Why Do Index Changes Have Price Effects?*
Burcu Hacıbedel†
This version: November 2007
Abstract The positive price effect of index inclusion has been well-documented in the literature. The underlying cause still remains in dispute, since this finding is consistent with a number of hypotheses. In this paper, I revisit this debate by examining the price effects in the emerging markets setting using MSCI EM index changes. I find the inclusions to have a permanent price effect, while this is not the case for the exclusions. This result contradicts the demand and new information hypotheses, but is consistent with the investor awareness hypothesis. By making use of analysts’ recommendations data, I am able to show that there is a significant increase in coverage for the included stocks. This is also significantly related to the change in price.
*
I am grateful to Magnus Dahlquist for helpful comments and discussions. I would also like to thank the seminar participants at Central Bank of Sweden, EBRD, FMA Europe Meetings 2006, Said Business School (University of Oxford) and particularly at SIFR, for their valuable suggestions and feedback. All omissions and errors are mine. † Swedish Institute for Financial Research (SIFR), Saltmätargatan 19A, SE-113 59, Stockholm, Sweden,
[email protected].
1
1 Introduction It is a well-documented finding in the finance literature that inclusion in a benchmark index results in significant increase of the stock price. While the empirical evidence is convincing, the underlying mechanism still remains in dispute. The observed price effects are consistent with a number of hypotheses documented in the literature. In this paper, I revisit this issue in a new market setting to disentangle the suggested hypotheses. Existing studies are clustered around the US market and S&P500 changes. The evidence shows that the stock price reacts instantly to the announcement of the index change. The overall effect is positive and permanent with a slight price reversal upon the actual index change.1 In the time period between the announcement and the change, the abnormal price effect persists and builds up.2 A number of hypotheses have been suggested in the literature to explain this phenomenon. These can be listed under two headings: the downward sloping demand curve (DSDC) and the information effect. The first group argues that the index change is an information free event.3 Thus, any change in price can be explained by the change in aggregate demand, given that the demand curve is downward sloped. Both Shleifer (1986) and Harris and Gurel (1986) explain the price changes around S&P500 inclusions by the increase in aggregate demand from index funds.4 Several following studies find evidence consistent with this argument.5 The second set of hypotheses questions the ‘information free’ assumption. A number of recent studies argue that this price effects is actually an adjustment to new price equilibrium upon the inclusion. The increase is caused by the revelation of information that directly affects the underlying stock value. Additionally, the informationhypothesis is consistent with the horizontal demand curve. Jain (1987), and Dhillon and Johnson (1991) argue that the index inclusions reveal information by some form of certification that is implicit in S&P’s selection of the stock. Beneish and Whaley (1996), and Hegde and McDermott (2003) show that the price reactions can be explained by changes in market liquidity. Denis, McConell, Ovtchinnikov and Yu (2003) propose that the index inclusion results in better monitoring and investment decisions, thus a price increase. Chen,
1
Among the existing papers, only Harris and Gurel (1985) document a full price reversal after the actual index inclusion. 2 In the post-1989 period, there is a lag between the announcement and the change. This lag varies between 1 day and 1 month. 3 ‘Information’ refers to any fundamental information that may affect the stock value. 4 Both papers suggest that this demand increase comes from the changes in institutional index fund holdings. 5 Lynch and Mendenhall (1997), Kaul et al (2000), and Wurgler and Zhuravskaya (2002) also document price increases due to demand changes.
2
Noronha and Singal (2004) attribute the permanent price effect to the increase in investor awareness following the inclusion. All of these hypotheses are consistent with the documented price effect. One method to disentangle these is to compare the price effects around inclusions and exclusions. Since some of these hypotheses predict symmetric responses, this approach is a plausible test of their validity.6 However, only a few of the existing papers analyse both events in a comparative study. The main problem with the S&P500 case is the small number of exclusions that is suitable for empirical analysis.7 Analyzing the issue in an alternative market/index setting can overcome this problem. This is the main motivation underlying this study. In this paper, I analyse the price effects of index changes to MSCI Emerging Markets index (MSCI EM). I am able to construct a clean sample of 280 inclusions and 264 exclusions, based on the index changes that occur between 1996 and 2006. This setting has a number of advantages. Unlike the S&P500 exclusions, most of MSCI EM exclusions occur voluntarily. This results in a larger and comparable sample of observations. Moreover, the announcement policies of both index companies are similar, i.e. the index change is preceded by the announcement.8 Both companies make it clear that the decision to included or exclude a stock from the index is solely based on publicly available information. Thus, there is no reason to assume that the index change has any new information content. One other advantage of MSCI EM data is that ‘investor awareness’ is a stronger candidate in explaining the asymmetric price effects, if that is the case. Investor awareness hypothesis is based on the model of asset pricing with imperfect information by Merton (1987). The investors refrain from investing in stocks which have high cost to access firm specific information.9 Therefore, these stocks trade at discount price. In S&P studies, the change in investor awareness refers to the change in the institutional shareholdings in addition to index funds. Additionally, the enhance awareness is closely related to the information content of the inclusion. However, in the true sense of the phrase, this does not cover all that has been implied in Merton’s model. In this sense, emerging markets (EM) provide a better setting to study the effect of changes in investor awareness on stock prices. Indirect barriers to cross6
For example, the demand based hypotheses requires the price reaction to be symmetric. Among the information based ones, certification and liquidity hypotheses also require some degree of symmetry. Harris and Gurel (1986), Lynch and Mendenhall (1997), and Beneish and Whaley (2002) examine also the exclusions in addition to the inclusions. 7 S&P500 exclusion data is usually quite small, since most of the exclusions occur involuntarily, i.e. around more stock specific events such as bankruptcy. This decreases the number of clean observations. On the other hand, MSCI EM exclusions are made mostly to rebalance and update the index. 8 MSCI changes are comparable with the post-1989 S&P changes. 9 In the literature, this has been referred to as shadow cost or background information cost.
3
border investment, including lack of information and awareness, have been well-documented for these markets. Therefore, the price effect of enhanced investor awareness, if any, ought to be more pronounced. Additionally, in this context, it is independent of any new information.10 It refers to becoming aware of the stocks once included as well as enhanced information availability. In this paper, I find the price effects of inclusions and exclusions to be asymmetric. While the inclusion in the index induces a permanent and significant price increase of 2.4%, the exclusion only has a significant temporary effect. The initial price decrease that occurs upon the exclusion announcement gradually disappears over the two-months following the actual index change. The asymmetric response is inconsistent with the DSDC, certification and liquidity hypotheses, but consistent with the investor awareness hypotheses. Since the shareholding data is not publicly available for the EM stocks, I suggest an alternative measure of investor awareness, following Arbel et al. (1983) and Merton (1987). I use the changes in the number of analyst recommendations to capture this effect. The results are consistent with the predictions of this hypothesis. I find a significant increase in the analyst coverage of newly included stocks. Additionally, stocks with higher increase in coverage experience larger price increases. This effect is more significant for stocks with no prior access to the foreign markets through ADRs. The rest of the paper is organized as follows: I describe the sample and methodology in Section 2. The hypotheses as well as the price and volume results are presented in Section 3. The issue of investor awareness is tested and discussed in Section 4. Lastly, Section 5 concludes.
2 Data and methodology 2.1
The Sample
2.1.1 The Index The sample consists of inclusions and exclusions from Morgan Stanley Capital International’s Emerging Markets index (MSCI EM). This is an internationally recognized and tracked index, 10
In some respects, the ‘investor awareness’ is defined differently from the existing studies. It has been used as part of the information hypothesis. However, in this paper, it is independent of any information revealed by the index change. In this sense, it captures the imperfect information effect in Merton’s paper much better than the existing studies. In some S&P studies, investor awareness increases because of the positive information revealed by the inclusion. Upon the inclusion, higher number of shareholders results in better monitoring of the company and earnings.
4
which is used among both practitioners and researchers as a benchmark for emerging equity markets.11 MSCI EM belongs to the family of MSCI Standard indices, and is thus constructed by MSCI standard index methodology. It aims to capture 85% of the total free float market capitalization in 26 emerging markets. 12 [INSERT TABLE I HERE] Table I shows the breakdown of the index and the sample. The index, as any other standard MSCI index, is maintained through three types of reviews; annual full country index reviews, quarterly index reviews (QIRs) and ongoing event-related changes. In this study, I focus on QIRs, which occur on only four dates throughout the year. These occur at the close of the last business days of February, May, August and November. The QIRs are announced at least two weeks (10 days) prior to these dates. QIRs announce a number of changes to the standard indices to ensure the accurate representation of the dynamic market place, and to avoid significant under- and over-representation of any industry group in a country index. One issue that frequently comes up in the index inclusion literature is the predictability of the event. With respect to this, MSCI does not have any dominant criteria like the size. Instead, the decision is based on a number of different factors, with an emphasis on the market representation.13 Therefore, it can be said that there are no significant individual criteria but a mixture of stock characteristics, none of which seem to dominate.14 Briefly, inclusions in QIRs occur because of several reasons. These include the change in size of the stock (free float adjusted), under-representation of one or more industry groups following mergers, acquisitions, restructuring and other major market events affecting that industry group and changes in industry classification
11
As of 2005, MSCI declares that it has a market share of more than 75% in the global equity indexing industry and that over USD 3 trillion of equity assets are benchmarked to its indices. 12 The index covers 26 EMs: Argentina, Brazil, Chile, China, Colombia, Czech Republic, Egypt, Hungary, India, Indonesia, Israel, Jordan, Korea, Malaysia, Mexico, Morocco, Pakistan, Peru, Philippines, Poland, Russia, South Africa, Taiwan, Thailand, Turkey and Venezuela. 13 On average 85% of free float market capitalization 14 MSCI applies minimum size guidelines for the inclusion of countries and industry groups in its standard indices. The selection criteria for countries include the overall free float-adjusted market capitalization of the market, distribution of free float-adjusted market capitalization in the country, level of market concentration, and marginal contribution to the market of the largest security at different percentiles of the free float-adjusted market capitalization distribution. Similarly, the selection criteria for stocks for country index inclusion can be grouped under three headings. First one is the ‘business activities’ of the company and the degree of diversification it will contribute to the index. The second criterion is the size (free float adjusted market capitalization) and liquidity of the stock. For a stock to be added to the index in the Quarterly Index Review, it has to meet double the minimum size guidelines. Regarding the liquidity criterion, though there is no definite measure, liquidity is evaluated based on trade volume or traded value, i.e. ATVR (annualized traded value ratio). The third criterion is the estimated free float for the company and its individual share classes. There is a minimum free float requirement of 15%, with certain exceptions to this rule.
5
When compared with S&P500, MSCI differs in numerous ways. First of all, the number of stocks in the index is not restricted, so there is no need to exclude a stock if a new one is being included. Additionally, it is possible to examine a large number of inclusions and exclusions, since the index is more recent compared to S&P, and has been subject to a large number of changes during the sample period. Based on the declared selection criteria, it seems to be harder to anticipate the changes to MSCI EM, since factors like market representation play a key role. The period between the announcement and the actual index change is 9 business days, for the complete sample, whereas this period may differ in length for S&P changes. Lastly, since the index is maintained through three types of reviews, it is easier to obtain a sample of stocks that is added to MSCI in QIRs, independent of on-going event related changes.
2.1.2 The Sample Construction The sample consists of stocks that have been added to or deleted from the index at the Quarterly Index Rebalancing announcements between 1996 and 2006. During this time period, the number of inclusions and exclusions are relatively high. Once the sample stocks are identified from the MSCI announcements, which are available on Bloomberg and MSCI Barra, the relevant stock price information is obtained from DataStream. All the data is in US dollars. The initial number of inclusions is 353; however I exclude the stocks for which the data is not available from 30 days prior to the announcement. This is not an issue with the exclusions. The final sample consists of 280 inclusions and 264 exclusions from 24 emerging markets.
2.2
Methodology
2.2.1 Abnormal returns I study the index changes using event study methodology. Thus, the tests are based on daily average abnormal returns (ARs) and cumulative abnormal returns (CARs). Abnormal returns are calculated as the local market-adjusted returns. This is consistent with the asset pricing differences in emerging markets.15 I focus on the equal weighted average abnormal returns
15
In the literature, it has been documented that local market factors are priced in emerging markets, whereas the global factors are much less significant in asset pricing. Adjusting for the local market return also controls for any macroeconomics factors that coincide with the timing of the inclusion or the exclusion.
6
while presenting the test results. I also calculate the value weighted averages for abnormal returns by using the market capitalizations at the time of the index change. The null hypothesis is that ‘the event does not have any significant impact’. If this holds, then ARs should have a mean of zero, in which case they shouldn’t differ significantly from zero and not have any tendency to build up or down. The market-adjusted ARs are calculated as the excess return on stock i on day t over the local market index. Rit = Rmt + ε it , ε i ∼ N (0,σ i2 ) ARit = Rit − Rmt
where Rmt is the return on the local market index on day t. Using these ARs, we calculate the average abnormal returns (ARt) for day t (event day being t=0) as well as cumulative abnormal returns (CARt) to test for the significance of ARs in each sub-event window. ARt =
1 N
N
∑ AR ,
ARt ∼ N (0, var( ARt ))
it
i =1
var( ARt ) =
1 N2
CAR (t1 , t2 ) =
N
∑σ
2 i
i =1
t2
∑ AR , t
CAR (t1 , t2 ) ∼ N (0, var(CAR (t1 , t2 ))
t = t1
var(CAR(t1 , t2 )) = where σ
2 i (t1 , t2 )
1 N2
N
∑σ
2 i (t1 , t 2 )
i =1
= (t2 − t1 + 1)σ i2
Therefore, CARs are calculated for the anticipation, run-up, price reversal, short-run and long-run event windows. These windows are explained in the next section. Each event window starts on t1 and ends on t2. I also need the variance to conduct the significance tests. The sample estimator of the variance is calculated as following.16 σˆ i2 =
−1 1 ( Rit − Rmt ) 2 L1 − 2 t =−29
∑
where L1 is the number of days in the estimation window. To calculate the variance of AR, we use the sample estimator of σi2; σˆ i2 . Following these, we test the null hypothesis of ‘no event related ARt and CAR(t1, t2)’ by using the following test statistics 16
I also use the cross sectional variance of Asquith (1983). The test results change very little when variance is estimated with this method. In the results section, we just report the t-stats with time series variance.
7
θ1 =
θ2 =
ARt var( AR )
∼ N (0,1)
CAR(t1 , t2 ) var((CAR (t1 , t2 ))
∼ N (0,1)
In addition to these parametric tests, we also check for the robustness of the results by using non-parametric sign and rank tests following the methodology suggested by Corrado (1989).
2.2.2 Event windows While analysing the price changes around the inclusions and exclusions, I employ a multiple event-window approach, following Lynch and Mendenhall (1997).17 Thus, the large event window, which starts 10 days before the announcement and runs until 60 days after the index change, is split into smaller windows.18 The main focus is on price changes on two main event dates; the announcement of the change (AD) and the date when the index change becomes effective (CD).19 There is a two-week (9 day) period between AD and CD, which is referred to as the run-up window. Figure 1 illustrates this timeline for the event window. [INSERT FIGURE 1 HERE] I. Anticipation (Pre-announcement) window runs through (AD-11) and (AD-1). II. Announcement day (AD) is our first event day. III. Run-up window covers the time period from the day after the announcement (AD+1)
through the day before change date (CD-1). IV. Inclusion day (CD) is our second event day, which is the actual index change date. V. Price reversal window covers the period between (CD+1) and (CD+10). 20
VI. Short-run event window covers the time period from (AD-10) until CD+10. VII. Long-run event window is between (AD-10) and (CD+50).
21
The anticipation window is used to test whether there is any anticipation of the announcement that can be detected from ARs. The cumulative average abnormal returns (CARs) in the run17
The multi-window framework was initially suggested by Lynch and Mendenhall (1997). They use 5 windows in their study, using also the release date to test for their hypothesis. The release-ending day is the day when the demand for the stock turns to its normal post-change level, i.e. when the index fund demand ends (so that the price release starts). Under the price pressure hypothesis, any price release ends with the completion of index fund trades. 18 Number of ‘days’ refers to number of ‘business days’. 19 AD is actually the day following the announcement since MSCI QIR announcements occur after 8pm GMT, when European and Asian markets are already closed. Therefore, the announcement effects will be reflected on the prices the following day. I adjust the data for the Western Hemisphere stocks in our sample, as the markets are still open during the announcement. The change date (CD) is the day of the announced inclusion date, since the investors are already aware of this before the inclusion takes place. 20 Alternatively, we also report results for the extended window between (AD-10) and (CD+10). 21 This can also be read as (AD-10) to (AD+70).
8
up window enable us to see if there is any significant price drift between the announcement and the actual change. Using the price-reversal window, I test whether the CARs that occur in the previous windows are due to temporary or permanent price changes. The last two windows are used to examine the temporary and permanent price changes induced by the index changes. The last window refers to the complete event window that we also use to examine the long-run price effects.
2.2.3 Calendar time approach The issue of clustering is important in event studies since this may indicate cross sectional correlation, which then may make the results spurious. To correct for this and check for the robustness of results, I calculate the abnormal returns using the calendar-portfolio approach.22 This methodology is based on forming calendar time portfolios at monthly frequency. For each calendar month, abnormal return is calculated for each stock that has been announced to be included in the index in that month. However, for this paper, I conduct the exact analysis at daily frequency, since I am interested in the significance of daily returns within the event windows. Thus, I compute the average abnormal returns on each day across these stocks, in order to obtain the average daily return on portfolio of included stocks. Abnormal returns are calculated as before. The portfolios are reformed each day. The stocks are included in the portfolio for two weeks, starting from the announcement day. If there has been no inclusion, the portfolio is the benchmark, thus the abnormal return is zero.
3 Hypotheses and Results This section presents the empirical analysis and results with a focus on volume and price effects. The results of inclusions and exclusions are discussed separately. The term ‘information’ refers to any fundamental stock information that can directly affect the underlying stock value.
3.1
Hypotheses
One of the well-known predictions in finance is that if markets are ‘semi-strong’ efficient, there should not be any profitable trading patterns or opportunities based on publicly available information. In the context of index changes, this translates into ‘no significant abnormal returns’. The first hypothesis tests this.
22
This was initially suggested by Jaffe (1974) and Mendelker (1974).
9
Hypothesis no.1: Efficient markets If the markets are efficient in the semi-strong form, the index change should not induce any significant abnormal returns. The validity of this hypothesis requires that there is no evidence for significant abnormal returns around the announcement and the actual index change. If the event has no information content, then semi-strong efficiency predicts that there should not be any significant price reactions. On the other hand, even if the event reveals any information, this should be incorporated into the stock price rapidly. Thus, there should not be any long lasting abnormal returns or trade patterns. Additionally, in the presence of horizontal demand curve, any demand pressure should be met by substitute stocks. Therefore, any result indicating abnormal returns is inconsistent with semi-strong form of market efficiency. However invalidity of this hypothesis is necessary to carry out further tests of downward sloping demand curve or information content. If there is any evidence of significant price effects induced by the index changes, this is consistent with a number of hypotheses. These may not be mutually exclusive, but yet to be empirically disentangled. The following hypotheses are structured around this. Hypothesis no.2: Pure DSDC or Pure Information The index change induces a significant price change either because of downward sloping demand curve or the release of new information. Both of these hypotheses predict symmetric price results around the index inclusion and exclusion. Downward sloping demand curve hypothesis (hereafter DSDC) predicts that the stock price changes due to the change in aggregate demand that stems from the index funds. Hence, the funds that start demanding these stocks upon inclusions would sell in a similar fashion upon exclusion. On the other hand, pure information hypothesis interprets the price change as adjustment to new price equilibrium.23 If the index inclusion reveals any new stock information to the market, so should the exclusion. The only difference is that the latter is negative information. If either of these is the sole underlying cause for the observed price change, the effect should be symmetric and opposite. Any other result suggests otherwise. Alternatively, the significant price increase upon the inclusion announcement may be 23
The information hypothesis includes certification, liquidity effects as well as the changes in expected future operating cash flows. While the first one refers to any signalling of private information y the index company, the second one attributes the price change to the change in market liquidity. Because of any of these effects, the underlying stock value changes. In the literature, investor awareness is usually considered as part of the information hypothesis. However, it refers to either increase in investor recognition following the information release or the increase in monitoring due to new shareholders. They all depend on the information preceding the new investor interest. Moreover, in the US setting, this has been tested by the changes in institutional shareholdings.
10
explained by a combination of demand and information effects or enhanced investor awareness. Particularly, in the emerging markets setting, the investor awareness story is a strong candidate. Therefore, the third hypothesis is designed to test for this effect. Hypothesis no.3: Investor Awareness The inclusion increases the investor awareness for the particular stocks. This then causes an increase in aggregate demand and decrease in the shadow cost, resulting in a permanent price increase. In this context, investor awareness does not increase due to any information release. It refers to foreign investors becoming aware of EM stocks once they are part of a benchmark EM index. Ex-ante, there is no reason to believe that the inclusion introduces new information to the market about the fundamental stock value. In the case of EMs, there is information already available about the particular stock, but not really visible to, i.e. in the radar screen of, the foreign investors. This is a case of indirect barriers to cross border investments and their impact on asset pricing. It may also be defined as increased visibility upon index inclusion. If this hypothesis is true, then there should be significant and permanent price effect due to adjustment to a new equilibrium. Though there is not any information release, the stock price decreases because of the decrease in the shadow cost.24 This results in a price increase, but different from the one initiated by the price pressure. This one should be permanent. Additionally, there is no requirement of symmetric price responses to inclusions and exclusions. Since the investors who become aware of the stock upon inclusion, cannot become unaware upon exclusion. Investor awareness does not implicate investor awareness. Overall the investor awareness hypothesis has both a demand and an information aspect to it. There is a demand increase from both index funds and investors who become aware of the stock. The latter includes other mutual funds and EM specific funds which are not only tracking MSCI EM. The information aspect does not refer to new information, but certain investors groups becoming aware of the existing information. Hence, there is no new information release or signalling by MSCI or the particular company.25
24
This decreases the required rate of return as well as the discount rate, which is used to discount the expected cash flows and the underlying stock price. 25 In some respects, this hypothesis is also consistent with behavioral finance; style investing.
11
3.2
Price and Volume Results
3.2.1 Index Inclusions The price changes around the inclusion are examined based on the abnormal return results. This analysis makes use of both the daily and cumulative abnormal returns over the specified event windows. [INSERT TABLE II and FIGURE 2 HERE] There is some evidence of anticipation of the event, as documented by the significant CARs in the 10 day period before the announcement. However, this effect disappears when the value-weighted ARs are used.26 On the announcement day, there is a significant price average abnormal return of 0.5%. Following the announcement, the stock price keeps on increasing, resulting in a visible drift until the actual inclusion. The CARs in the run-up window are significant, and contribute to the overall price increase by 0.6%. On the day of the index change, there is a 0.2% price increase; much lower than the one at the announcement. By the time the stock is included in MSCI EM, there is a significant price increase of 2.2%. There is no evidence for any price reversal afterwards. Though there is a 0.3% decrease over the ten days following the inclusion, this is not significant. The price increase that occurs upon the inclusion announcement is found to be permanent. 60 days after the announcement, the average CAR is around 2.4%. This result is robust to alternative measures of abnormal return. When the ARs are averaged by market capitalization, the price increase becomes larger and more significant. While there is no significant price change before the index announcement, there are significant abnormal price changes at the announcement, over the run-up window and at the change date. The permanent price change, when value weighted averages are used, is around 5.6%. The overall picture, as illustrated in Figure 2, shows that stock price increases significantly upon the announcement. Then this trend continues until the actual addition. After the change, the price stays at this new price level, as documented by the CARs. This indicates a mean shift, i.e. a gradual adjustment to new price equilibrium that can be attributed to the index inclusion. The index inclusion also directly affects the trade volume. The changes in ETV show that the volume peaks just after the announcement and at the inclusion. Upon the index change,
26
This increase is particularly due to the Chinese stocks, which have relatively low market capital. Results are available upon request.
12
volume goes back to its pre-announcement levels. Precisely, the excess demand disappears four days after the change, CD+4.
3.2.2 Index Exclusions While the impact of inclusion on stock prices is well-documented, the evidence for the case of exclusion seems to be more ambiguous. Studying exclusions is also more cumbersome because of the relatively small number of events. However, the effects of exclusions remains important, since validity of the suggested hypotheses require symmetrical price changes around inclusions and exclusions. While the S&P and US changes are biased towards inclusions, testing in alternative index and market settings enables us to study the case of index exclusions more thoroughly. The price and volume effects of index exclusion are analysed following the same methodology. There is no evidence documenting any anticipation of the announcement. CARs prior to the announcement, though negative, are not significant at any acceptable level. On the day of the announcement, the returns are negative, but insignificant. However, the ARs on the following three days are highly significant and negative. This is observable from the CAR results in the run-up window. Between the announcement and the deletion from the index, the stock prices decrease by 2.15% in excess of the market. This price drop persists through the index change, by an additional drop of 1.05%. Both CARs in the run-up window and ARs on the index change day are found to be highly significant. There is no significant price reversal in the ten-day period after the exclusion. This is the price reversal window that has been initially chosen for the inclusion case. Yet, the reversal may occur more slowly in the postexclusion period. This is indeed what the results show. In the short run window, price reaction to exclusion cumulates to -3.44 %. However, in a larger window, there is evidence for significant price reversal. It is clear that there is a correction in the stock price after the exclusion, but it occurs more gradually than expected. By the 60th day after the exclusion, the cumulative price change is around -1%. However, this is not significant. Moreover, this permanent price effect upon exclusion completely disappears once value weighted ARs are used. The magnitude of the price changes around the exclusions is similar to the equal-weighted AR results. While the initial impact is significant, the price goes back to the pre-announcement levels over the 60 day period following the index change. The results show that the average price changes around the inclusions and exclusions are different. The price starts to increase gradually upon the inclusion announcement, and stabilizes at a higher level after the actual addition to the index. On the other hand, the price
13
reaction is much sharper and larger upon the exclusion announcement. The price reversal is significant, but gradual after the actual exclusion, while there is no reversal after inclusion.
3.2.3 Volume Effects of Index Changes The analysis of the abnormal volume changes around both the inclusion and the exclusion provides further evidence on the demand hypothesis. Volume is measured by two alternative variables. The first one is based on the dollar trade volume ($Vol) which is measured as; $Voli ,t =
log(1 + ETVi ,t ) log(1 + MCapi ,t ) N
$Volt =
∑ $Vol
i
i =1
N
$Voli,t represents the dollar trade volume for stock i on day t. ETVi,t is the total value of equity traded in US$ for stock i on day t. MCapi,t stands for the total market value of the stock i in US$ on day t. N is the number of stocks in the cross section. I first calculate the $Vol series for each stock, and then take the cross sectional average over each day in the event window. Additionally, I also standardized this measure (s$Vol), to obtain a clearer measure of volume changes between pre-announcement and post-announcement. This is done by scaling the $Vol by the average $Vol in the 41 day period prior to the announcement. s$Voli ,t =
$Voli ,t ⎡⎛ ⎤ ⎞ ⎢⎜ ∑ $Voli ,t ⎟ / 41⎥ ⎠ ⎣⎝ t =−60 ⎦ −20
In addition to the dollar trade volume, I also calculate the volume series by using the turnover ratio. The turnover ratio ( Ti ,t ), which is calculated as:
Ti ,t =
Qi ,t NOSH i ,t
where NOSHi,t denotes the daily number of shares outstanding for stock i and Qi,t is the number of shares of company i traded on day t.27 I also scale this measure, by the average turnover, between event days -60 and -20. The volume analysis presents interesting results. First, the trade volume is always higher for included stocks than excluded ones. This is actually consistent with trade volume being
27
Turnover ratio is also a widely-used measure of liquidity.
14
one of the inclusion/exclusion criteria. Thus, causality may be in both directions; from inclusion to volume or vice versa. [INSERT TABLE III AND FIGURE 3 HERE] Additionally, I find that the effect of exclusion on volume is stronger and more significant than of the inclusion. There is a more rapid and larger increase upon the exclusion announcement. The volume is significantly higher for almost all the days in the run up window. Highest volume is observed on the actual index change date. In the days following the actual index change, volume stays at a higher level than pre announcement levels. Upon the inclusion announcement, volume increases significantly on days 1, 2 and 10. Volume increases on other days in the run-up window, but at a smaller scale. There is a significant drop in the volume four days after the inclusion, which is potentially consistent with the release from index fund demand. Finally, both measures provide consistent results.
3.2.4 Robustness Checks To check for the robustness of these price results, I conduct a number of tests, using alternative methodologies and measures. These also target some methodological issues that may be problematic in an event study. The abnormal return results explained so far are based on parametric tests. Thus, there is a possibility that these may be biased due to outliers. The significance of abnormal returns around both events is also tested using non-parametric methodology. This is done by sign and rank tests. The significant event-induced returns are robust to this change of methodology. Additionally, the significance tests are done using both cross-sectional and time series variance measures. Since the results are found to be consistent, and almost the same, I only report the test results with the cross-sectional variance measure. One issue that often comes up in event studies is the appropriateness of the return generating model. In this paper, the abnormal returns are calculated as excess over the local market returns. However, different return generating models exist in the literature, such as the market model.28 Therefore, the abnormal returns are recalculated using MSCI EM and world market as the benchmarks. The significance and the pattern of the observed returns are consistent, though slightly higher in both cases than the original findings.29 Additionally, I also consider the market model with the world and EM indices as the market portfolio. Again, 28
Brown and Warner (1980, 1985) show that in short-run event studies, the results are robust to the choice of the return generating model. 29 Findings: 1.9% permanent increase around inclusions, in excess of MSCI EM.
15
the significant abnormal returns and price effects are consistent. Since this difference is negligible, I choose to use local market adjusted returns, due to the peculiar nature of emerging market stocks.30 Another common problem in event studies is the clustering of event dates. This would then result in cross sectional correlation between the synchronized stock returns, thus spurious abnormal returns. One way to control for this issue is to use the calendar portfolio approach. This methodology has been explained in detail in Section 2 of this paper. I find the abnormal returns on the constructed portfolios to be significant on the announcement date and on the index change date. Similarly, the returns are significant for most days in the run-up window. These findings verify the robustness of the event-time results, and clarify the issue of clustering. Last but not the least, I run cross country tests. Since the sample consists of stocks from 24 different markets, the question of whether any country group results dominate the overall results comes up. I do not find any particular country or countries dominating the overall findings. Moreover, some emerging markets are more integrated with the developed markets. This may have a significant impact on the magnitude of the price changes induced by the index inclusion.31 To test for this, I first rank the markets in the sample based on the level of integration and foreign investor recognition. Since there is no established ranking, I base it on the level of US Holdings of equities in each market.32 Then I scale this by the total stocks market capitalization of each market.33 The results show that Korea, Brazil and Israel are the most integrated while China, India and Malaysia are the least integrated markets in the sample. As expected, I find that less integrated markets experience larger and more permanent price increases.34
3.2.5 Discussion The price and volume analyses reveal a number interesting findings on the issue of price effects of index changes. In this section, I focus mainly on the following four results obtained from this new EM data;35 30
The segmented nature of these markets is well documented in the literature. It has been shown that local market factors have a much stronger influence on the asset prices than global factors. 31 If a market is more integrated, foreign investors should already be informed about these markets. Thus, I would expect to see a lower price impact on stocks from more integrated EMs. 32 This data is available form the US Department of Treasury website. 33 Both of these measures are in US dollars, and the domestic market capitalization data is available from the website of World Federation of Exchanges. 34 The detailed country results are available upon request. However, I do not report these in the paper for the sake of brevity. 35 The ‘events’ in this section refers to the inclusion and the exclusion.
16
1. Stock prices react immediately to both inclusion and exclusion announcements. 2. While the inclusions result in a permanent price change, this is not true for the exclusions. 3. The price dynamics within the event window are different for inclusion and exclusion. a. This is also true for the volume changes. The exclusion has a stronger impact on the trade volume.
4. While there is a significant price reversal upon the exclusion, there is none in the inclusion case. I also find that the price effects for EM stocks are smaller than those reported for S&P500 stocks.36 However, this is of secondary importance, since the main aim of this study is to test for the demand and information hypotheses. The first result documents that both events induce some kind of abnormal price reaction. This contradicts with the semi-strong form of market efficiency. If the index change does not introduce any new information, in the presence of market efficiency there should not be any significant abnormal reaction to the announcement. Even if the index change reveals some stock specific information, the stock prices should incorporate this information rapidly. In an efficient market, there should not be any profitable trade pattern based on this. However, I find convincing evidence for an abnormal price reaction to the index change announcement. Most importantly, this finding enables me to carry on with tests of the suggested hypotheses. The second result is not consistent with the predictions of the ‘downward sloping demand curve’ hypothesis (DSDC). DSDC predicts that if demand is the underlying cause for the price changes, both of these index changes should result in a similar price change. The intuition is that the investors who demand the stocks upon inclusion would sell these upon exclusion, everything else being the same. However, this also assumes that the same type of investors react to both types of index changes. In the case of MSCI EM inclusions, I do not observe a symmetrical price effect. While the index inclusions results in an average of 2.4% increase in price, the exclusion effect seems to disappear in the next two months following the index change. Yet, this does not necessarily mean that the demand curves are horizontal. I explore this further in the next section. The third result is similar to the findings of S&P500 studies. I observe the price dynamics in each case to be noticeably different. The exclusion has a sharper and larger impact on the
36
They are also smaller than those reported for other US index changes and developed markets; like Japan (Nikkei), UK (FTSE), Canada (TSE). For S&P500 inclusions, the permanent price increase is reported to be around 3-4%.
17
stock price. Upon the announcement, price drops sharply until the actual exclusion. Once the stock is out of the index, there is a reversal, and price increases gradually. On the other hand, price moves in a different way upon the inclusion announcement. Immediately after the announcement, there is a significant and positive reaction. This upward trend persists until the actual addition to the index with a total increase of 1.9%. Two months after the inclusion, the price is table at around 2.4%. The volume changes are also consistent with these different price dynamics. I find the volume changes around the exclusion to be larger than the inclusion. The larger price change is, therefore, associated with higher volume. There is also high correlation between the volume and CARs; for inclusions, this is 0.73. This is consistent with the explanation that demand drives the price. The fourth result is a key divergence from the existing literature regarding the documented price changes around S&P500 inclusions.37 While earlier studies document a price reversal, I do not find any evidence for this in the EM context. This result is consistent with two stories. First, all of the price change upon inclusion may be resulting from new information. In this case, the observed change would be an adjustment to new price equilibrium. However there is no prior reason to assume that the decision to be included in or excluded from the index contains any new information. Additionally, any fundamental stock information hypothesis also predicts symmetric responses to both index changes. Thus, if new information is defined as any fundamental information that should change the underlying stock price, the results in this paper do not support this explanation. On the other hand, in the presence of market segmentation, as in the EM case, the existing information may not be visible to all the investors, particularly the foreign ones. Thus, the second potential explanation is the change in ‘investor awareness’ around the event. In the literature, this term has been used as an alternative explanation to demand-based ones. In this paper, I use it to refer to increased investor recognition and decreased shadow cost. This is similar to a radar-screen effect, which means that more investors start following these stocks. This does not necessarily require the event of inclusion to have any information content. The investor awareness story, in this case,
37
The findings of this paper are comparable to those of post-1989 index changes to S&P500. Prior to 1989, the index changes were announced and implemented on the same date, whereas there are at least two weeks between these two events since 1989. MSCI’s announcement policy for the EM index is compatible with that of S&P, post-1989. The documented price effects around pre-1989 and post-1989 inclusions are different. While there is a full permanent price increase during the earlier period, there is a significant price reversal upon the index change in the latter case.
18
is more emerging markets specific, i.e. arises due to market segmentation.38 It refers to international investors increasing their exposure to the emerging market stock which are recognized internationally by this inclusion. This is not necessarily mutually exclusive with the demand story, unlike what has been so far discussed in the literature. The excess demand around S&P inclusions has been explained by the heavy index fund trading upon the announcement. When this is the case, the index change is followed by a price reversal once this excess demand is met. Although I find a significant relationship between the abnormal returns and volume, the absence of any price reversal suggests that not all of the price increase can be attributed to the index fund trading. Even after the inclusion, there is no significant price decrease. This is consistent with increase in investor awareness. If investors, other than the index funds, start demanding these newly included stocks, this additional demand will have a different impact on the price. In the S&P case, all the included stocks are already in investors’ set of investable assets, so the event does not have a direct and major effect on the stock visibility. The inclusion affects mainly the holdings of index funds, which are forced to invest in the constituents of the tracked index. In the EM case, the underlying price mechanism is quite different.39 Unlike the US market, foreign investors are subject to direct and indirect barriers to investment while investing into the emerging markets. While the direct barriers are usually abolished by the official capital market liberalizations, indirect barriers, like poor information availability, still keep the foreign investors away from these stocks. Inclusion in a benchmark global index is a good candidate for an event, around which information availability increases. This then would make these stocks more visible to the international investors. I find that the findings in this study are consistent with this argument. Similarly, they are also consistent with the ‘investor recognition hypothesis’ of Merton (1987), ‘pricing of neglected stocks’ by Arbel et al. (1983), ‘pricing of internationally restricted/segmented stocks’ by Errunza and Losq (1985) and Hietala (1985). All these studies show that in the 38
This hypothesis is still consistent with the DSDC, while it predicts a different demand schedule throughout the event window. Actually, it is also similar to the ‘market segmentation’ that Shleifer (1986) mentions as an additional explanation to inclusion-induced price increases. However, he does not test for this directly. In his case, the market segmentation refers to the new demand from new investors who base their investment decisions on inclusion in certain asset categories, e.g. S&P500 vs. non-S&P500 stocks. While this change in investment strategy is more mechanical in the S&P case, it would refer to international capital market segmentation /integration in the EM case. (The segmented nature of EMs and its asset pricing implications have been well documented in the literature) Though foreign investors initially refrain from investing in EM stocks, inclusion in a global index may put these stocks in the investable set of stocks. However, in this case, it is not an issue of being part of an index of specific interest, but becoming more visible and practically more eligible to foreign investors. This is what MSCI EM inclusion may be implying for the segmented EM stocks. 39 Particularly since the asset pricing is different in these markets, which are subject to imperfect information, market segmentation, and investment restrictions.
19
presence of market imperfections such as poor information availability, restricted investors/stocks, stocks are traded at a risk premium. Any decrease in these imperfections should then result in a correction in price. In this study, I am able to document significant abnormal price reaction induced by index rebalancings. However, this result cannot be attributed to any pure demand (DSDC type) or new information effect. While the basic results are similar to those of S&P500, the permanent effect of inclusion suggests that in the EM context, there is an additional effect caused by either release of new information or increased investor awareness. In the remainder of this paper, I conduct further tests to identify the underlying cause.
4 Alternative Explanation: Investor Awareness It is relatively harder to obtain a plausible measure for investor awareness for the emerging markets. The data on institutional or any other kind of share holding is not publicly available. Moreover, any available data is more restricted than those for the US markets. On the other hand, the results in this paper indicate an ‘investor awareness’ story, particularly for the inclusions. Therefore, I suggest an alternative measure for investor awareness to test for the validity of this argument. Though different from the shareholder data, it should still detect and capture any increase in investor awareness. The validity of the investor awareness hypothesis does not require symmetrical price effects between the inclusion and exclusion cases. Since my results confirm this, I only focus on the inclusion cases in this part of the analysis. Following Merton (1987) and Arbel et al. (1983), I use the analyst coverage as a proxy for investor awareness. I study the number of recommendations issued each month; obtained from I/B/E/S database. The monthly data for each stock is synchronized for the whole sample. This involves data starting from 12 months prior to the event, until 24 months after the event. This data is available for 223 stocks in the inclusion sample. If there is any increase in investor awareness, this should result in a higher number of recommendations. The statistical significance is measured using non-parametric rank and sign tests. They all show that there is a significant increase in the number of recommendations upon the inclusion. Out of the 223 stocks, 26 do not have any recommendations issued prior to the inclusion. The percentage increase is the highest for these. First I compare the monthly average for number of recommendations issued in the 5month period prior to the inclusion and 5-month period after the inclusion. The month in which the inclusion occurs is included in the post-inclusion period. On average, I find the 20
change to be 157%.40 The recommendations per month has increased for 78% of the stocks, while this number stayed the same for 3%, and decreased for 18%. I also test for this change using 12-month windows. The results are robust, though there is a larger increase (221%) for a larger percentage (87%) of the stocks in the later exercise. This increase is significant when tested using non-parametric tests; sign and rank tests. Additionally, I regress the cumulative abnormal returns on the change in number of recommendations. The results show that the size of the price change is positively and significantly related to the size of the change in stocks recommendations.41 The recommendation results support the argument that the investor awareness significantly increases upon the index inclusion. This increase, then, contributes to the increase in demand, and thus in stock price. Thus the inclusion increases the visibility of stocks as well as of available stock information, in this particular market setting. The investors, who refrain from investing in these stocks due to indirect barriers such as lack of information, start include these emerging market stocks in their global portfolio of investable assets. Though there is no new information arrival which could change the underlying stock value, this increased awareness results in lower shadow costs, and revaluation of the stock price.
4.1
ADR vs. no-ADR Stocks
American Depositary Receipts (ADRs) are quite common for EM stocks, which want to raise capital in foreign markets. The existence of ADRs can significantly affect the change in investor awareness of the particular stock. If the included stocks already have ADRs, then they should already have some analyst coverage. Thus, the increase in the number analyst recommendations issued should be lower than, if any, the one no-ADR stocks experience. To see if this is true for the EM case, I compare the recommendation changes between ADR and no-ADR stocks that are being included in MSCI EM. Only 27 stocks have active ADRs at the time of the inclusion. 11 of the ADR stocks are among the largest 20 stocks in the sample. The stocks are dispersed among markets. I find that on average these stocks experience only a 30% increase in issued recommendations. Additionally, I do not find any significant relationship between the magnitude of the price change and of the number of recommendations. While there is a significant change in analyst 40
The minimum change is -37% and the maximum is 2500%. In the latter case, there is a shift from norecommendations to an average of 25 recommendations per month, upon the inclusion. 41 The coefficient is 0.003 with t-stats of 1.76. This is estimated by using White heteroskedasticity-consistent standard errors.
21
coverage for the no-ADR stocks, this is quite small for the ADR stocks. This finding strengthens the validity of the investor awareness hypothesis. I find that less visible EM stocks experience a larger change in investor awareness upon the inclusion.
4.2
Small vs. Large Stocks
It is well documented in the literature that there is an increase in aggregate demand around index inclusions. The natural suspects are the index funds, which are obliged to track any changes in the respective index. Given that there is a substantial amount of index tracking associated with MSCI EM, the index funds are responsible for some of the price increase. However, as we see in the case of S&P500 inclusions, if the index funds are the major source of new demand, then there is partial price reversal once the index change takes place. In the case of MSCI inclusions, I document that the price increase in permanent with no price reversal. This suggests that part of the increase in demand may be coming from investors other than the index funds. This is also intuitive, once the change in stock visibility around the inclusion is taken into consideration. Therefore, in this section, I test for any evidence in the observed price dynamics that can be attributed to other institutional investors, such EM funds. To test for this, I use one of the major investment criteria for EM funds. In practice, while these funds are interested in MSCI EM stocks, they are not allowed to invest into small stocks. I rank the included stocks in the sample based on size as measured by the total market capitalization at the time of the inclusion. Then, I compare the largest 10% with the smallest 10%. The results show that the price patterns are noticeably different around these two groups of stocks. While there does not seem to be any significant price increase upon the inclusion for smallest stocks, there is a significant price increase of 5% for largest 20 stocks. One interesting observation is that for the small stocks, there is a price increase between day (-10) and AD. This is around 3%, and is caused mainly by the Chinese stocks. The small stocks experience a price increase around CD, not AD.42 However, this gradually disappears in the 20 days following the inclusion.
5 Conclusion In this paper, I revisit the debate on the underlying cause of significant price effect of the index inclusions. My aim is to test for the validity of the existing hypotheses in a different
42
The maximum CAR occurs three days before the actual index change (CD-3) at 1.8%.
22
market setting. To serve this purpose, I analyse the changes in MSCI EM index. The sample consists of 280 inclusions and 264 exclusions that occur between 1996 and 2006. This enables me to conduct stronger tests of the DSDC and information hypotheses. First, I test if there are any abnormal price effect induced by the announcement and the index change. After documenting that there are significant price effects around both inclusions and exclusions, I compare the price changes around these two events. Any asymmetric response is not consistent with DSDC or new information hypotheses. I find this to be the case for MSCI EM inclusions. The initial reaction to both events is significant and opposite as expected. However, while the index inclusion results in a permanent price increase, the price decrease upon the exclusion gradually decreases over the two months after the actual exclusion. The total price effect of the inclusion is around 2.4% with no price reversal upon the actual addition to the index. The last result is a major difference from the existing findings in the literature. The permanent effect is consistent with the investor awareness hypothesis. Thus, the final part of the empirical analysis focuses on this. Especially the segmented nature of the emerging markets makes the investor awareness hypothesis a strong candidate for explaining the observed price effect. The intuition is that though there is no information release by the index company or the company of the stock, investors become aware of the already existing information once the stock is part of this global index. In short, the stock becomes part of the global investable portfolio of these investors. The price adjustment upon the inclusion is caused by this. I am able to document empirical evidence for increased investor awareness using the analyst recommendation data which is an alternative measure of investor awareness suggested in the literature. The results show that analyst coverage of the included stocks increases significantly upon the announcement. Furthermore, I also find a significant and positive crosssectional relationship between the abnormal returns and the change in number of recommendations. While the results support the investor awareness story, I do not reject the downward sloped demand curve hypothesis. I do find evidence for the price pressure that is possibly caused by index funds; however, the findings suggest that the total price effect cannot only be explained by DSDC. The enhanced investor awareness seems to have a significant role as well. Though subject to caveats, this study provides convincing evidence on the role of imperfect information on asset pricing by providing evidence from index changes in EMs. It also enlightens some market specific issues in emerging markets, which so far have been only 23
dealt within the asset pricing framework. The findings are consistent with the strong impact of ‘lack of information availability’ and ‘foreign investor awareness’ on asset pricing in these markets.
24
References Arbel, Avner, Steven Carvell, and Paul Strebel, 1983, Giraffes, Institutions and Neglected Firms, Financial Analysts Journal May/June 1983, 2-8. Asquith, Paul, 1983, Merger Bids, Uncertainty, and Stockholder Returns, Journal of Financial Economics 11, 51-83. Beneish, Messod D., and Robert E. Whaley, 1996, An Anatomy of the "S&P Game": The Effects of Changing the Rules, Journal of Finance 51(5), 1909-1930. Brown, Stephen J., and Jerold B. Warner, 1980, Measuring Security Price Performance, Journal of Financial Economics 8(3), 205-258. Brown, Stephen J., and Jerold B. Warner, 1985, Using Daily Stock Returns: The Case of Event Studies, Journal of Financial Economics 14(1), 3-31. Chen, Honghui, Gregory Noronha and Vijay Singal, 2004, The Price Response to S&P 500 Index Additions and Deletions: Evidence of Asymmetry and a New Explanation, Journal of Finance 59 (4), 1901-1929. Corrado, Charles J., 1989, A Nonparemetric Test for Abnormal Security-Price Performance in Event Studies, Journal of Financial Economics 23, 385-395. Denis, Diane K., McConnell, John J., Ovtchinnikov, Alexei V. and Yu, Yun, 2003, S&P 500 Index Additions and Earnings Expectations, Journal of Finance 58(5), 1821-1840. Dhillon, Upinder, and Herb Johnson, 1991, Changes in the Standard and Poor's 500 List, Journal of Business 64(1), 75-85. Errunza, Vihang, and Etienne Losq, 1985, International Asset Pricing under Mild Segmentation: Theory and Test, Journal of Finance 40(1), 105-124. Harris, Lawrence, and Eitan Gurel, 1986, Price and Volume Effects Associated with Changes in the S&P 500 List: New Evidence for the Existence of Price Pressures, Journal of Finance 41 (4), 815-829. Hegde Shantaram P. and John B. McDermott., 2003, The Liquidity Effects of Revisions to the S&P500 Index: An Empirical Analysis, Journal of Financial Markets, 6(3), 413-459. Hietala, Pekka T., 1985, Asset Pricing in Partially Segmented Markets: Evidence from the Finnish Market, Journal of Finance 44(3), 697-718. Jaffe, Jeffrey F., 1974, Special Information and Insider Trading, Journal of Business, 47(3), 410-428. Jain, Prem C., 1987, The effect on stock price from inclusion in or exclusion from S&P 500, Financial Analysts Journal 43, 58-65. Kaul, Aditya, Vikas Mehrotra, and Randall Morck, 2000, Demand Curves for Stocks Do Slope Down: New Evidence from an Index Weights Adjustment, Journal of Finance 55 (2), 893–912. Lynch, Anthony W., and Richard R. Mendenhall, 1997, New Evidence on Stock Price Effects Associated with Changes in the S & P 500 Index, Journal of Business 70(3), 351-383. Mandelker, Gershon, 1974, Risk and return: The Case of Merging Firms, Journal of Financial Economics, 1(4), 303-335. Merton, Robert C., 1987, A Simple Model of Capital Market Equilibrium with Incomplete Information, Journal of Finance 42(3), 483-510 Shleifer, Andrei, 1986, Do Demand Curves for Stocks Slope Down?, Journal of Finance 41 (3), 579590. Wurgler, Jeffrey and Ekaterina Zhuravskaya, 2002, Does Arbitrage Flatten Demand Curves for Stocks?, Journal of Business 75, 583–608.
25
Country Argentina Brazil Chile China Colombia Czech Republic Egypt Hong Kong Hungary India Indonesia Israel Jordan Korea Malaysia Mexico Morocco Pakistan Peru Philippines Poland Russia South Africa Taiwan Thailand Turkey Venezuela Total
MSCI EM Index (as of August 2005) Index Total No of Minimum size weights MCAP stocks guidelines 0.3% 4,523 9 75 10.8% 149,732 47 200 1.9% 26,775 22 100 4.0% 55,881 45 200 0.3% 3,690 6 75 0.8% 11,390 6 100 0.8% 11,057 15 75 3.4% 47,654 28 200 1.4% 20,005 6 100 5.7% 78,716 62 200 1.4% 19,922 26 150 3.6% 49,263 39 150 0.3% 4,259 13 75 17.8% 246,621 73 450 3.4% 46,835 75 150 6.3% 87,844 22 200 0.2% 3,399 11 75 0.3% 3,995 14 75 0.3% 4,259 6 75 0.5% 6,520 18 75 1.8% 25,346 22 100 5.2% 72,487 19 200 9.7% 134,832 49 450 14.2% 196,845 103 450 1.9% 26,872 43 150 2.0% 27,926 35 100 0.1% 1,415 5 75 100% 1,387,304 823 Sample weights 0.30% 4.04% 2.17% 8.58% 0.53% 5.19% 5.44% 0.07% 6.03% 1.80% 2.01% 15.44% 1.16% 1.03% 0.36% 0.16% 0.20% 0.38% 9.70% 5.52% 19.59% 10.10% 0.18% 0.02% 100%
Inclusion sample Total No of MCAP stocks 1,020 2 13,670 15 7,359 5 29,053 31 0 1,799 3 17,550 2 18,425 8 235 3 20,396 24 6,086 20 6,789 8 0 52,266 37 3,914 13 0 3,480 3 1,204 1 550 10 686 5 1,287 5 32,829 12 18,691 7 66,314 22 34,182 41 618 2 60 1 338,464 280
Exclusion sample Sample Total No of weights MCAP stocks 0.09% 148 2 10.49% 16,678 14 1.36% 2,158 7 1.99% 3,161 15 0 0.00% 7 1 2.65% 4,218 5 13.08% 20,799 13 0.00% 0 12.22% 19,423 19 1.38% 2,187 21 0.03% 43 5 0 9.47% 15,061 34 0.83% 1,313 6 2.40% 3,823 4 0.58% 921 3 0.20% 325 10 0.07% 104 4 0.48% 763 12 1.65% 2,629 5 13.66% 21,720 5 1.91% 3,030 7 2.56% 4,070 27 22.48% 35,732 31 0.38% 611 7 0.03% 54 7 100% 158,977 264
26
This table presents country breakdown of Morgan Stanley International, Inc’s Emerging Markets Index (MSCI EM) and of the inclusion and exclusion samples, used in this paper. The table is divided into three main groups: MSCI EM (based on the information as of August 2005), the sample of stocks added to the index, and the sample of stocks deleted from the index. The first column presents the countries that constitute MSCI EM, and the weights of each country group in the index (with respect to the total market capitalization of the index) are shown in the second column. Similarly, columns six and nine show the weights of each country group in the samples. In column three, the total market capitalization (MCAP) for each country in MSCI EM is shown, in millions of USD. Columns seven and ten show the MCAP for each country, using free float market value as of 2005. Numbers of stocks from each country are presented in columns four, eight and eleven. Eligible minimum size thresholds for inclusions of new securities in MSCI country indices are presented in column five, as free-float adjusted MCAP in millions of USD, as of May 2005.
Table I Sample and MSCI EM Index Description
Table II Cumulative Abnormal Returns (CARs) This table illustrates the CARs over seven event windows, within the large event window. Panel A shows the results for the inclusions (total of 280) while Panel B shows these for the exclusions (total of 264). The first column specifies the event window of interest. The actual start and end dates of these windows are shown in the second column. AD stands for the ‘Announcement date’ and CD stands for ‘Change date’. CARs are cumulated separately within these windows, and for each window are presented in the third column. While the third and fourth columns illustrate the equal-weighted average CARs and t-statistics, the last two columns present the value weighted averages for the sample. The significance of the CARs is calculated using t-statistics using cross-sectional variance, and these are reported in the fourth column. *, ** and *** denote significance at the 10%, 5%, and 1% level, respectively.
Panel A: Inclusions Event Window
Event Dates
equal-weighted CAR t-stats
I. Anticipation
AD-10, AD-1
0.83%**
2.10
0.25%
0.67
II. Announcement day
AD
0.47%***
3.99
1.10%***
8.77
III. Run up window
AD+1, CD-1
0.55%**
2.18
2.66%***
7.48
IV. Index change day
CD
0.22%*
1.89
1.10%***
9.28
V. Price reversal
CD+1, CD+10
-0.27%
-0.77
-0.17%
-0.46
VI. Short-run event window
AD-10, CD+20
1.89%**
2.48
4.44%***
5.91
VII. Long-run event window
AD-10, CD+60
2.36%**
2.36
5.56%***
5.60
Panel B: Exclusions Event Window
Event Dates
I. Anticipation
AD-10, AD-1
-0.66%
-1.05
-0.60%
-1.00
II. Announcement day
AD
-0.19%
-0.98
-0.35%*
-1.77
III. Run up window
AD+1, CD-1
-2.15%***
-5.35
-1.77%***
-2.98
IV. Index change day
CD
-1.05%***
-5.26
-0.95%***
-4.79
V. Price reversal
CD+1, CD+10
0.61%
0.97
0.32%
0.53
VI. Short-run event window
AD-10, CD+20
-3.44%***
-3.16
-1.99%
-1.62
VII. Long-run event window
AD-10, CD+60
-0.98%
-1.11
-1.09%
-0.66
equal-weighted CAR t-stats
value-weighted CAR t-stats
value-weighted CAR t-stats
27
Table III Volume Changes around MSCI Inclusions and Exclusions This table presents the daily average volume results obtained from two different measures. In the first column, the event days are shown, the announcement day AD is day zero, and the index change day CD is day 10. The second and third columns present the standardized average volume, for included and excluded stocks respectively. The turnover rate for stock i on day t is measured as Qi ,t Ti ,t = NOSH i ,t number of shares traded of stock i on day t, divided by the number of shares outstanding of stock i on day t. the fourth and fifth columns show the average volume measured as the dollar trade volume: $Voli,t log(1 + ETVi ,t ) $Voli ,t = log(1 + MCapi ,t ) ETVi,t stands for the total equity value traded for stock i on day t, and MCapi,t is the total market capitalization for the same stock on day t. The $volume sample consists of 254 inclusions and 217 exclusions, while the turnover ratio is available for 243 inclusions and 219 exclusions. Both variables are in US dollars. *, ** and *** denote significance at the 10%, 5%, and 1% level, respectively.
TURNOVER Inclusions Exclusions Event day -3 -2 -1 AD=0 1 2 3 4 5 6 7 8 9 CD=10 11 12 13 14 15 16 17 18 19 20
1.33 1.20 1.29 1.42 2.74*** 2.15*** 1.34 1.62 1.41 1.46 1.49 1.34 1.44 2.15*** 1.72 1.49 1.56 1.12 1.47 1.46 1.33 1.36 1.35 1.38
1.09 1.23 1.47 1.34 1.30 2.05 1.86 2.35* 1.97 2.86** 2.06 2.46* 2.82** 3.27*** 2.34* 2.52** 1.95 1.51 1.86 2.35* 1.68 1.70 1.51 1.63
$VOLUME Inclusions Exclusions 1.12 1.08 1.07 1.09 1.08 1.13* 1.09 1.13 1.07 1.21** 1.04 1.08 1.06 1.15* 1.11 1.13 1.11 0.86 1.07 1.12 1.10 1.34*** 1.08 1.10
0.97 1.17 1.15 1.07 1.17 1.28** 1.30** 1.33*** 1.33*** 1.22* 1.23* 1.26** 1.33*** 1.39*** 1.37*** 1.22* 1.25** 1.29** 1.18 1.24** 1.29** 1.31*** 2.15** 1.29**
28
AD-10
I. Anticipation window
AD-1
AD
III. Run-up window
AD+9
VII. Long-run event window
CD CD+1 (=AD+10)
IV. Index composition change date
VI. Short-run event window
AD+1
II. Announcement date
V. Price reversal window
CD+10
29
CD+60
Figure 1. Timeline and Multiple Event Windows. This timeline illustrates the event windows used in this paper. AD stands for the announcement date. CD stands for the actual index change (inclusion and exclusion) date. There is one large ‘long-run’ event window (VII), which covers the period 10 days before the announcement date until 60 days after the change date that is between AD-10 and CD+60. In order to analyse, the price changes within the event window, the large window is divided into five smaller event windows: I. Anticipation window (between AD-10 and AD-1), II. Announcement day, AD, III. Run-up window (between AD+1 and CD-1), IV. Index composition change day, CD, V. Price reversal window (between CD+1 and CD+10). The short-run window VI that is used to analyse the temporary price effect of the index change runs from AD-10 until CD+20.
-4.00%
-3.00%
-2.00%
-1.00%
0.00%
1.00%
2.00%
3.00%
4.00%
-10
I
-5
0
VI
III
5
10
V
15
20
25
VII
30
35
40
45
50
55
60
70 event day
65
Exclusions
Inclusions
30
Figure 2. Cumulative Abnormal Returns (CARs) for Inclusions in and Exclusions from MSCI EM. The figure illustrates CARs for both inclusions and deletions. Daily sample averages of local-market adjusted abnormal returns (ARs) are used to calculate the CARs; AARs are accumulated from day (-10) onwards, until day 70. The upper line represents the CARs for inclusions, while the lower line shows the CARs for the exclusions. The numbers I to VII stand for the event windows that are shown in Figure 1, on the time line. I denotes the announcement window, III: the run-up window, V: price reversal window, VI: short-run event window and VII: long-run event window. The three vertical dashed lines, starting from the left, represent the announcement (AD, event window II), change dates (CD, event window IV) and CD+10 respectively.
CAR
31
