The Disappearing Index Effect* 

 

 

Robin Greenwood & Marco Sammon 

Harvard Business School 

 

March 2024 

 

 

Abstract 

The abnormal return associated with a stock being added to the S&P 500 has fallen from an average 
of 7.4% in the 1990s to less than 1% over the past decade. This has occurred despite a significant increase in 
the share of stock market assets linked to the index. A similar pattern has occurred for index deletions, with 
large negative abnormal returns during the 1990s, but only 0.1% between 2010 and 2020. We investigate the 
drivers of this surprising phenomenon and discuss implications for market efficiency. Finally, we document a 
similar decline in the index effect among other families of indices.  

  

 
* We thank Robert Ialenti for excellent research assistance and Paige Frasier for editorial assistance.  We received helpful 
comments from Ian Appel (discussant), Lloyd Blankfein, Alex Chinco, Caitlin Dannhauser (discussant), Adam Denny, 
Taro Hornmark, Bill Jacques, Owen Lamont, Ananth Madhavan, Victor Martin, Phil Mackintosh (discussant), Adam 
Morrissey, Lubos Pastor, David Rabinowitz, John Shim, Andrei Shleifer, Erik Stafford, Nogie Udevbulu, and seminar 
participants at Harvard.  

Electronic copy available at: https://ssrn.com/abstract=4294297



2 
 

One of the early and persuasive challenges to the efficient markets hypothesis is the observation that 

stock prices react to investor demand unrelated to fundamentals.  Shleifer (1986) and Harris and Gurel (1986) 

showed that stocks added to the S&P 500 index experienced abnormal returns of approximately 3 percent 

around the announcement of an index change.  Since then, an extensive literature has documented similar 

price impact in other stock indices such as the Russell and the MSCI, as well as many other settings in which 

investors buy or sell for reasons unrelated to fundamentals, including mutual fund inflows, mechanical 

reinvestment of dividends, price pressure around mergers, and Treasury auctions.1  

The initial studies of S&P 500 changes were performed during a time when index investing was 

nascent. Shleifer (1986) notes that S&P 500 announcement returns were smaller prior to 1976 than in the 

1976-1983 period he focuses on, consistent with more dollars tracking the index leading to more price 

pressure. Over the past 40 years, driven by inflows into passive mutual funds and ETFs, index tracking has 

continued to grow at a rapid pace. We estimate that funds tracking the S&P 500 in the form of mutual funds 

or ETFs have grown from essentially zero in the 1980s to approximately 7 percent of market capitalization in 

recent years. Other estimates, based on trading volume (Chinco and Sammon 2022) or sell-side research, 

suggest even higher levels of investor indexation to the S&P 500 today.  

What has happened to the price impact associated with being added to or removed from the S&P 

500? A natural starting point would be to assume a demand curve with a constant elasticity, hit by a shock 

that has been growing in magnitude over time: 

𝑃𝑟𝑖𝑐𝑒 𝐼𝑚𝑝𝑎𝑐𝑡 =  𝑀 × 𝐷   (1) 

Where 𝑃𝑟𝑖𝑐𝑒 𝐼𝑚𝑝𝑎𝑐𝑡  denotes the percentage change in price, 𝐷  refers to the percentage of capitalization 

of stock 𝑖 bought upon index addition or sold upon index deletion, and the multiplier M denotes minus 1 

over the demand elasticity. Given the rise of indexation, this logic would predict substantially growing price 

 
1 See e.g., Warther (1995), Mitchell et al. (2005), Ben-Rephael et al. (2010), Lou et al. (2013) and Hartzmark and Solomon 
(2022).  

Electronic copy available at: https://ssrn.com/abstract=4294297



3 
 

impact from the 1980s onwards. Conforming to this intuition, we show that the average price impact grew 

from the 1980s to the 1990s, from an average total return of 3.4% in the 1980s to 7.4% in the 1990s. 

Surprisingly, however, and consistent with Bennett et al. (2020), we show that the average price impact 

decreased somewhat in the 2000s to 5.2%, and then fell further to below 1.0% in the most recent decade, 

statistically indistinguishable from zero, even though indexation has continued to tick upwards. A similar 

pattern has occurred with index deletions. The average effect of being removed from the S&P 500 was -4.6% 

in the 1980s, -16.1% in the 1990s, -12.4% from 2000-2009, and -0.6% from 2010-2020. Again, the average 

return in the past decade is not statistically distinguishable from zero. Taken at face value, this would seem to 

imply that a multiplier M was declining substantially over the sample. 

Why did the S&P 500 index effect seemingly disappear? And if so, can we interpret this change from the 

lens of market efficiency and a lower multiplier, as an initial look at the data would suggest?  We consider four 

broad classes of explanations:  

1) Changing composition of additions and deletions: One possible explanation for the disappearing 

index effect is that the nature of firms added or dropped from the S&P 500 has changed over our sample.  

We quickly rule out that the effects we document are driven by changes in the characteristics of additions and 

deletions since the 1980s, although such shifts do account for some of the changes. For example, the size of 

additions and deletions relative to the total capitalization of the S&P 500 has been shrinking over time.  This 

could partially explain the disappearance of the index inclusion and deletion returns, because empirically, the 

size of the added or dropped firm is strongly related to the magnitude of the index effect. We use a simple 

regression-based approach to show that changes in the composition (as measured by Wurgler and 

Zhuravskaya (2002)’s “arbitrage risk”, trading volume, size relative to total index capitalization and coverage 

by sell-side analysts) account for only a small portion of changes in the average index addition and deletion 

returns that we have observed over our sample. 

2) Migrations: A second class of explanation is that the net demand shock D experienced by the typical 

index addition or deletion is smaller than it appears. We show that in recent years, an increasing percentage of 

index additions and deletions are “migrations” from the S&P MidCap index. When these stocks are added to 

Electronic copy available at: https://ssrn.com/abstract=4294297



4 
 

the S&P 500 index, they simultaneously leave the S&P MidCap.  In these cases, forced buying by S&P 500-

tracking funds is simultaneously matched with forced selling from S&P MidCap-tracking funds, leading to a 

smaller net demand shock. From the 1990s to the present day, migrations went from about 50% of additions 

to over 70%. The trend toward more migrations is even stronger among S&P 500 index deletions.  

The differences in returns for migrations reflects the increasing importance of the S&P MidCap index 

over time. In the mid-1990s, migration and non-migration additions had average returns of 10.2% and 6.7%, 

respectively.  By the 2010s, however, direct additions had returns of 5.4%, while migrations had returns of -

1.8%.  This divergence coincides with the rise of MidCap-focused funds.   

3) Front running: A third class of explanation is that index additions and deletions have become more 

predictable over time, attracting arbitrageurs who front-run index demand. In this explanation, sophisticated 

market participants who anticipate index changes purchase additions and sell deletions before the 

announcement day, leading the price to move before the official announcement. In the extreme case, in which 

index changes could be perfectly anticipated, we would expect no abnormal returns at all during the window 

of time between announcement and when the index change occurs. While we do not have position-level data 

to measure front-running activity, we can measure pre-event returns, finding mixed evidence to support this 

hypothesis.  In recent years, returns in the three months leading up to the announcement have become 

somewhat stronger, although the reason for this is subject to interpretation.  In addition, a simple rule of 

selecting the largest eligible firm has become a better indicator of future S&P 500 addition, further suggestive 

evidence of predictability.  That said, predicting which precise stocks get added is still difficult, and we find 

very little change in the returns experienced in the 20-trading day period before announcement, where one 

would expect to see higher returns if there were substantial anticipation of index changes. Overall our 

assessment is that this explanation has played only a limited role in the decline of the index effect. 

4) Increased liquidity: The last explanation is that the stock market has simply become more efficient in 

the context of providing liquidity to S&P 500 index additions and deletions. Or, in the context of Equation 

(1), that M has declined. We show that even after accounting for the increased migrations changing our 

estimate of the average demand shock D, M has indeed declined by a factor of approximately 20 for index 

Electronic copy available at: https://ssrn.com/abstract=4294297



5 
 

additions, and even more so for index deletions. We then evaluate this hypothesis in conjunction with all the 

other drivers, reaching the same conclusion. 

Why has the market for index changes become so much more elastic? Part of the answer is that the 

stock market is more liquid today than in the past, in the sense that trading costs in other settings have also 

declined since the 1980s and 1990s. However, this is only part of the story, as trading costs have not fallen by 

enough between the 1990s and the late 2010s to explain the decline of the index effect. Moreover, the decline 

in market-wide trading costs predates the disappearance of the index effect. 

  Several market changes are likely to have facilitated the greater provision of liquidity around index 

events. First, over the past 15 years, Wall Street trading desks have increased personnel and computing 

resources devoted to index trading, with several large players (UBS, Goldman Sachs) having specialized sell-

side teams. Large passive investors also employ large teams to study and improve liquidity around 

rebalancing. Second, the distribution of trading volume has become more concentrated around index change 

events, facilitating liquidity provision (Chinco and Sammon, 2022). Third, despite the large size of the 

demand shock experienced by additions and deletions, much of it appears to be accommodated by other 

institutions. Specifically, although index trackers now buy about 7-8% upon index addition, total institutional 

ownership barely moves around index changes. We interpret this as professional active investors providing 

liquidity to passive buyers and sellers.  

Overall, the findings suggest an account along the following lines. In the 1980s, index changes were 

unanticipated, index funds were small, and there was mispricing in the market. As index funds grew larger, 

the mispricing deepened and turned into an opportunity. As a result, the market adjusted to take advantage of 

this opportunity, in part by better anticipating inclusions, but mostly by creating arrangements where other 

institutions stood ready to sell to indexers upon inclusions. This worked to eliminate the anomaly on average, 

despite demand shocks that continued to grow in magnitude over the 2000s and 2010s. In this sense, the 

decline of the index effect is much like the evidence for other anomalies: they decline once they are well 

recognized by the market (McLean and Pontiff, 2016). 

Electronic copy available at: https://ssrn.com/abstract=4294297



6 
 

Our results raise the obvious question of whether the index effect has declined in other settings, such 

as for the Russell 1000 and other well-tracked indices.  Many other papers, including Madhavan (2003) and 

Petajisto (2011), have shown index addition effects outside of the S&P 500.  To this end, we apply our 

methodology to several other families of indices, specifically the Russell 1000 and 2000, as well as the Nasdaq 

100 and other S&P mid- and small-cap indices. Overall, we find similar results for these indices in the sense 

that addition and deletion returns have declined over the past decade. However, the details and magnitude 

vary by index, and the statistical and economic significance is generally weaker than the decline we have 

described for the S&P 500. For the Russell Index additions, the change is not statistically significant. We leave 

a more complete analysis of these indices to future work.  

There is a long and vibrant literature on downward sloping curves and price pressure for individual 

stocks. Beginning with Shleifer (1986), Harris and Gurel (1986), and Lynch and Mendenhall (1997), dozens of 

studies analyze the implications of index changes for stock returns.2 Cai (2007) and Shahrbabaki (2022) 

distinguish between fundamental news and price pressure. Most closely related to our paper is Bennett et al. 

(2020), who first noted the decline in the index inclusion effect among additions between 1997 and 2017. 

Preston and Soe (2021) also document the decline in index inclusion effect among additions and deletions 

beginning in 1995.  Finally, Vijh and Wang (2022) document the lower absolute returns associated with 

migrations from the S&P MidCap to the S&P 500. 

A more recent literature has studied the effects of rising passive ownership, including Qin and Singal 

(2015), Bond and Garcia (2018), Garleanu and Pedersen (2018), Kacperczyk et. al. (2020), Buss and 

Sundaresan (2020), Ernst (2020), Malikov (2023), Lee (2021), Coles et. al. (2022). Koijen and Yogo (2019) and 

Gabaix and Koijen (2022) study implications for inelastic demand curves for stock prices and the aggregate 

market. Several papers focus specifically on estimating a price elasticity, i.e., how much prices can be expected 

to move for a demand shock of a given size (e.g., Scholes (1972), Loderer et al. (1991), Bagwell (1992), 

 
2 For additional evidence on S&P 500 index addition/deletion effects, see e.g., Wurgler and Zhuravskaya (2002), Yang 
and Morck (2002) and Petaijsto (2011). For evidence on addition/deletion effects in other indices see e.g., Kaul et al. 
(2000), Madhavan (2003), Greenwood (2005), Chang et al. (2015), Patel and Welch (2017) and Madhavan et. al. (2022).   

Electronic copy available at: https://ssrn.com/abstract=4294297



7 
 

Kandel et al. (1999) Kaul et al. (2000), Wurgler and Zhuravskaya (2002), Petajisto (2011), Chang et al. (2015)), 

which we do as well, although there is currently no consensus in the literature.3  

The paper proceeds as follows. In Section 1, we lay out the puzzle, documenting both the increase in 

mechanical demand driven by index changes, as well as the puzzling disappearance, on average, of an effect 

on returns. Section 2 considers, in turn, each of the potential explanations. Section 3 documents the decline 

of addition and deletion effects in other families of indices. Section 4 concludes. 

 

1. Index tracking and the index inclusion effect 1980-2020 

In this section we present the main facts. We first describe how we assemble a list of additions and 

deletions, before turning to how we identify funds that track the S&P 500. We then present statistics on 

announcement, effective date, and total returns associated with index changes. Last, we examine whether 

there is any correlation between net purchases by mutual funds and ETFs tracking the S&P 500 and the 

returns we observe.  

1.1 Data 

 The S&P 500 index is designed to track the performance of US large market capitalization stocks and 

be representative of the US economy.  To be added, the firm must file 10-K reports and be primarily based in 

the US.  Further, as of 2023, the firm must have a market capitalization of $12.7B or greater, a float-adjusted 

market capitalization that is at least 50% of this threshold, and an investible weight factor of at least 0.1 i.e., 

10% of the shares outstanding must count toward the float.  Finally, the firm must meet criteria for minimum 

float-adjusted liquidity, have the sum of the most recent four consecutive quarters of GAAP earnings be 

positive, and have positive GAAP earnings in the most recent quarter.4  At any given point in time, several 

stocks may meet these inclusion criteria, but a committee inside S&P decides which stock is added to the 

 
3 For example, Wurgler and Zhuravskaya (2002) highlight that estimates from these papers can range from -3,000 to -
1.65.  More recent papers including Petajisto (2011) and Chang et. al. (2015) find estimates between -1.5 and -0.84. 
4 For more details, see e.g., Petajisto (2011) and S&P’s documentation on their index methodology. 

Electronic copy available at: https://ssrn.com/abstract=4294297



8 
 

index.  According to S&P documentation, “Stocks are added to make the index representative of the U.S. 

economy and is not related to firm fundamentals (S&P, 2017).”  

We obtain data on S&P 500 additions and deletions between 1980 and 2020 from Siblis Research.  

For each index change, Siblis provides the date the change was announced (announcement date) as well as the 

date the change was implemented (effective date).  If the index changes occur on a weekend or trading 

holiday, we mark the next trading day in CRSP as the announcement or effective date.  We merge these 

events to CRSP on date and ticker, and hand match cases on names when either (1) there are multiple CRSP 

permnos associated with that ticker or (2) there are no CRSP permnos associated with that ticker.  Using this 

method, we can match 732 of the 736 additions and 726 of the 731 deletions between Siblis and CRSP for 

the years 1990 to 2020.  Before 1990, Siblis does not provide information on announcement dates, so for the 

pre-1990 additions and deletions, we use data from Barberis et al. (2005).  

 For purposes of measuring returns, the full sample we just described is sufficient. But, to have a 

consistent sample to perform all of our analysis, we remove observations which cannot be matched to the 

Thompson S12 mutual fund holding data on CUSIP, either the quarter before or the quarter after the index 

change.  We also exclude cases where the firm was either listed, acquired, delisted for reasons other than an 

acquisition or was an acquirer (i.e., had an ACPERM in CRSP) within 100 days of the index change.5  These 

filters exclude e.g., spin-offs, where a security can be added to the index and then quickly removed. Appendix 

Table A1 describes in more detail the original sample as well as the final sample size after applying all our 

filters. Appendix Table A2 verifies that our conclusions about absolute addition and deletion returns 

declining are not sensitive to sample selection criteria.  

1.2 Identifying S&P 500 index trackers 

 
5 We exclude cases where the firm is an acquirer because in the case of a stock merger, there could be significant effects 
on the number of shares outstanding, which could contaminate our estimates of mechanical buying and selling by S&P 
500 index-tracking funds.   

Electronic copy available at: https://ssrn.com/abstract=4294297



9 
 

To quantify the amount of money tracking the S&P 500 index, we leverage the Thompson S12 data on 

the quarterly holdings of mutual funds and ETFs.  Our goal is to identify funds that tend to buy additions, or 

sell deletions, around the time of S&P 500 index change.  To this end, for each fund, we count the number of 

times an added stock is not held by the fund the quarter before the addition, but held by that fund the quarter 

after addition.  Similarly, we count the number of times a deleted stock is held by the fund the quarter before 

the addition, and the stock is not held by that fund the quarter after the addition.  We divide the sum of these 

counts by the total number of additions and deletions each year to compute the fraction of index-tracking 

trades made by each fund. We classify funds as S&P 500 trackers if, on average across all years that they are 

present, they perform at least 50% of index-tracking trades each year. According to this methodology, the 

largest S&P 500 trackers in 2020 were the Vanguard 500 Investor Shares (VFINX) and Admiral Shares 

(VFIAX), the SPDR S&P 500 ETF Trust (SPY), the Fidelity 500 Index Fund (FXAIX) and the iShares Core 

S&P 500 ETF (IVV).6  To allay concerns of overreaching with our classification of S&P 500 funds, in 

Appendix Figure A1 we show that we obtain a slightly larger estimate for the size of the S&P 500 tracking 

industry when we identify funds based on their objective codes and names, instead of changes in holdings.  

Having identified the S&P 500 tracking funds, we measure net buying and selling by these funds around 

index changes by adding up the shares held by all trackers the quarter before and after the index change.  

Then, we define net trading by trackers as: 𝑁𝑒𝑡 𝑇𝑟𝑎𝑑𝑖𝑛𝑔 , = 100 × (Δ𝑆ℎ𝑎𝑟𝑒𝑠 ℎ𝑒𝑙𝑑 ,( , ))/

𝑆ℎ𝑎𝑟𝑒𝑠 𝑜𝑢𝑡𝑠𝑡𝑎𝑛𝑑𝑖𝑛𝑔 ,  , where both shares held, and change in shares outstanding are split-adjusted 

using the CRSP cumulative factor to adjust shares outstanding.  

Figure 1 shows the average net trading by trackers across additions and deletions each year.  Consistent 

with the aggregate rise of passive ownership, in the early 1990s, net buying by trackers of additions was close 

 
6 To identify funds based on objective codes we use CRSP objective codes SP and SPSP.  To identify funds based on 
names we follow Appel et. al. (2016) and use variants of “S&P 500”, “S and P 500” and “SP 500”.  We prefer our 
method of identifying S&P 500 trackers based on changes in holdings to this alternative method because before 1999, 
the CRSP objective codes (and more broadly, the flag for index funds) are sparsely populated. 

Electronic copy available at: https://ssrn.com/abstract=4294297

To quantify the amount of money tracking the S&P 500 index, we leverage the Thompson $12 data on
the quarterly holdings of mutual funds and ETFs. Our goal is to identify funds that tend to buy additions, or
sell deletions, around the time of S&P 500 index change. To this end, for each fund, we count the number of
times an added stock is not held by the fund the quarter before the addition, but held by that fund the quarter
after addition. Similarly, we count the number of times a deleted stock is held by the fund the quarter before
the addition, and the stock is not held by that fund the quarter after the addition. We divide the sum of these
counts by the total number of additions and deletions each year to compute the fraction of index-tracking
trades made by each fund. We classify funds as S&P 500 trackers if, on average across all years that they are
present, they perform at least 50% of index-tracking trades each year. According to this methodology, the
largest S&P 500 trackers in 2020 were the Vanguard 500 Investor Shares (VFINX) and Admiral Shares
(VFIAX), the SPDR S&P 500 ETF Trust (SPY), the Fidelity 500 Index Fund (FXAIX) and the iShares Core
S&P 500 ETF (IVV).° To allay concerns of overreaching with our classification of S&P 500 funds, in
Appendix Figure Al we show that we obtain a slightly larger estimate for the size of the S&P 500 tracking

industry when we identify funds based on their objective codes and names, instead of changes in holdings.

Having identified the S&P 500 tracking funds, we measure net buying and selling by these funds around
index changes by adding up the shares held by all trackers the quarter before and after the index change.
Then, we define net trading by trackers as: Net Trading;, = 100 x (AShares heldj(¢+1,-1))/

Shares outstanding; +41 , where both shares held, and change in shares outstanding are split-adjusted

using the CRSP cumulative factor to adjust shares outstanding.

Figure 1 shows the average net trading by trackers across additions and deletions each year. Consistent

with the aggregate rise of passive ownership, in the early 1990s, net buying by trackers of additions was close

6 To identify funds based on objective codes we use CRSP objective codes SP and SPSP. To identify funds based on
names we follow Appel et. al. (2016) and use variants of “S&P 500”, “S and P 500” and “SP 500”. We prefer our
method of identifying S&P 500 trackers based on changes in holdings to this alternative method because before 1999,
the CRSP objective codes (and more broadly, the flag for index funds) are sparsely populated.




10 
 

to 0% of shares outstanding, while now it is over 6%.  This pattern is mirrored for deletions, going from 

approximately 0% to almost 8% of shares outstanding.  

We believe our estimate of net trading by index trackers is a lower bound for several reasons.  First, our 

estimates are based only on S12 data i.e., mutual funds and ETFs.  There are surely institutions of other types, 

such as pension funds and endowments, with assets that directly replicate the index, and which are not 

included in our calculations. In fact, as argued by Chinco and Sammon (2022), the direct replication industry 

(i.e., investors who internally replicate indices rather than buy index funds) may be larger than the AUM of 

explicitly passive funds. Another reason our estimates may be too small is that there are shadow indexers who 

closely track the S&P 500, but not often enough to be classified as index trackers by our method (Mauboussin 

et al., 2017).  Sell-side research estimates the size of S&P 500 index tracking industry in 2022 to be 

approximately 13%, about fifty percent higher than our estimate.7   

1.3. Addition and deletion returns 

Figure 2 presents statistics on average returns for S&P 500 index additions and deletions in 10-year 

rolling windows. Table 1 presents statistics by year. We define the cumulative abnormal return as:  

𝐶𝐴𝑅 = 𝑅 − 𝑅 &  ,      (2) 

For announcement returns, R is measured as the cumulative return between the trading day before the 

announcement and the trading day after the announcement. Effective date abnormal returns are also defined 

according to (2) as the cumulative abnormal return between the day before the implementation of the index 

change and the trading day after the change. For additions, the average period between the announcement 

and the effective date is 4.8 days; for deletions it is 5.8 days.  Our main interest is the 𝐶𝐴𝑅 , defined as the 

cumulative abnormal return from the last trading day before the announcement to the first trading day after 

 
7 Authors’ calculation obtained by dividing predicted net purchases by market capitalization for index additions in the 
2022 UBS S&P 500 report.  We obtain estimates of a similar magnitude to the UBS estimate when we apply our 
methodology but use a more generous classification of index trackers. Specifically, if we identify tracking trades as funds 
increasing the number of shares they hold of index additions, or decreasing the numbers of shares they hold of index 
deletions, instead of requiring holding no shares before or after the index change. 

Electronic copy available at: https://ssrn.com/abstract=4294297

to 0% of shares outstanding, while now it is over 6%. This pattern is mirrored for deletions, going from

approximately 0% to almost 8% of shares outstanding.

We believe our estimate of net trading by index trackers is a lower bound for several reasons. First, our
estimates are based only on S12 data Le., mutual funds and ETFs. There are surely institutions of other types,
such as pension funds and endowments, with assets that directly replicate the index, and which are not
included in our calculations. In fact, as argued by Chinco and Sammon (2022), the direct replication industry
(Le., investors who internally replicate indices rather than buy index funds) may be larger than the AUM of
explicitly passive funds. Another reason our estimates may be too small is that there are shadow indexers who
closely track the S&P 500, but not often enough to be classified as index trackers by our method (Mauboussin
et al., 2017). Sell-side research estimates the size of S&P 500 index tracking industry in 2022 to be

approximately 13%, about fifty percent higher than our estimate.”

1.3. Addition and deletion returns

Figure 2 presents statistics on average returns for S&P 500 index additions and deletions in 10-year

rolling windows. Table 1 presents statistics by year. We define the cumulative abnormal return as:

CARit = Rit — Rsap 500,t (2)

Por announcement returns, R is measured as the cumulative return between the trading day before the
announcement and the trading day after the announcement. Effective date abnormal returns are also defined
according to (2) as the cumulative abnormal return between the day before the implementation of the index
change and the trading day after the change. For additions, the average period between the announcement
and the effective date is 4.8 days; for deletions it is 5.8 days. Our main interest is the CAR, defined as the

cumulative abnormal return from the last trading day before the announcement to the first trading day after

7 Authors’ calculation obtained by dividing predicted net purchases by market capitalization for index additions in the
2022 UBS S&P 500 report. We obtain estimates of a similar magnitude to the UBS estimate when we apply our
methodology but use a more generous classification of index trackers. Specifically, if we identify tracking trades as funds
increasing the number of shares they hold of index additions, or decreasing the numbers of shares they hold of index
deletions, instead of requiring holding no shares before or after the index change.

10




11 
 

the implementation. In principle, the 𝐶𝐴𝑅  captures the price impact resulting from the market absorbing 

net demand from index traders. For most of our sample, index changes are pre-announced with much of the 

return (to the extent that there is a return) occurring on announcement. In the early part of our sample, 

however, index changes are not pre-announced. 

Figure 2 plots the average index inclusion and deletion effect in 10-year rolling windows, while Table 

1 reports the year-by-year averages. For additions, the index inclusion effect was 3.4% in the early 1980s, 

increasing to 7.4% by the 1990s.  This is where the effect peaked, as it declined to 5.2% by the 2000s before 

declining to a statistically insignificant 1.0% in the 2010s.  Our results mirror those in Bennett et al. (2020), 

who show a decline in the index addition effect between 1997 and 2017. We find the deletion effect has 

followed a similar trend toward zero, albeit in a less smooth way.  In the 1980s, firms removed from the S&P 

500 had abnormal returns of -4.6%, while in the 90s, they had returns of -16.1%.  The deletion effect fell in 

magnitude to -12.4% in the 2000s and disappeared in the 2010s, with an average abnormal return of -0.6%.  

One data point that stands out in Figure 2 is the increase in the inclusion effect in 2020, which drove 

the uptick in the overall inclusion effect in the late 2010s and 2020.  This is due to Tesla being added to the 

index in November 2020, which, as a fraction of the S&P 500’s total market capitalization, was the largest 

addition of all time (Arnott et al., 2021).  Excluding Tesla, the average inclusion effect in 2020 was -3 basis 

points.  In Section 2.1, we examine whether characteristics e.g., a firm’s size relative to the total index 

capitalization can explain cross-sectional and time series variation in the index inclusion effect. 

Table 1 also breaks the total index inclusion and deletion effect into the announcement return and 

the implementation return.8 The 2nd column shows that, for additions, the announcement return has been 

declining over time.  In the 1980s, it was 3.4%, rising to 4.9% in the 1990s and then falling to 4.1% by the 

2000s and to 1.0% by the late 2010s.  The 3rd column shows that this pattern is mirrored for effective day 

 
8 Note that in this table, the announcement return, and implementation return do not necessarily add up to the total 
return, as there are typically over 6 days between the announcement and implementation i.e., not all these days are 
included in the t-1 to t+1 window around each event. 

Electronic copy available at: https://ssrn.com/abstract=4294297



12 
 

returns. Specifically, in the 1980s, the effective date return was around 3.4%, increasing to 3.6% in the 1990s 

and then declining to 1.4% by the 2000s and close to zero by the 2010s. The 2nd to last row reports the 

difference in average returns between the 2000-2009 and 2010-2020 periods. Across the total, announcement, 

and implementation returns, this difference is highly statistically significant. 

Columns 5-8 of Table 1 replicate Columns 1-4, but for firms dropped from the S&P 500.  Like the 

results for additions, the implementation and announcement returns became indistinguishable from zero by 

the 2010s.  Also like the additions, this difference is strongly statistically significant.  

At this point, the puzzle is clear: returns to index changes grew in the 1990s in a manner that is 

consistent with the growing importance of index funds, but then declined slightly in the 2000s and 

disappeared, on average, in the 2010s, despite a growing index fund industry. Another potentially more direct 

way to illustrate the puzzle is to compare, event-by-event, the return to the size of assets tracking the index. 

We show this in Figure 3, which plots the index inclusion return against net purchases by index trackers, 

defined as the total change in split-adjusted shares held by index trackers between the quarter before and the 

quarter after the index change, divided by split-adjusted shares outstanding (multiplied by 100). There is no 

apparent relationship between net purchases and the index inclusion effect, for either additions or deletions. 

Specifically, for additions, a regression of inclusion returns on net purchases has a negative slope and an R-

squared of 2%, while for deletions the same regression has a negative slope and an R-squared of 6%. Not 

only has the index effect disappeared, but there does not appear to be a stable relationship between net 

purchases and returns, as hypothesized by Equation (1).   

 

2. Explanations  

In this section we explore four explanations for the declining index effect in the face of increased 

index tracking.   

Electronic copy available at: https://ssrn.com/abstract=4294297



13 
 

2.1 Explanation 1: Changing composition of additions and deletions 

We have shown that the average returns of S&P 500 additions have been declining over time.  A 

natural first hypothesis is that this trend is driven by a change in the composition of added and deleted firms, 

rather than a decline in the nature of the index inclusion effect.  For example, larger firms typically experience 

larger inclusion returns, a phenomenon perhaps driven by benchmarked investors being more likely to buy 

additions that form a large share of the index, as doing so helps them avoid tracking error (this should not 

apply to index trackers or ETFs, which should aim to track the index perfectly).  As a specific application of 

this, Tesla was the largest ever firm added to the S&P 500 index, relative to the S&P 500’s total market 

capitalization (over 2%).  And, as mentioned above, in 2020 Tesla drove a positive overall average addition 

effect, experiencing a cumulative abnormal announcement return of 5.2% and implementation return of 

4.5%.  

Another potential change in composition is related to the notion of “arbitrage risk” associated with 

added and deleted stocks, as discussed by Wurgler and Zhuravskaya (2002). They show that stocks with closer 

substitutes, and thus lower arbitrage risk, should experience lower price pressure. This is closely related to the 

idea that price impact should be correlated with fundamental volatility (Kyle, 1985, Chacko et al., 2008). 

There have been changes in arbitrage risk over time, such as the addition of high-risk internet stocks in the 

late 1990s.  

A third potential change in composition has to do with the amount of investor attention associated 

with index changes. In the past, index inclusions may have been a large shock to institutional investor 

attention, leading to increases in sell-side analyst coverage.  In more recent years, however, with greater 

average information production, the effect of index changes on attention may be relatively less important.  

To quantify the effect of firm characteristics on the inclusion and deletion effects, we run the 

following regression separately for additions and deletions:  

𝐶𝐴𝑅 =  𝑏 𝑡𝑢𝑟𝑛 , + 𝑏  𝑠𝑖𝑧𝑒 , + 𝑏 𝑊𝑍 , + 𝑏 𝑐𝑜𝑣𝑒𝑟 , + ∑ 𝛾 1  + 𝑒  (3) 

Electronic copy available at: https://ssrn.com/abstract=4294297

2.1 Explanation 1: Changing composition of additions and deletions

We have shown that the average returns of S&P 500 additions have been declining over time. A
natural first hypothesis is that this trend is driven by a change in the composition of added and deleted firms,
rather than a decline in the nature of the index inclusion effect. For example, larger firms typically experience
larger inclusion returns, a phenomenon perhaps driven by benchmarked investors being more likely to buy
additions that form a large share of the index, as doing so helps them avoid tracking error (this should not
apply to index trackers or ETFs, which should aim to track the index perfectly). As a specific application of
this, Tesla was the largest ever firm added to the S&P 500 index, relative to the S&P 500’s total market
capitalization (over 2%). And, as mentioned above, in 2020 Tesla drove a positive overall average addition
effect, experiencing a cumulative abnormal announcement return of 5.2% and implementation return of

4.5%.

Another potential change in composition is related to the notion of “arbitrage risk” associated with
added and deleted stocks, as discussed by Wurgler and Zhuravskaya (2002). They show that stocks with closer
substitutes, and thus lower arbitrage risk, should experience lower price pressure. This is closely related to the
idea that price impact should be correlated with fundamental volatility (Kyle, 1985, Chacko et al., 2008).
There have been changes in arbitrage risk over time, such as the addition of high-risk internet stocks in the

late 1990s.

A third potential change in composition has to do with the amount of investor attention associated
with index changes. In the past, index inclusions may have been a large shock to institutional investor
attention, leading to increases in sell-side analyst coverage. In more recent years, however, with greater

average information production, the effect of index changes on attention may be relatively less important.

To quantify the effect of firm characteristics on the inclusion and deletion effects, we run the

following regression separately for additions and deletions:

CAR: = byturnjy_1 + bz size; ¢—1 + bsWZj4-1 + bacover;s—1 + Lka1 Velera=k + eit (3)

13




14 
 

where 𝐶𝐴𝑅  is the cumulative abnormal return from the day before the announcement to the day after the 

implementation.  𝑡𝑢𝑟𝑛 ,  is the abnormal turnover in stock i over the month before the index changes.  

This is defined as turnover (volume/shares outstanding) minus the value-weighted average turnover across all 

ordinary common shares (share codes 10 and 11) traded on major exchanges (exchange codes 1, 2 and 3) in 

CRSP over the same month. 𝑠𝑖𝑧𝑒 ,  is the firm’s market capitalization on the last day before the 

announcement of the index change divided by the total market capitalization of the S&P 500 on the same 

day.  𝑊𝑍 ,  is the Wurgler and Zhuravskaya (2002) measure of arbitrage risk, computed as the variance of 

CAPM regression residuals over the year before the announcement of index addition or deletion.  𝐶𝑜𝑣𝑒𝑟 ,  

is the number of analysts covering the stock in the last earnings announcement before the index change, 

based on data from IBES.  1  are dummy variables for 10-year periods e.g., 1980-1989. We demean 

𝑡𝑢𝑟𝑛 , , 𝑠𝑖𝑧𝑒 , ,  𝑊𝑍 , and 𝑐𝑜𝑣𝑒𝑟 , . Note that because we include separate dummies for each era, 

there is no constant term in the regression. 

We start by running the regression in Equation 3 without controls for past turnover, size, arbitrage 

risk, and analyst coverage. This recovers average returns for each of the decades. Results are shown for 

additions and deletions in Columns 1 and 3, respectively, of Table 2. Note that the averages differ slightly 

from those shown in Table 1 because we lose a handful of observations where we do not have information 

on all the characteristics.  In the last 3 rows, we compare the coefficients between the various decades.  

Column 2 of Table 2 shows the full regression results, including controls.  The first quantities of 

interest are the 𝛾  i.e., the residual average index inclusion effect not explained by the past intensity of trading 

volume or relative firm size.  For the 1980s, 1990s, 2000s, and 2010s, these coefficients are positive, and 

statistically significant. The coefficient for the 1990s stays essentially unchanged at 7.4 when including all the 

firm-level characteristics, suggesting that the index inclusion effect in the 1990s was not entirely explained by 

these firm-level characteristics.  Further, consistent with our previous results, these coefficients shrink from 

the 1990s to the 2000s, and shrink again from the 2000s to the 2010s.  Comparing the difference between 

e.g., the 1990s and 2010s, we can see that the decline in the index effect is slightly smaller (in absolute terms) 

Electronic copy available at: https://ssrn.com/abstract=4294297

where CARj; is the cumulative abnormal return from the day before the announcement to the day after the
implementation. turn; —4 is the abnormal turnover in stock 7 over the month before the index changes.
This is defined as turnover (volume/shares outstanding) minus the value-weighted average turnover across all
ordinary common shares (share codes 10 and 11) traded on major exchanges (exchange codes 1, 2 and 3) in
CRSP over the same month. size; ;_4 is the firm’s market capitalization on the last day before the
announcement of the index change divided by the total market capitalization of the S&P 500 on the same
day. WZ; +-1 is the Wurgler and Zhuravskaya (2002) measure of arbitrage risk, computed as the variance of
CAPM regression residuals over the year before the announcement of index addition or deletion. Cover; p—4
is the number of analysts covering the stock in the last earnings announcement before the index change,
based on data from IBES. leygax ate dummy variables for 10-year periods e.g., 1980-1989. We demean
turn; t—1, SIZ€; 4-1, WZ; 4-1and cover; ,_,. Note that because we include separate dummies for each era,

there is no constant term in the regression.

We start by running the regression in Equation 3 without controls for past turnover, size, arbitrage
risk, and analyst coverage. This recovers average returns for each of the decades. Results are shown for
additions and deletions in Columns 1 and 3, respectively, of Table 2. Note that the averages differ slightly
from those shown in Table 1 because we lose a handful of observations where we do not have information

on all the characteristics. In the last 3 rows, we compare the coefficients between the various decades.

Column 2 of Table 2 shows the full regression results, including controls. The first quantities of
interest are the Vx Le., the residual average index inclusion effect not explained by the past intensity of trading
volume or relative firm size. For the 1980s, 1990s, 2000s, and 2010s, these coefficients are positive, and
statistically significant. The coefficient for the 1990s stays essentially unchanged at 7.4 when including all the
firm-level characteristics, suggesting that the index inclusion effect in the 1990s was not entirely explained by
these firm-level characteristics. Further, consistent with our previous results, these coefficients shrink from
the 1990s to the 2000s, and shrink again from the 2000s to the 2010s. Comparing the difference between

e.g., the 1990s and 2010s, we can see that the decline in the index effect is slightly smaller (in absolute terms)

14




15 
 

once we condition on the changing characteristics (6.6% vs. 5.6%).  The difference, however, is still 

economically large, and statistically significant, suggesting that changing characteristics cannot explain our 

results.  

Next, we turn to the role of the firm characteristics themselves.  The logic of including turnover in 

the regression is that more liquid firms (i.e., firms with more past trading volume) would potentially have 

relatively smaller index inclusion effects, because the demand shock upon inclusion is a smaller fraction of 

average past volume.  The first row shows that the effect of past turnover for additions has the expected sign 

i.e., the estimated coefficient is negative, but is statistically insignificant. The second row shows that the size 

of the firm being added to the index matters, and the magnitude is economically large.  Specifically, being 1% 

larger relative to total index capitalization would imply an 18% larger inclusion effect.  Admittedly, an 

addition of this size is rare, because the average addition is 9 basis points of total index capitalization while 

the average deletion is 3 basis points of total index capitalization.  Further, additions shrank in relative size 

between 1980 and 2019, going from an average of 12 basis points to 7 basis points of total index 

capitalization.  But our regression estimates imply this would only explain roughly 50 basis points of the 

decline in the index inclusion effect.  

The third characteristic is arbitrage risk, WZ,  which is related to stock-level volatility. Consistent 

with theory and Wurgler and Zhuravskaya (2002), WZ attracts a positive and significant coefficient.  Finally, 

we find that the effect of analyst coverage is negative and insignificant, suggesting prior sell-side attention is 

not an important driver of index addition effects. 

Column 4 of Table 2 replicates column 2 for S&P 500 index deletions.  The coefficients are negative 

each decade and statistically significant in the 1990s and 2000s.  The difference between the 1990s and 2010s 

is virtually identical once we control for characteristics (14.6% vs. 14.6%), suggesting that changing 

characteristics of deletions cannot explain the decline in the index removal effect.  Turning to the 

characteristics, turnover surprisingly attracts a negative coefficient, which is not the expected sign, however it 

is not statistically significant. Size is marginally significant and positive.  A possible explanation for this is that 

Electronic copy available at: https://ssrn.com/abstract=4294297



16 
 

large firm deletions are likely to migrate to the S&P MidCap index, which we discuss in the next subsection.  

Finally, we find that arbitrage risk and analyst coverage have the expected sign, but are insignificant. 

The bottom line from Table 2 is that the decline in index effects is not explained by a simple shift in 

the composition or characteristics of the firms being added or deleted from the index. While not the focus of 

our paper, composition effects are important for a handful of years, such as the anomalous large average 

return in 2020 driven by the Tesla inclusion. 

2.2  Explanation 2: Index Migrations  

A second explanation is that we have mismeasured the net demand D (in Eq. 1), and that properly 

measured demand for additions has fallen, with similar results for deletions. A notable type of index change 

for which this holds are so-called index “migrations”. An index change is a migration when the stock moves 

from the S&P MidCap index to the S&P 500, or vice versa.  An example of this would be Targa Resources, 

which was dropped from the MidCap and added to the S&P 500 on October 6, 2022.  This differs from 

direct additions, where a firm is added to the S&P 500 from outside the MidCap or SmallCap universe.  An 

example of this is PG&E, which was directly added to the S&P 500 on October 3, 2022. 

When a stock migrates from the S&P MidCap to the S&P 500, MidCap-tracking funds sell, and 500-

tracking funds buy. Further, over the last 30 years, the passive ownership of mid-cap stocks (i.e., the fraction 

of these stocks’ shares outstanding) has grown dramatically.  As mid-cap focused funds have grown, we 

would expect this to reduce the price impact of a firm being added to the S&P 500.  We hypothesize that 

migrations should have smaller index inclusion effects than direct additions, and that the difference between 

migrations and non-migrations should be increasing over time.9 

To quantify differences between migrations and direct additions, we start by obtaining data on S&P 

MidCap index changes from Siblis Research.  Unlike our dataset on S&P 500 index changes, which starts in 

1980, the MidCap changes dataset starts in 1995.  We follow a similar procedure to the one described in 

 
9 A similar phenomenon is documented by Burnham et al. (2018) in the context of country reclassifications in the MSCI 
index. 

Electronic copy available at: https://ssrn.com/abstract=4294297



17 
 

Section 1.1 to match these observations to CRSP. We define a migration as any stock which is added to 

(dropped from) the S&P 500 and dropped from (added to) the S&P MidCap within 5 calendar days. Figure 4 

shows that migrations have become a steady share of additions and especially deletions.  In the 1990s, 

migrations accounted for about 69% of additions and less than 10% of deletions.  In recent years, they both 

make up over 70% of index changes.  

To fully understand the impact of migrations, we need to estimate capital linked to the S&P MidCap 

index. This is more challenging than identifying funds tracking the S&P 500 because the largest S&P MidCap 

funds infrequently report their holdings for much of the sample period. Instead, each year, we identify S&P 

MidCap 400 index funds based on names and correlations.  To identify funds based on names, we require 

that the fund name contains either variants of “S&P”, “SPDR” or “S and P” as well as variants of “400” or 

“MidCap”.  To identify funds based on correlations, we first restrict to the universe of mid-cap focused 

equity funds (those with either CRSP objective code “EDCM” or a Lipper objective code that starts with 

MC) that do not include variants of “Vanguard” or “Russell” in the name.  Among these funds, we classify 

them as S&P MidCap 400 trackers if their returns have a correlation of at least 99.5% with the index itself for 

three consecutive years. According to this methodology, the largest S&P MidCap trackers in 2020 were the 

iShares Core S&P Mid-Cap ETF (IJH), the SPDR S&P MidCap 400 ETF (MDY), and the iShares S&P Mid-

Cap 400 Growth and Value ETFs (IJK and IJJ).  

Figure 5 shows the sum of these funds’ AUM, scaled by the total market capitalization of the S&P 

MidCap 400 index. As can be seen, dollars tracking the MidCap have grown substantially over time, reaching 

about 6% of capitalization, slightly more than that of the S&P 500, which we showed earlier. All of this 

suggests that in recent years, stocks that were added to the S&P 500 from the MidCap experienced a 

significantly smaller net buying pressure than firms directly added to the S&P 500 from outside the S&P 1500 

universe. 

We now turn to Table 3, which compares the average returns of migrations and direct additions, by 

decade, for additions and deletions. We find that, in each decade, the returns to migration additions are 

Electronic copy available at: https://ssrn.com/abstract=4294297



18 
 

statistically significantly lower than direct additions, and this gap has been growing over time.  Specifically, for 

direct additions, the index inclusion effect was 10.2% in the late 90s, 8.8% in the 2000s, and 5.4% in the 

2010s.  The large positive return in the last period is driven by Tesla, which, as previously discussed, had a 

massive return of 53.35% between its announcement and effective dates.10 For migrations, the index 

inclusion effect was 6.7% in the late 1990s and 2.7% in the 2000s .  By the 2010s, this effect became negative, 

at -1.8%.  These results are consistent with those in Vijh and Wang (2022), who document smaller absolute 

returns for migrations between the S&P 500 and the S&P MidCap indices. 

Based on the results for migration additions, we would expect that deletions to the MidCap would 

have returns which become less negative over time.  This is true, but the effect is smaller in magnitude than 

for additions.  For direct deletions, the index removal effect was -12.8% in the late 1990s, -15.7% in the 2000s 

and -6.9% in the 2010s. For deletions which are also migrations, the index removal effect was -17 basis points 

in the late 1990s, -4.1% in the 2000s, and 6 basis points in the 2010s.   

We visualize these trends in Figure 6.  The top left panel shows that  direct additions to the S&P 500 

experienced a decline in the index inclusion effect over the past 25 years.  The top right panel shows that, 

consistent with the increased size of the offsetting demand shock due to the rise of MidCap funds, there has 

been an even larger decline in the index inclusion effect for migrations. The bottom left panel shows the 

returns to direct deletions steadily declined in magnitude, while the results in the bottom right panel are 

noisier for migration deletions.   

To sum up, migrations are helpful for understanding some of the decline in the returns associated 

with index changes, especially for additions. We return to the migrations in section 2.4 where we account for 

the effect of migrations in assessing the impact for price elasticity.  

2.3  Explanation 3: Predictability of Index Changes 

 
10 Another way that Tesla’s addition was unusual is that there was a 32-day gap between the announcement of its 
addition and the implementation of the index change, while the typical gap is fewer than 10 days.  So, the total 
cumulative abnormal return to Tesla may also be high because other good news about Tesla was released between the 
announcement date and effective date. 

Electronic copy available at: https://ssrn.com/abstract=4294297



19 
 

As the amount of money tracking various indices has grown, so has the industry of investors trying 

to take advantage of the trades they make. For example, there are widespread press reports about Wall Street 

teams earning large sums betting on index additions and deletions across a variety of indices (see Bloomberg). 

This behavior is not restricted to proprietary trading desks.  In fact, many investment banks (e.g., UBS) 

publish short-lists of stocks they think will be added to various indices for their wealth-management clients.  

If index additions have become more predictable, we should see certain patterns emerge in pre- and 

post- addition returns.  Suppose, to start, that index changes were completely unpredictable.  In this case 

prices should rise around announcement, followed by lower returns over very long horizons. Alternatively, if 

additions become more predictable, we should see an increase in price before announcement, coupled with 

reversion in the long run. It is hard to test this scenario, however, because of the endogeneity of index 

additions. Namely, non-index stocks that increase in value are more likely to be added to the index in the first 

place. 

We start by looking at cumulative abnormal pre-addition returns.  To this end, we calculate the 

cumulative abnormal returns starting 100 trading days before the announcement of the index change to 14 

trading days after the announcement. Figure 7 shows that, over the past 30 years, the total price change over 

this period has been roughly equal (in the 1980s, the total price change was less).  The difference, however, is 

that the price spike on the announcement of the index change has become less sharp over time.  Specifically, 

the cumulative abnormal return up to the day before the announcement was 9.6% in the 1990s, 14.2% in the 

2000s and 18.7% in the 2010s.  Then, the cumulative abnormal return up to 10 days after the announcement 

was 17.2% in the 1990s, 17.4% in the 2000s and 19.8% in the 2010s.    

As we noted, one issue with Figure 7 is that all of this is defined ex-post i.e., we are looking at the 

firms that ended up getting added.  It could be, however, that S&P has become more likely to add firms which 

went up a lot in the pre-announcement period over time.  In short, while the evidence is consistent with 

higher predictability, it is not dispositive, because one could equally interpret this evidence as saying that S&P 

500 has become better at adding the best performing stocks, as we show below. 

Electronic copy available at: https://ssrn.com/abstract=4294297



20 
 

Table 4 summarizes the pre-announcement returns shown in Figure 7. For each era, and separately 

for additions and deletions, we show average abnormal returns for the window beginning nk days before 

announcement and ending one day before the announcement, for n = 10, 20, 50, and 100. For purposes of 

discussion, we focus on the shorter windows because they are less confounded by selection, and because 

front-running activity seems more likely in the immediate window before the event. As can be seen, additions 

had average [-20,-1] abnormal returns of 1.54% in the 1990s compared to 2.35% in  the 2010s. Deletions had 

average [-20,-1] abnormal returns of -5.56% in the 1990s compared to -3.03% in the 2010s. In the immediate 

window preceding the event, then, there is no evidence of changes in front running for additions and the 

evidence goes the other way for deletions. At longer windows, however, the picture is murkier. Additions had 

average [-100,-1] abnormal returns of 9.6%  in the 1990s compared to 18.7% today.  

Another way to test whether additions have become more predictable is to see whether, in fact, we 

can predict which stocks are added to the index. Here we focus on the most salient characteristics of index 

additions, namely that they are large stocks that are not in the index and develop a simple model of S&P’s 

index inclusion rule. To quantify this, each month, we compute the market capitalization rank of all ordinary 

common shares traded on major exchanges outside the index.   

Then, in Figure 8, we plot the average rank of firms that end up getting added, as well as the 25th and 

75th percentile of these ranks.11  While this is an imperfect ranking system12, Figure 8 shows that over time, 

the S&P has moved to picking larger firms, with a smaller interquartile size range. This, however, does not 

mean that it is easy to predict which particular stocks will be added at any given time, as the average rank of 

firms added in the last 10 years is around 50.  Further, as can be seen, the interquartile range can be quite 

large, spanning about 40 ranks, suggesting significant randomness in which firms end up getting added.13  

 
11 In a small subset of years, the mean is above the 75th percentile – these are years where one or two extremely low 
ranked firms were added to the index. 

12 This ranking system does not account for other index criteria such as (a) the float-adjustment made by S&P (b) the 
fact that S&P may add non-ordinary common shares and (c) S&P’s other rules such as profitability, size, liquidity, and 
insider ownership. 
13 One concern with the results in Figure 8 is that the number of publicly listed firms has been declining (Doidge et al., 
2017).  This trend could mechanically increase the rank of added firms if S&P always chose firms in the same part of the 

Electronic copy available at: https://ssrn.com/abstract=4294297



21 
 

 

To sum up, the evidence on front-running is mixed. Index changes are more forecastable in the last 

decade than they were in the past. But we do not observe meaningful changes in returns in the short windows 

leading up to announcement, leading us to be skeptical that index change predictability is entirely responsible 

for the patterns that we see. We return to the predictability section in the next section, when we consider it in 

conjunction with the other explanations. 

2.4 Explanation 4: Higher liquidity  

A fourth class of explanation is that the market is more efficient today at accommodating the 

required changes in ownership associated with index additions and deletions. Or, in the context of Equation 

1, that the multiplier M on demand shocks has declined.  

2.4.1 Preliminary estimates 

To estimate how much M has changed, we take means of Equation 1 by decade, and rewrite it to 

reflect the fact that the average net demand shock D varies for index migrations compared to non-migrations:  

𝐶𝐴𝑅 =  𝑀 × 𝐷 = 𝑀 × (𝑤 ∙ 𝐷 + (1 − 𝑤) ∙ 𝐷 )  (4) 

Our goal is to estimate how much M has declined. This exercise is only possible beginning in 1995, when we 

have information on index migrations. The top 6 rows of Table 5 show these results, separately for index 

additions and deletions. For example, in the 2010-2020 period, the average abnormal return for 153 additions 

was 0.80%. 63% of additions were migrations with an average net demand shock of 0.3% of market cap; the 

remaining 37% of additions experienced a net average demand shock of 5.6%. This yields an estimate of M of 

0.37, or equivalently, a demand elasticity of -2.72. Repeating this exercise era by era, the multiplier M has 

fallen by a factor of roughly 20, from 6.75 in the late 1990s to 0.37 in the last decade. A similar pattern 

 
firm-size distribution.  To address this concern we replicated the results in Figure 8 but used the percentile rank of added 
firms instead of the numerical rank.  We found a similar time-series trend to that in Figure 8, suggesting that the decline 
in the universe of public firms does not explain this result. 

Electronic copy available at: https://ssrn.com/abstract=4294297

To sum up, the evidence on front-running is mixed. Index changes are more forecastable in the last
decade than they were in the past. But we do not observe meaningful changes in returns in the short windows
leading up to announcement, leading us to be skeptical that index change predictability is entirely responsible
for the patterns that we see. We return to the predictability section in the next section, when we consider it in

conjunction with the other explanations.
2.4 Explanation 4: Higher liquidity

A fourth class of explanation is that the market is more efficient today at accommodating the
required changes in ownership associated with index additions and deletions. Or, in the context of Equation

1, that the multiplier Mon demand shocks has declined.
2.4.1 Preliminary estimates

To estimate how much / has changed, we take means of Equation 1 by decade, and rewrite it to

reflect the fact that the average net demand shock D varies for index migrations compared to non-migrations:
CAR = MxD=Mx (w: Duigrations +(1-w): Dyonmigrations) (4)

Our goal is to estimate how much M has declined. This exercise is only possible beginning in 1995, when we
have information on index migrations. The top 6 rows of Table 5 show these results, separately for index
additions and deletions. For example, in the 2010-2020 period, the average abnormal return for 153 additions
was 0.80%. 63% of additions were migrations with an average net demand shock of 0.3% of market cap; the
remaining 37% of additions experienced a net average demand shock of 5.6%. This yields an estimate of M of
0.37, or equivalently, a demand elasticity of -2.72. Repeating this exercise era by era, the multiplier M has

fallen by a factor of roughly 20, from 6.75 in the late 1990s to 0.37 in the last decade. A similar pattern

firm-size distribution. To address this concern we replicated the results in Figure 8 but used the percentile rank of added
firms instead of the numerical rank. We found a similar time-series trend to that in Figure 8, suggesting that the decline
in the universe of public firms does not explain this result.

21




22 
 

appears for index deletions, with M falling from 10.76 in the late 1990s to 0.7 in the last decade. In summary, 

even after accounting for the impact of index migrations, liquidity has increased substantially.   

 As we noted, our estimate of the multiplier can be converted into an elasticity by taking −1/𝑀.  We 

can benchmark this against the large range of different price elasticity estimates from the literature on index 

changes.  For example, Wurgler and Zhuravskaya (2002) provide a summary table of these estimates, which 

range from -1 (Shleifer, 1986) to -37.2. (Kandel et al., 1999). Our estimate is -0.15 to -0.09  in the 1990s, -0.28 

to -0.17 in the 2000s and -2.72 to -1.44 in the 2010s.   

One reason that our calculations of price elasticity are smaller in magnitude than previous estimates 

in the literature (i.e., for a given demand shock, we find higher price impact) is because our estimates for the 

demand shock are smaller in magnitude.  For example, Petajisto (2011) assumes a 10% demand shock, which 

is larger than the average demand shock we measure at the end of our sample.14 

In Table 6, our primary estimates of the multiplier M are based on the event return, computed 

throughout the paper as the abnormal return beginning a day before the announcement of an index change 

and ending the day after implementation. In doing so, it leaves out changes in the index effect that might be 

driven by higher predictability. For this reason, the bottom 6 rows of Table 6 also provide estimates of M 

based on a longer horizon return, computed as the cumulative abnormal return of each addition/deletion, 

beginning 20 days before the announcement and ending one day after implementation. Here, the results are 

similar, albeit slightly smaller, with M falling by a factor of roughly 7 for additions and deletions between the 

1990s and 2010s. In other words, even if we allow for increased predictability of index changes to play a role, 

we cannot escape the conclusion that the multiplier has changed.   

2.4.2 Estimates using controls 

 
14Another paper that computes elasticities from index changes is Chang, et. al. (2014), who obtain an elasticity of -1.5, 
relying on estimates of index linked assets provided by Russell.  We believe it is important to focus on index trackers 
rather than all benchmarked investors, because only index trackers appear to buy additions and sell deletions near the 
change date. We have investigated this issue and provide a fuller discussion in the Internet Appendix. 

Electronic copy available at: https://ssrn.com/abstract=4294297



23 
 

 Our estimates above suggest that M has fallen over time. We now show that we reach the same 

conclusion, even after controlling for the other drivers of returns that we have discussed earlier in the paper. 

To do so, we present regressions of the form: 

𝐶𝐴𝑅 =  𝑏 𝑡𝑢𝑟𝑛 , + 𝑏  𝑠𝑖𝑧𝑒 , + 𝑏 𝑊𝑍 , + 𝑏 𝑐𝑜𝑣𝑒𝑟 , + 𝛾 1 × 𝐷  + 𝑒  

(6) 

where 𝐷  is an estimate of the average net demand shock each decade.  As in Equation 3, we demean 

𝑡𝑢𝑟𝑛 , ,  𝑠𝑖𝑧𝑒 , ,  𝑊𝑍 ,  and 𝑐𝑜𝑣𝑒𝑟 , .  D denotes the average size of the demand shock for that decade, 

just as we had in Table 5, taking into account the blend of index migrations and non-migrations. In this 

specification, the slope coefficients 𝛾 , 𝛾  and 𝛾  can be interpreted as the multiplier M for that decade of 

additions or deletions, after controlling for the other drivers of differences in returns.   

These results are presented in Table 6. We follow a parallel structure to Table 5 and present estimates 

based on both the cumulative abnormal return from the day before the announcement to the day after the 

implementation, as a well from 20 days before the announcement to the day after the implementation.  The 

estimates in column 1 mirror those in Table 5, with a decline of the multiplier from 6.8 to 0.3 (the differences 

are due to a slightly smaller sample that we can match to the characteristic variables).  Column 2 shows that 

adding controls for characteristics shrinks the decline of the multiplier, but it is still significant, going from 6.6 

to 0.7.   

In column 3, we instead swap the left-hand side variable to a longer-window return that starts 20 

days before the announcement, rather than a day before the announcement.  Here, we see the level of the 

multiplier is higher, but it also decreased significantly from 9.1 to 1.3 between the 1990s and the 2010s.  In 

column 4, we again add all our characteristics, and find that the decline of the multiplier M is still significant, 

going from 8.7 to 1.7.  Columns (5)-(8) mirror the analysis from columns 1-4, except they use deletions 

Electronic copy available at: https://ssrn.com/abstract=4294297

Our estimates above suggest that M has fallen over time. We now show that we reach the same
conclusion, even after controlling for the other drivers of returns that we have discussed earlier in the paper.

To do so, we present regressions of the form:

3
CARi_ = byturnj yy + bz size; ¢_1 + b3WZj 4-1 + Dacover;,~_1 + », Vileraj=k X Dera=k + eit
k=1

(6)

where Deyq= x is an estimate of the average me¢ demand shock each decade. As in Equation 3, we demean
turnjr—1, SiZ€;4-1, WZ;,4-1 and cover;,_,. D denotes the average size of the demand shock for that decade,
just as we had in Table 5, taking into account the blend of index migrations and non-migrations. In this
specification, the slope coefficients 1, Y2 and y3 can be interpreted as the multiplier / for that decade of

additions or deletions, after controlling for the other drivers of differences in returns.

These results are presented in Table 6. We follow a parallel structure to Table 5 and present estimates
based on both the cumulative abnormal return from the day before the announcement to the day after the
implementation, as a well from 20 days before the announcement to the day after the implementation. The
estimates in column 1 mirror those in Table 5, with a decline of the multiplier from 6.8 to 0.3 (the differences
are due to a slightly smaller sample that we can match to the characteristic variables). Column 2 shows that
adding controls for characteristics shrinks the decline of the multiplier, but it is still significant, going from 6.6

to 0.7.

In column 3, we instead swap the left-hand side variable to a longer-window return that starts 20
days before the announcement, rather than a day before the announcement. Here, we see the level of the
multiplier is higher, but it also decreased significantly from 9.1 to 1.3 between the 1990s and the 2010s. In
column 4, we again add all our characteristics, and find that the decline of the multiplier M is still significant,

going from 8.7 to 1.7. Columns (5)-(8) mirror the analysis from columns 1-4, except they use deletions

23




24 
 

instead of additions.  Again, we find that the decline in the multiplier is robust to both controlling for 

characteristics and including a larger pre-event return, to account for increased predictability.15 

2.4.3 Why did the market become more efficient at accommodating index changes? 

Our analysis in 2.4.1 and 2.4.2 suggests that, even after controlling for all the potential drivers of the 

index effect (characteristics, migrations, and predictability), the market has become significantly more efficient 

in accommodating index changes. How and why did this happen? While this exercise is more speculative, 

below we explore some of the potential forces associated with these changes, including: (1) overall increases 

in market liquidity and reductions in trading costs in other settings (2) coordination of trading around index 

changes, and (3) accommodation of index changes by active managers (4) accommodation by firms issuing 

stock. We find evidence for all but the last. 

Increases in overall market liquidity 

The US stock market has become more liquid overall since the 1980s and 1990s, in the sense that it 

has gotten cheaper to trade without moving the price. We consider two ways of measuring trading costs. 

Amihud and Mendelson (1986) suggest the bid ask spread as a simple measure of trading costs. To quantify 

this, we use the WRDS intraday indicators suite to obtain measures of the bid-ask spread based on high 

frequency data.  This dataset uses the method in Holden and Jacobsen (2014) to compute the percent 

effective spread.  The percent effective spread is the percent distance away from the midpoint that the (value-

weighted) average trade occurs at each day.  Given that our study spans 1980-2020, we need to leverage both 

the second-based version of TAQ, which runs from 1993-2014, and the millisecond-based version of TAQ, 

which runs from 2003-present. We do not have a good measure of trading costs before 1993. 

 
15 We have experimented with several versions of Table 7, including allowing for a separate estimate of M for migrations 
and non-migrations, as well as allowing the effect of the control variables to vary by decade.  See the Internet Appendix 
for additional details.  

 

Electronic copy available at: https://ssrn.com/abstract=4294297



25 
 

The red line in Figure 9 plots the value-weighted average effective spread for all ordinary common 

shares traded on major exchanges, computed using TAQ data.  Value-weighted average effective spreads have 

experienced a large time-series decline, from 60bp to 6bp.  The decline is similar when examining an equal-

weighted average of the smallest 100 stocks by market capitalization in the S&P 500. Note that the timing of 

the increase in overall market liquidity (which is most pronounced in the 1995-1998 and 2002-2005 periods) 

does not closely match the pattern of changes we observe in the index effect. In other words, while the 

changes in the index effect are occurring over a period of enhanced overall liquidity, this is not the whole 

story. 

The bid-ask spread captures costs associated with small trades near the midpoint.  But the type of 

trades executed on days of index changes likely don’t fit this description: the fraction of shares that need to be 

purchased by index funds are enormous, now making up over 7% of total shares outstanding. For this reason, 

we also examine implementation shortfall collected from Virtu Financial. The implementation shortfall is the 

difference between the arrival price and the execution price for a trade. The implementation shortfall, plotted 

as a blue line in Figure 9, fell significantly less than the average bid-ask spread over our sample. 

Coordination and specialization of trading 

Over the past 15 years, Wall Street trading desks have increased personnel and computing resources 

devoted to index trading. Large players, including UBS and Goldman Sachs have specialized sell-side teams. 

Passive fund managers also employ large teams to study and improve liquidity around rebalancing.  We have 

confirmed this in discussions with several large index investors, including Blackrock, which alone employs a 

team of over 60 people in index research. 

Further, trading around index changes has become increasingly concentrated (Chinco and Sammon, 

2022). We interpret this using the logic of Admati and Pfleiderer (1988): to coordinate on this “sunspot”, 

index providers have moved to a system of disclosing ahead of time which stocks they are going to trade and 

when (Li, 2021).  

Accommodation by other institutions 

Electronic copy available at: https://ssrn.com/abstract=4294297



26 
 

Liquidity around index changes has increased, but who provides it?  Below we show that liquidity is 

provided, on average, by other institutions exiting their positions, rather than accommodation by other parties 

such as retail investors. In other words, while a large, dedicated group of mutual funds and ETFs must buy, on 

average, total institutional ownership changes very little around these events.  To quantify this, we obtain data 

on institutional ownership from Thompson 13F.  Specifically, we examine changes in 13F ownership from 

the quarter before to the quarter after the index change.  These changes are tabulated in Table 7, which 

compares the changes in ownership by S&P 500 trackers to the total change in institutional ownership.  

Despite the rise in the change in tracker ownership, there has been (if anything) a decline in the change in 13F 

ownership.  For example, the table shows that for the average addition in 2020, index trackers buy 7.53% of 

shares outstanding. However, institutional ownership only increases by 0.65% of shares outstanding. This 

suggests that other institutions have stepped up to meet the buying and selling pressure from S&P 500 funds. 

The same pattern appears for deletions: in 2020, index trackers sell an average of 8.45% of shares outstanding 

upon deletion, but institutional ownership overall falls by only 1.42%.  

We also examine the net trading behavior of active and passive mutual funds and ETFs around S&P 

500 index additions and deletions.  We identify passive funds in the S12 data using the methodology in Appel 

et. al. (2016) and define active funds as all remaining funds.  The “Passive” column of Table 7 shows that, on 

average, passive funds buy a smaller percentage of additions’ shares outstanding than S&P 500 trackers.  This 

is consistent with some additions being migrations, where MidCap-tracking passive funds sell shares to S&P 

500-tracking funds.  The “Active” column shows that active mutual funds are not the group providing 

liquidity, as their average net demand is roughly zero.  This implies that non-S12 filing institutions (e.g., hedge 

funds, pension fund, endowments) are the primary liquidity providers around addition events.   

Accommodation by corporate share issuance? 

A last hypothesis is that index additions are partially accommodated by firms that issue shares into 

the rebalance. A few high-profile examples of this, such as CoStar upon its addition to the S&P 500 in 2018, 

where it concurrently issued $750 million of new equity, suggest the possibility of this phenomenon being 

Electronic copy available at: https://ssrn.com/abstract=4294297



27 
 

important. We have investigated changes in split-adjusted shares outstanding around all S&P 500 index 

additions and deletions beginning in 1980. The fraction of firms that have issued stock near S&P 500 

membership changes increased somewhat in the 2000s, but then declined in the 2010s. Large stock issuances 

around index changes are rare. Corporate issuance is not a major force driving the improvement in liquidity 

around index rebalances.  

 

3. Discussion and extension to other indices  

Our findings suggest the following sequence of events. In the 1980s, index changes were 

unanticipated, but index funds were small, so addition and deletions effects were relatively modest.  As index 

tracking grew larger throughout the 1990s, the mispricing deepened and turned into an opportunity. As a 

result, the market adjusted to take advantage of this opportunity, in part by better anticipating inclusions, and 

in part by creating arrangements where other institutions stood ready to sell to indexers upon inclusion. At 

the same time, the S&P 500 grew to rely more on index migrations (especially for deletions), which helped to 

reduce overall price impact, and benefitted from increasing assets tracking midcap indices. Together, these 

forces worked to eliminate the index addition and deletion anomalies on average, despite demand shocks that 

continued to grow in magnitude over the 2000s and 2010s.  

Our interpretation raises the question: do we observe similar reductions in addition and deletion 

returns associated with other indices, such as the Russell 2000, or the Nasdaq 100? A complete analysis of this 

question would have to acknowledge the myriad of differences between these indices, including how index 

changes are announced, how predictable they are, and estimates of how much capital has tracked these 

indices over time. Still, it is useful to understand whether the broad patterns we have described for the S&P 

500 – and, as originally highlighted by Bennett et al. (2020) – exist in other settings.  

Below we expand our event return analysis to several other families of indices, choosing to focus on 

indices for which we can track changes for a 20-year or more period. To preview our results, we show that, 

like our findings for the S&P 500, such effects grew from the 1990s to the 2000s, and diminished thereafter 

Electronic copy available at: https://ssrn.com/abstract=4294297



28 
 

for the Russell 1000 and 2000, as well as the S&P MidCap, SmallCap and Nasdaq 100.  We would like to 

emphasize, however, that there are differences in magnitudes across the different indices and few of these 

differences are individually statistically significant.  

3.1 Russell indices 

Each May, Russell ranks firms according to their float-adjusted market capitalization ahead of their 

annual index reconstitution, which typically occurs on the fourth Friday of June.  Before 2007, the 1000 

largest firms would comprise the Russell 1000, while the next 2000 largest firms would make up the Russell 

2000.  After 2007, Russell switched to a banding rule, where a firm needed to be more than 2.5% of the total 

Russell 3000E’s capitalization away from the 1000th ranked stock to switch indices (see e.g., Coles et al. (2022) 

for more details). 

For the Russell 1000 and 2000, there are six types of index changes to consider: (1) firms that switch 

from the Russell 1000 to the Russell 2000 (2) firms that switch from the Russell 2000 to the Russell 1000 (3) 

additions to the Russell 2000 from outside the Russell 3000 universe and (4) additions to the Russell 1000 

from outside the Russell 3000 universe (5) deletions from the Russell 2000 to outside the Russell 3000 

universe and (6) deletions from the Russell 1000 to outside the Russell 3000 universe.  We focus on additions 

to the Russell 1000 and 2000 from outside the Russell 3000 universe, as these are most comparable to direct 

additions to the S&P 500. We do not consider deletions from the Russell 3000 to outside the Russell 3000 

universe, as these are often firms that are delisting and therefore do not pass our sample selection criteria.  

We obtain data on Russell index membership from 1990 to 2020 directly from FTSE Russell.  To 

quantify index addition effects, we apply the same methodology we used for S&P 500 index changes with the 

following minor change.  Rather than use the cumulative abnormal return from the day before the 

announcement date to the day after the effective date, we instead use the cumulative market-adjusted return 

from the day before the ranking date in May to the day after the implementation, typically in June (Madhavan, 

2003).  The market adjusted return is defined as the return on the stock minus the return on the CRSP value 

weighted index over the same period.  As with the S&P 500, we exclude firms that were either listed, 

Electronic copy available at: https://ssrn.com/abstract=4294297



29 
 

acquired, delisted for reasons other than an acquisition and firms which were acquirers (i.e., had an ACPERM 

in CRSP) within 100 days of the index change. 

In Table 8, we show that, consistent with Madhvan (2003) and Petajisto (2011), there were positive 

returns associated with direct additions to the Russell 1000 and 2000 indices in the 1990s and large returns 

associated with these index changes in the 2000s.16  Mirroring our results for the S&P 500, these effects grew 

from the 1990s to the 2000s.  This increase may have been driven by an increase in the demand shock 

associated with index addition, as IWB and IWM, the largest Russell 1000 and 2000 ETFs, were launched in 

May 2000.  In terms of magnitudes, for stocks added to the 2000, addition cumulative market-adjusted 

returns increase from an average of 2.2% in the 1990s, to 8.3% in the 2000s and thereafter declined to 3.1% 

in the 2010s.  For stocks added to the 1000, the effect increased from 1% in the 1990s, increasing to 17.0% in 

the 2000s and then decreasing to 8.53% in the 2010s. As the bottom rows of the Table show, however, none 

of these individual declines between the 2000s and 2010s are statistically significant.    

Unlike changes to the S&P 500, changes to the Russell indices are based on mechanical rules and 

may therefore be easier to predict.  This makes the similarity between the time-series trends for the Russell 

1000 and 2000 indices and S&P 500 striking, as changes in predictability are not likely the drivers of a weaker 

Russell index inclusion effect in recent years.  Other factors, such as increased event-specific liquidity arising 

from pre-arranged trades, may be more important in explaining the smaller returns to Russell index additions 

in the 2010s relative to the 2000s. 

3.2 S&P SmallCap and MidCap 

The broader S&P 1500 universe includes all firms in the S&P 500, the S&P MidCap 400 and the S&P 

SmallCap 600.  These indices are not as widely tracked as the S&P 500, but still attract considerable attention 

and have mutual funds and ETFs that track them. The same rules of profitability and liquidity for the S&P 

500 also apply to their other indices, with the main difference being the size threshold for index inclusion. 

 
16 We would like to highlight, however, that the large returns in the 2000s are largely driven by the year 2000 itself, where 
both direct additions to the Russell 1000 and 2000 had returns greater than 30% on average.  

Electronic copy available at: https://ssrn.com/abstract=4294297



30 
 

For the MidCap, there are six types of index changes: (1) additions from outside the S&P universe (2) 

deletions to outside the S&P universe (3) additions from the S&P 500 (4) deletions to the S&P 500 (5) 

additions from the SmallCap (6) deletions to the SmallCap.  For the SmallCap, there are also 6 types of index 

changes: (1) additions from outside the S&P universe (2) deletions to outside the S&P universe (3) additions 

from the S&P 500 (4) deletions to the S&P 500 (5) additions from the MidCap (6) deletions to the MidCap.  

As with the Russell family of indices, we focus on direct additions to the MidCap and SmallCap from outside 

the S&P 1500, as well as direct deletions to outside the S&P 1500 universe.   

We obtain data on MidCap changes between 1995 and 2020 and SmallCap changes from 1997 to 

2020 from Siblis Research.  We apply an almost identical methodology for computing index inclusion and 

deletion effects, with one minor change:  Siblis does not provide announcement dates for the MidCap and 

SmallCap, so we use a window from 10 days before to one day after the index change, as changes are typically 

announced 5 business days in advance. As with the S&P 500, we drop firms that were either listed, acquired, 

delisted for reasons other than an acquisition and firms which were acquirers (i.e., had an ACPERM in CRSP) 

within 100 days of the index change.  Finally, for consistency across benchmarks, and as with the Russell 

index changes, we use cumulative market-adjusted returns. 

Table 8 shows that addition effects for the MidCap increased from the 1990s to the 2000s, and 

decreased thereafter, going from 5.6% in the 1990s, to 8.3% in the 2000s and 5.7% in the 2010s.  Similarly, 

deletion effects grew and then shrunk from -4.5% in the 1990s, to -18.3% in the 2000s and -1.2% in the 

2010s.  The effect of being added to the SmallCap hasn’t changed much over the past 30 years, consistently 

hovering around 6%, but the effect of being deleted from the SmallCap increased and then shrunk, from -

6.7% in the 1990s to -24.6% in the 2000s and finally -12.2% in the 2010s.  As with the changes to the Russell 

indices, except for the increase in deletion returns for the S&P SmallCap index, none of the changes between 

the 2000s and 2010s were statistically significant. 

 

3.3 Nasdaq 100 

Electronic copy available at: https://ssrn.com/abstract=4294297



31 
 

The Nasdaq 100 comprises the 100 largest non-financial companies with a primary listing on the 

Nasdaq exchange, with a few exceptions as stipulated by index rules. We obtain data on Nasdaq 100 index 

changes between 1995 and 2020 from Siblis Research.  As with the S&P 500, we drop firms that were either 

listed, acquired, or delisted for reasons other than an acquisition and firms which were acquirers (i.e., had an 

ACPERM in CRSP) within 100 days of the index change.  Again, as with the MidCap and SmallCap, we do 

not have announcement dates for the Nasdaq 100, but these changes are typically announced 5 business days 

in advance, so we use a window of 10 days before the change to one day after when computing index 

addition and deletion effects.  Finally, as we describe above, we focus on market-adjusted returns for our 

sample of Nasdaq index changes. 

Table 8 shows that there has been a steady decline in the effect of being added to the Nasdaq 100, 

going from 3.9% in the 1990s to 2.6% in the 2000s and 2.0% in the 2010s.  The deletion effects are 

consistently minimal, hovering near zero for the past 30 years.  And once again, none of these declines from 

the 2000s to the 2010s are statistically significant. 

3.4 Summary 

Overall, the results for the Russell 1000 and 2000, S&P MidCap and SmallCap, and Nasdaq 100 paint 

a picture consistent with our main results: index addition and deletion effects grew into the 2000s and shrunk 

thereafter, but the statistical significance is weak. As a final test, in Column 9 of Table 8, we pool together all 

the additions from the Russell 1000, Russell 2000, S&P MidCap, S&P SmallCap and the Nasdaq 100.  These 

results mirror the patterns documented in the individual indices, showing a decline of 4.2% in addition effects 

across all indices, significant at the 5% level.  Finally, in Column 10 of Table 8, we pool together all the 

deletions from the S&P MidCap, S&P SmallCap and the Nasdaq 100.  Here, the deletion effect shrinks by an 

average of 10.3 percent points between the 2000s and 2010s, statistically significant at the 1% level. 

4. Conclusion 

According to efficient markets theory, if a class of investors were to buy or sell a stock for reasons 

unrelated to fundamentals, well-capitalized arbitrageurs should respond aggressively to provide liquidity, 

Electronic copy available at: https://ssrn.com/abstract=4294297



32 
 

limiting the price impact. The well-known index effect, whereby a stock added to an index such as the S&P 

500 goes up in price, is often held up as an example of market inefficiency. The notion of a downward 

sloping demand curve is a key ingredient in most behavioral finance models of the stock market.  

Over the past decade, the well-known index effect for the S&P 500 has disappeared, with the average 

addition or deletion experiencing abnormal returns near zero. In this paper, we consider four explanations for 

why this happened. To sum up, our assessment is that the declining index effect is driven primarily by two 

factors: in the role of migrations from the S&P MidCap Index, as well as an overall increase in the market’s 

ability to provide liquidity to index changes. A third factor, increased predictability of index changes, also 

plays a small role. Our overall conclusion is that, when demand shocks become regular and repeated, 

competitive markets adapt over time to minimize price impact, in the spirit of Lo’s (2004) adaptive markets 

hypothesis. 

While our findings uncover and quantify the forces driving the decline of the index effect, they leave 

open another puzzle. Namely, S&P 500 index changes have been around since 1957, but it took until the 

2010s for the market to evolve to provide meaningful liquidity around index changes and neutralize the price 

impact. Why did it take so long? And why did it take much longer for the market to adapt than in the context 

of other anomalies related to informational efficiency? We expect subsequent work to explore these 

questions.   

Electronic copy available at: https://ssrn.com/abstract=4294297



33 
 

References 

Admati, Anat R., and Paul Pfleiderer, 1988, A Theory of Intraday Patterns: Volume and Price Variability, The 
Review of Financial Studies 1, 3-40.  

Amihud, Yakov, and Haim Mendelson, 1986, Asset Pricing and the Bid-Ask Spread, Journal of Financial 
Economics 17, 223-249. 

Appel, Ian R., Todd A. Gromley, and Donald B. Keim, 2016, Passive Investors, Not Passive Owners, Journal 
of Financial Economics 121, 111-141. 

Arnott, Rob, Vitali Kalesnik, and Lillian Wu, 2021, Revisiting Tesla’s Addition to the S&P 500: What’s the 
Cost, Before and After?, Research Affiliates. 

Bagwell, Laurie S., 1992, Dutch auction repurchases: An analysis of shareholder heterogeneity, The Journal of 
Finance 47, 71-105. 

Barberis, Nicholas, Andrei Shleifer, and Jeffrey Wurgler, 2005, Comovement, Journal of Financial Economics 75, 
283-317. 

Ben-Rephael, Azi, Shmuel Kandel, and Avi Wohl, 2010, The Price Pressure of Aggregate Mutual Fund Flows, 
Journal of Financial Quantitative Analysis 46, 585-603. 

Bennett, Benjamin, Rene Stulz, and Zexi Wang, 2020, Does Joining the S&P 500 Index Hurt Firms? Working 
paper. 

Bond, Philip, and Diego Garcia, 2018, The Equilibrium Consequences of Indexing, Working paper. 

Burnham, Terrence, Harry Gakidis, and Jeffrey Wurgler, 2018, Investing in the Presence of Massive Flows: 
The Case of MSCI Country Reclassifications, Financial Analysts Journal 74, 77-87. 

Buss, Adrian, and Savitar Sundaresan, 2020, More Risk, More Information: How Passive Ownership Can 
Improve Informational Efficiency, Working paper. 

Cai, J., 2007, What’s in the news? Information content of S&P 500 additions, Financial Management 36, 113–
124. 

Chacko, George, Jakub Jurek, and Erik Stafford, 2008, The Price of Immediacy, Journal of Finance 63, 1,253-
1,290. 

Chang, Yen-Cheng, Harrison Hong, and Inessa Liskovich, 2015, Regression Discontinuity and the Price 
Effects of Stock Market Indexing, The Review of Financial Studies 28, 212-246.  

Chinco, Alex, and Marco Sammon, 2022, The Passive-Ownership Share Is Double What You Think It Is, 
Working paper. 

Coles, Jeffrey L., Davidson Heath, and Matthew C. Ringgenberg, 2022, On index investing, Journal of Financial 
Economics 145, 665-683. 

Doidge, Craig, G. Andrew Karolyi, and René M. Stulz, 2017, The U.S. listing gap, Journal of Financial 
Economics 123, 464-487. 

Ernst, Thomas, 2022, Stock-specific price discovery from ETFs, Working paper, University of Maryland, 
College Park. 

Electronic copy available at: https://ssrn.com/abstract=4294297



34 
 

Gabaix, Xavier, and Ralph S. J. Koijen, 2022, In Search of the Origins of Financial Fluctuations: The Inelastic 
Markets Hypothesis, Working paper. 

Garleanu, Nicolae, and Lasse Heje Pedersen, 2018, Efficiently Inefficient Markets for Assets and Asset 
Management, The Journal of Finance 73, 1,663-1,712. 

Greenwood, Robin, 2005, Short- and Long-Term Demand Curves for Stocks: Theory and Evidence on the 
Dynamics of Arbitrage, Journal of Financial Economics 75, 607-649. 

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, The Journal of Finance 41, 815-829. 

Hartzmark, Samuel M. and David H. Solomon, 2022, Reconsidering Returns, The Review of Financial Studies 35, 
343-393. 

Holden, Craig W., and Stacey E. Jacobsen, 2014, Liquidity Measurement Problems in Fast, Competitive 
Markets: Expensive and Cheap Solutions, Journal of Finance 69, 1,747-1,785. 

Kacperczyk, Marcin, Savitar Sundaresan, and Tianyu Wang, 2020, Do Foreign Investors Improve Efficiency?, 
The Review of Financial Studies 34, 1,317-1,367. 

Kandel, Shmuel, Oded Sarig, and Avi Wohl, 1999, The demand for stocks: An analysis of IPO auctions, The 
Review of Financial Studies 12, 227-247. 

Kaul, Aditya, Vikas Mehrotra & Randall Morck. 2000. Demand Curves for Stocks Do Slope Down, Journal of 
Finance 55, 893-912. 

Koijen, Ralph S. J., and Motohiro Yogo, 2019, A Demand System Approach to Asset Pricing, Journal of 
Political Economy 127, 1,475–1,515. 

Kyle, Albert S., 1985, Continuous auctions and insider trading, Econometrica 53, 1315–1335. 

Lee, Jeongmin. 2021, Passive investing and price efficiency, Working paper, Toulouse School of Economics. 

Li, Sida, 2021, Should Passive Investors Actively Manage Their Trades?, Working paper. 

Lo, Andrew W., 2004, The Adaptive Markets Hypothesis, The Journal of Portfolio Management 30, 15-29. 

Loderer, Claudio, John W. Cooney, and Leonard D. Van Drunen, 1991, The price elasticity of demand for 
common stock, The Journal of Finance 46, 621-651. 

Lou, Dong, Hongjun Yan, and Jinfan Zhang, 2013, Anticipated and Repeated Shocks in Liquid Markets, The 
Review of Financial Studies 26, 1,891-1,912. 

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, 351-83. 

Madhavan, Ananth, 2003, The Russell Reconstitution Effect, Financial Analysts Journal, 59, 51-64. 

Madhavan, Ananth, Jason Ribando, and Nogie Udevbulu, 2022, Demystifying Index Rebalancing: An 
Analysis of the Costs of Liquidity Provision, The Journal of Portfolio Management 48, 171-184. 

Malikov, George, 2023, Information, participation, and passive investing, Working paper. 

Mauboussin, Michael J., Dan Callahan, and Darius Majd, 2017, The Incredible Shrinking Universe of Stocks, 
Credit Suisse. 

Electronic copy available at: https://ssrn.com/abstract=4294297



35 
 

Mitchell, Mark, Todd Pulvino, and Erik Stafford, 2005, Price Pressure Around Mergers, The Journal of Finance 
59, 31-63. 

McLean, David, and Jeffrey Pontiff, 2016, Does Academic Publication Destroy Stock Return Predictability?, 
Journal of Finance, 71, 5-31.  

Patel, Nimesh, and Ivo Welch, 2017, Extended Stock Returns in Response to S&P 500 Index Changes, The 
Review of Asset Pricing Studies 7, 172–208. 

Petajisto, Antti, 2011, The index premium and its hidden cost for index funds, Journal of Empirical Finance 18, 
271-288. 

Preston, Hamish, and Aye Soe, 2021, What Happened to the Index Effect? A Look at Three Decades of S&P 
500 Adds and Drops, Research S&P Dow Jones Indices. 

Qin, Nan, and Vijay Singal, 2015, Indexing Stock Price Efficiency, Financial Management 44, 857-904. 

Scholes, Myron, 1972, The Market for Securities: Substitution versus Price Pressure and the Effects of 
Information on Share Prices, The Jou