1
Price and Volume Effects Associated with Index Composition
Changes:
Evidence from the Euronext Amsterdam Exchange Index
Patrick Klink (s2016680)University of Groningen
Abstract
This paper investigates AEX index composition changes. The event of an index composition change has two important days: the announcement day and the effective day. Additions to the index are expected to experience a stock price increase. Symmetric effects are expected for deletions. Empirical tests are done to find price and volume effects. The results show a large increase in stock price around the announcement day for additions, which is consistent with the certification and liquidity hypothesis. The price effect is not accompanied by a large volume effect. The largest volume effect is found around the effective day, which is consistent with the downward sloping demand curve and price pressure hypothesis. The price and volume effects are temporal and revert to their original levels.
JEL Classification: G11, G14
Keywords: AEX index, stock market index, price pressure, price reversal, liquidity, event study
Master thesis MSc Finance June 26, 2015
2 1. Introduction
After the financial crisis that started in 2008, an investment trend started that caused a shift in portfolio management from active to passive (Walker, 2013). Large fees were asked by active managers that did not deliver positive alphas. Therefore, wealthy people invested more of their money in passive investment funds. One of the advantages of passive investment funds is low costs. Operation and management fees are lower and fewer stock specific information is needed, resulting in lower costs of information. Further, the number of trades is low, resulting in low transaction costs. Stocks only have to be bought or sold when the weighing of a company in the index changes or when companies are added to and deleted from the index. However, trading on changes in index composition has some implication. Since it is the intention of passive portfolio managers to have a low tracking error, passive managers have to buy all companies added to the index and sell all companies deleted from the index. This will lead to an increase in demand and consequently might result in a price effect. Short term investors can anticipate on this obligation of passive investors and buy companies announced to be added to the index before the change becomes effective. Passive investors can end up buying index additions at a higher price and short term investors can earn a return by trading on this.
Index composition changes do not only cause short term effects. Literature provides theories that predict long term price effects for companies added to an index. There are serveral reason for this. First, information availability will be higher for companies after they are added to an index which reduces investors’ costs of information. With lower costs, investors are prepared to pay a higher stock price. Further, the principal-agent information asymmetry is reduced leading to more value enhancing decisions from the management, resulting in improved corporate performance. Finally, a long term effect is expected due to passive investors. The demand in stocks of added companies will increase permanently resulting in a permanent increase in stock price. One would expect that a stock represents the fundamental value of a company. Although literature predicts short and long term effects, the fundamentals of the company do not change. This inconsistency makes trading on index composition changes interesting and for passive investors something to bear in mind when tracking an index.
3 attention in equity markets, making profits by trading on under- and overpricing becomes more difficult. However, equity market mechanisms, such as index composition changes, are still able to create opportunities for investors to generate profits. Until now, mainly large indices with high trading volumes have been studied. Little research is done on the profitability of trading on composition changes in small indices. Compared to large indices, small indices might be tracked by fewer passive investment fund what might lead to a smaller price effects when the composition is changed.
In this paper short and long term price effects of composition changes in a small index are studied. The index that is studied is the AEX index. This index is the most important Dutch stock index and contains the 25 companies with the highest trading volume. The effect of additions and deletions in the AEX index, has never been studied before. Compared to the popular S&P 500 index, composition changes of the AEX index are much easier to predict. This paper tests if the price and volume effects found for large indices, are also observed in the AEX index. It is interesting to study if price effects occur in small indices with high composition predictability and whether the effects are temporal or permanent. Price effects on the announcement day are interesting for investors trying to make a short term profit. A permanent price effect would affect the company itself. For added companies market capitalization rises, resulting in a higher company value and lower cost of capital. In this paper I do an event study to test for short term abnormal returns resulting from changes in AEX index composition. To test for long term effects, I use the Carhart (1997) four factor model. As a last test I do a regression analysis to test if company characteristics determine the price effect.
The results of the event study show large positive cumulative abnormal returns (CAR) for added companies in the days before the announcement day. The positive CAR of the additions revert to zero within 50 days. A large increase in volume is found for additions and deletion around the effective day. Inconsistent with the theory, long term effects are not found in the four-factor model. However, a combined portfolio of a short position in additions and a long position in deletion generates a significantly large long term return.
4 2. Literature
The efficient market hypothesis predicts that securities reflect all publicly available information. The market is ‘’semi-strong-efficient’’ (Fama, 1970), meaning that investors have all publicly available information about the security and this information is at all times reflected in the security price. For investors it is impossible to beat the market on expert stock selection or market timing, since it is impossible to buy undervalued securities and sell overvalued securities. The only way for an investor to beat the market is by taking on more risk. Securities are assumed to be near substitutes of each other. The index where the security is listed is irrelevant. The hypothesis is highly controversial and often disputed. Multiple studies reject the efficient market hypothesis. Harris and Gurel (1986) provide evidence that the efficient market hypothesis does not hold for additions to and deletions from the S&P 500. They find positive significant price effects after the announcement day. The authors explain these results by either price pressures or by imperfect substitutions among securities. Chen et al. (2004) provide four hypotheses why the efficient market hypothesis should be rejected and why there should be a stock price effect for index composition changes. The hypotheses are: the certification hypothesis, liquidity hypothesis, downward sloping demand curve hypothesis and price pressure hypothesis.
2.1 Certification hypothesis
5 in indices where the level of indexing is small or nonexistent, companies that are added experience abnormal returns similarly large to those for additions to the S&P 500. These two studies together provide evidence that index changes are consistent with the certification hypothesis (Chen et al, 2004).
Denis et al. (2003) test the hypothesis whether the event of an index inclusions convey new company specific information. They find that companies added to the S&P 500 have significantly better corporate performance and higher earnings per share. The study concludes that an index inclusion is not an information-free event. However, they stay inconclusive whether the improved corporate performance is the result of the index inclusion or if the index inclusion is the result of an improved corporate performance.
Goetzmann et al. (1986) study an event in 1983 where 7 telecommunication companies are removed from the index and replaced by 7 new telecommunication companies. In particular the stock returns of the deleted companies where analyzed. The study finds that deleted companies have significant negative returns after the deletion. The researchers argue that this decrease in stock price is not caused by an immediate loss of information but by an anticipation on the quality and quantity of future information about the company. Investors expect that analysts will focus less on these companies resulting in less reliable predictions about future performance. In other words, the deletion is a clear signal to the market regarding the quantity and quality of future analysis and disclosure.
6 significantly higher in the year before the deletion, compared to one year after the deletion. In contrast, added companies have a richer information environment one year before and one year after the inclusion in the Dow Jones Industrial Average compared to their matched counterparts. Beneish and Gardner (1995) attribute this result to the editors of the Wall Street Journal adding large, more widely followed companies are added to the index. No significant differences in the quantity of information in the period one year before and one year after the addition to the index are found. Given the results from this study, the assumption from investors that information availability will be larger and of higher quality for companies added to an index is not supported. Results do show support for this assumption for deleted companies.
2.2 Liquidity hypothesis
The liquidity hypothesis predicts that adding a company to an index will result in higher liquidity of the stock, resulting in a lower discount rate and consequently a higher stock price. The first explanation for an increase in liquidity starts with the same reasoning as the certification hypothesis. Both hypotheses assume that an added company will be followed more closely by investors and analysts in the future. The liquidity hypothesis predicts that the monitoring on the management increases and hence reduces information asymmetry between agent (managers within the company) and principal (investors). As a consequence the stock is traded more actively resulting in a higher trading volume. More trade makes it possible to buy and sell stocks more easily and lowers the bid-ask spread. A lower bid-bid-ask spread can be regarded as a reduction in trading cost, because a lower price gap has to be breached before making a return. In order words, due to a lower bid-ask spread a lower required rate of return is perceived by investors. Deleted companies are expected to be followed less closely by investors and analysts, resulting higher agency costs and consequently lower trading volume, higher bid-ask spread and a permanently lower stock price.
7 in a permanently higher stock price. For deletions higher shadow costs are expected resulting in a higher required rate of return and a lower stock price.
The literature provides multiple theoretical reasons that support the liquidity hypothesis. The increase in attention from analysts and investors will lower the asymmetry component in the bid-ask spread (Amihud and Mendelson, 1986). Trading volume will increase for companies new in the index due to lower expected information asymmetry. The increase in trade volume results in higher liquidity due to a reduction in the inventory component of market makers (Sofianos, 1993). However, additions to an index suffer from a long-term shift in ownership structure. Equity is more possessed by index traders that are uninformed and have a buy and hold strategy causing a drop in trades. This can result in an increase of the information asymmetry component of the bid-ask spread (Hedge and McDermott, 2003).
Evidence for the liquidity hypothesis is not very strong. Multiple studies do find an increase in liquidity (Chen et al. 2004; Wooldridge and Ghosh, 1986) for companies added to an index. Hedge and McDermott (2003) find a significant reduction in bid-ask spread for added companies due to a decrease in direct transaction costs, where no significant changes in bid-ask spread are found for deleted companies. The change in bid-ask spread for added companies is primarily driven by changes in the inventory cost rather than the information asymmetry cost. However, multiple studies (Collins et al. 1995; Beneish and Gardner 1996; Edminster et al. 1994; Erwin and Miller, 1998) on the effect on the bid-ask spread do not find significant results. Woolridge Ghosh (1986) find an increase in liquidity after a company is added to the S&P 500, however this increase is not accompanied by excess trading volume.
8 2.3 Downward sloping demand curve hypothesis
Investment portfolios can be actively and passively managed. Active managers try to pick stocks that outperform the market. They try to exploit market inefficiencies by buying stocks when they are underpriced and sell when overpriced. Passive investors try to create a portfolio that follows their benchmark as close as possible. They only trade when the composition of the benchmark changes or when individual stock are reweighted. The benchmark of index investors is often a large index. When an company is added to an index, passive investors that track that index will have to adopt this company in their portfolio to keep a low tracking error.
The downward sloping demand curve hypothesis (also known as the imperfect substitute hypothesis) is first described by Shleifer (1986). The hypothesis predicts a permanent price and volume effect for added companies, caused by passive investors tracking the index. The hypothesis assumes that no private information is involved in the event of an index addition or deletions. Another assumption of the hypothesis is that stocks have imperfect substitutes, which results in less than perfectly elastic long term demand curves. This assumption is supported by the fact that passive investors adjust their portfolios when index composition changes are made. The effect of an addition to an index results in excess trade volume. This increase in demand will be compensated by a permanent shift in the demand curve. Given the hypothesis that the demand curve is downward sloping, the stock price of the added company will increase. The new supply-demand equilibrium is not expected to revert to its old equilibrium in the long term, resulting in permanent price effects. For deletions a symmetric effect is expected, where an increase in trade volume results in a lower stock price.
Several researchers provide support for a downward sloping demand curve. Shleifer (1986) describes in his research several results that support the downward sloping demand curve. The first result is found in research on stock buybacks of companies, this event is accompanied by an price increase. The hypothesis is also strongly corroborated by the tax-loss explanations of the January effect. The results of Shleifer (1986) show abnormal announcement returns for additions of 2,79% on the announcement day.
9 price increase (decrease) for index additions (deletions). As the post-event period in most studies is small, the results of these studies cannot be interpret as evidence of no price reversal. However, the these results can neither be interpret as evidence against a full price reversal.
2.4 Price pressure hypothesis
The price pressure hypothesis predicts an increase in stock prices when there is a positive shock in demand. The hypothesis presumes that investors that accommodate demand shifts, such as index inclusions, should be compensated for the portfolio risk and transaction costs. They bear these costs when they immediately buy or sell stocks which they otherwise would not trade. For example, providers of liquidity may demand a premium to absorb a demand or supply shock. Immediately after the addition to the index the trading volume will sharply rise and stock prices will rise above its equilibrium. The market is not able to absorb a sudden demand shock which leads to an immediate price reaction. In the long term, the stock price of an added company a will return to its original equilibrium. The price pressure hypothesis assumes a short term less than perfect elastic demand curve, which is similar to the downward sloping demand curve hypothesis. However, opposite to the downward sloping demand curve hypothesis, the price pressure hypothesis assumes a perfectly elastic long term demand curve, causing a full price reversal. For deleted companies the hypothesis predicts a symmetric effect. Trading volume will increase due to the need for index investors to track the index and sell the stock, this demand shock lead to a temporal price decrease.
10 to absorb the demand shock following an index inclusion. Their results show a partly absorbed demand shock from day one to day six. This effect is only observed around the effective day. Evidence from Lamoureux and Wansley (1987) provides additional support for the existence of the price pressure hypothesis. They find excess returns of 2.3% for additions and -2.6% for deletions on the announcement day. These price effects completely reversed within 20 days after the announcement.
2.5 Summarizing the hypothesis
In table 1 a summary of the predicted price and volume effects is shown. Where every hypothesis predicts a positive price effect for additions and negative price effect for deletions they stay inconclusive if the effects are temporal or permanent. Trading volume is expected to increase in all hypotheses, except for the liquidity hypothesis that predicts a decrease in trading volume for deleted companies. The hypotheses are rather vague in their prediction if the price and volume effects happen at the announcement day or effective day. In general, information driven events are predicted to happen on the announcement day and more mechanical (information free) events, like portfolio adjustments made by index funds, are predicted to happen on the effective day. The certification and liquidity hypothesis assume that an event of an index addition or deletion contain information. Therefore, price effects are predicted on the announcement day. The downward sloping demand curve and price pressure hypothesis assume that a composition change of the index is an information free event. Price effects will therefore happen at the effective day. Note that this is a rather general classification. In the studied literature this classification is not used consistently. Some price effects observed at the announcement day are attributed to the downward sloping demand curve or price pressure hypothesis. Other literature finds support on the certification and liquidity hypothesis on the effective day.
Table 1. Predicted effects for price and volume effects for companies added and deleted from an index Hypothesis Addition/Deletion Price effect Duration Trading
volume
Duration Day of price effect Certification
hypothesis
Additions Increase Permanent Increase Temporal Announcement Deletions Decrease Permanent Increase Temporal Announcement Liquidity
hypothesis
Additions Increase Permanent Increase Permanent Announcement Deletions Decrease Permanent Decrease Permanent Announcement Downwards
sloping demand curve hypothesis
Additions Increase Permanent Increase Temporal Effective
11 3. AEX index
The AEX index reflects the performance of the 25 most actively traded stocks listed on Euronext Amsterdam. The day-to-day management of the index is done by Euronext. The AEX index is the most wildly used indicator for the Dutch stock market and first introduced on January 3 1983. The weighting of individual companies in the index is based on market capitalization. The AEX index serves as an underlying for several derivative instruments such as, index tracker funds, ETFs, options and futures. The index composition changes are made annually in March. However Euronext has created the option for fast additions and deletions four times a year in March, June, September and December.
3.1 Adoption criteria
For adding a company to the index, Euronext has set some rules which determine the composition of the index. The rules are executed by the AEX Steering committee. This committee consist of at least 3 natural persons that are not employed at Euronext and are appointed for 3 years. In March the AEX Steering committee ranks all companies, that fulfill the velocity and the free float requirements, to their trading volume. It is beyond the scope of this paper to mention all requirements, though one of the most important requirements is an annual stock turnover of at least 10% of the total outstanding stocks. The 23 companies with the highest trade volume are automatically included in the AEX index. The last two companies will be selected from the companies ranked from place 24 to 27, where companies currently included in the index are preferred over companies not currently included. The rules for fast index adjustments are much more strict. For a company to be added to the AEX index, needs his trading volume to rank at least 15th. In addition to this, the company has to be listed for at least 40 trading days.(Euronext, 2014). The AEX Steering committee has the option to make index weighing adjustments when corporate events (takeovers, mergers, acquisitions, liquidations or bankruptcies) affect the value of one or more stocks in the index. The committee has to make these adjustments to provide an up to date reflection of the value of the underlying stocks. (Euronext, 2014).
3.2 Differences between the S&P 500 and the AEX index
12 the S&P 500 seeks to reflect the American stock market with sector representation. Companies in the S&P 500 are considered to be the leading companies in their industry. Second, the rules used by Euronext to manage the composition of the AEX are publicly known, whereas composition changes in the S&P 500 are much harder to predict. Additions have to fulfilling an number of criteria but the company is only added with the approval of the U.S. Index Committee. Composition changes happen as needed and not on an annual or semi-annual basis.
3.3 Expected price effects in the AEX index
AEX investors are, compared to S&P 500 investors, much more able to anticipate on index additions and deletion. In a stock market with the characteristics described by the efficient market hypothesis, no price and volume effects are expected when a company is added to or deleted form an index. The market knows the composition change is coming and make trades on this information in advance. Literature provides examples which convince me the efficient market hypothesis can be rejected for composition changes in small indices. Duque and Madeira (2004) study the main Portuguese stock index which has similar addition and deletion criteria and is also managed by Euronext. The results of their paper show positive announcement effects for added companies and negative announcement effects for deletions. In the study of Chakrabati et al. (2004) 29 MSCI country indices are tested on price effects by index composition changes. MSCI does not provide clear rules for index additions and deletions. The country indices are a mix of large and small indices. The results show significant positive abnormal returns for additions and significant negative returns for deletions at the announcement day. The results of these studies justify the research in this paper in using the AEX index, which is a relatively small and has publicly known rules for additions and deletions.
13 4. Data and methodology
4.1 Sample
The data on AEX index additions and deletions is obtained from Euronext. The AEX index is one of the indices managed by Euronext. On the website of Euronext press releases are used to obtain index additions, deletions and the reason for the composition change. From the total sample of index additions or deletions only involuntary index changes are included in the sample. Voluntary additions or deletions, caused by mergers, takeovers, spin-offs and bankruptcies, are not included. These composition changes may disturb the results. The announcement and effective date are also obtained from these documents. In the press releases announcement and effective dates are not available for composition changes before 2009. Effective days are obtained from the file ‘’The composition of the Amsterdam Exchange-index (AEX) from 1983’’ on their website and the corresponding announcement days are obtained from financial newspapers. Over a period from January 1983 until April 2015 a sample of 56 added and 32 deleted companies is made. Data on stock prices and trading volume are obtained from DataStream. For several companies no data is available which reduced the sample size to 46 additions to the index and 26 deletions. For an event study this is a relatively small sample. However, compared to other studies on index compositions changes, a small sample is not uncommon (Kaul et al. 2002, 31 additions. Goetzmann et al. 1986, 7 deletions. Duque and Madeira 2004, 17 additions, 22 deletions). The sample in this study has a large difference in the number of companies added to the index and deleted from the index in the sample period. There are two reasons for this difference. First, the AEX index started with 15 companies in 1983 and currently consist of 25 companies. So, at least 10 additional companies had to be added, whereas no companies had to be deleted. Second, many deleted companies are removed from the sample due to mergers and takeovers. An overview of the sample with corresponding announcement and effective days is presented in appendix 1 and 2.
4.2 Methodology
14 downward sloping demand curve and price pressure hypothesis predict a price effect on the ED. The majority of the literature finds price effects on the AD, however a number of studies also finds price effects at the ED. This paper will study both days and treat them as separate events. The announcements of compositions changes in AEX index are made after the stock markets are closed. Because the initial reaction of the market on the news is one day later, this will be the event day (T=0).
The event study will test whether the companies in the sample have abnormal returns surrounding the event day. As an event period this paper uses a period of 20 days before the event until 60 days after the event. Compared to other studies a relatively long period before the event day is taken. Composition changes in the AEX are more predictable, therefore price effects before the AD possibly show up. A long time period after the event is taken to test for price reversions, as predicted by the price pressure hypotheses. Multiple sub-periods within the event period will be tested to isolate ex post and ex ante price effects. Returns are calculated as arithmetic returns.
𝑅𝑡 =𝑃𝑃𝑡
𝑡−1− 1 (1)
Where 𝑅𝑡 is the simple daily return at time 𝑡 and 𝑃𝑡 is the stock price at day 𝑡. I use the methodology
of Brown and Warner (1985) in this study. The authors use three models to calculate abnormal returns. The first model the authors use in their study is the Mean Adjusted model. Abnormal returns are calculated as the difference between the daily stock return and the average stocks return of an estimation period. 𝐴𝑖,𝑡= 𝑅𝑖,𝑡− 𝑅̅𝑖 (2) Where 𝑅̅𝑖 = 1 223 ∑ 𝑅𝑖,𝑡 −21 𝑡=−244 (3)
𝐴𝑖,𝑡 is the abnormal returns of stock 𝑖 at day 𝑡. and 𝑅̅𝑖 is the average daily return of stock 𝑖 from day
-244 until day -21. The second model to calculate abnormal returns Brown and Warner (1985) use is the Market Adjusted model. Abnormal returns are calculated as the difference between the daily stock return and the return of a benchmark on the same day.
15 Where 𝑅𝑚,𝑡 is the return of the AEX at day 𝑡. The third method of calculating abnormal returns is
with an the OLS market model. Abnormal returns are calculated as the difference between the daily stock return and the daily return predicted by an OLS model. Abnormal returns are calculated as:
𝐴𝑖,𝑡= 𝑅𝑖,𝑡− 𝛼̂ − 𝛽̂𝑅𝑚,𝑡 (5)
𝛼̂ and 𝛽̂ are OLS coefficients from the estimation period day -244 until day -21.
4.3 Market bias and event clustering
Abnormal returns in the Market Adjusted model are returns in excess of the market. The AEX index is used as benchmark because it is the most widely known Dutch index and serves as underlying for serval derivative instruments. However, there is a potential problem worth mentioning when using the AEX index as benchmark. Companies that are announced to be deleted still contribute (together with 24 other companies) to the return of the AEX index (=𝑅𝑚,𝑡). The index holds a relatively low
number of companies, so the impact of a single company on the performance of the index is large. This could bias the results and lower the significance. Another possible bias in the results is caused by event clustering. Especially in the later years of the sample Euronext tend to add and delete up to three companies at the same time. Results can be biased because possible endogenous events that affect the whole benchmark. This will affect the results of the event study. These two biases combined, could affect the abnormal returns of deleted companies, especially in the later year of the sample. Individual stock returns of these companies contribute for a relatively large part to the return of the AEX index. To test whether these biases affect the results, the Market adjusted returns and the OLS market model are tested again using a different benchmark: the MSCI Netherlands. This index consist of 48 Dutch companies. The higher number of companies in the index leads to a lower impact of a single company compared to the AEX index. When the initial results (with the AEX as benchmark) are biased the new benchmark will reduce the bias, resulting in possibly larger price effects.
4.4 Abnormal trading volume
The hypotheses described in section 2 predict an increase in the trading volume for additions and deletions. To test this prediction, the methodology of Beneish and Whaley (1996) is used to calculate abnormal trading volume (ATV). The ATV is calculated as:
ATV =V𝑖,𝑡
𝑉̅𝑖 (6)
16
𝑉̅𝑖 =2231 ∑−21𝑡= −244V𝑖,𝑡 (7)
V𝑖,𝑡 is the number of traded stocks of stock 𝑖 on day 𝑡 and 𝑉̅𝑖 is the average trading volume of stock 𝑖
for day -244 until -21. Note that in this formula there is abnormal positive trading volume when ATV is larger than one and negative abnormal trading volume when ATV is lower than one.
4.5 Additional tests
4.5.1 Calendar-Time Portfolio approach
The certification, liquidity and downward sloping demand curve hypothesis all predict a long term price effect for additions and deletions. Whereas the price pressure does not predict a long term price effect. The Calendar-Time Portfolio analysis used by Døskeland and Hvide (2011) tests if price effects are long term. Consistent with this method I create two portfolios, one portfolio for added companies and one portfolio for deleted companies. Stocks are bought one month after the AD and held in the portfolio for one or two months. Every company is equally weighted. Monthly returns are continuously compounded and calculated as:
𝑅𝐶𝐶𝑖,𝑞 = 𝐿𝑁(𝑃𝑃𝑖,𝑞
𝑖,𝑞−1) (8)
Where 𝑅𝐶𝐶𝑖,𝑞 is the continuously compounded return of stock 𝑖 in month 𝑞 and 𝑃𝑖,𝑞 the stock price
of stock 𝑖 in month 𝑞. The portfolio return in month 𝑞 is calculated as the average monthly return of all stocks included in the portfolio in month q and is denoted as: 𝑅𝐶𝐶̅̅̅̅̅̅𝑞 . To obtain a risk adjusted
alpha the monthly returns of both portfolios are regressed on the loading of specific types of risk. The model of is Carhart (1997) is used as time-series regression equation:
𝑅𝐶𝐶
̅̅̅̅̅̅𝑞− 𝑅𝑟𝑓,𝑞= 𝛼 + 𝛽1(𝑅𝑚,𝑞− 𝑅𝑟𝑓,𝑞) + 𝛽2𝑆𝑀𝐵𝑞+ 𝛽3𝐻𝑀𝐿𝑞+ 𝛽4𝑊𝑀𝐿𝑞+ 𝜀𝑞 (9)
Where 𝑅𝑚,𝑞 is the monthly market return in month 𝑞. The monthly AEX returns will be used as
market return. 𝑅𝑟𝑓,𝑞 is the risk free rate in month 𝑞. The SMB, HML and WML are returns on the
17 the additions portfolio and regressed on the risk factors. The alpha of this tests indicates the differences in returns between both portfolios controlling for the four factors.
The factors in the four-factor model are obtained from the website of Kenneth French. Market factors specifically for the Dutch market are not reported on his website. Therefore, market factors are chosen that fit the data as close as possible. The ‘’Fama/French European Factors’’ are used. These factors are obtained from 16 developed European countries (including the Netherlands). Although not complete representative for the Dutch market, I expect similar results with this data since the Dutch market is highly correlated with its surrounding countries (Song et al. 2011).
4.5.2 Bid-ask spread
According to the liquidity hypothesis an index addition will yield a higher stock price and a deletion will yield a lower stock price. A reason for this prediction is a change in trading volume resulting in a change in bid-ask spread. Consistent with the method of Amihud and Mendelson (1986) the difference between the bid-ask spread before and after the AD is tested to find support for the liquidity hypothesis. The difference between the bid and ask price will be calculated as a percentage:
𝑆𝑝𝑟𝑒𝑎𝑑𝑖,𝑡 =𝐴𝑃𝐵𝑃𝑖,𝑡
𝑖,𝑡− 1 (10)
Where 𝑆𝑝𝑟𝑒𝑎𝑑𝑖,𝑡 is the percentage difference between the bid and ask price of company 𝑖 at day 𝑡.
𝐴𝑃𝑖,𝑡 and 𝐵𝑃𝑖,𝑡 are the ask price and bid price of stock 𝑖 at day 𝑡. The sample period will be split up in
4 periods of 20 days. The first period contains the 20 days prior to the AD and the sub-periods 2, 3 and 4 the sub-periods after the AD. Using a mean comparison test, the difference between the first period and rest of the sub-periods is tested.
4.5.3 Company characteristics
18 analysis is done. The test helps in understanding what types of companies experience a large price effect when an index addition or deletion happens. The company characteristics (independent variables) can roughly be divided into two groups. The first group of variables reflect company specific characteristics. These variables are the result of the company’s operating activities and corporate decisions. The company is able to influence these variables, where investors are not. The variables in this group are: Leverage, size, profitability, change in profitability and dividend payout. The second group consist of variables mainly driven by stock trades in the secondary stock market. The company has little impact on these variables. This group consists of: stock price volatility and bid-ask spread. As a dependent variable the time period with the largest CAR is used. The equation used in the regression analysis is:
𝐶𝐴𝑅𝑖 = 𝛼 + 𝛽1𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒𝑖+ 𝛽2𝑆𝑖𝑧𝑒𝑖+ 𝛽3𝑃𝑟𝑜𝑓𝑖𝑡𝑎𝑏𝑖𝑙𝑖𝑡𝑦𝑖+ 𝛽4𝛥 𝑃𝑟𝑜𝑓𝑖𝑡𝑎𝑏𝑖𝑙𝑖𝑡𝑦𝑖 + 𝛽5𝐷𝑖𝑣𝑖𝑑𝑒𝑛𝑑 𝑝𝑎𝑦𝑜𝑢𝑡𝑖+ 𝛽6𝑉𝑜𝑙𝑎𝑡𝑖𝑙𝑖𝑡𝑦𝑖+ 𝛽7𝐵𝑖𝑑 𝑎𝑠𝑘 𝑠𝑝𝑟𝑒𝑎𝑑𝑖+ 𝜖𝑖
Where 𝐶𝐴𝑅𝑖 is the cumulative abnormal return of stock 𝑖 around the AD. In appendix 9 a detailed
description of the independent variables is presented. All observations are downloaded from DataStream. The observations are, except for the Change in profitability variable, all made in the period before the AD.
4.6 Descriptive statistics
The descriptive statistics of the sample are presented in table 2. After removing companies without available data the sample decreased to 46 additions and 26 deletions. For a number of companies the daily stock prices are not obtained for the entire sample period. This is caused by companies added and deleted a few weeks before the data was downloaded. The post-event period, up to 60 days after the event, was partly in the future at the moment the data was downloaded. This resulted in 3624 individual daily stock price observations for additions and 2026 for deletions. The mean daily
Table 2. Descriptive statistics
Additions Deletions
Number of companies 46 26
Daily stock price observations 3624 2003
Mean daily return 0.036% 0.024%
Standard deviation 3.10% 2.93%
Median return 0.024% 0.000%
Maximum return 21.30% 37.98%
Minimum return -48.09% -27.03%
19 return for both samples is positive. Although the mean return for deletions is lower than the mean return for additions, a positive mean is not expected. One would expect that deleted companies experience a decrease in stock price (see table 1), resulting in lower returns. The median return of the deletions is also not negative (0.000).
In appendix 3 the tests for normality are presented. A sample can be considered to have a normal distribution with a skewness value below one and kurtosis a value around three. The kurtosis and skewness statistics for abnormal stock returns do not have normal distributions in most models. In the statistics literature, Schmid and Trede (2003) report that in their sample, kurtosis and skewness statistics are extremely sensitive to outliers. I tested this statement on my data and concluded that it is applicable to my data set. For example, when the kurtosis and skewness are re-calculated for additions, with abnormal returns calculated using the Mean Adjusted model, excluded from outliers the skewness drops from -1.218 to -0.044 and kurtosis drops from 21.822 to 2.606. Excluding outliers in other models leads to similar values.
5. Results
In this section the results of the empirical tests described in section 4 are presented. The first two subsections present the results of the event studies that test for short term price effects. In subsection 3 the robustness of the results is tested. Subsection 4 tests for volume effects. The existence of long term price effects is tested in subsection 5. The size of the bid-ask spread before and after the event day are compared with each other in subsection 6. Finally, the effects of company characteristics is tested in subsection 7 by regressing them on the abnormal returns. 5.1 Price effects for additions
20 AD yet no price effect on the AD, is inconsistent with all other research on index composition changes. None of the studies find price effects before the AD. Although the moment of the effect is not consistent with earlier research, the size of the effect is comparably large. The mean daily return observed surrounding the announcement date is approximately 3%. Harris and Gurel (1986) find 3,13% in the later period of their sample. Shleifer (1986) finds abnormal returns of 2,79%. Cusick (2002) find a price effect of 4.34%. Kaul et al. (2002) find 2.07% abnormal return during the
Table 3. Cumulative abnormal returns AEX index
Panel A: Additions
Mean Adjusted Market Adjusted OLS model
Days relative to AD CAR (%) T-statistic CAR (%) T-statistic CAR (%) T-statistic
-20, +60 -0.19 -0.570 2.88 0.912 -1.78 -0.538 -20, 0 2.66 1.059 3.35 1.512 2.30 0.912 -3, -1 2.62 3.529*** 2.23 3.019*** 2.48 3.281*** 0 (AD) 0.03 0.075 0.27 0.816 -0.06 -0.184 +1, +3 0.77 1.275 0.51 0.918 0.63 1.032 0, +20 2.17 1.331 3.02 1.957* 1.78 1.082 0, +60 -4.57 -1.221 -0.35 -0.084 -4.15 -1.107
Days relative to ED CAR (%) T-statistic CAR (%) T-statistic CAR (%) T-statistic
-20, +60 -9.59 -1.789 -5.46 -1.165 -0.67 -1.468 -20, 0 2.53 1.369 3.98 2.314** 1.88 1.061 -3, -1 -0.32 -0.368 0.32 0.427 -0.52 -0.623 0 (ED) -0.51 -1.271 -0.32 -0.936 -0.59 -1.590 +1, +3 0.47 0.647 0.80 1.322 0.44 0.635 0, +20 -4.59 -2.206** -2.38 -1.320 -4.59 -2.272** 0, +60 -11.84 -2.458** -7.61 -1.993** -9.17 -2.134** Panel B: Deletions
Mean Adjusted Market Adjusted OLS model
Days relative to AD CAR (%) T-statistic CAR (%) T-statistic CAR (%) T-statistic
-20, +60 5.63 1.205 -3.20 -0.652 5.88 1.272 -20, 0 2.99 1.368 -0.05 -0.023 2.64 1.234 -3, -1 0.49 0.472 -0.01 -0.121 0.62 0.620 0 (AD) -0.27 -0.409 -0.20 -0.371 -0.236 -1.106 +1, +3 -0.37 -0.350 -0.04 -0.379 -0.37 -0.352 0, +20 -0.33 -0.138 -1.46 -0.484 1.50 0.749 0, +60 2.37 0.666 -3.35 -0.862 2.81 0.710
Days relative to ED CAR (%) T-statistic CAR (%) T-statistic CAR (%) T-statistic
-20, +60 -8.93 -1.150 -9.21 -1.751 -5.37 -0.653 -20, 0 1.50 0.442 1.65 0,554 1.12 0.317 -3, -1 -0.16 -0.116 -0.13 -0.088 -0.21 -0.150 0 (ED) 0.49 0.741 0.53 0.656 0.46 0.675 +1, +3 -0.09 -0.085 -0.60 -0.707 -0.16 -0.150 0, +20 0.69 0.261 0.19 0.090 0.48 0.174 0, +60 -9.93 -1.321 -10.32 -1.781 -6.09 -0.846
21 first week after the AD. The pre-announcement effect observed in table 3 suggest that investors know which companies are added to the index and trade on this knowledge before the announcement is made. An explanation could be the transparency of the rules Euronext uses to construct the AEX index. Compared to the other research (on large U.S. indices) where trading in advance is more difficult, it is possible to anticipate on announcements of AEX index inclusions. The bottom part of panel A in table 3 shows the results for price effects around the ED. No clear price effects are found around and on this day. There is however a post-ED effect. In the periods ‘’0, +20’’ and ‘’0, +60’’ significant negative abnormal returns can be observed. These negative values are a price reversion after the price increase in the days before the AD. As illustration the CAR values for additions around the AD are presented in graph 1 in appendix 4. Inconsistent with other studies, the price effect starts at day -4 and ends around day +12. Approximately 50 days after the AD the positive CAR completely disappeared and becomes negative in two of the three estimation models. This full price reversion is consistent with the price pressure hypothesis. Appendix 4 graph 3 shows the CAR values for additions around the ED. The effects in graph 3 show the significant negative abnormal returns after the ED in the long term as presented in table 3. Note that the average time between the AD and ED is 30 days, so a large part of the price effect in graph 1 and 3 of appendix 4 contain the same information.
5.2 Price effects for deletions
22 evidence for this asymmetry by looking at the number of shareholders. Added companies have significantly more shareholders after the company is added. Deleted company do not have significantly less shareholders.
The bottom part of panel B in table 3 shows the price effects around the effective day. The signs around the ED of all CARs are negative, this could imply that investors trade on a deletion from the index. However, price effects are small with CAR values between -0,05% and -0,7% the effect is statistically and economically not significant. As an illustration, CAR values around the announcement and effective day are presented in graph 2 and 4 in appendix 4. Around the announcement day the Mean Adjusted model and OLS model show positive CAR values in graph 2, what is opposite of the predictions of the hypotheses. The Market Adjusted model does become negative, however no clear effect can be observed around the announcement day. In graph 4 of appendix 4 all models show a similar effect for deletions on the effective day. Negative CAR values are observed 35 days after the effective day. This is much later than the four hypotheses predict and price do not revert to original values as the price pressure hypothesis predicts. Section 5.5 will test if these effects are the start of a long term price effect.
5.3 Robustness tests
23 sample period, investor behavior has not drastically changed regarding their reaction to index additions and deletions.
The second test checks whether the results are affected by outliers. Daily abnormal returns larger than 10% and smaller than -10% are deleted. Again, only the results of the Mean Adjusted model are presented. In Appendix 6 the results of the event study without outliers are presented. For additions the period ‘’-3, -1’’ is highly significant, which is similar to the results in table 3. A difference between the results of table 3 and appendix 6 is that the period ‘’+1,+3’’ also has a significant return. The effect in appendix 6 is however not much larger (1.07%) compared to the effect in table 3 (0.77%). For deleted companies the results are consistent with the findings in table3, no significant announcement effects are found. The approximately similar results between appendix 6 and table 3 imply that the results are not driven by outliers.
The last robustness test checks the validity of the AEX index as a benchmark in the Market Adjusted and OLS model. The AEX is a relatively small index and deletions from the index still contribute to the return of the index between the AD and ED. The tests done in table 3 are repeated with the MSCI Netherlands as a benchmark. This index contains of 48 companies implying the impact of a single company on the performance of the index is much lower. The results are presented in appendix 7 and are very similar to the results in table 3. For additions the announcement effects are similar and equally large. After the effective day negative CARs are found for additions which are approximately equally large compared to table 3. No announcement effects is found for deletions. The similarity between the results in table 3 and appendix 7 imply that the results are robust against the use of a different benchmark. Therefore, using the AEX index as benchmark does not lead to problems. 5.4 Trading volume
24 Panel B of table 4 shows the ATV for additions and deletions at the ED. The results show large and significant abnormal trading volume prior to and after the ED. The peak is day -1 where the trade volume is 329% higher than normal trade volume. Deleted companies also experience large and significant abnormal trading volume with a peak on day -1 of a 250% increase in trading volume. After the ED on days +1 and +2 the ATV for deletions is statistically insignificant, however the effects are economically large enough to consider it above normal trading volume. The results in panel B showing a peak in abnormal trading volume for both additions and deletions at day -1 can be explained by index funds. Index funds have to buy additions and sell deletions to keep their tracking error as low as possible. This is consistent with the downward sloping demand curve and price pressure hypothesis who predict an increase in trading around the ED. Trading volume does not appear to stay high after the ED indicating the volume effect is short term. As an illustration, appendix 7 shows two graphs of the trading volume around the AD and ED. The significantly higher trading volume around the ED is observed in the bottom graph as a large peak for both additions and deletions.
Table 4. Trading volume
Panel A: Days relative to the announcement day
Additions Deletions
ATV T-statistic ATV T-statistic
-3 1.188 1.334 0.960 -0.668 -2 1.096 0.855 1.130 0.745 -1 1.084 0.732 0.015 0.150 0 (AD) 1.073 0.607 1.067 0.358 1 1.476 2.017** 1.102 0.581 2 1.333 2.348** 1.245 1.246 3 1.230 1.926* 0.961 -0.308 4 1.120 1.196 0.959 -0.306 5 1.075 0.754 1.114 0.565
Panel B: Days relatives to the effective day
Additions Deletions
ATV T-statistic ATV T-statistic
-3 1.438 3.173*** 1.131 1.024 -2 1.437 3.048*** 1.481 2.703** -1 4.292 7.207*** 3.504 4.501*** 0 (ED) 1.774 3.706*** 2.195 2.217** 1 1.367 3.287*** 1.651 1.389 2 1.245 2.471*** 1.470 1.391 3 1.236 2.489** 1.098 0.440 4 1.045 0.578 1.083 0.348 5 1.104 1.109 1.023 0.183
25 Combining the results of abnormal returns from table 3 and abnormal trading volume from table 4, a somewhat contradicting view emerges. Both prices and volumes effects are found. According to predictions of some hypotheses an increase in stock prices is accompanied by an increase in trading volume. However, on the days where the largest price effects are found (period ‘’-1, -3’’) around the AD, no increase in trading volume is found. A significantly higher trading volume is found the period after the announcement day (period ‘’+1, +3). Around the ED significant abnormal trading volume is found (period ‘’-3, +3), however this is not accompanied by a price effect. In general, price effects are mainly observed around the AD, whereas the volume effects are mainly observed around the ED. 5.5 Calendar-Time Portfolio Analysis
The results of table 3 indicate that CARs of additions around the AD reverse to zero in the 60 days after the AD. For deleted companies no clear price effect is observed. Until now there is only been tested for short term effects. According to certification, liquidity and downward sloping demand curve hypothesis, index composition changes also have long term price effects. The certification, liquidity and downward sloping demand curve hypothesis predict a long term positive price effect for added companies and a long term negative price effect for deleted companies. To test the long term effects I use the calendar-time portfolio analysis from the paper of Døskeland and Hvide (2011). Monthly continuously compounding is used to calculate returns starting one month after the AD for a period of two years. The monthly returns of the additions and deletions portfolio are regressed on the loadings of specific types of risk. The four-factor model from Carhart (1997) is used. The factors in the four-factor model are obtained from the website of Kenneth French. Due to data unavailability of the risk factors, the start of the portfolio will be in February 1990.
26 alpha indicating a risk adjusted loss on the portfolio. The size of this alpha is not large enough (-1,94%) to be statistically significant. Though, the alpha is economically large enough to consider it a significant effect. The results of panel A show insignificant alphas with signs opposite to the predicted signs. The certification, liquidity and downward sloping demand curve hypothesis, who predict a long term price effect are not supported by these results. The price pressure hypothesis cannot be rejected based on the results of panel A, since no price effect is found.
Panel B of table 5 shows the results of the same portfolio but the stocks are now hold on to for two years. For additions the results are comparable to the one year holding period. The alpha is equally large (-0,63%) and has again the opposite sign as predicted by the hypotheses. The HML factor is again significant. For deleted companies, compared to the one year holding period, smaller results are found. Alpha is small (0,04%) and none of the other factors is significant, except for the market Table 5. Calendar-Time Portfolio Analysis
Additions Deletions Additions - deletions
Panel A: 1 year holding period
Variable Coefficient T-statistic (probability) Coefficient T-statistic (probability) Coefficient T-statistic (probability) Alpha -0.006 -1.331 (0.184) 0.006 0.836 (0.404) -0.019 -1,406 (0,161) Market return risk
free rate 0.660 7.068 (0.000) 0.585 5.287 (0.000) 0.951 3,498 (0,001) SMB -0.014 -0.073 (0.942 0.494 1.654 (0.099) -1.572 -2,805 (0,006) HML -0.490 -2.545 (0.011) 0.304 1.171 (0.243) -1.464 -2,805 (0,006) WML -0.186 -1.581 (0.115) -0.348 -2.278 (0.024) -0.126 -0,401 (0,689)
Panel B: 2 year holding period
Variable Coefficient T-statistic (probability) Coefficient T-statistic (probability) Coefficient T-statistic (probability) Alpha -0.006 -1.249 (0.212) 0.001 0.0587 (0.953) -0.029 -2,200 (0,029) Market return –
risk free rate
27 factor. The long-short portfolio presented in the last two columns has a statistically and economically significantly negative alpha (-2,95%). This indicates that added companies have a significantly lower return compared to deleted companies in the two years following the AD. Opposite to the predictions of the hypotheses, which predicted that this long-short portfolio would have the largest positive return, the portfolio has the largest negative return. When reversing the long and short positions a positive return can be made when taking a short position in the stocks of companies added to the index and a long position in stocks of companies deleted from the index. The certification, liquidity and downward sloping demand curve hypothesis are not supported by the results of panel B.
5.6 Bid-ask spread
The liquidity hypothesis predicts for index additions lower information asymmetry, resulting in lower trading costs reflected by a lower bid-ask spread. Opposite effects are predicted for deletions. Table 6 shows the results of the mean comparison tests for bid-ask spreads between different time periods. The sample period of the event study in section 5.1 is divided in four sub-periods. Period 1 is the period before the AD and period 2, 3 and 4 are after the AD. The periods are sequential and contain 20 days each. Column 2 shows the mean spread of each sub-period. There is a clear
Table 6. Bid-ask spread analysis Period Mean spread
(%)
Mean difference with period 1 (%) T-statistic (Probability) Panel 1: Additions 1 0.39 0 - (-) 2 0.37 -0.026 1.717 (0.094) 3 0.36 -0.031 1.991 (0.053) 4 0.38 -0.013 0.671 (0.506) Panel B: Deletions 1 0.90 0 - (-) 2 0.92 0.014 -0.370 (0.713) 3 0.98 0.081 -1.641 (0.109) 4 0.97 0.066 -1.430 (0.161)
28 difference between the additions in panel A and deletions in panel B. The spread of deletions is approximately 0,6% larger in every time period. Column 3 shows the mean differences between sub-period 1 and the other sub-sub-periods in absolute terms. The statistical mean difference of sub-sub-period 1 and the other sub-periods with corresponding probabilities is shown in column 4. The signs of the values in column 3 are in line with the liquidity hypothesis, showing a drop in bid-ask spread for additions and a larger spread for deletions. The effect is only weakly significant for additions in sub-period 2 and 3 at a 10% confidence level. The bid-ask spread decreases (in sub-sub-periods 2 and 3) but does not stay permanently low. Sub-period 4 is not statistically different from sub-period 1 indicating the bid-ask spread has increased from sub-period 3 to 4 to its pre-event level. A temporal decrease in bid-ask spread is also found by Beneish and Whaley (1996). The decrease in their study is found in the first 10 days after the effective day. From day 10 until day 40 after the effective day, the spread slowly returns to its original level. The results for table 6 partly support the liquidity hypothesis. The bid-ask spread is lower for additions after the AD. However, the effect on the bid-ask spread appears to be temporal, which does not support the liquidity hypothesis. Another inconsistency with the liquidity hypothesis is the insignificant change of the bid-ask spread for deletions.
5.7 Regression analysis
In this section the results of the cross-sectional regression are presented which build on the results of the event study in section 5.1 and tests if company characteristics have effect on CARs around the AD. In table 3 I observe that the largest abnormal returns around the AD are from day 3 until day -1. Therefore, the dependent variable in this regression will be the CAR value of the period ‘’-3,-1’’. The independent variables are explained in detail in appendix 9. The correlation matrix is presented in table 7. Panel A shows the correlation values for the sample of additions and panel B shows the correlations for the sample of deletions. Multiple high correlations are observed in both panels. Especially ‘’volatility’’ and ‘’bid-ask spread’’ in panel A and ‘’volatility’’, ‘’ΔProfitability’’ and ‘’profitability’’ in panel B have high correlation value. The strength of the relations (mostly around 0.5) can be classified as ‘’moderate’’ from a value of 0.4 up to a value of 0.7 (Dancey and Reidy, 2004) or to a value of 0.6 (Evans, 1996) depending on what classification standard are used. Although the moderately strong correlations, I decided to not exclude variables from the regression analysis. All variables measure a different company characteristic and contribute to the model.
29 variance in the models is explained by the independent variables. The Durbin-Watson tests in both regressions have values around 2, indicating the absence of autocorrelation.
Column 2 and 3 show the results for added companies. Weakly significant values are found for ‘’Size’’ and ‘‘ΔProfitability’’. Company size has a negative effect on the abnormal returns when added to the AEX index. In other words, small companies have larger abnormal returns when added to the AEX compared to large companies. A reason for this difference is provided by the certification hypothesis. In general, large companies disclose more information that is easier to access. Therefore, the prospects of large companies are easier to estimate. For small companies relatively less is known and future prospects are harder to predict. Both the addition of a small and a large company to the AEX result in the expectation of more and better disclosed information. However, large companies disclose already more information, the increase in information availability will have a relatively low impact. Small companies disclose less information, the impact of an increase in information will be relatively large. This difference will result in a larger price effect around the announcement day for small companies.
In panel B the ‘’dividend payout’’ coefficient is negative and significant at a 5% confidence level. This implies that companies with high dividend payout ratios have lower (or more negative) abnormal returns compared to companies with low dividend payout ratios. A reason for this might be that Table 7. Correlations matrix
Panel A: Additions
CAR Leverage Size Profitabil ity ΔProfitabili ty Dividend payout Volatility Bid-ask spread CAR 1.00 Leverage -0.35 1.00 Size -0.35 0.16 1.00 Profitability 0.22 -0.59 -0.20 1.00 ΔProfitability 0.31 -0.35 0.14 0.12 1.00 Dividend payout -0.30 0.26 -0.34 -0.25 -0.12 1.00 Volatility 0.29 0.13 -0.04 0.03 -0.28 -0.29 1.00 Bid-ask spread 0.43 -0.33 -0.32 0.50 0.12 -0.45 0.45 1.00 Panel B: Deletions
CAR Leverage Size Profitabil ity ΔProfitabili ty Dividend payout Volatility Bid-ask spread CAR 1.00 Leverage -0.06 1.00 Size 0.28 -0.57 1.00 Profitability 0.38 0.07 0.32 1.00 ΔProfitability -0.51 -0.17 -0.22 -0.61 1.00 Dividend payout -0.42 0.36 -0.28 0.33 -0.42 1.00 Volatility -0.53 -0.03 -0.49 -0.49 0.39 -0.24 1.00 Bid-ask spread 0.18 0.20 0.18 -0.16 0.12 -0.18 0.12 1.00
30 relatively more value already left the company resulting in a lower company value. Further, high dividend payout can work as a signal that the investment opportunities of a company are slim and do not create additional shareholder value. Note that in table 3 no significant price effects for deleted companies are found are the AD and ED. This makes interpretation and implementation of the results in panel B difficult.
Table 8. Cross sectional regression analysis
Additions Deletions
Variable Coefficient T-statistic
(probability) Coefficient T-statistic (probability) Alpha 0.356 1.856 (0.096) 0.269 1.280 (0.291) Leverage -0.008 -0.424 (0.682) -0.002 -0.730 (0.518) Size -0.021 -1.863 (0.095) -1.245 -1.083 (0.358) Profitability -0.019 -0.122 (0.905) -0.185 -1.503 (0.230) ΔProfitability 0.001 1.596 (0.145) -0.001 -1.780 (0.173) Dividend payout -0.101 -1.228 (0.250) -0.100 -3.377 (0.043) Volatility 0.152 1.293 (0.228) -0.146 -1.012 (0.386) Bid-ask spread -1.644 -0.499 (0.630) 0.907 1.008 (0.387) R-squared 0.530 0.894 DW-test 2.247 1.826
31 6. Conclusion
This paper analyzes the impact of changes in the AEX index between 1983 and 2015. The AEX index is the main Dutch stock exchange which contain the 25 most frequently traded companies. Additions to and deletions from the index are tested. Event studies are done around the announcement and effective day to test for short term price and volume effects. A calendar-time portfolio analysis is done to test for risk adjusted long term price effects. The last test is a cross-sectional regression analysis that tests for differences in company characteristics.
Literature provides four hypotheses that predict a positive price effect for companies added to an index and a negative price effect for companies deleted from the index. This price effect is predicted to be accompanied by an increase in trade volume according to most hypotheses. The hypotheses are inconsistent in predicting whether the price effect occurs at the announcement or effective day. Events that are information free are assumed to happen on the effective day. Events that contain information (about the future performance of the company) are assumed to happen on the announcement day. The hypotheses are also inconsistent in the prediction whether the price and volume effects are temporal or permanent.
The results of the event studies show a large price effect for additions in the three days up to the announcement day. This result is never been found by earlier research. The majority of earlier research finds a large price effect on the announcement day. This is the first study that finds a large price effect before the announcement day. The main price effect is not accompanied by higher trading volume. However, the trading volume does increase in the days after the announcement is made. The largest increase in volume is found around the effective day. An analysis of the increase in stock price shows that the cumulative abnormal returns reverse to zero in the 3 months following the announcement day. For companies deleted from the index, no price and volume effects are found around the announcement day. Around the effective day there are volume effects, however not accompanied by a price effect. Long term price effects are predicted by three of the four hypotheses. The results of the calendar-time portfolio analysis do not show long term price effects. The analysis of the bid-ask spread finds that in the 40 days after the announcement day companies do have a lower bid-ask spread which return to its original level after 40 days. The results of cross sectional analysis show there is not a clear differences in reactions to an index addition or deletion, between companies with different characteristics.
32 downward sloping demand curve and price pressure hypothesis. The price and volume effects of additions are both reverting to original levels. Long term price effects are not observed, indicating temporal price and volume effects. This is consistent with the price pressure hypothesis. The results for deleted companies do not provide much support for the hypotheses. The only result, which is support for the downward sloping demand curve and price pressure hypothesis, is the temporal volume effect found at the effective day
The results in this paper have implication for multiple types of investors. First, the results have implications for short term investors. Since the AEX composition is based on trade volume, they can make a return by predicting what composition changes will be made and when the announcement days are. The investor can buy stocks added to the index four days before the announcement is made and gain from the abnormal returns in the following days. Second, the results have implications for passive long term investors. These investors will buy an index addition at a high price at the effective day. In the days following the effective day, stock prices of additions will decrease. Therefore passive investors will have to take into account a negative abnormal return shortly after index additions are bought. Finally, the results have implications for investors trading on the ‘’new information’’ that a stock is announced to be added to the AEX index. These investors expect a long term improvement in return. The results do not show long term price effects. The price effects observed at the announcement day fully reverse to zero. The decrease in bid-ask spread is also temporarily. The only long term return that can be made is creating a portfolio that holds stocks for two years and consist of a short position in companies added to the index and a long position in companies deleted from the index.
33 Fourth, the stock price gradually returns to its original price level (consistent with the price pressure hypothesis).
34 7. References:
Barber, B. M., Lee, Y. T., Liu, Y. J., & Odean, T. (2009). Just how much do individual investors lose by trading?. Review of Financial studies, 22(2), 609-632.
Beneish, M. D., & Gardner, J. C. (1995). Information costs and liquidity effects from changes in the
Dow Jones Industrial Average list.Journal of Financial and Quantitative Analysis,30(01),
135-157.
Beneish, M. D., & Whaley, R. E. (1996). An anatomy of the “S&P Game”: The effects of changing the rules. The Journal of Finance, 51(5), 1909-1930.
Blouin, J. L., Raedy, J. S., & Shackelford, D. A. (2000). The impact of capital gains taxes on stock price reactions to S&P 500 inclusion (No. w8011). National Bureau of Economic Research.
Brown, S. J., & Warner, J. B. (1985). Using daily stock returns: The case of event studies. Journal of financial economics, 14(1), 3-31.
Carhart, M. M. (1997). On persistence in mutual fund performance. The Journal of finance, 52(1), 57-82.
Chakrabarti, R., Huang, W., Jayaraman, N., & Lee, J. (2005). Price and volume effects of changes in MSCI indices–nature and causes. Journal of Banking & Finance, 29(5), 1237-1264.
Chan, K., Kot, H. W., & Tang, G. Y. (2013). A comprehensive long-term analysis of S&P 500 index additions and deletions. Journal of Banking & Finance, 37(12), 4920-4930.
Chen, H., Noronha, G., & Singal, V. (2004). The price response to S&P 500 index additions and deletions: Evidence of asymmetry and a new explanation. The Journal of Finance, 59(4), 1901-1930.
Collins, M. C., Wansley, J. W., & Robinson, B. (1995). Price and volume effects associated with the creation of Standard & Poor's MidCap index. Journal of Financial Research, 18(3), 329-350. Curran, M. M., & Moran, D. (2007). Impact of the FTSE4Good Index on firm price: An event study.
Journal of environmental management, 82(4), 529-537.
Cusick, P. A. (2002). Price Effects of Addition or Deletion from the Standard & Poor's 500 Index. Financial Markets, Institutions & Instruments, 11(4), 349-383.
Dancy, C. P., & Reidy, J. (2004). Statistics without maths for psychology. Harlow: Pearson Education Limited.
Denis, D. K., McConnell, J. J., Ovtchinnikov, A. V., & Yu, Y. (2003). S&P 500 index additions and earnings expectations. The Journal of Finance, 58(5), 1821-1840.
Dhillon, U., & Johnson, H. (1991). Changes in the Standard and Poor's 500 List. Journal of Business, 75-85.
Dhillon, U. S., & Johnson, H. (1994). The effect of dividend changes on stock and bond prices.The
