The effects of an index change on a firm : the price effect and the co-movement effect

(1)

The effects of an index change on a firm

The price effect and the co-movement effect

Student: Hugo van den Biggelaar Student Number: 10270507 Msc Business Economics: Finance track Faculty of Economics and Business University of Amsterdam

Date: 06/2016

First Supervisor: dr. L. Zou

(2)

Statement of Originality This document is written by student Hugo van den Biggelaar who declares to take full responsi-bility 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)

Abstract The purpose of this study is to test what the price effects and the co-movement effects are of an addition to (or a deletion from) an index for a stock by using an event study and running several univariate and bivariate regressions. The three indices of interest are the S&P500 Index, the FTSE 100 Index and the Hang Seng Index. The main results for the S&P 500 are that there is a permanent price effect after an index change and a co-movement effect does exists, however, only when considering the rest of the market. The results of the FTSE 100 show a negative non-permanent price effect for both additions and deletions and a significant co-movement effect after a change to the FTSE 100. For the Hang Seng Index, the sample size was too small; therefore, the results possess little explanatory value. For different indices, the price effect and the co-movement effects are different.

(4)

Index 1. Introduction ………... 5 2. Theoretical Framework………..………...………... 7 2.1 Price effect an index change has on a stock………...………. 7 2.2 Co-movement of a stock when added to/deleted from an index……… 9 3. Hypotheses and Methodology………...………...……….. 11 3.1 Price effect………...………...………... 11 3.1.1 Hypotheses…………..………...………...………. 11 3.1.2 Methodology………...………...……… 11 3.2 Co-movement………...………...……….... 13 3.2.1 Hypotheses………...………...………13 3.2.2 Methodology………...………...……… 13 4. Data………...………...………... 16 4.1 Indices………...………...………... 16 4.2 Dataset………...………...………... 17 4.3 Descriptive statistics………...………...………. 19 5. Results………...………...………...………. 20 5.1 Price effect an index change has on a stock………...………. 20 5.1.1 Additions………...………...……….... 20 5.1.2 Deletions………...………...……….... 23 5.2 Co-movement of a stock when added to/deleted from an index………... 26 5.2.1 Event study………...………...………... 26 5.2.1.1 S&P 500 Index………...………...………… 26 5.2.1.2 FTSE 100 Index………...………...………. 28 5.2.1.3 Hang Seng Index………...………...……... 29 5.2.3 Calendar time approach………...………...………… 30 5.2.3.1 S&P 500 Index………...………...………… 31 5.2.3.2 FTSE 100 Index………...………...………. 33 5.2.3.3 Hang Seng Index………...………...……... 34 6. Discussion………...………...………..…... 36 6.1 Price effect an index change has on a stock………...………. 36 6.1.1 Additions………... 36 6.1.2 Deletions……… 38 6.2 Co-movement of a stock when added to/deleted from an index………... 39 6.2.1 Univariate regressions..………...……… 39 6.2.2 Bivariate regressions..………...………... 40 7. Conclusion………. 41

(5)

1. Introduction The best-known financial index in the world is the S&P 500 Index. This index consists of 500 large firms, which have been chosen to reflect the U.S. market. Every quarter, the index is revalued to capture corporate actions. These actions differ from mergers and acquisitions to issuing shares or paying dividends. The index is also adjusted to remain an appropriate reflec-tion of the U.S. market. Such a revaluation consists of additions or deletions of firms from the index. One might expect that such an index change has several effects. For an addition, one might expect that this has a positive effect on the return of a stock and the stock price. Another effect of an addition could be the increase of the co-movement of a firm with the index it is added to. The firm becomes better known to investors, it is followed by more analysts and is traded more since it is incorporated in index funds. Similarly, one might expect that a deletion has negative effects on the return, the co-movement and the stock price of a firm. This thesis aims to test the effects of an index change on a firm. In the literature on this topic, two papers stand out, i.e. the paper by Chen et al. (2004) and the one by Barberis et al. (2005). Chen et al. (2004) conducted their research to determine whether there is a price impact following a revaluation of the S&P 500 Index. The authors found an asymmetric price effect. At an addi-tion, the price effect is a permanent positive one and at a deletion, the price effect is a temporary negative one. Barberis et al. (2005) investigated whether there is a change in the co-movement of a stock with the S&P 500 Index when a revaluation is performed. These authors found that the co-movement of a stock increases with the index when listed on the index and decreases when delisted from the S&P 500 Index. By combining both papers, the following research question is formulated for a study of other indices: “What is the effect of an index change on a stock on the price impact caused by such a change and the co-movement of a stock with that index?” To answer this question, hypotheses are formulated, based on the work of Chen et al. (2004) and Barberis et al. (2005). It is expected that there is a positive permanent price impact at an addition and a temporarily negative price impact at a deletion. It is also expected that the co-movement of a stock with an index increases when listed to an index, while simultaneously the co-movement of that stock with the rest of market decreases. For deletions, the opposite is ex-pected. This thesis will contribute to existing literature in three ways. First, both studies are combined to one study. Secondly, the research is extended to two other indices, i.e. the FTSE100 Index and the Hang Seng Index. By adding these two indices, it can be tested whether the conclusions in

(6)

both papers apply worldwide. The FTSE100 Index is one of the most renowned indices in Eu-rope and the Hang Seng index is one of the largest indices used in Asia; therefore, both hold a similar position as the S&P 500 Index in the U.S. market. Thirdly, this thesis contributes to existing literature by extending the research of Chen et al. (2004) and Barberis et al. (2005) with an-other time period, i.e. January 2004-December 2014 for all three indices. The time frame used in the two previous mentioned papers was January 1962-December 2000; thus, in the research presented here, the validity of these results can be assessed for a more recent period. In order to answer the research question, two separate studies have to be completed. The first study is an event study to determine the price effect of an index change on a stock. In order to do this, an estimation window is created around an event. The event is the announcement of the index change. This window will consist of 61 days, with the event date being day 0. The next step is to determine the cumulative average abnormal return (CAAR), which is found by determining the abnormal return (AR), followed by the average abnormal return (AAR). After determining the CAAR, a conclusion can be drawn whether there is a significant price effect when an index change is announced. The second study will test the co-movement effect. The co-movement is presented by the β of a stock. At first, a univariate regression is run in order to determine if there is a significant change in the β of a stock when listed on an index / delisted from an index. Secondly, a bivariate regression is run to test whether there is a significant change in the βindex

of a stock and a significant change in the βnon-index of a stock when listed on/delisted from an index. In order to determine the βnon-index of a stock, another index is chosen to depict the rest of the market. To test for robustness, the calendar time approach is applied. In this way, the corre-lation of returns across events, which causes impure results when performing an event study, is eliminated. In this approach, two portfolios have to be created. These portfolios are a pre-event and a post-event portfolio. Hereafter, a univariate regression and a bivariate regression are run. The data necessary to perform these studies are retrieved from DataStream. This thesis proceeds as follows. Chapter 2 discusses the two most important papers on index mutations and places them in perspective by comparing the results with results from other stud-ies. In chapter 3, the methodology is described and the hypotheses are formulated. Chapter 4 is on data and describes which indices are used in different parts of the world, what data are used and why, the criteria for a regular index change and some descriptive statistics. In chapter 5, the results are presented which are discussed and interpreted in chapter 6. Finally, in chapter 7 conclusions are presented and some recommendations for further research are discussed.

(7)

2. Theoretical Framework The research by Chen et al. (2004) and the research by Barberis et al. (2005) form the basis of this thesis. In paragraph 2.1 the research by Chen et al. (2004) is reviewed and thereaf-ter papers that support or contradict their conclusions are discussed. Chen et al. (2004) focus on the price impact of a mutation of the S&P 500 Index on a firm. The research by Barberis et al. (2005) is discussed in paragraph 2.2. This study focuses on index changes in the S&P 500 Index and what effect these mutations have on the co-movement of a firm with the S&P500 Index and the β of that firm. The aim of this thesis is to extent the research by Chen et al. (2004) and Bar- beris et al. (2005) by testing the validity of their results for a different time period and for differ-ent indices used in Europe and Asia. 2.1 Price effect an index change has on a stock Central to the study of Chen et al. is the relation between stock price and listing or delist-ing of a firm on an index. They test for a (significant) difference in the excess return of a stock at addition to/deletion from the S&P 500 Index. Their main findings are: when a firm is added to the S&P 500 Index, that firm experiences a permanent positive price effect; when a firm is deleted from the S&P 500 Index, that firm experiences a smaller, non-permanent negative price ef-fect. In most cases, the firm recovers from this negative effect in less than three months after deletion. Elaborating further on the results by Denis et al. (2003), Chen et al. (2004) explain this price asymmetry as follows: When a firm is added to the S&P 500 Index, investor awareness and monitoring by analysts increases. Therefore, firms may be forced to pursue a more cost-efficient business model and make better business decisions. Another effect of an addition is increased access to capital, since investors and institutions are more willing to lend to listed firms. These consequences cause a permanent positive price impact for the firm after addition to the index. However, when a firm is deleted, they find that this awareness does not diminish and thus a de-letion has a non-permanent price effect. In order to include only regular additions and deletions, Chen et al. (2004) define some criteria for index changes. In this way, they prevent that the returns, used in their calculations, would be less significant because of noise caused by non-regular changes. Firms that are deleted/included because of spin-offs, mergers, acquisitions, restructuring, etc. are excluded from the dataset. In this thesis, the same criteria are used to eliminate irregular changes. The sample period of Chen et al. started July 1962 and ended December 2000. Because of some major changes regarding the S&P 500, they distinguished three sub-periods: July 1962-August 1976, September 1976-September 1989 and October 1989-December 2000.

(8)

The main difference between the first sub-period and the last two sub-periods was the way an-nouncements were made public. Until August 1976, no public announcements were made of a modification in the S&P 500 Index. Therefore, investor awareness was not expected to change after an index mutation and a price effect was unlikely to occur, whereas these would be differ-ent in the latter two sub-periods. Chen et al. (2004) found evidence supporting their theory about the price asymmetry by additions/deletions. This asymmetry exists due to enlarged investor awareness when a stock is added to the S&P 500 Index. When a stock is deleted, this awareness does not decrease. Therefore, the price effect after additions has a more permanent charac-ter than the non-permanent character after deletions from the index. Shleifer (1986) supports the results of Chen et al. (2004). He also found that an addition to the S&P 500 Index is accompanied by a permanent price increase of the added stock. The results of Chen et al. are further supported by Jain (1986), who showed that an addition to the S&P 500 Index was accompanied by a 3% increase in stock price, while a delisting from the index is ac-companied by a 1% price decrease. Mase (2007) contradicts the theory on increased investor awareness. He tested whether inves-tor awareness is a key factor in explaining the positive price effect when a stock is added to an index. Mase compared the increased return of a new entrant firm with the increased return of a long listed firm on the FTSE 100 Index. He found no significant difference between returns of both firms. In the thesis presented here, the FTSE 100 Index is one of the three indices used to analyze the hypotheses regarding the price effect, co-movement and the β of a firm. The re-search of Mase (2007) is extended to a different time period in this thesis. Masse et al. (2000) investigated the price effect on a stock by a listing on/delisting from the TSE 300 Index. The TSE 300 Index was a Canadian index consisting of the 300 most influential stocks traded on the Toronto Stock Exchange, but the S&P/TSX Composite Index replaced this index in 2002. Masse et al. (2000) found a significant positive reaction for stocks added to the TSE 300 Index. For stocks deleted from the index, the authors found an insignificant negative reaction. For the S&P500 Index, Haris and Gurel (1986) corroborated the results of Masse et al. (2000), thus strengthening the theory that the price effect of adding a firm to an index is significantly positive and the delisting of a firm causes a temporary, insignificant negative one.

(9)

2.2 Co-movement of a stock when added to/deleted from an index Barberis et al. (2005) distinguish two general theories about the co-movement of a stock with its index. Their first theory is the traditional theory: co-movement in prices reflects co- movement in fundamental values. Traditional theory dictates that economies are without frictions and all investors are rational, and thus a co-movement in price has to reflect a co- movement in fundamental value. The second theory is that the co-movement in prices is de- tached from the co-movement in fundamental values, due to economies with frictions and irra-tional investors. This theory consists of a set of theories ‘sentiment-based’ and ‘friction-based’ theories. Barberis et al. (2005) divide this set of theories into three. The first view is the category view as defined by Barberis and Schleifer (2003). Investors do not allocate funds at an individual level, but prefer to allocate funds at a category-based level, group-ing assets into categories. By moving funds from one category to another, the assets within these categories face a coordinated demand shock, increasing the correlation between these assets. This increased correlation cannot be explained by the co-movement of fundamental values and according to Barberis and Schleifer (2003) this is evidence of the existence of the general class of non-traditional based theories. The second view Barberis et al. (2005) is the habitat view. This view means that investors prefer a subset of the available securities and therefore have a preferred ‘habitat’. When an investor changes his/her preferred habitat because of risk aversion, liquidity distress or sentiment, the result is a co-movement in the returns of the securities in the specific preferred habitat of that investor. The third view is defined as the information diffusion view, which states that -because of market friction- some information is more easily incorporated in some stock prices than in others. The co-movement of stocks that incorporate information at the same pace is therefore larger than the co-movement of fundamental values would predict. These three views differ from the traditional view, because the traditional view states that the listing or delisting of a firm to/from an index should not change the correlation of that firm’s return with the return of other companies in that index. The inclusion of a firm does not signal an opinion, but is merely an act of making the index as representative as possible for the econo-my. The correlation of a stock’s fundamental value with the fundamental value of the other stocks should therefore not change.

(10)

The first two views, the category view and the habitat view, predict that the addition of a stock to an index causes an increase in co-movement in the return of that stock with the return of other stocks in that index. By adding a stock to an index, it moves in the preferred habitat or catego-ry for many investors. The information diffusion view supports the idea that adding a stock to an index increases the co-movement with the other stocks in the index, while indices are better monitored and thus news is incorporated at a faster pace. To test whether this is the case, Barberis et al. (2005) choose to test whether the β of a stock, with regards to the index, changes after a stock is add-ed/deleted, since the β of a stock with an index is a parameter of the co-movement of that stock with an index. Their dataset contains data from 1976 till 2000 and focuses on mutations in the S&P 500 Index. In order to create a useful sample consisting of solely regular additions and dele-tions, they use the same criteria as Chen et al. (2004). Barberis et al. (2005) found that an addition (deletion) causes a significant increase (decrease) in the β of a stock and thus a significant increase (decrease) in the co-movement of that stock with the S&P 500 Index. They also found that when adding (deleting) a stock from the index, the β of that stock with that index increases (decreases), while the β of that stock with the non-market index decreases (increases). Therefore, their results support the ‘friction-based’ and ‘sentiment-based’ views and not the traditional theory. This thesis will elaborate further on their results and test whether the validity of the second class of theories can be extended to other parts of the world. Coakley and Kougoulis (2004) extended the research by Barberis et al. (2005) for the FTSE 100 Index for the period of 1992-2002, and found supporting evidence for the second general class of theories of Barberits et al. (2005). In this thesis, the research of Coakley and Kougolis (2004) is extended for the time period 2004-2014.

(11)

3. Hypotheses and Methodology In this chapter, the hypotheses are given and the methodology is explained to investigate the effect of an index change on a stock. In paragraph 3.1 a light is shed on the hypotheses and the methodology regarding the price effect of a change. In paragraph 3.2 the hypotheses and methodology in establishing the effect of an index change on the co-movement are clarified. For both the price and the co-movement effect, the date of announcement is chosen as the event date. The efficient market hypothesis dictates that all available information is incorporated into the price of a stock. If one chooses to use the effective date of the inclusion or deletion of a stock, the ‘shock’ and the information provided by this ‘shock’ are already incorporated in the price of the stock. Therefore, the date of announcement is viewed as a more appropriate event date, since this is the day the information is incorporated in the price of the stock. 3.1 Price Effect 3.1.1 Hypotheses The research by Chen et al. (2004) is simulated and the validity of their research is tested for different continents and for a different time period. The hypotheses regarding the price effect of an index change are: Hypothesis 1: The regular addition of a firm to an index has a positive and permanent price effect on that firm. Hypothesis 2: The regular deletion of a firm from an index has negative non-permanent price effect on that firm. 3.1.2 Methodology In this thesis an event study is performed to determine the price effect of an index change on a stock. This type of study is chosen, since it is an adequate way to isolate an event at a specif-ic date. An event window of 61 days is created around the event date, with the event date being day 0. In this way, the difference in returns can be determined. One suitable variable to determine this difference is the Cumulative Average Abnormal Return (CAAR), which was introduced by Fama et al. in 1969. They see the CAAR as a suitable variable because it shows the aggregate effect of abnormal returns. To determine the CAAR, several steps have to be taken.

(12)

The first step is to determine the normal return. This return is determined by using the market model based on the Capital Asset Pricing Model (CAPM) as introduced by Sharpe (1964) and Lintner (1965). The main difference between both models is that the intercept is a constant in-stead of the risk-free rate. This constant, b0, and another market model parameter, b1, can be determined by Ordinary Least Squares (0LS) regression. To retrieve the normal return, DataStream is used. Since a continuous process is assumed, the natural logarithmic return is derived from the normal return. In this way, the returns are consistent with the notion of continuous compounding. The market return is determined by using DataStream and for the afore-mentioned notion the natural logarithmic market return is used. This return is then employed to derive the expected market return, which is the market return minus the risk free rate. Howev-er, since the used market returns are established on a daily basis, the risk free rate is so small that it does not significantly affect the expected market return. The formula for the market model is:

E(R

i,t

) =

b

0

+

b

1

(E(R

m,t

)

(1)

First, the expected return for stock i between time t and t-1, displayed by

E(R

i,t

)

is obtained. Then, with the use of the market return between time t and t-1, which is given as

E

(

R

m,t

)

, the

Abnormal Return (AR) can be calculated. The daily AR is the difference between the normal re-turn and the actual return; it often reveals an event that is a non-regular market movement. Therefore, the AR is a useful variable when performing an event study. The formula for the daily AR is:

AR

i,t

= R

i,t

– E(R

i,t

)

(2)

For stock i between time t and t-1 the AR is

AR

i,t

and the actual return is

R

i,t,

. After the AR is de-termined, the Average Abnormal Return (AAR) is calculated for each day in the event window. This aggregates the abnormal returns for all N stocks to determine the AAR at each time t. In this way, measurement outliers are deleted. To determine the AAR, the following formula is used:

(13)

1 N

AAR

i,t

=

-

Σ

AR

i,t

(3)

N

i=1

The AAR in the formula above is

AAR

i,t

. The last step is to determine the CAAR with the formula

to define the CAAR:

T

CAAR

T

= Σ AAR

t (4)

t=1

In formula 4, the CAAR is

CAAR

T and is obtained by summing the average abnormal returns over

T days in the event. 3.2 Co-movement 3.2.1 Hypotheses On the basis of Barberis et al. (2005), the following hypotheses are formulated and tested for global validity and for validity when studying a different time period: Hypothesis 3: The addition (deletion) of a firm to (from) an index has a positive (negative) co-movement effect of that stock with that index. Hypothesis 4: The addition (deletion) of a firm to (from) an index has a positive (negative) effect on the β of that firm with that index and a negative (positive) effect with the β of that firm with non-index market. 3.2.2 Methodology To test the effect of an index change on the co-movement of a stock with that particular index, two different methodologies are applied to test if there is a significant change in co-movement of a stock. The first methodology is an event study. The second methodology is the ‘calendar time approach’. This second methodology is applied as a robustness check, because often correlation of returns across events occurs when performing an event study (Barberis et al., 2005). To test whether the co-movement of a stock changes significantly after an index change, a uni-variate regression is run. For each event stock j, the model is:

R

j,t

=

α

j

+ β

j

R

index,t

+

ε

j,t

(5)

(14)

The univariate model, displayed above, is run separately for each event and will estimate the co-movement for the pre-event period and the post-event period.

R

j,t

is the return of stock j be-tween time t-1 and t.

R

index,t

is the return of the index for the same period, time t-1 and t. The

slope coefficient of stock j is given by

β

j. The regressions are conducted for daily returns, weekly returns and monthly returns, where the first two returns are run for the same time-window, while for the latter a larger window is chosen. For the daily and weekly data, a pre-event regres-sion is run, starting thirteen months before the event and ending one month prior to the event. The post-event regression is run for a period starting one month after the announcement is made and ending twelve months later. For monthly data, a time-window is chosen of 36 months. The effect on the co-movement is measured by estimating the change in

β

, as

Δβ

.

Next, a bivariate regression is run to test whether a distinction can be made between the two general classes of theories (see theoretical framework): the fundamentals-based theory and the friction- or sentiment-based theories. This regression consists of two

β

’s, namely the

β

index and

the

β

non-index. In this way it can be tested whether an index change causes an increase (decrease)

of its

β

with that index, while simultaneously the

β

of that firm decreases (increases) with the rest of the market.

The bivariate regression is given by:

R

j,t

=

α

j

+ β

j,index

R

index,t

+ β

j,non-index

R

non-index,t

+

ε

j,t (6)

In formula 6, the slope coefficients of the index betas and the non-index betas are represented by

β

j,index and

β

j,non-index.

R

non-index,t is the return of non-index stocks between time t-1 and time t.

As was the case for the univariate regressions, the bivariate regressions are run for a pre-event window and a post-event window. The chosen time-windows are the same as for the univariate regressions and the frequencies are also the same, i.e. daily, weekly and monthly. The difference is that not only

Δβ

index

is registered, but also

Δβ

non-index

.

To test whether the obtained results might be biased because of a correlation of returns across events, a calendar time approach is used. This type of correlation can occur when two events are

(15)

This return is the equal-weighted average return at time

t

for all stocks that are added to an in-dex within the time-window after time

t

. The second portfolio should be a post-event portfolio and have a return of

R

post,t at time

t

. This return is the equal-weighted average return at time

t

for all stocks that have been added to an index within the time-window after time

t

. In the calen-dar time approach, two univariate regressions have to be run to test whether the co-movement of a stock is affected by an index change:

R

pre,t =

α

pre +

β

pre

R

index,t +

ε

pre,t

(7)

And

R

post,t

=

α

post

+ β

post

R

index,t

+

ε

post,t

(8)

Just as in the event study, three time frequencies are run, i.e. daily, weekly and monthly. For the first two frequencies a twelve-month time-window is chosen and for the latter frequency, a time-window of 36 months is chosen. After the regressions, a conclusion can be drawn whether there is significant increase/decrease in the

β

of the stock with the index of interest after an in-dex change. This is the case when

β

post is significantly different from the

β

pre.

The bivariate regressions to test whether the

β

of a firm changes due to an index change when performing a calendar time approach are:

R

pre,t

=

α

pre

+

β

pre,index

R

index,t

+ β

pre,non-index

R

index,t

+

ε

pre,t

(9)

And

R

post,t

=

α

post

+ β

post,index

R

index,t

+ β

post,non-index

R

index,t

+

ε

post,t

(10)

(16)

4. Data In this chapter, data is described. First, the three indices of interest are outlined. Second-ly, the construction of the final sample set is explained. Finally, the descriptive statistics of the data are given. 4.1 Indices One way this thesis elaborates on the work of Barberis et al. (2005) and Chen et al. (2004) is by extending their research to other parts of the world, i.e. Europe and Asia. The three indices of interest are Standards & Poor 500 Index (S&P 500 Index, USA), the Financial Times Stock Exchange 100 Index (FTSE 100 Index, Europe) and the Hang Seng Index (Asia), which are often seen as a benchmark for the equity market of the continent the indices are located. These three indices are introduced in this paragraph. The S&P500 index is an abbreviation for the Standard’s and Poor 500 index and includes 500 large companies that have common stock and which are listed on the New York Stock Exchange or on the NASDAQ. These companies are chosen to ensure that the S&P 500 is a representation of the U.S. market. To determine the weight of the market capitalization of a firm, this index uses free-float weighting. A float factor is assigned to each stock; this factor is determined by making a distinction between shares held by the public and shares held by government, royalty or firm ‘insiders’. After determining this factor, the market capitalization of a firm is multiplied by this factor and thus the weighted value of the firm is determined. A special committee determines whether a firm should be added/deleted on the basis of eight criteria, with as the most im-portant market capitalization and liquidity. The companies have to be representative for the industries in the United States and have to satisfy three liquidity requirements: a market capitalization equal to or greater than $5.3 billion, an annual dollar traded to float-adjust market capi-talization greater than 1.0 and a minimum monthly trading volume of 250,000 shares in each of the six months prior to the evaluation date. In order to calculate the correct value of the index, the market capitalization is divided by the ‘Divisor’. This is a factor that corrects for corporate actions like share issuance, share dividends, mergers, et cetera. In this way, the index is only influenced by stock price changes, not by a change in market capitalization. The FTSE 100 Index is an index that consists of the 100 companies with the highest market capi-talization that are listed on the London Stock Exchange (LSE). These 100 companies represent

(17)

rope. The stocks in the index are weighted by market capitalization assigned by using a free-float weighting system. The market capitalization is corrected by a Divisor in the same way the Divi-sor works regarding the S&P500 Index. Three important criteria to determine whether a firm is eligible to be added to the FTSE 100 Index are liquidity, nationality and free-float. Every quarter the FTSE Group determines whether stocks should be added to/deleted from the index to re-balance the index. This is done every first Friday of the month in March, June, September and December. The Hang Seng Index is an index consisting of the 50 companies with the highest market capital-ization, representing 58% of the market capitalization of the entire Hong Kong Stock Exchange (HKSE). Just as the other two indices, it is a free float-adjusted market capitalization weighted stock market index. To qualify for addition to the index, a firm has to comply with three criteria. The first criterion is that the firm must comprise top 90% of the total market value of all ordi-nary share, the second criterion is it must comprise top 90% of the total turnover on the HKSE and the last criterion is that the firm has a listing history of at least 24 months. The index is re-viewed and rebalanced on a quarterly basis. 4.2 Dataset In order to obtain the datasets for this thesis, several steps were taken. First, it was nec-essary to examine whether an adjustment of an index was a regular one. This was a necessity to exclude the returns of the stocks that were influenced by other factors than that of the announcement of the inclusion or dissolution of a stock. The criteria for selecting, which adjust- ments are regular, were those used by Barberis et al. (2005) and Chen et al. (2004). These crite-ria demand that for an inclusion of a firm to the index as regular, the firm must not be a spin-off or a restructured form of a firm that already resided in the index, the firm must not be increased in size due to a merger or acquisition or the event of inclusion is too close to the end of the sam-ple, making it impossible to determine the post-event

β

of that firm. For the deletion of a stock to be regular, the firm must not be engaged in a takeover, a merger or bankruptcy proceedings or the event of the deletion is too close to the end of the sample, making it impossible to determine the post-event

β

of that firm. To determine the regularity of an inclusion or deletion, several esteemed and well-known news websites were reviewed, for example Bloomberg and Reuters. Information about the effective date and the announcement date of these index changes was also found on these websites. The final dataset covers a period starting from 01-01-2001 till 31-12-2015. This time-window is

(18)

chosen to make it possible to run regressions that cover a period from 01-01-2004 till 31-12- 2014. Since the time-window for monthly frequencies is 36 months, index changes after 30-11-2012 have to be excluded. In table 1 a summary is displayed regarding the final sample set. Table 1

Inclusions & Deletions

The initial sample consists of all deletions from and addition to all the S&P 500 Index, the FTSE 100 In-dex and the Hang Seng Index from January 2004 till December 2014. To be enclosed in the final sample, a change from the index has to be regular and sufficient return data has to be available on DataStream. In appendix 1 and appendix 2, more information is provided about the deletion of non-regular index chang-es. Inclusions & Deletions

The S&P 500 Index The FTSE 100 Index The Hang Seng Index

Inclusions Initial Sample Final Sample Deletions Initial Sample Final Sample 258 152 254 56 125 52 125 35 33 26 17 13 The International Securities Identification Number (ISIN) is a unique identification number of every tradable security and can be used to retrieve the daily, weekly and monthly returns of these securities. In order to obtain these returns, DataStream is used. To obtain the returns of the indices, DataStream is used. In order to run the bivariate regressions and determine the effect of an inclusion/deletion on the

β

index of a stock with a specified index, an alternative index that depicts the rest of the market has

to be used in order to estimate the

β

non-index of a stock. For the S&P 500, the Dow Jones Industrial

Average (DJIA) is chosen to represent the rest of the U.S. market, since the DJIA is often used as a proxy for the U.S. market. The FTSE 250 Index is chosen to represent the rest of the market for the FTSE 100 Index, since a promotion to the FTSE 1OO means a firm leaves the FTSE 250. For a deletion, this is the other way around. Therefore, the FTSE 250 is a good proxy to measure the co-movement effect of an addition to or deletion from the FTSE 100. The proxy chosen for the Hang Seng Index is the Hang Seng China Enterprises Index, since a promotion to the Hang Seng Index often means a departure from the Hang Seng China Enterprises Index for a firm and vice versa. These indices are thus useful when trying to determine whether the

β

index and the

β

(19)

non-4.3 Descriptive Statistics In table 2 the descriptive statistics are shown for all the index changes in the three indi-ces of interest, the S&P 500 Index, the FTSE 100 Index and the Hang Seng Index. The statistics are taken for a period of 61 days, beginning 30 days prior to the event, the date of announcement, and ending 30 days after the event. The total sample period is from 01-01-2004 till 31-12-2014 for all three indices. Table 2

Descriptive Statistics

The descriptive statistics are shown for the abnormal returns regarding the regular changes made to the three indices of main interest from 01-01-2004 till 31-12-2014. The time-window is 61 days with day 0 being the announcement date of index change. Descriptive Statistics

The S&P 500 Index The FTSE 100 Index The Hang Seng Index

Inclusions Mean Median Maximum Minimum Std. Deviation Skewness Kurtosis Observations Deletions Mean Median Maximum Minimum Std. Deviation Skewness Kurtosis Observations 0.009 -0.003 3.076 -2.889 1.497 0.127 2.691 3477 -0.068 0.000 3.946 -4.302 1.875 -0.127 3.438 1342 -0.435 -0.060 2.903 -2.725 1.433 0.133 2.620 1525 -0.082 0.001 5.037 -5.316 2.469 -0.060 3.075 1403 -0.055 -0.061 3.907 -3.701 1.880 -0.149 2.807 1464 0.012 0.042 3.147 -2.850 1.481 0.243 2.930 244

(20)

5. Results In this chapter the results are described. First, the price effect of an index change is de-picted. Secondly, the effects of an index change on the co-movement of a stock are presented. In chapter 6 these results are interpreted and discussed. 5.1 Price effect of an index change on a stock In paragraph 5.1 the results are shown of an event study on the price effect of an index change. In tables 3 and 4, the price effects of an index change are displayed for additions and deletions, respectively. The CAAR is shown for two time-periods and the AAR for three different dates. 5.1.1 Additions In table 3, the CAARs and the AARs are displayed for the three indices for different time- windows. The S&P 500 shows that the CAAR for the time-window previous to the announce-ment is significant at a 1% level: This means that before the inclusion, the stock returns are overall positive. At the dates around the announcement, the AARs are positive, even though not significantly. For the post-event period, the CAAR is highly significant, implying that after an inclusion, the cumulated returns were -on average- negative. The second column, representing the FTSE 100, shows that the CAAR prior to announcement is positive at a 1% significance level. Nonetheless, on the announcement date and both the post-event dates, the AARs and the CAAR are all negative at different significance levels. The results obtained for the Hang Seng show a different pattern. For the pre-event period a highly significant negative CAAR is found, followed by an insignificant positive AAR at the announcement date and an insignificant post-event CAAR.

(21)

Table 3

Abnormal Returns around Additions to the S&P 500 Index, the FTSE 100 In-dex and the Hang Seng Index

The results for additions for the S&P 500 Index, the FTSE 100 Index and the Hang Seng Index. The time period is from 01-01-2004 till 31-12-2014. The initial sample consists of all deletions from and all additions to the S&P 500 Index, the FTSE 100 Index and the Hang Seng Index from January 2004 till Decem-ber 2014. To be enclosed in the final sample, a change from the index has to be regular and sufficient return data has to be available on DataStream. [-30,-1] depicts the CAAR before the announcement of the index change, which is the event date. [0,0] shows the AAR at the event date. [0,1] displays the AAR at the event date plus the AAR of the day after. [1,1] represents the AAR at the day after the announcement. [2,30] portrays the CAAR after the announcement of the index change. The first line in each cell shows the CAAR or the AAR for the specified time period. The second line in each cell displays the standard error in parentheses of the specific CAAR or the specific AAR. The significance of the CAARs and the AARs is tested with a standard t-test. Daily Additions

The S&P 500 Index The FTSE 100 Index The Hang Seng Index Initial Sample Final Sample CAAR [-30,-1] AAR [0,0] AAR [0,1] AAR [1,1] CAAR [2,30] 258 152 0.586*** (0.037) 0.059 (0.198) 0.195 (0.148) 0.135 (0.222) -0.219*** (0.040) 125 52 0.295*** (0.067) -0.599* (0.369) -0.654** (0.284) -0.055 (0.433) -1,270*** (0.075) 33 26 -2.623*** (0.068) 0.2018 (0.485) 0.207 (0.324) -0.469 (0.429) 0.035 (0.072) *, ** and *** denote significance at the 10%, 5% or 1% level, respectively

Figures 1, 2 and 3 show the CAAR and the AARs for a time-window of 61 days for the three indices. In the FTSE 100, the CAAR is positive and starts to fall until three days prior to the an-nouncement of the inclusion, when it starts to increase. When the announcement is made, the CAAR starts to fall again until two weeks after the announcement. At that point, the CAAR starts to increase. In the S&P 500, the effect of an inclusion is positive, but very volatile. During the window of 61 days, the CAAR drops and rises multiple times. The effect of an addition of a stock to the Hang Seng Index seems to be negative, with a small peak right before the event date.

(22)

Figure 1: The AAR and the CAAR for additions to the FTSE100 Index Figure 2: The AAR and the CAAR for additions to the S&P 500 Index -2,5 -2 -1,5 -1 -0,5 0 0,5 1 1,5 -3 0 -2 8 -2 6 -2 4 -2 2 -2 0 -1 8 -1 6 -1 4 -1 2 -1 0 _-8 _-6 _-4 _-2 0 2 4 6 8 ₁₀ ₁₂ ₁₄ ₁₆ ₁₈ ₂₀ ₂₂ ₂₄ ₂₆ ₂₈ ₃₀ CAAR (%) Days

FTSE 100

CAAR AAR -0,6 -0,4 -0,2 0 0,2 0,4 0,6 0,8 1 1,2 1,4 -3 0 -2 8 -2 6 -2 4 -2 2 -2 0 -1 8 -1 6 -1 4 -1 2 -1 0 _-8 _-6 _-4 _-2 0 2 4 6 8 ₁₀ ₁₂ ₁₄ ₁₆ ₁₈ ₂₀ ₂₂ ₂₄ ₂₆ ₂₈ ₃₀ CAAR (%) Days

S&P 500

AAR CAAR

(23)

Figure 3: The AAR and the CAAR for additions to the Hang Seng Index 5.1.2 Deletions In table 4, the CAARs and AARs are displayed for different time periods and the three indices. For the S&P 500, the period prior to the announcement is characterized by a highly significant negative CAAR, just as in the post-event window. Around the announcement date, the re-sults are insignificant and do not have any explanatory value. Column 2 presents the CAARs and AARs found in the FTSE 1OO. Just as in the S&P 500, the pre- and post-event periods show a highly significant negative CAAR. One difference between both indices is that the positive AAR for the day after the event is of small significance for the FTSE 100. The results for the Hang Seng index differ strongly from the other indices. The pre-event CAAR is negative and significant at a 1% level; however, the post-event CAAR is positive at a 1% significance level. Another difference with the two indices is that the AAR for the day after the event is positive instead of negative. -5 -4 -3 -2 -1 0 1 2 -3 0 -2 8 -2 6 -2 4 -2 2 -2 0 -1 8 -1 6 -1 4 -1 2 -1 0 _-8 _-6 _-4 _-2 0 2 4 6 8 ₁₀ ₁₂ ₁₄ ₁₆ ₁₈ ₂₀ ₂₂ ₂₄ ₂₆ ₂₈ ₃₀ CAAR (%) Days

Hang Seng

AAR CAAR

(24)

Table 4

Abnormal Returns around Deletions to the S&P 500 Index, the FTSE 100 In-dex and the Hang Seng Index

The results are shown for deletions for the S&P 500 Index, the FTSE 100 Index and the Hang Seng Index. The time period of interest is from 01-01-2004 till 31-12-2014. The initial sample consists of all deletions from and all additions to the S&P 500 Index, the FTSE 100 Index and the Hang Seng Index from January 2004 till December 2014. To be enclosed in the final sample, a change from the index has to be regular and sufficient return data has to be available on DataStream. [-30,-1] depicts the CAAR before the announcement of the index change, which is the event date. [0,0] shows the AAR at the event date. [0,1] dis-plays the AAR at the event date plus the AAR of the day after. [1,1] represents the AAR at the day after the announcement. [2,30] portrays the CAAR after the announcement of the index change. The first line in each cell shows the CAAR or the AAR for the specified time period. The second line in each cell displays the standard error in parentheses of the specific CAAR or the specific AAR. The significance of the CAARs and the AARs is tested with a standard t-test. Daily Deletions

The S&P 500 Index The FTSE 100 Index The Hang Seng Index Initial Sample Final Sample CAAR [-30,-1] [0,0] [0,1] [1,1] [2,30] 254 56 2.035*** (0.078) -0.215 (0.497) 0.016 (0.329) 0.232 (0.412) -6.579*** (0.095) 126 35 -3.086*** (0.093) 0.336 (0.576) 0.535* (0.39) 0.267 (0.539) -2.623*** (0.156) 17 13 -0.102*** (0.140) -0.526 (0.454) -1.361** (0.602) -0.835 (1.21) 2.209*** (0.126) *, ** and *** denote significance at the 10%, 5% or 1% level, respectively

Figures 4, 5 and 6 show the CAAR and the AARs for a time period of 61 days for the three indices for a deletion. The FTSE 100 shows a mainly negative CAAR, with some positive peaks. To weeks after the announcement, the CAAR starts to ascend. In figure 4, the CAAR of the S&P 500 shows a different pattern. The CAAR is ascending prior to the event and stars to fall shortly after. At the end of the post-event window, the CAAR starts to decline rapidly. Figure 6 shows that the CAAR

(25)

Figure 4: The AAR and the CAAR for deletions from the FTSE100 Index Figure 5: The AAR and the CAAR for deletions from the S&P 500 Index -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 -3 0 -2 8 -2 6 -2 4 -2 2 -2 0 -1 8 -1 6 -1 4 -1 2 -1 0 _-8 _-6 _-4 _-2 0 2 4 6 8 ₁₀ ₁₂ ₁₄ ₁₆ ₁₈ ₂₀ ₂₂ ₂₄ ₂₆ ₂₈ ₃₀ CAAR (%) Days

FTSE 100

AAR CAAR -5 -4 -3 -2 -1 0 1 2 3 -3 0 -2 8 -2 6 -2 4 -2 2 -2 0 -1 8 -1 6 -1 4 -1 2 -1 0 _-8 _-6 _-4 _-2 0 2 4 6 8 ₁₀ ₁₂ ₁₄ ₁₆ ₁₈ ₂₀ ₂₂ ₂₄ ₂₆ ₂₈ ₃₀ CAAR (%) Days

S&P 500

AAR CAAR

(26)

Figure 6: The AAR and the CAAR for deletions from the Hang Seng Index 5.2 Co-movement of a stock when added to/deleted from an index 5.2.1 Event study In paragraph 5.2.1 the results are analyzed of the event study on the effect of an index change on the β of a firm with that index. First, the univariate regression of this effect is ex-plained. Hereafter a bivariate regression is given of the increase/decrease in the βindex and its simultaneous decrease/increase in the βnon-index. 5.2.1.1 S&P500 Index In table 5 the results are given for the S&P500 Index. For the daily returns, the changes in β for additions are all insignificant and thus are of no value when testing the validity of hy-potheses 3 and 4. However, when regarding deletions for daily returns, the results show a small significant negative ∆βindex and a significant positive ∆βnon-index . These results support hypothe-sis 4. For weekly returns, the changes after additions are in favour of hypothesis 4, showing a small significant positive ∆βindex and a significant positive ∆βnon-index at a 1% level. For deletions and the weekly returns, the ∆β is positive and significant at a 1% level, which is in contrast with -3 -2 -1 0 1 2 3 4 5 -3 0 -2 8 -2 6 -2 4 -2 2 -2 0 -1 8 -1 6 -1 4 -1 2 -1 0 _-8 _-6 _-4 _-2 0 2 4 6 8 ₁₀ ₁₂ ₁₄ ₁₆ ₁₈ ₂₀ ₂₂ ₂₄ ₂₆ ₂₈ ₃₀ CAAR (%) Days

Hang Seng

AAR CAAR

(27)

turns are most important. Though the results for daily returns are the most important, the pre-sent results are either insignificant or support hypothesis 4. Table 5

Changes in Co-movement of a Stock added to and deleted from the S&P 500

Index

The results of an event study that makes use of both univariate regressions and bivariate regressions are presented. The main goal of the study was to determine the effect on the co-movement of stock when re-garding a change in the listing to the S&P 500 Index. The final sample consists of all deletions from and all additions to the S&P 500 Index from January 2004 till December 2014. To be included in the final sam-ple, an addition to or deletion from the S&P 500 has to be regular (described in text) and have sufficient

return data on DataStream. For each event stock j, the univariate model Rj,t = αj + βjRindex,t + εj,t. and the

bivariate model Rj,t = αj + βj,indexRindex,t + βj,non-indexRnon-index,t + εj,t are separately estimated for the pre- and

post-event period. Returns from the S&P 500 are obtained from DataStream, as are the returns obtained

for the ‘rest of the market’. For the univariate model, the change in slope across the event date ∆β is

looked at. For the bivariate model, the changes in slopes across the event date ∆βindex and ∆βnon-index are

examined. The results are shown for three frequencies, respectively daily, weekly and monthly. The standard error is depicted in the second line of each cell in parentheses. The significance of the changes in slopes is tested with a standard t-test.

S&P 500 T Univariate Bivariate

∆β ∆βindex ∆βnon-index Daily returns Additions 2004-2014 14144 -0.003 0.058 -0.072 (0.020) (0.058) (0.062) Deletions 2004-2014 5261 -0.358 -0.274* 0.332** (0.054) (-0.150) (0.160) Weekly returns Additions 2004-2014 2817 0.012 0.116* -0.229*** (0.049) (0.065) (0.078) Deletions 2004-2014 1417 0.368*** 0.268 0.133 (0.121) (0.167) (0.197) Monthly returns Additions 2004-2014 1568 -0.055 0.316 -0.289 (0.064) (0.253) (0.270) Deletions 2004-2014 560 0.330 0.331 0.005 (0.205) (0.208) (0.222) *,** and *** denote significance at the 10%, 5% or 1% level

(28)

5.2.1.2 FTSE 100 Index

In table 6, the results are displayed for three frequencies in the FTSE 100 Index. For the daily returns, both hypotheses are supported. The daily returns support hypothesis 4, since there is a highly significant positive ∆βindex and a highly significant negative ∆βnon-index for addi-tions. For deletions, hypothesis 4 is also supported by a highly significant negative change of the

∆β

index and a highly significant positive ∆βnon-index. Both results show that an index change

causes an increase/decrease in co-movement of the stock with that index and a decrease/increase in co-movement with the rest of the market. For deletions, hypothesis 3 is sup-ported since ∆β is significantly negative at a 1% level. The weekly returns of the FTSE 100 Index are of small use when testing both hypotheses, while only the ∆βnon-index 1supports hypothesis 4

at a 10% significance level. For the monthly returns, the conclusion can be drawn that hypothesis 3 is rejected since the ∆β shows a highly significant decrease and that hypothesis 4 is sup-ported by a highly significant decrease of ∆βindex and a highly significant increase of ∆βnon-index for deletions.

(29)

Table 6

Changes in Co-movement of a Stock added to and deleted from the FTSE 100

Index

The results of an event study that makes use of both univariate regressions and bivariate regressions are presented. The main goal of the study was to determine the effect on the co-movement of stock when re-garding a change in the listing to the FTSE 100 Index. The final sample consists of all deletions from and all additions to the FTSE 1OO Index from January 2004 till December 2014. To be included in the final

sample, an addition to or deletion from the FTSE 100 has to be regular (described in text) and have suffi-cient return data on DataStream. For each event stock j, the univariate model Rj,t = αj + βjRindex,t + εj,t. and the

bivariate model Rj,t = αj + βj,indexRindex,t + βj,non-indexRnon-index,t + εj,t are separately estimated for the pre- and

post-event period. Returns from the FTSE 100 are obtained from DataStream, as are the returns obtained

for the ‘rest of the market’. For the univariate model, the change in slope across the event date ∆β is

looked at. For the bivariate model, the changes in slopes across the event date ∆βindex and ∆βnon-index are

examined. The results are shown for three frequencies, respectively daily, weekly and monthly. The standard error is depicted in the second line of each cell in parentheses. The significance of the changes in slopes is tested with a standard t-test.

FTSE 100 T Univariate Bivariate

∆β ∆βindex ∆βnon-index Daily returns Additions 2004-2014 11000 -0.022 0.324*** -0.363*** (0.021) (0.041) (0.042) Deletions 2004-2014 8373 -0.107*** -0.206*** 0.394*** (0.031) (-0.061) (0.063) Weekly returns Additions 2004-2014 2197 -0.010 0.018 -0.107* (0.051) (0.076) (0.062) Deletions 2004-2014 1676 0.010 0.003 0.009 (0.076) (0.077) (0.010) Monthly returns Additions 2004-2014 1152 -0.348*** -0.016 -0.281 (0.076) (0.150) (0.127) Deletions 2004-2014 771 -0.047 -0.067*** 0.069*** (0.109) (0.217) (0.193) *,** and *** denote significance at the 10%, 5% or 1% level, respectively 5.2.1.3 Hang Seng Index In table 7 the results are shown for the three frequencies run in the Hang Seng Index. For additions, for daily returns, hypothesis 3 is supported by a highly significant increase in ∆β. On the other hand, for deletions, hypothesis 3 is rejected because of a significant increase of ∆β. The other results for daily returns are of no significance. When regarding weekly returns, both hypothesis 3 and 4 are rejected. Hypothesis 3 is rejected again by a significant increase of ∆β for deletions, this time at a 1% level. Hypothesis 4 is defied by a highly significant decrease in

(30)

∆β

index and a highly significant positive increase ∆βnon-index for additions. The results for monthly returns have no significant value. Table 7

Changes in Co-movement of a Stock added to and deleted from the Hang

Seng Index

The results of an event study that makes use of both univariate regressions and bivariate regressions are presented. The main goal of the study was to determine the effect on the co-movement of stock when re-garding a change in the listing to the Hang Seng Index. The final sample consists of all deletions from and all additions to the Hang Seng Index from January 2004 till December 2014. To be included in the final sample, an addition to or deletion from the Hang Seng Index has to be regular (described in text) and have sufficient return data on DataStream. For each event stock j, the univariate model Rj,t = αj + βjRindex,t +

εj,t. and the bivariate model Rj,t = αj + βj,indexRindex,t + βj,non-indexRnon-index,t + εj,t are separately estimated for the pre- and post-event period. Returns from the Hang Seng are obtained from DataStream, as are the returns obtained for the ‘rest of the market’. For the univariate model, the change in slope across the event date

∆β is looked at. For the bivariate model, the changes in slopes across the event date ∆βindex and ∆βnon-index

are examined. The results are shown for three frequencies, respectively daily, weekly and monthly. The standard error is depicted in the second line of each cell in parentheses. The significance of the changes in slopes is tested with a standard t-test.

Hang Seng T Univariate Bivariate

∆β ∆βindex ∆βnon-index Daily returns Additions 2004-2014 6223 0.185*** 0. 050 0.107 (0.058) (0.147) (0.109) Deletions 2004-2014 1437 0.0252** -0.184 0.311 (0.114) (0.297) (0.224) Weekly returns Additions 2004-2014 2197 -0.037 -.523*** 0.439*** (0.111) (0.159) (0.120) Deletions 2004-2014 1676 0.509** 0.491 0.006 (0.229) (0.336) (0.247) Monthly returns Additions 2004-2014 648 -0.026 0.120 -0.146 (0.081) (0.166) (0.119) Deletions 2004-2014 160 -0.297 -0.764 0.366 (0.183) (0.478) (0.353) *, ** and *** denote significance at the 10%, 5% or 1% level, respectively 5.2.3 Calendar time approach In this paragraph the results are analyzed that were obtained performing a calendar time

(31)

5.2.3.1 S&P 500 Index

In table 8, the results are displayed for the three frequencies run with the S&P 500 Index. The results of the daily returns show that hypothesis 4 is supported by a highly significant de-crease of the ∆βindex and a highly significant increase of the ∆βnon-index . This shows that, as sug-gested by hypothesis 4, a deletion of a firm from an index indeed results in less co-movement of that firm with the that index and more co-movement with that of the rest of the market. Hypoth-esis 4 is also supported by the results obtained by running regressions for weekly returns. The

∆β

index increases significantly when added to a particular index and the ∆βnon-index decreases significantly due to the addition of the firm. Also, for weekly returns hypothesis 3 is rejected because of the increase in the β at a small significant level. However, for monthly returns, hy-pothesis 3 is supported because of the significant increase of β for additions.

(32)

Table 8

Calendar Time Estimates of Changes in Co-movement of a Stock added to

and deleted from the S&P 500 Index

The results of an event study that makes use of both univariate regressions and bivariate regressions are presented. The main goal of the study was to determine the effect on the co-movement of stock when re-garding a change in the listing to the S&P 500 Index. The final sample consists of all deletions from and all additions to the S&P 500 Index from January 2004 till December 2014. To be included in the final sam-ple, an addition to or deletion from the S&P 500 Index has to be regular (described in text) and have sufficient return data on DataStream. For each event stock j, two separate univariate models are run for

each portfolio on the S&P 500: Rpre,t = αpre + βpreRindex,t +εpre,t and Rpost,t = αpost + βpostRindex,t + εpost,t. The

difference between the pre- and post-event regressions is shown as ∆β.

For each even stock j, two sepa-rate bivariate models are run:

Rpre,t = αpre + βpre,indexRindex,t + βpre,non-indexRindex,t + εpre,t

and

Rpost,t = αpost + βpost,indexRindex,t + βpost,non-indexRindex,t + εpost,t. The difference between the pre- and post-event

regressions is shown as ∆βindex and ∆βnon-index. Returns from S&P 500 are obtained from DataStream, as

are the returns obtained for the ‘rest of the market’. The results are shown for three frequencies, respec-tively daily, weekly and monthly. For daily and weekly, the pre-event (post-event) portfolio’s contain stocks that were added or deleted within one year prior to (in the year past) the event. For monthly re-turns, these windows are extended to three years. The standard error is depicted in the second line of each cell in parentheses. The significance of the changes in slopes is tested with a standard t-test.

S&P 500 T Univariate Bivariate

∆β ∆βindex ∆βnon-index Daily returns Additions 2004-2014 2937 -0.013 -0.002 -0.102 (0.028) (0.062) (0.066) Deletions 2004-2014 5023 -0.018 -0.434*** 0.483*** (0.055) (-0.055) (0.150) Weekly returns Additions 2004-2014 605 0.034 0.152** -0.217** (0.066) (0.123) (0.090) Deletions 2004-2014 506 0.197* 0.261 -0.024 (0.123) (0.164) (0.164) Monthly returns Additions 2004-2014 136 0.170** 0.309 -0.154 (0.090) (0.327) (0.350) Deletions 2004-2014 136 0.365 0.360 0.078 (0.232) (0.235) (0.252) *,** and *** denote significance at the 10%, 5% or 1% level, respectively

(33)

5.2.3.2 FTSE 100 Index In table 9 the results are displayed for the three frequencies run regarding the FTSE 100 Index. For daily returns, both hypothesis 3 and 4 are supported by highly significant results. Hypothesis 3 is supported by a highly significant decrease in ∆β, thus showing that a deletion indeed causes less co-movement of a stock with the index it is deleted from. Hypothesis 4 is sup-ported by the results for both additions and deletions. For additions there was a positive ∆βindex and a negative ∆βnon-index at a 1% significance level, while for deletions there was a highly signif-icant negative ∆βindex and a highly significant positive ∆βnon-index. The results for weekly returns are not of any explanatory value, except for the fact that there is small significant negative ∆β non-index . When observing the results for monthly returns, the conclusion can be drawn that hypoth-esis 3 is rejected by a highly significant negative ∆β when regarding additions.

(34)

Table 9

Calendar Time Estimates of Changes in Co-movement of a Stock added to

and deleted from the FTSE 100 Index

The results of an event study that makes use of both univariate regressions and bivariate regressions are presented. The main goal of the study was to determine the effect on the co-movement of stock when re-garding a change in the listing to the FTSE 100 Index. The final sample consists of all deletions from and all additions to the FTSE 100 Index from January 2004 till December 2014. To be included in the final sample, an addition to or deletion from the FTSE 100 Index has to be regular (described in text) and have sufficient return data on DataStream. For each event stock j, two separate univariate models are run for

each portfolio on the FTSE 100: Rpre,t = αpre + βpreRindex,t +εpre,t and Rpost,t = αpost + βpostRindex,t + ε

post,t. The differ-ence between the pre- and post-event regressions is shown as ∆β. For each event stock j, two separate

bivariate models are run:

Rpre,t = αpre + βpre,indexRindex,t + βpre,non-indexRindex,t + εpre,t

and

Rpost,t = αpost + βpost,indexRindex,t + βpost,non-indexRindex,t + ε

post,t. The difference between the pre- and post-event re-gressions is shown as ∆βindex and ∆βnon-index. Returns from FTSE 100 are obtained from DataStream, as are

the returns obtained for the ‘rest of the market’. The results are shown for three frequencies, respectively daily, weekly and monthly. For daily and weekly, the pre-event (post-event) portfolio’s contain stocks that were added or deleted within one year prior to (in the year past) the event. For monthly returns, these windows are extended to three years. The standard error is depicted in the second line of each cell in pa-rentheses. The significance of the changes in slopes is tested with a standard t-test.

FTSE 100 T Univariate Bivariate

∆β ∆βindex ∆βnon-index Daily returns Additions 2004-2014 2807 0.033 0.357*** -0.404*** (0.027) (0.050) (0.051) Deletions 2004-2014 2663 -0.148*** -0.193*** 0.411*** (0.036) (-0.065) (0.067) Weekly returns Additions 2004-2014 561 -0.064 0.042 -0.148* (0.075) (0.086) (0.080) Deletions 2004-2014 532 -0.051 -0.078 0.023 (0.102) (0.103) (0.023) Monthly returns Additions 2004-2014 144 -0.273*** 0.057 -0.251 (0.103) (0.178) (0.147) Deletions 2004-2014 136 -0.047 -0.372 0.0454* (0.171) (0.308) (0.258) *,** and *** denote significance at the 10%, 5% or 1% level, respectively 5.2.3.3 Hang Seng Index