Market Celebrate Holiday?

- A Comprehensive Analysis of Holiday Effect on Equity and Bond Market

A thesis presented for Master in International Finance program

University of Amsterdam, Amsterdam Business School

Xuyao Cao 11610859

Supervisor: Florian Peters

(2)

Acknowledgements

I would first like to thank my thesis supervisor Mr. Florian Peters, who is responsible for the Master in International Finance program at the University of Amsterdam. His lecture on be-havioral finance triggered the interest that led to this thesis, and his knowledge has guided me through the whole writing process.

I would also like to thank my colleagues in the AFS Group for their support and guidance. They have provided me with access to valuable databases and with advice on data collection. Finally, I must express my most profound gratitude to my girlfriend for providing me with unfailing support and continuous encouragement throughout my years of study. It is, indeed, a difficult challenge to have a full-time trader job and a full-time master’s program running simultaneously. I could not have accomplished this without her.

Thank you all! Xuyao Cao

(3)

Abstract

This research investigates both the pre-holiday and post-holiday effect on major equity and bond markets. Daily reruns and trading volumes of equity indexes and daily yields of bond yield indexes are collected in a full range of history. By applying an OLS linear regression model, this research finds that both the pre-holiday effect and post-holiday effect exist in major equity markets, while such effects in bond markets are relative insignificant. This research continues the test that current scholarship has suggested as to the power of sentiment in the relationship between holidays and market performance; this research

ascertains that sentiment is not able to fully explain the presence of the holiday effect. It also finds that massive financial crises diminish the effect of holidays both on equity markets and bond markets.

(4)

Section 1: Introduction

This research focuses on one specific aspect of the calendar effect on financial markets—the holiday effect on equity and bond markets. The calendar effect has been considered as a chal-lenge to classical finance theories, in particular, the market efficiency hypothesis (MEH). This hypothesis states that asset pricing should fully reflect all available market information and assumes that market participants are rational (Fama, 1997). The holiday effect, typically the pre-holiday effect, has been studied in existing literatures. Pettengill’s (1989) finding of an abnormally high stock return earned in the pre-holiday period has provided evidence against market efficiency. In addition, Ariel (1990) has found that from 1963 to 1982, the av-erage return earned during pre-holiday trading was fourteen times higher than that during other trading days.

Recent studies have shown more detailed discoveries regarding the holiday effect in markets. Vergin and McGinnis (1999) found that from 1987 to 1996, excess pre-holiday returns van-ished for large companies and were diminvan-ished substantially for small companies. Further-more, Keef and Roush (2005) have investigated the period between 1930 and 1987, and they have found that the pre-holiday effect was strong during this time, while after 1987 it was greatly diminished. Both studies were conducted on the U.S. stock market.

Jacobs and Levy (1988) believed that psychological reasons best explain the pre-holiday ef-fect. Boyle (2004) has also supported the notion that holiday euphoria is the underlying be-havioral explanation for the pre-holiday effect. With regard to this bebe-havioral explanation, Chui and Wei (1998), Sias and Starks (1997), Teng and Liu (2013), and Al-Khazali (2014) have supported the notion that there are significant sentiment changes between holiday and non-holiday periods, that these are responded to by investors, and that they can cause market anomalies.

Hong and Yu’s (2009) study of market seasonality has shown that stock turnover, mean stock return, and trading volume are lower in summer when market participants are on vaca-tion. A recent study by Matthew Hood and Vance Lesseig (2017) has also observed the inat-tention of investors around stock-market holidays.

(6)

It is quite clear that holidays affect stock markets, especially in pre-holiday trading days. However, less studies have focused on the post-holiday effect: for example, data from the American stocks has been used to analyze the post-holiday effect (Pettengill, 1989), while American market indexes have been used to analyze the pre-holiday effect (Ariel, 1990). This research fills in these methodological gaps in the existing literature on the holiday effect.

Previous studies have mainly used data from single equity markets to analyze the holiday effect, while in this research data from major financial markets is collected. The comparison of the holiday effect among markets with different cultural and religious backgrounds is also conducted because it is believed that certain holidays often stem from culture and religions. In contrast, there is no clear interpretation as to how holidays would impact bond markets and how the holiday effect has changed after the 2008 crisis—especially since the 1987 crisis impacted the pre-holiday effect. Therefore, the holiday effect on bond markets and the

impact from the most recent crisis are both analyzed in this research.

This paper aims to discover more with regard to how holidays affect stock markets and bond markets. It intends to explore market movements over time—both before and after

holidays—and to compare three major religion-segmented markets: the Asian, Middle Eastern, and Western markets. This paper also furthers the discussion on market sentiment and inattention in our understanding of the holiday effect. Moreover, the change of holiday effect over a large time horizon is studied in order to describe the specific impact of the 2008 financial crisis.

1.1 Research Questions

The central question of this research can be formulated as follows:

How do stock markets and bond markets react to public holidays?

In order to answer this question and to enhance this research area, four sub-questions were designed. Sub-question 1, 2, and 3 evaluate the holiday effect from a macro perspective. From a micro perspective, sub-question 4 explores if sentiment or inattention can explain the holiday effect, while sub-question 5 views the change of the holiday effect from the

(7)

1. How does the market perform differently in a period close to a holiday compared to nor-mal trading days?

Although it is a widely accepted that a holiday does affect the stock market, this sub-question evaluates the holiday effect on the bond market price and stock market price.

2. How does the market trading volume change over time before and after public holidays?

This question helps us to understand how trading volume changes with time when we are getting close to or far away from public holidays. The comparison between trading volume during pre-holiday trading days and post-holiday trading days can reveal more detailed holi-day effects on the market.

3. What is the holiday effect on markets from a cultural and geographic perspective?

This question aims to provide a comprehensive view as to how Asian, Middle Eastern, and Western markets react differently to holidays. A cross-sector analysis is performed based on geographic and cultural segments in both market and holiday types.

4. How to explain the holiday effect on the market?

Previous literature has given two major explanations for the holiday effect: there are changes in invest sentiment or mood, and there is investor inattention. This sub-question explores which explanation of the holiday effect is more convincing.

5. How does the holiday effect change in correspondence to changes in the market?

This question discovers how markets react differently on the same holiday when a market it-self is experiencing different conditions—such as in a crisis condition or when the market turns into a bull market, and a bear market. For bond markets, the change also includes dif-ferent monetary policy implementations as well as changes in credit ratings.

Since this paper explores and determines the relationship between holiday and market performance, a quantitative research method has been adopted. Data was collected to the maximum extent from the Bloomberg Terminal. According to data availability, in total, 42 major country equity indexes were collected—6 from the America, 12 from Asia, 19 from Europe, and 5 from the Middle East—moreover, 19 major sovereign bond yield indexes were also collected.

(8)

Section 2: Literature Review

2.1 Market Efficiency

Eugene Fama (1997) and Kenneth French (2012) have identified three forms of efficient market hypothesis: weak, semi-strong, and strong. Under the original efficient market hypothesis, no matter the forms, investors cannot earn abnormal returns—that is to say, it is impossible to beat the market when the prices fully reflect all the information. In the “weak” efficient market hypothesis, the future return should not be predictable based on past

information.

2.2 Calendar Effect

The idea of the calendar effect arose when researchers discovered that certain time periods had an effect on market performance, which was a challenge to the EMH. Among those studies, the earliest and most famous one is called the “January Effect”—in short, markets generally experience a drop in prices in January (Dyl, 1977). Roll’s (1983) study points out that small market cap stocks, especially, experience significant losses during January, whereas Keim (1983) has explained that this is due to the Christmas shopping season and to corporations’ bonus plans.

Recent studies have expanded the “January Effect” to the broader issue of “Seasonality”. Kamstra, Kramer, and Levi (2003) have challenged the EMH by referring to the seasonal affective disorder: in winter, due to the lack of sunlight, investors tend to be more

risk-averse. Kamstra, Kramer, and Levi analyzed equity price data from 9 countries. Hong and Yu (2009) have studied stock-trading activity in summer vacations with respect to trading

volume and expected returns. They have used air travel and hotel occupancy as sources to define summer vacations, and they have set the summer vacation as a dummy variable in their regression. In total, Hong and Yu collected not only price but also trading volume from 51 countries, and they concluded that stock turnover, mean stock returns, and trading volume are lower in summer when market participants are on vacation. They explained this on account of investors’ limited attentions. Hood and Lesseig (2017) have also observed investors’ inattention when adjacent to holiday. They included nearly all the public equity data in the U.S. and found out that most stocks are not able to earn abnormal returns during

(9)

holidays—and even those with extreme abnormal returns will eventually drift on the subsequent post-holiday trading days.

2.3 The Holiday Effect

Pettengill (1989) has documented abnormally high stock returns earned around the holiday closings. Regardless of the closing of week, year, or holiday, the pre-holiday trading days often experience unusually high returns when compared to other trading days. Pettengill has also included post-holiday return patterns that only show abnormally high return when they occur on Fridays. Just as with the January effect (Roll, 1983), a small cap stock typically shows the highest abnormal return.

The result of Pettengill’s study have been confirmed by Ariel (1990). Instead of analyzing each stock data, Ariel has chosen to collect and analyze market indexes and market portfolio performances. Just as with the previous study, a dummy variable that represents pre-holiday is placed in the regression formula. The study of Ariel has firstly confirmed that nearly two thirds of market indexes experienced positive returns on pre-holiday trading date from 1963 to 1982, while only 54% of the normal trading days experienced positive return. During that same period, the overall return earned by market portfolios on pre-holiday trading days were on average fourteen times higher than on other trading days. Ariel has thus helped to explain the holiday effect on market indexes and market portfolios. Positive investor moods because of the expectation of holidays makes investors (including short sellers) want to close position early instead of keeping it open through the holiday period.

Only a few studies have discussed both pre- and post-holiday effects on the markets, and they have shown different results. Chui and Wei (1998) have revealed that investors

participate in important buying activities during both pre- and post-holiday periods and that they have higher pre-holiday stock returns than post-holiday returns. Dzhabarov and

Ziemba’s (2010) study has shown positive pre-holiday stock return anomalies and negative post-holiday return anomalies, while Picou (2006)’s study has shown that post-holiday positive reactions continue when stock markets reopen.

The holiday effect has also been studied from a religious angle recently because public holiday are often linked to cultural and religious ideology. Religion has been identified in some studies as a key factor affecting economic and financial environments, and it has been

(10)

found to affect individual investors’ decisions to invest and their risk attitudes (Miller and Hoffmann, 1995; Hilary and Hui, 2009). Chung (2014) believes that cultural influences play a determining role in investors’ decision-making. Several studies have been done separately on the effect of major religious holidays on stock returns, such as Chinese New Year,

Ramadan, and the Easter-period holidays. These holiday periods have been found to increase stock returns (Al-Khazali, 2014; Kaplanski and Levy, 2012; Teng and Liu, 2013). For

example, it has been identified that there is a lower trading volume in the U.S. market during Jewish holidays and Ramadan (a Muslim holiday) since these days create a religious

sentiment, lead to higher stock returns, and lower volatility (Frieder and Subrahmanyam, 2004). Gavriilidis, Kallinterakis, and Tsalavoutas (2015) have discovered a herding significance from Ramadan that reveals a variation in levels across markets, both

domestically and internationally. Similarly, another study that investigates the effect of the Chinese Lunar New Year (CLNY) holiday on major Asian stock markets has shown high pre-CLNY returns for China, Japan, Malaysia, South Korea, and Taiwan. (Yuan and Gupta, 2014) Some literature has also revealed a possible impact on the holiday effect from the market itself, which brings different results. Vergin and McGinnis (1999) have found that from 1987 to 1996, excess pre-holiday returns vanished for large companies and substantially

diminished for small companies. Also, Keef and Roush (2005) found that from 1930 to 1987 the pre-holiday effect was strong, while after 1987 it was greatly diminished. Both studies were conducted on the U.S. stock market. And in 1987, the U.S. stock market experienced “Black Monday”, a noteworthy stock crisis.

2.4 The Source of the Holiday Effect

There are different views regarding the source of the holiday effect. Ariel (1990) believed that investors have a positive mood when they expect a holiday; therefore, they often close their positions before a given holiday. Hirshleifer (2001), Baker and Wurgler (2007), and Dellavigna and Pollet (2009) have supported the notion that mood—a transient state of feel-ing at a particular time—can affect the expectation of future fundamentals or can interact with risk preferences. Both equity and bond prices correlate positively with investor moods, and a better investor mood leads to higher asset prices (Hui-Chu Shu, 2010). However, Hong and Yu (2009) have explained the calendar effect from an investor inattention point of view, and they have found that Friday announcements can have a 15% lower immediate response

(11)

and a 70% higher delayed response. In addition, Dellavigna and Pollet (2009) have observed that trading volume is 8% lower around Friday announcements. They have explained this post-earnings announcement drift on account of limited attention.

More literature can be reviewed to show different explanations. Debataa, Dashb, and Ma-hakud (2017) have found a positive (and negative) effect of investor sentiment on liquidity (or illiquidity). Furthermore, stock-market liquidity affects expected returns, market effi-ciency, and overall financial stability, according to Chordia’s (2001) study. Jacobs and Levy (1988) have used the notion of holiday euphoria to explain the abnormally high returns dur-ing the turn of the week, month, and year. They believe that psychological reasons best ex-plain the pre-holiday effect. However, they have also questioned the existence of the holiday effect, and have answered it by anticipating that portfolio managers would not liquidate their entitle portfolio on pre-holiday days merely in order to obtain short-run high returns. Boyle (2004) has also supported the notion that holiday euphoria is the behavioral explanation for the pre-holiday effect.

Based on the observation of Shu (2010), the higher the complexity of a decision and the un-certainty of its outcome, the higher the impact of sentiment in decision-making. Chui and Wei (1998) and Al-Khazali (2014) have detailed how during holidays markets show buy-side sentiments among investors. They believe that the significant sentiment changes between holiday and normal trading days cause market anomalies.

2.5 The Holiday Effect on Bond Market

There are not many studies on the holiday effect on bond markets. Zaremba and Schabek (2016) have investigated the calendar effect in bond returns. They focused on the January effect and the “sell-in-May-and-go-away” anomaly in government bond returns. They collected government bonds from 25 countries for the years 1992–2016 and concluded that both bond returns and the factor premium from credit risk and momentum are not affected by the January effect and the “sell-in-May-and-go-away” effects. They believe that seasonal patterns in government bond markets tend to be merely a statistical coincident. Bethke, Trapp, and Kempf (2014), in contrast, have investigated how the bond performance is related to investor sentiment. Their data zooms in on U.S. corporate bonds. Their study has shown that the risk-factor correlation increases with negative investor sentiments, whereas positive sentiment causes risk factor and bond correlation to be negative.

(12)

Previous literature has confirmed that holidays have an effect on stock markets—and especially in the pre-holiday trading days, equity market experiences abnormally high returns. However, the sampling method chosen and research direction have varied. For example, Pettengill (1989) has analyzed American equity performance in both pre- and post-holiday trading days, while Ariel (1990) has focused on market index performance in pre-holidays only. Fewer studies have been conducted on post-holiday effects and holiday effects on bond markets. There is no clear interpretation as to how the holiday effect has changed after the 2008 crisis because, as previous study has shown, the 1987 crisis impacted the pre-holiday effect. Besides this, there is no comprehensive study that compares the holiday effect on markets in countries with different cultural and religious backgrounds. And the explanation of holiday effect, in general, is still a debated topic.

As this study aims to fill in those gaps in the previous literature and to answer the remaining questions, the following section outlines the data sampling and research methodology undertaken in this research.

(13)

Section 3 Data and Methodology

3.1 Data

Both quantitative data (equity and bond yield indexes) and qualitative data (information on holidays, weekends, and financial crises) were collected. The holiday and weekend data were reverted into dummy variables in the regression, and financial crisis was defined as a control variable. The daily closing prices and trading volume of 42 equity indexes and daily closing prices of 19 bond indexes were obtained in the full range of the time horizon from the Bloomberg Terminal. The description of these equity and bond indexes are shown in Table 1 and 2.

Table 1 Descriptive statistics – Equity Index

Note: Table 1&2 provide descriptive statistics for indexes of equity market and bond market in this research. It aims to give an overview about the data that has been used in the research, therefore some statistics show economic meanings, for example: % change per year. Some indexes are segmented based on cultural differences instead of geographical distance, for example: Malaysia and Turkish are included in Middle East Sector. This research also collects trading volume for each equity index, please see appendix.

(14)

Table 2 Descriptive statistics – Bond Index

The above market indexes were chosen based on the significance and representativeness of local financial markets. More than one equity indexes was chosen to represent the U.S. equity market because of the country’s significance in the global market. In order to fully reflect the holiday effect, all the above indexes were collected over the largest time period possible. Three popular maturities (1 year, 10 years, and 30 years) of sovereign bonds were chosen in order to explore the holiday effect on bonds with different maturities. Some indexes were adjusted: for instance, some indexes only had weekly closing prices in the first 5 years; accordingly, only time ranges that had sequential daily closing prices were chosen in those cases.

In order to test the market sentiment explanation of the holiday effect, the CBOE Volatility (VIX) index, the first benchmark index to measure a market’s expectation of future volatility, was collected.

Table 3 Descriptive statistics – VIX Index Descriptive Statistics

N Mean Std. Deviation Skewness Kurtosis Begin Date Begin Value End Data End Value

(15)

The qualitative data in this study is the information about financial crises. It is generally known that the global financial crises in history can be listed as follow:

1. “Black Monday.” On the 19th_{of October 1987, after a deep “jump” in the Dow Jones}

industrial average, stock markets around the world crashed and caused global panic in financial markets.

2. “The 1997 Asian Financial Crisis.” From July to October in 1997, initiated by Thailand giving up its pegged exchange rate system towards the US dollar, most East Asian financial markets joined the slump, and this phenomenon eventually raised fears of a worldwide economic meltdown due to financial contagion.

3. “The 2008 Financial Crisis.” This crisis started with the subprime mortgage crisis in the U. S. and developed into a full-blown international banking crisis with the collapse of the

investment bank Lehman Brothers on the 15th of September 2008. (Mark, 2010).

4. “The European Debt Crisis.” This describes a multi-year debt crisis in the EU since the end of 2009. Several Eurozone member countries, including Greece, Portugal, Spain, and Ireland were unable to repay or refinance their government debt and asked for bailouts.

3.2 Hypothesis

Just as the previous studies in this area have done, this research applies the ordinary least squared (OLS) linear regression model. It may be considered to have flaws of assumptions; for example, there is no autocorrelation. Therefore, for this research, all the dependent variables were translated into log return of indexes instead of daily closing prices, whereas dependent variables of trading volume were translated into percentage increases compared to the last trading date.

In order to answer sub-question 1, 2, and 3, the OLS method has been specified as below:

Log Return+= c + β0123D0123+ β056+3D056+3+ β01278D01278+ β056+78D056+78+ ε+ (1) ∆% Volume₊= c + B₀₁₂₃D₀₁₂₃+ B₀₅₆₊₃D₀₅₆₊₃+ B₀₁₂₇₈D₀₁₂₇₈+ B₀₅₆₊₇₈D₀₅₆₊₇₈+ ε₊ (2)

In Equation 1, Log Returnt is the log return of daily closing prices of equity indexes and bond

(16)

varia-bles that are equal to 1 if the returns are on one trading day before public holidays and week-end, respectively. D₀₅₆₊₃ and D₀₅₆₊₇₈ represent the dummy variables that are equal to 1 if returns occur one trading day after public holidays and weekend, respectively.

This research included the dummy variable of “weekend” in order to distinguish the holiday effect from the weekend effect on the market.

Equation 2 is a transformation of Equation 1 in order to apply volume data. Ceteris paribus, the dependent variable ∆% Volume₊ represents the percentage change of volume at time t compared to that of volume at time t-1. ε+ in both equations indicates error terms, which is

assumed to be white-noise processes.

If the coefficient of the dummy is tested to be significant, this suggests that the dummy varia-ble has an effect on the market price or trading volume. If the sign of the coefficient is signif-icantly positive, the dummy has a positive effect (and vice versa). For example, if D₀₁₂₃ in Equation 1 is significantly positive this suggests that market returns on days before a holiday are significantly higher than those returns on normal trading days. If D₀₁₂₃ in Equation 2 is insignificant, then this suggests that returns on days before a holiday are indifferent when compared with those on normal trading days, regardless of a positive or negative sign. And if the coefficient is significant, the number of the coefficient represents the expected percentage change in market price and volume in holiday time range compared to normal trading days. The significance of coefficient was tested on three different confidence levels (90%, 95%, and 99% confidence levels).

To answer sub-question 4, this research needed to explore the possible causal relationship be-tween market sentiment and the holiday effect—whether sentiment mediates the relationship between holiday and market performance. The VIX index, which is also known as the “fear index”, was chosen to represent the market sentiment since the VIX index measures market expectation of future volatility and since market anticipants bear market when the VIX index is high.

Judd and Kenny (1981) have proposed a way to test the mediation effect. To estimate the in-direct coefficient, they have suggested calculating the difference between two coefficients (see Table 4):

(17)

Table 4. Coefficients Approach

Therefore, by applying the method of OLS regression and the approach of Judd and Kenny, the following 6 equations were designed to answer sub-question 4:

Log Return₊= c + β_@D₃+ ε₊ (3) Log Return+= c + γBCDVIX++ βGD3+ ε+ (4) β1_62I+CJ2I+= β@− βG (5) ∆% Volume+= c + B@D3+ ε+ (6) ∆% Volume+= c + γBCDVIX++ BGD3+ ε+ (7) BB_62I+CJ2I+= B@− BG (8)

The data used here was modified from the original data. Because the VIX index was calcu-lated from option prices in the S&P500 market index, only the most relevant equity market indexes in each region were calculated here. And in order to analyze the impact of the VIX index on holiday effect, the time range of equity indexes were cut to match that of the VIX index. As this sub-question does not focus on the difference between pre- and post-holidays and the difference between holidays and weekends, D3 represents the market return or

vol-ume on the days both before and after holiday. VIX+ is the VIX index at time t. All of the

above dependent variables are same as those in Equation 1and 2. β1_62I+CJ2I+ and BB_62I+CJ2I+ are

the coefficients of the mediation effect an are based on the Judd-and-Kenny approach. The qualitative data regarding the financial crises was used as a control variable to segment the data into different time periods. The time slot data was collected and used to segment price and volume data into two groups: (1) “before crisis” and (2) “after crisis”. And the data was matched with the most relevant crisis (see Table 5):

(18)

Table 5. Index Match with Financial Crisis

By applying Equation 1and 2 and then comparing the coefficient results, this research ex-plored the mediation of general market conditions in the holiday effect.

Time Slot Equity Index Bond Index

Black Monday 1987/10/19 America

Asian Financial Crisis 1997/7/1 Asia Asia

2008 Financial Crisis 2007/1/2 America, Asia

(19)

Section 4 Empirical Results

This study used an OLS linear regression model to examine the presence of the holiday effect in both equity markets and bond markets. A total of 42 major country equity market indexes and 19 sovereign bond yield indexes were included. Both equity indexes and bond indexes were segmented into four market divisions (America, Europe, Asia, and the Middle East), based on geographic and religious aspects. Table 6 to 10 shows the results of the estimated coefficients in regression formula 1and 2, and it answers sub-question 1, 2, and 3.

Table 6 documents the results of six equity market indexes in America. Regarding index return over the history of each index, most of the results confirm the presence of positive pre-holiday effects on market return (including the U.S., Canada, and Brazil); this means that the market was expected to experience higher positive returns on pre-holiday dates than on other trading dates. Especially, two market indexes in the United States—the S&P 500 and the Dow Jones Industrial Average—appeared to have positive pre-holiday effects, with a

significance on the 99% confidence level. Table 6 also shows results from the market indexes of Mexico and Argentina—but in these regions only the estimated coefficient of

“post-weekend” has a significance on at least the 95% level.

Only a few market indexes had significant post-holiday effects. The S&P 500 showed negative post-holiday effects; this means that it is expected to have lower returns when investors return from holidays compared to other trading days. Brazil, instead, showed the opposite results; this means that the equity market in Brazil tends to have higher returns on both pre-holiday and post-holiday trading days when compared to normal trading days. As for trading volume, all six indexes had negative estimated coefficients, and five of them were significant at the 99% confidence level. This indicates that the market is expected to have less trading volume on days before a holiday in the U.S., Canada, Argentina, and Brazil with a 99% confidence level, and in Mexico with a 95% confidence. Lower trading volume was also observed on post-holiday days in the U.S. market indexes.

(20)

Table 6. The Holiday Effect on America Equity Indexes

Table 7 documents the results from 12 Asian equity market indexes. A significantly positive holiday effect was observed in equity indexes from Hong Kong, China, Taiwan, Thailand, Malaysia, Vietnam, and the Philippines. The coefficients of pre-holidays from indexes from Japan, Korea, Australia, India, and Singapore do not pass the robustness test; this means that their results cannot be tested on at least a 90% confidence level. A post-holiday effect was also observed. Japan, Hong Kong, India, Singapore, and the Philippines have higher market index returns after holidays, while Korea is expected to have lower market index returns after holidays.

(0.0000)

United States United States Canada Mexico Argentina Brazil S&P Dow Jones S&P/TSX MEXBOL MERVAl IBOV Index Return c 0.0004*** 0.0003*** 0.0003** 0.0009*** 0.0023*** 0.0015*** (0.0001) (0.0007) (0.0019) (0.0005) (0.0000) (0.0004) βpreH 0.0024*** 0.002*** 0.0011** -0.0012 0.0014 0.0035** (0.0000) (0.0000) (0.0074) (0.2637) (0.4187) (0.0323) βpostH -0.0009** -0.0003 -0.0001 0.0006 -0.0019 0.0028* (0.0326) (0.3598) (0.747) (0.5914) (0.2831) (0.0823) βpreWE 0.0000 0.0002 0.0002 -0.0003 -0.0004 0.0017** (0.8529) (0.3281) (0.2086) (0.4849) (0.7102) (0.0476) βpostWE -0.0013*** -0.001*** -0.0005** -0.0014** -0.0035*** -0.0017** (0.0000) (0.0000) (0.0097) (0.0036) (0.0002) (0.0443) Trading Volume c 0.0493*** 0.0344*** 0.0739*** 0.1187*** 0.0655*** 0.0621*** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) BpreH -0.2349*** -0.1568*** -0.2445*** -0.1224** -0.1281*** -0.1055*** (0.0000) (0.0000) (0.0000) (0.002) (0.0000) (0.0001) BpostH -0.0052 -0.0366** -0.0418** -0.0536 -0.0277 -0.056** (0.8684) (0.0369) (0.0226) (0.1765) (0.3274) (0.0373) BpreWE -0.0455** 0.0331*** -0.1256*** -0.1848*** -0.118*** -0.114*** (0.0021) (0.0001) (0.0000) (0.0000) (0.0000) (0.0000) BpostWE -0.1803*** -0.1836*** -0.2309*** -0.4132*** -0.208*** -0.1749*** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000)

Notes. Table 6 to 10 presents results from the OLS linear regression model. Specifically, the model is given as follows. Log Return+= c + β0123D0123+ β056+3D056+3+ β01278D01278+ β056+78D056+78+ ε+

∆% Volume+= c + B0123D0123+ B056+3D056+3+ B01278D01278+ B056+78D056+78+ ε+

The variables of interest in this study are the four dummy variables in the equation: DpreH and DpostH, representing days before and after a holiday, respectively; and DpreWE and DpostWE, representing days before and after a weekend, respectively. This was designed in order to distinguish the holiday effect from the weekend effect. The null hypothesis of holiday effect in the stock market was tested against the alternative hypothesis that there is a significant pre/post holiday effect on the stock market. If it is found that the coefficient for the dummy beta is significantly different from 0, then the null is rejected and one can conclude that the stock market exhibits a significant pre/post holiday effect. The sign * represents the rejection of the null hypothesis at a 10% significance level. The sign ** represents the rejection of the null

hypothesis at a 5% significance level. The sign *** represents the rejection of the null hypothesis at a 1% significance level. Values in parentheses are p-values.

(21)

Regarding trading volume, most countries in Asia had negative coefficients on pre-holidays; this means that most market indexes in Asia are expected to have lower trading volumes before holiday. Only the market index from the Philippines had a positive coefficient for pre-holidays; this confirms that the market there experienced high trading volumes on days before holidays. On the post-holiday side, results varied. Regardless, those did not pass the robustness check. Hong Kong, China, Thailand, Singapore, and Malaysia were estimated to have a higher trading volume when a holiday finished, while Japan, Australia, India,

Vietnam, and the Philippines were estimated to have lower trading volumes after a holiday.

Table 7. The Holiday Effect on Asian Equity Indexes

The results for indexes from the Middle East are shown in Table 8. Five market indexes were analyzed, and only Abu Dhabi and Egypt presented significant higher returns for days before holidays and days after holidays. The market index of South Africa has a significantly positive post-holiday effect, while its coefficient for pre-holiday returns did not pass the robustness test. Both South Africa and Turkey experienced lower trading volumes before a holiday. Abu Dhabi and Egypt also had significantly lower trading volumes on days after a holiday. In particular, Turkey was estimated to have a higher trading volume after holidays.

Japan Hong Kong China Taiwan Korea Australia Thailand India Singapore Malaysia Vietnam Philippine

NIKKEI HANGSENG SHSZ300 TWSE KRX100 AS51 SET SENSEX STI FBMKLCI VNINDEX PCOMP

Index Return c 0.0001 -0.0001 -0.0003 0.0000 0.0006** 0.0003 0.0001 0.0003 0.0000 0.0003* 0.0000 0.0001 (0.8159) (0.658) (0.4662) (0.9607) (0.0256) (0.1077) (0.8216) (0.1678) (0.9000) (0.0574) (0.9302) (0.6146) βpreH -0.0001 0.0018* 0.0036** 0.0021** 0.0008 0.0008 0.0025** 0.0006 0.0001 0.0012* 0.0041** 0.0032*** (0.9055) (0.0824) (0.0379) (0.0013) (0.5455) (0.342) (0.0019) (0.265) (0.8686) (0.0656) (0.008) (0.0004) βpostH 0.0029** 0.0019* 0.0021 0.0004 -0.0025* 0.0011 0.0002 0.0023*** 0.0028** 0.0002 0.0029* 0.0022** (0.0013) (0.0698) (0.2391) (0.5358) (0.0532) (0.1931) (0.8154) (0.0000) (0.002) (0.7514) (0.0638) (0.017) βpreWE -0.0006 0.0007 0.0006 0.0003 -0.0008 -0.0004 0.0022*** 0.0002 0.0008* 0.0009** 0.0017** 0.0009** (0.2523) (0.2133) (0.4199) (0.6489) (0.2132) (0.2282) (0.0000) (0.662) (0.0641) (0.0108) (0.0037) (0.0478) βpostWE -0.0004 0.0000 0.0014* -0.001* -0.0005 -0.0001 -0.002*** -0.0001 -0.0007* -0.0015*** -0.0006 -0.0008 (0.4714) (0.9521) (0.0577) (0.0842) (0.4502) (0.7378) (0.0000) (0.7612) (0.0871) (0.0000) (0.3179) (0.1011) Trading Volume c 0.0389*** 0.0353*** -0.0037 0.0299*** 0.041*** 0.089*** 0.0512*** 0.058*** 0.0461*** 0.0378*** 0.021** 0.0573*** (0.0000) (0.0000) (0.4683) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0017) (0.0000) BpreH -0.0189 -0.0752*** -0.0509** -0.0805*** -0.0146 -0.1202*** -0.1393*** -0.1803*** -0.0601* -0.0134 -0.0488 0.0953** (0.1611) (0.0005) (0.0417) (0.0000) (0.3823) (0.0000) (0.0000) (0.0000) (0.0782) (0.4854) (0.1719) (0.0016) BpostH -0.0444** 0.1352*** 0.118*** -0.0052 0.027 -0.1141*** 0.0573** -0.1884*** 0.0465 0.0504** -0.0921** -0.1653*** (0.001) (0.0000) (0.0000) (0.6288) (0.106) (0.0000) (0.0132) (0.0000) (0.1755) (0.0086) (0.0105) (0.0000) BpreWE 0.0212** -0.042*** -0.0184* -0.0398*** -0.0831*** -0.1295*** -0.082*** -0.0309 -0.0689*** -0.0496*** -0.0051 -0.0711*** (0.0058) (0.0002) (0.0793) (0.0000) (0.0000) (0.0000) (0.0000) (0.2068) (0.0000) (0.0000) (0.7031) (0.0000) BpostWE -0.2194*** -0.1553*** 0.0321** -0.1248*** -0.1346*** -0.3094*** -0.1767*** -0.1814*** -0.1786*** -0.1622*** -0.0794*** -0.2302*** (0.0000) (0.0000) (0.0022) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000)

(22)

Table 8. The Holiday Effect on Middle Eastern Equity Indexes

Table 9 includes 19 market indexes from European countries. All of them confirm that

market indexes had higher returns on days both before and after holidays—expect for the fact that the UK, Switzerland, the Netherlands, Belgium, Luxembourg, Sweden, Poland, and Russia did not have a significance level over 10% for the pre-holiday effect, nor the UK, Spain, Finland, Sweden, Austria, and Greece for the post-holiday effect.

As for trading volume, most of the market indexes experienced lower trading volumes before holidays, expect Luxembourg and Austria that experienced insignificant coefficients. The results for trading volume on days after holidays are complex. Despite the fact that Germany, France, Portugal, Luxembourg, Austria, Greece, and Poland had insignificant coefficients of post-holiday trading volumes, the UK, Spain, Denmark, and Finland had lower trading volumes after a holiday, while the rest of the indexes had a higher volume.

Abu Dhabi South Africa Egypt Qatar Turkey

ADX TOP40 HERMES DSM XU100

Index Return c 0.0000 0.0003 0.0001 0.001** 0.0012** (0.8584) (0.2031) (0.7969) (0.0012) (0.0012) βpreH 0.0019** -0.0001 0.0012* 0.0022 0.0000 (0.0296) (0.946) (0.0734) (0.2037) (0.9891) βpostH 0.0024** 0.0029** 0.0015** 0.0006 0.0014 (0.0063) (0.002) (0.0174) (0.7432) (0.479) βpreWE 0.0009* -0.0004 0.0017** -0.0001 0.0013 (0.0722) (0.4238) (0.0035) (0.812) (0.1038) βpostWE 0.0003 0.0005 0.0005 -0.0033*** -0.0014* (0.5241) (0.327) (0.3683) (0.0000) (0.0709) Trading Volume c 0.0142 0.0985*** 0.0509*** 0.037*** 0.0545*** (0.2255) (0.0000) (0.0000) (0.0003) (0.0000) BpreH 0.0738 -0.2517*** 0.0093 0.0359 -0.3441*** (0.2205) (0.0000) (0.7427) (0.4566) (0.0000) BpostH -0.1958** 0.039 -0.0271 -0.0899* 0.2316*** (0.0012) (0.3402) (0.3427) (0.0643) (0.0000) BpreWE 0.0004 -0.196*** -0.0322** -0.0517** -0.12*** (0.9882) (0.0000) (0.0378) (0.0128) (0.0000) BpostWE -0.0623** -0.2838*** -0.2346*** -0.1376*** -0.1453*** (0.0087) (0.0000) (0.0000) (0.0000) (0.0000)

(23)

Table 9. The Holiday Effect on European Equity Indexes

The analysis of bond indexes is shown in Table (10). Most coefficients had a significance level of over 10% regarding pre- and post-holiday effects on yield; this means that on a 90% confidence level, there could be no holiday effect on yield at all (regardless of whether it is post-holiday or pre-holiday). The U.S. treasury bonds with all short-term, middle-term, and long-term maturities are expected to have lower yields on days before a holiday. Regarding the post-holiday effect, only a 10-years Chinese government bond had a negative effect on yields, while a 30-years UK government bond had a positive effect on yields.

Europe UK Germany France Spain Switzerland Italy Portugal Netherlands Belgium

STOXX 50 FTSE100 DAX CAC IBEX SMI FTSEMIB BVLX AEX BEL20

Index Return c 0.0002 0.0002 0.0002* 0.0003 0.0000 0.0002 0.0001 0.0000 0.0003* 0.0002 (0.226) (0.1442) (0.0799) (0.1739) (0.8524) (0.1608) (0.637) (0.8237) (0.0995) (0.3202) βpreH 0.0026 0.0012 0.0023*** 0.0019* 0.003** 0.0012 0.0037** 0.0015** 0.0004 0.0003 (0.1256) (0.1538) (0.0002) (0.0752) (0.0032) (0.1607) (0.0278) (0.0146) (0.736) (0.7949) βpostH 0.0054** 0.0008 0.0015** 0.0024** 0.0014 0.0026** 0.0045** 0.0013** 0.0035** 0.0025** (0.0019) (0.347) (0.0148) (0.0274) (0.1682) (0.0024) (0.0069) (0.0314) (0.0012) (0.0121) βpreWE 0.0000 0.0005 0.0004 -0.0001 0.0004 0.0000 -0.0003 0.0003 0.0001 0.0001 (0.899) (0.1186) (0.1185) (0.8974) (0.3166) (0.9438) (0.5728) (0.3532) (0.6984) (0.7804) βpostWE -0.0006 -0.0006** -0.0009*** -0.001** -0.0002 -0.0005 -0.0012** -0.0003 -0.0006 -0.0003 (0.1403) (0.0375) (0.0006) (0.0133) (0.6307) (0.1579) (0.0374) (0.425) (0.1255) (0.3706) Trading Volume c 0.0638*** 0.1624*** 0.0755*** 0.0867*** 0.0549*** 0.0692*** 0.0625*** 0.067*** 0.08*** 0.0751*** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) BpreH -0.7016*** -0.5405*** -0.173*** -0.2431*** -0.062* -0.1485*** -0.1591*** -0.2616*** -0.318*** -0.1004** (0.0000) (0.0000) (0.0000) (0.0000) (0.0941) (0.0000) (0.0000) (0.0000) (0.0000) (0.0087) BpostH 0.5492*** -0.5558*** -0.043 -0.0167 -0.0849** 0.0582* -0.0658** 0.0341 0.1432*** 0.0901** (0.0000) (0.0000) (0.1213) (0.5643) (0.0222) (0.0506) (0.0468) (0.568) (0.0000) (0.0187) BpreWE -0.0605** -0.1893*** -0.0523*** -0.1009*** -0.0097 -0.0418*** -0.0724*** -0.0771*** -0.0785*** -0.047*** (0.0012) (0.0000) (0.0000) (0.0000) (0.487) (0.0003) (0.0000) (0.0002) (0.0000) (0.0007) BpostWE -0.2552*** -0.5538*** -0.3132*** -0.3206*** -0.2536*** -0.3061*** -0.231*** -0.2407*** -0.3152*** -0.3168*** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) Luxembourg Denmark Finland Norway Sweden Austria Greece Poland Russia

LUXX KFX HEX OBX OMX ATX ASE Wig Imoex

Index Return c -0.0001 0.0001 -0.0001 0.0001 0.0002 0.0001 -0.0001 0.0000 -0.0001 (0.7837) (0.7405) (0.7138) (0.6074) (0.4535) (0.6783) (0.6243) (0.8837) (0.8174) βpreH 0.0011 0.0018** 0.0047*** 0.0027** 0.0007 0.0015** 0.0037** 0.0006 0.0023 (0.3321) (0.0491) (0) (0.0415) (0.504) (0.0458) (0.0018) (0.5714) (0.2547) βpostH 0.0023** 0.0023** 0.0016 0.0043** 0.0011 0.0007 0.0007 0.0026** 0.0069*** (0.0421) (0.0113) (0.1439) (0.0013) (0.2827) (0.3542) (0.5379) (0.0153) (0.0005) βpreWE 0.0008* 0.0005 0.0009* 0.0007 0.0005 0.0002 0.002*** 0.0003 0.001 (0.0895) (0.1752) (0.0553) (0.2056) (0.2287) (0.5499) (0.0007) (0.5524) (0.3082) βpostWE -0.0004 0.0003 0.0002 (0.0000) 0.0001 -0.0001 -0.0006 0.0007 0.0012 (0.3609) (0.3962) (0.7011) (0.9977) (0.834) (0.7939) (0.2718) (0.1745) (0.1918) Trading Volume c 0.0424** 0.085*** 0.0946*** 0.0772*** 0.0832*** 0.0536*** 0.0479*** 0.0544*** 0.0584*** (0.0341) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) BpreH -0.0316 -0.0602** -0.1153*** -0.1855*** -0.3642*** -0.0068 -0.1384*** -0.1251*** -0.2863*** (0.7426) (0.0444) (0.0000) (0.0000) (0.0000) (0.8193) (0.0000) (0.0003) (0.0000) BpostH 0.0029 -0.1344*** -0.0605** 0.0608* 0.1728*** 0.0146 -0.0222 -0.0002 0.2396*** (0.9763) (0.0000) (0.0267) (0.0607) (0.0000) (0.6206) (0.4962) (0.9963) (0.0000) BpreWE -0.0436 -0.1172*** -0.1382*** -0.1204*** -0.146*** 0.0032 -0.0355** -0.0766*** -0.1481*** (0.2755) (0.0000) (0.0000) (0.0000) (0.0000) (0.8341) (0.0233) (0.0000) (0.0000) BpostWE -0.1751*** -0.3021*** -0.3268*** -0.2626*** -0.2603*** -0.2849*** -0.1905*** -0.1828*** -0.1601*** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000)

(24)

Table 10. The Holiday Effect on Bond Yield Indexes

Table 11.1 and 11.2 summarize the results of the holiday effect on the 42 equity market indexes:

Table 11.1. A Coefficient Summary on Equity Indexes Returns Table 11.2. A Coefficient Summary on Equity Indexes Volume

By simply looking at the portion of estimated coefficients, there is not a strong holiday effect on equity index returns in all the countries. Over half of the market indexes experienced a positive holiday effect. An exceedingly small portion of indexes were tested to have a

negative post-holiday effect on returns. The rest (over one third) of the indexes were tested to have a statistically insignificant coefficient; this means that the holiday effect on returns were hardly obvious.

Positive Negative Statistically Insignificant

Asia Pre Holiday 7 5

Post Holiday 6 1 5

America Pre Holiday 4 2

Post Holiday 1 1 4

EU Pre Holiday 10 9

Post Holiday 13 6

Middle East Pre Holiday 2 3

Post Holiday 3 2

Return Positive Negative Statistically Insignificant

Pre Holiday 54.76% 45.24%

Post Holiday 54.76% 4.76% 40.48%

Equity Return Positive Negative Statistically

Insignificant

Asia Pre Holiday 1 7 4

Post Holiday 4 5 3

America Pre Holiday 6

Post Holiday 3 3

EU Pre Holiday 17 2

Post Holiday 8 4 7

Middle East Pre Holiday 2 3

Post Holiday 1 2 2

Volume Positive Negative Statistically Insignificant

Pre Holiday 2.38% 76.19% 21.43%

Post Holiday 30.95% 33.33% 35.71%

Equity Trading Volume

US US US US Germany France France France UK UK Turkish USGG2YR USGG5YR USGG10YR USGG30YR GDBR30 Index GFRN10 GFRN30 GUKG1 GUKG10 GUKG30 IECM1 Yield c 0.0003 0.0001 -0.0001 0.0000 0.0001 -0.0002 -0.0001 -0.0004 0.0002 0.0001 -0.0002 (0.4794) (0.7668) (0.6774) (0.8886) (0.8218) (0.6106) (0.4845) (0.7406) (0.5529) (0.6586) (0.6839) βpreH -0.0026 -0.0017* -0.0014* -0.0019** -0.0005 -0.0004 0.0007 0.0002 -0.0008 -0.0005 -0.0006 (0.2399) (0.0922) (0.0598) (0.0356) (0.9026) (0.9074) (0.651) (0.9911) (0.8026) (0.879) (0.6792) βpostH -0.0004 -0.0006 0.0002 0.0003 0.002 0.0012 0.0002 -0.0034 0.0025 0.0035 -0.0012 (0.842) (0.5639) (0.8086) (0.7339) (0.6526) (0.7256) (0.9133) (0.8173) (0.4351) (0.2871) (0.413) βpreWE -0.0014* -0.0003 0.0000 -0.0005* -0.0016** -0.0018** -0.0013** 0.0009 -0.0011** -0.0013** 0.0018* (0.0529) (0.5262) (0.8731) (0.0807) (0.0201) (0.0342) (0.0025) (0.712) (0.0291) (0.0023) (0.0546) βpostWE -0.0002 0.0001 0.0004 0.0003 -0.0002 0.0011 0.0006 -0.0006 -0.0005 -0.0005 0.0001 (0.8311) (0.7731) (0.1758) (0.3235) (0.8128) (0.1801) (0.173) (0.7952) (0.3451) (0.2228) (0.9194)

Turkish Malaysia Malaysia Malaysia China China China Japan Singapore Singapore Singapore IECM10 MAGY3 MAGY10 MAGY20 GCNY1 GCNY10 GCNY30 GJGB30 MASB2 MASB10 MASB30 Yield c -0.0002 -0.0003 0.0000 -0.0003 0.0004 0.0002 0.0009 -0.0005 -0.0013* -0.0004 -0.0006 (0.5935) (0.4195) (0.9676) (0.4627) (0.3936) (0.3325) (0.1507) (0.401) (0.0748) (0.2148) (0.1932) βpreH 0.0002 0.0012 -0.0001 0.0004 -0.0011 -0.0005 0.0000 0.0011 0.0001 -0.0002 -0.0003 (0.928) (0.3597) (0.9396) (0.6861) (0.4073) (0.4853) (0.9991) (0.8583) (0.9801) (0.8583) (0.8569) βpostH 0.0004 -0.0013 0.0003 0.0006 0.0000 -0.003*** -0.0018 -0.0042 0.0035 0.0027* -0.0002 (0.8152) (0.3524) (0.7408) (0.5417) (0.9906) 0.0000 (0.2421) (0.4889) (0.2202) (0.0541) (0.9126) βpreWE 0.0012* 0.0003 -0.0005 0.0019** -0.0003 -0.001** -0.0029** -0.0008 0.0029** 0.0002 0.0006 (0.0982) (0.7413) (0.2789) (0.018) (0.733) (0.0461) (0.0277) (0.5102) (0.0383) (0.7178) (0.5027) βpostWE 0.001 0.0005 -0.0002 -0.0009 -0.0002 0.0005 0.0012 0.002* 0.0026* 0.0007 0.0028** (0.1728) (0.5492) (0.6881) (0.2773) (0.8081) (0.341) (0.3607) (0.0794) (0.0604) (0.3001) (0.0025)

(25)

Table 11.2 reveals a clearer picture. Almost 80% of market indexes experienced lower trading volumes before a holiday, while the rest of the indexes had insignificant coefficients. As for post-holiday effects on volume, an approximately equal number of indexes had positive and negative coefficients; however, there were still over one third of indexes with statistically insignificant coefficients.

Table 12.1 to 12.4 present the results of sentiment mediation in the holiday effect on equity markets. The major representative market indexes from the 42 countries were selected based on their GDP ranking. In this case, the VIX index was representative of bad sentiment. Therefore, under the hypothesis that sentiment can explain the relationship between the holiday effect and market performance, coefficient 2 was expected to be smaller than

coefficient 1, and the results of coefficient 1 minus coefficient 2 were expected to be positive.

Table 12.1 Sentiment Test for Asian equity indexes

Table 12.2 Sentiment Test for European equity indexes

Index Return Japan HongKong Taiwan Thailand India

β1 0.0001 0.0016* 0.0018* 0.0025* 0.0015* (0.8769) (0.0817) (0.0661) (0.0033) (0.0475) β2 0.0002 0.0016* 0.0018* 0.0024* 0.0016* (0.8223) (0.0813) (0.0636) (0.0044) (0.0392) β1-β2 -0.0001 0.0000 0.0000 0.0001 -0.0001 Volume B1 -0.021450463 -0.0165 -0.002 -0.1089*** 0.1509*** (0.1359) (0.423) (0.9028) (0.0000) (0.0000) B2 -0.0214 -0.0164953 -0.002 -0.1091*** 0.1512*** (0.1361) (0.423) (0.9026) (0.0000) (0.0000) B1-B2 -0.0001 0.0000 0.0000 0.0001 -0.0003

Index Return EU UK Germany France Spain NL

β1 0.0033* 0.0017* 0.0028* 0.0024* 0.0021* 0.0019 (0.0305) (0.0322) (0.0145) (0.0221) (0.0493) (0.1019) β2 0.0033* 0.0015* 0.0023* 0.0022* 0.002* 0.0017 (0.0295) (0.0501) (0.0375) (0.0351) (0.0672) (0.1441) β1-β2 0.0000 0.00015 0.00043 0.0002 0.00016 0.00022 Volume B1 -0.1081* -0.6739*** -0.1092*** -0.0825* -0.0167 -0.0784* (0.0708) (0.0000) (0.0000) (0.0018) (0.636) (0.0052) B2 -0.108* -0.6738*** -0.109*** -0.0823* -0.0167 -0.0783* (0.0711) (0.0000) (0.0000) (0.0019) (0.6346) (0.0053) B1-B2 -0.00011 -0.00015 -0.00027 -0.0002 0.00007 -0.00013

Notes. Table 12.1 to 12.4 present results from Judd & Kenny Difference of Coefficients approach. Specifically, the model is given as follows.

Log Return+= c + β@D3+ ε+

Log Return+= c + γBCDVIX++ βGD3+ ε+

β1_62I+CJ2I+= β@− βG

The variables of interest of this study are the four dummy variables in the equation, DH representing days before and after holiday; VIXt representing VIX index on time t. The null hypothesis of holiday effect in the stock market will be tested against the alternative hypothesis

that there is a significant pre/post holiday effect in the stock market. If it is found that the coefficient for the dummy beta is significantly different from zero, then we will reject the null. As β@ is expected to be positive on returns and negative on volume. Therefore, β1_62I+CJ2I+

being negative in returns meaning positive VIX (sentiment) mediation power; and being positive in trading volume meaning positive VIX (sentiment) mediation power. * represents the rejection of the null hypothesis at 10% significance level. ** represents the rejection of the null hypothesis at 5% significance level. *** represents the rejection of the null hypothesis at 1% significance level. Values in parentheses

(26)

Table 12.3 Sentiment Test for Middle Eastern equity indexes

Table 12.4 Sentiment Test for American equity indexes

Regarding market return, Hong Kong, Taiwan, and India showed statistically significant coefficients results. However, Hong Kong and Taiwan showed zero mediation power for sentiment. Thailand showed a positive and India a negative mediation power for the VIX index; this means correspondingly negative and positive explanation power of sentiment. As for trading volume, only Thailand and Indian were tested to have statistically significant coefficients, and the results of the two were different: Thailand had a positive explanation power for sentiment while India had a negative one.

Among selected European market indexes, the UK, Germany, France, and Spain presented a positive mediation power; this means that sentiment could explain at least a portion of the holiday effect on market returns. However, the STOXX 50, the overall market index for European as a whole, presented zero results, which indicates that there was zero sentiment explanation power. Regarding trading volume, only the Netherlands had a statistically insignificant coefficient, while other indexes had negative results. Under the hypothesis that sentiment contributes positive effect on negative relationship between holiday and trading volume, the results shows positive sentiment explanation power on trading volume aspect.

Index Return Abu Dhabi Saudi Arabia Egypt Qatar Turkey

β1 0.002* 0.0005 0.0009 0.0023 0.0001 (0.0203) (0.5847) (0.1742) (0.22) (0.9588) β2 0.0025* 0.0005 0.0008 0.0022 0.0001 (0.0041) (0.611) (0.2216) (0.2238) (0.9547) β1-β2 -0.00048 0.00004 0.00009 0.00002 -0.00001 Volume B1 0.1561* -0.0535 0.0562* 0.0344 -0.2515*** (0.0148) (0.1875) (0.0872) (0.5002) (0.0000) B2 0.1546* -0.0534 0.0564* 0.0358 -0.2514*** (0.0159) (0.1879) (0.0867) (0.483) (0.0000) B1-B2 0.00145 -0.00004 -0.00012 -0.00143 -0.00013

Index Return S&P INDU Canada Mexico

β1 0.0001 0.0001 0.0008* -0.0005 (0.8085) (0.7921) (0.0962) (0.4815) β2 0.0001 0.0001 0.0007 -0.0004 (0.9154) (0.8943) (0.1106) (0.5626) β1-β2 0.00007 0.00006 0.00004 -0.00009 Volume B1 -0.0973*** -0.0666*** -0.1005*** -0.0441* (0.0000) (0.0000) (0.0000) (0.0953) B2 -0.0973*** -0.0665*** -0.1005*** -0.044* (0.0000) (0.0000) (0.0000) (0.0965) B1-B2 0.00004 -0.00009 -0.00001 -0.00014

(27)

The market index of Abu Dhabi had the opposite results compared to other indexes. It had negative results for market return and positive results for trading volume. In contrast, the rest of the indexes from the Middle East had a lesser significance for the results of the returns. Egypt and Turkey presented negative coefficient results that align with results from the European indexes.

As for the U.S. market, none of the selected indexes had a statistically significant value on returns, while each of the indexes presented significant results for trading volume. Both Canada and Mexico had negative coefficients results; this indicates a positive explanation power of sentiment in the relationship between holidays and market volume. The two indexes representing the U.S., the S&P 500 and Dow Jones Industrial Average, however, showed the opposite results: positive coefficients for S&P500 and negative coefficients for the Dow Jones Industrial Average.

Table 13.1 to 13.3 shows the results of market condition changes on the holiday effect. By comparing the differences between coefficients from periods before and after crises, the market condition changes on the holiday effect can be estimated. The blank cell in the tables means the absence of data in that time range. Each segment has two columns: the left one represents the estimated coefficients for the corresponding period, and the right one represents the differences between the left results and the results from Table 6 to 10, thus indicating the estimated coefficients for the whole recorded history. Negative results mean a diminished effect from a crisis, whereas positive results mean a strengthened effect from a crisis.

Equity indexes from U.S. markets were segmented into three periods: “Before 1987” (before the Black Monday crisis), “1987–2007”, and “After 2008 (after the 2008 financial crisis). The crisis in 1987 caused the holiday effect on American equity market indexes to be weakened. The coefficient differences are negative; this means that the holiday effect

diminished after that crisis, especially for the post-holiday effect. Similar results apply to the 2008 crisis. However, diminishing coefficients after the 2008 crisis still showed statistical significance; this means that even after 2008, the returns and trading volume of equity indexes in America were still affected by holidays.

(28)

Table 13.1 Changes of holiday effect on American equity indexes

Table 13.2 Changes of holiday effect on European bond indexes

S&P βpreH 0.0033*** 0.0009 0.0002 (0.0022) 0.0008 (0.0016) βpostH -0.0013*** (0.0004) -0.0001 0.0008 0.0004 0.0013 S&P-V BpreH -0.3269*** (0.0920) -0.1168*** 0.1181 BpostH -0.0021 0.0031 -0.0074 (0.0022) INDU βpreH 0.0026*** 0.0006 0.0021*** 0.0001 0.0006 (0.0014) βpostH -0.0006 (0.0003) -0.0004 (0.0001) 0.0003 0.0006 INDU_V BpreH -0.2236*** (0.0668) -0.0765*** 0.0803 BpostH -0.0429** (0.0063) -0.0278 0.0088 SPTSX βpreH 0.0021** 0.0010 0.0012* 0.0001 0.0000 (0.0011) βpostH -0.0018** (0.0017) 0.0007 0.0008 0.0000 0.0001 SPTSX_V BpreH -0.2244*** 0.0201 -0.2756*** (0.0311) -0.1989*** 0.0456 BpostH -0.167*** (0.1252) -0.07*** (0.0282) 0.0300 0.0718 MEXBOL βpreH -0.0011 0.0001 -0.0012 0.0000 βpostH 0.0007 0.0001 0.0004 (0.0002) MEXBOL_V BpreH -0.1168** 0.0056 -0.1275** (0.0051) BpostH -0.0799 (0.0263) -0.0175 0.0361 MERVAL βpreH 0.0015 0.0001 0.0017 0.0003 βpostH -0.0059** (0.0040) 0.0034* 0.0053 MERVAL _V BpreH -0.2295*** (0.1014) -0.0386 0.0895 BpostH -0.0057 0.0220 -0.0427 (0.0150) IBOV βpreH 0.0042 0.0007 0.0026 (0.0009) βpostH 0.0035 0.0007 0.002 (0.0008) IBOV_V BpreH -0.2017*** (0.0962) -0.0112 0.0943 BpostH -0.0473 0.0087 -0.064** (0.0080) Before 1987 1987-2007 After 2008 GDBR30 βpreH 0.0003 0.0008 -0.0185 (0.0180) βpostH 0.0017 (0.0003) 0.0000 (0.0020) GFRN10 βpreH -0.0003 0.0001 -0.0077 (0.0073) βpostH 0.0009 (0.0003) 0.0005 (0.0007) GFRN30 βpreH 0.0002 (0.0005) 0.0091 0.0084 βpostH 0.0009 0.0007 -0.0173 (0.0175) GUKG1 βpreH 0.0036 0.0034 -0.0239 (0.0241) βpostH -0.0067** (0.0033) 0.0311 0.0345 GUKG10 βpreH -0.0001 0.0007 -0.0057 (0.0049) βpostH 0.0004 (0.0021) 0.0344* 0.0319 GUKG30 βpreH -0.0003 0.0002 -0.001 (0.0005) βpostH 0.0001 (0.0034) 0.0245** 0.0210 IECM1 βpreH 0.0039 0.0045 -0.002 (0.0014) βpostH 0.0002 0.0014 -0.0014 (0.0002) Before 2009 After 2009

Notes. Table 13.1 to 12.3 present results from OLS linear regression method. Specifically, the model is segmented based on before/after financial crisis.

The variables of interest of this study are the four dummy variables in the equation, DpreH and Dpost H representing days before and after holiday. The blank cell in the tables means the absent of data in that time range. Each segment has two columns, the left one presents the estimated coefficients for the corresponding period and the right one presents the differences between the left results and the results from Table 6 to 10. Negative results (mark in red) meaning diminish effect from crisis and positive result

meaning strengthen effect from crisis.

* represents the rejection of the null hypothesis at 10% significance level. ** represents the rejection of the null hypothesis at 5% significance level. *** represents the rejection of the null hypothesis at 1% significance level. Values in parentheses are p-values.

(29)

Table 13.3 Changes of holiday effect on Asian equity indexes Nikkei βpreH 0.0005 0.0006 0.0007 0.0008 -0.0009 (0.0008) βpostH 0.0012 (0.0017) 0.0041*** 0.0012 0.0024* (0.0005) Nikkei_V BpreH -0.0171 0.0018 -0.0408* (0.0219) -0.0004 0.0185 BpostH -0.0729* (0.0285) -0.0287 0.0157 -0.0462*** (0.0018) HangSeng βpreH 0.0016 (0.0002) 0.0023 0.0005 0.0014 (0.0004) βpostH -0.0045* (0.0064) 0.0048*** 0.0029 0.0012 (0.0007) HangSeng_V BpreH -0.0614 0.0138 -0.1259*** (0.0507) -0.0367 0.0385 BpostH 0.1338*** (0.0014) 0.1462*** 0.0110 0.1268*** (0.0084) SHSZ300 βpreH 0.0014 (0.0022) 0.0043** 0.0007 βpostH -0.0005 (0.0026) 0.0029 0.0008 SHSZ300_V BpreH -0.0314 0.0195 -0.0518** (0.0009) BpostH 0.0331 (0.0849) 0.1263*** 0.0083 TWSE βpreH 0.0027** 0.0006 0.0011 (0.0010) 0.0017 (0.0004) βpostH 0.001 0.0006 -0.0011 (0.0015) 0.0005 0.0001 TWSE_V BpreH -0.1162*** (0.0357) -0.058*** 0.0225 -0.074*** 0.0065 BpostH 0.0657** 0.0709 -0.1005*** (0.0953) 0.0988*** 0.1040 KRX100 βpreH -0.0008 (0.0016) 0.0011 0.0003 βpostH -0.008* (0.0055) -0.0017 0.0008 KRX100_V BpreH -0.0202 (0.0056) -0.0108 0.0038 BpostH 0.0502* 0.0232 0.015 (0.0120) AS51 βpreH -0.0003 (0.0011) 0.0006 (0.0002) 0.0019 0.0011 βpostH -0.0004 (0.0015) 0.0009 (0.0002) 0.0024 0.0013 AS51_V BpreH -0.1249*** (0.0047) -0.1183*** 0.0019 BpostH -0.2551*** (0.1410) -0.0354 0.0787 SET βpreH 0.0026* 0.0001 0.0047*** 0.0022 0.0008 (0.0017) βpostH -0.0009 (0.0011) 0.0004 0.0002 0.0009 0.0007 SET_V BpreH -0.0638 0.0755 -0.1087*** 0.0306 -0.1896*** (0.0503) BpostH 0.02 (0.0373) -0.0195 (0.0768) 0.1342*** 0.0769 SENSEX βpreH 0.0008 0.0002 -0.0022 (0.0028) 0.0023** 0.0017 βpostH 0.0025*** 0.0002 0.0006 (0.0017) 0.0026** 0.0003 SENSEX_V BpreH -0.1605* 0.0198 -0.2117*** (0.0314) BpostH -0.7063*** (0.5179) 0.1212*** 0.3096 STI βpreH 0.0004 0.0003 0.0000 (0.0001) βpostH 0.0018 (0.0010) 0.0034*** 0.0006 STI_V BpreH -0.0601* 0.0000 BpostH 0.0465 0.0000 FBMKLCI βpreH 0.0017* 0.0005 0.0006 (0.0006) 0.0005 (0.0007) βpostH -0.0001 (0.0003) 0.0007 0.0005 0.0004 0.0002 FBMKLCI_V BpreH -0.0794* (0.0660) -0.0066 0.0068 0.0039 0.0173 BpostH 0.0122 (0.0382) 0.0428 (0.0076) 0.0684*** 0.0180 VNINDEX βpreH 0.0018 (0.0023) 0.0052*** 0.0011 βpostH 0.0026 (0.0003) 0.003 0.0001 VNINDEX_V BpreH -0.1411 (0.0923) -0.0368 0.0120 BpostH -0.217* (0.1249) -0.0784** 0.0137 PCOMP βpreH 0.0053*** 0.0021 0.0052*** 0.0020 0.0005 (0.0027) βpostH 0.0012 (0.0010) 0.0047*** 0.0025 0.0012 (0.0010) PCOMP_V BpreH 0.2669** 0.1716 0.1393** 0.0440 0.0546 (0.0407) BpostH -0.4434*** (0.2781) -0.208*** (0.0427) -0.1096*** 0.0557 Before 1997 1997-2008 After 2008

(30)

Indexes from the Asian markets were segmented into three periods as well: “Before 1997” (before the Asian financial crisis), “1987–2007”, and “After 2007” (after the 2008 financial crisis). It is easy to observe that after the Asian financial crisis in 1987, the holiday effect (both pre- and post-holiday effects), diminished index returns and trading volumes. To be more specific, the consequence from financial crisis on trading volumes is more obvious. The differences in coefficients of returns are much smaller than that of volume. In contrast, the degree of diminishing on holiday effect was less after the 2008 crisis compared to the period after the 1987 crisis.

Table 13.2 presents the impact of the European debt crisis on the relationship between the holiday effect and bond yield indexes. Typically, European sovereign bond indexes were chosen for the testing. The results suffered the same issue that can be found in Table 10— namely, that only few results showed statistical significance. If the robustness check is not regarded, then 80% of the bond indexes present diminished results for the holiday effect on its yield after the Eurozone debt crisis. The pre-holiday effect on yield was weakened more than the post-holiday effect.

(31)

Section 5: Conclusion

This study employed a full range of data from both equity market indexes and bond yield indexes to research market performance during holidays. It is believed that holidays have often developed from cultural and religious origins. Forty-two equity indexes and 19 bond-yield indexes, each including trading volume, were segmented into four parts based on geographic and religious aspects: America, Asia, Europe, and the Middle East.

This study shows that over half of the equity indexes have positive pre-holiday effects and post-holiday effects; this indicates that equity markets generate higher returns on days around holidays. Nearly 40% of the equity indexes presented statistically insignificant results; this means that the holiday effect is not obvious for every equity market. None of those equity indexes had negative pre-holiday effects, and only a small portion of them had negative post-holiday effects; this means that the post-holiday effect is either positive for equity market returns or not statistically significant. Among those indexes, indexes from America and Asia

presented the highest portion of positive holiday effects on returns compared to the portions of statistically insignificant effect. In contrast, equity markets from Europe and the Middle East showed significantly positive post-holiday effects; this ensured that those indexes generated higher returns on days after holidays. America had the least portion of indexes that had significant post-holiday effects.

Regarding equity index trading volume, over two thirds of the total equity indexes showed negative pre-holiday effects, and the rest presented either statistically insignificant results or a relatively small portion of positive pre-holiday effects. Results were equally distributed for positive post-holiday effects, negative post-holiday effects, and statistically insignificant post-holiday effects (one third each). Asia, America, and Europe presented the most indexes that had negative pre-holiday effects on trading volume, while indexes from the Middle East did not present any outstanding holiday effects on trading volume.

This research used the VIX index to represent sentiment—specifically, bad sentiment when the VIX index is high. The VIX index was included into the regression formula as a control variable, and the coefficient results of the holiday effect were compared between one with and one without the VIX index in the regression. A result of a negative mediation power for

(32)

the VIX index explained the positive power of sentiment. Analysis was also conducted on market returns and trading volumes.

Among equity indexes from the Asian markets, only the indexes from Thailand and India presented results with statistical significance. The index from Thailand showed a negative explanation power of sentiment on returns and a positive explanation power of sentiment on trading volume, whereas the index from India presented the opposite results—positive on returns and negative on trading volume. Results from the European market were clearer than from the Asian market. Despite a few indexes that had insignificant results, indexes from each European country presented a positive explanation power for sentiment in the holiday effect, both on market returns and on trading volume. However, the STOXX 50 index showed zero explanation power for sentiment on the holiday effect. Just as with the Asian market indexes, indexes from the Middle East market presented results that did not align with each other. As for the American market indexes, only a positive explanation power for sentiment on trading volumes was observed, while the rest of the results were statistically insignificant.

Both equity indexes and bond indexes were matched with the corresponding financial crisis in order to exam the impact of market conditions on the holiday effect. For example, the Asian equity indexes were segmented based on two relevant financial crises: the 1997 Asian financial crisis and the 2008 global financial crisis. The American equity indexes were segmented based on the 1987 “Black Monday” crisis and the 2008 global financial crisis. Bond indexes from Europe were segmented based on the 2009 Eurozone debt crisis. By comparing the holiday coefficients before and after crises, this study reveals that financial crises impact the holiday effect on returns and trading volume. Each crisis makes the holiday effect on returns diminish afterwards—and, in particular, the 2008 financial crisis makes more diminishing consequence on the holiday effect that on returns than on trading volume. This research shows results that partially align with the existing literature: for example, higher returns and lower trading volume in equity indexes can be observed on days before holidays; a financial crisis diminishes the holiday effect; and the holiday effect is considered to be consistent among markets with different cultural and religious background. This research also brings added value to this field. The post-holiday effect on returns and trading volume and both the pre-holiday and post-holiday effect on bond yields were detailed and

(33)

explored. This research continues the discussion of the source of the holiday effect. In this regard, the test to explain the power of sentiment is significantly positive on European equity indexes, while it remains unclear for indexes from Asian, American, and Middle Eastern markets.

This research represented most of the equity indexes from the major financial markets in the world, and each of the indexes was obtained for its full range. However, market indexes are not always the perfect representation of the stocks in each market. Also, only government bond yield indexes were collected, while data on corporation bonds was not. These omissions were mainly due to limited data availability and resources. Furthermore, this research

methodology may not be the optimal one since, even though the OLS regression method has been chosen by several scholars, some current scholarship has suggested that the ARMA (1,1)-GARCH (1,1) model might overcome the problem of OLS regression assumptions. The choice of the VIX Index might also be a potential limitation as there are no clear subjects available that can be used to measure investors’ sentiment. Moreover, this research tested the explanation power of sentiment; however, the results cannot be used to testify whether investor inattention could explain the holiday effect.

This research contributes value to both theoretical and practical fields. It strengthens and confirms the existing literatures on pre-holiday effect and the changes of such an effect with financial crises. It also offers new insights on post-holiday effects and the effect on bond markets. In a practical way, the results may not be able to persuade portfolio managers to shift their entire position according to the presence of a holiday because of insignificant results and relatively lower coefficients. However, the results can used to assist traders to decide the time for making trades when a given position has already been established with regard to a specific change.

In the future, when more advanced computation technologies can be applied, further research should be able to conduct analysis on individual stocks and bonds in major financial markets. Ideally, an accurate quantitative measurement of investor sentiment and inattention could also be invented and used in such research.

How Does Market Celebrate Holiday? : a Comprehensive Analysis of Holiday Effect on Equity and Bond Market