S&P 500 Affiliation and Stock Price Informativeness

Abstract

When firms are added to a stock index, more information should be discovered, traded on, and incorporated into their stock prices, making them more informative. We test this hypothesis using a large sample of additions to the S&P 500 index. Using two alternative statistical tests, we find that the stocks added experience more random, less predictable return and, thus, appear to be priced more efficiently information-wise. We further find concurrent increases in institutional ownership and investor awareness, which tend to contribute to the higher pricing efficiency, adding to the literature. These findings should be of interest to academics and practitioners.

Keywords:

• S&P 500
• Affiliation
• Informational efficiency
• Price informativeness
• Return predictability
• Institutional ownership
• Investor awareness

JEL classification:

• G14
• G15

Notes

Acknowledgements

We wish to thank an anonymous referee and the editor for helpful comments and suggestions, which have led to substantial improvements in the paper. As usual, we are solely responsible for any errors or omissions remaining.

Notes

1 See Afego (2017) for a comprehensive literature review. A working paper by Kasch and Sarkar (2011), not covered in Afego (2017) but kindly suggested by a referee, shows that there is no permanent S&P500 inclusion effect on market value and co-movements in return, as measured by factor betas (market, size, value, and momentum), using matched control samples. Nonetheless, the working paper does not touch upon the weak-form efficiency effect investigated in the current study. If anything, their results could be related to the semi-strong form efficiency explored in prior studies, in view of the factor betas used.

2 For instance, efficiency should vary with liquidity, considering that higher liquidity tends to decrease first-order autocorrelation in stock return (Campbell et al. 1993). Presumably, changes in information environment should affect informational efficiency.

3 Liu (2009) provides evidence of more press coverage for stocks added to the Nikkei 225. Yet, more press coverage per se may not lead to investment; better investor awareness as a result of more press coverage may. That is why we investigate investor awareness, instead of press coverage per se, in this paper, let alone the absence of reliable data on press coverage for a much larger sample.

4 Fama (1970, 1991) provides a general review of the literature on market efficiency, which necessitates information being incorporated into stock prices fully and promptly. The current study is germane to “fully”.

5 For instance, the Nikkei 225 is price weighted, while the S&P 500 is value weighted. As such, the Nikkei 225 may be subject to more manipulation around the membership changes While there is a strong momentum effect in the U.S. market, there seems to be no such effect in the Japanese market. (p.237, Afego 2017) These differences between indices and markets, among others, might give rise to different effects of index rebalancing, as illustrated by the new findings in the current study.

6 Conceivably, press coverage could increase investors’ awareness. That is, press coverage could arguably be a means to the end, investor awareness. As such, one could easily argue that they would be strongly collinear if used concurrently in a regression. It is preferable to use the end (investor awareness) rather than a possible means (press coverage), for closer relation to trade and for better data.

7 As pointed out by a referee, return predictability may not be synonymous with a measure of weak-form efficiency (randomness) used in this study, i.e., abnormal runs (or trends) in past price data, a reduction in which may “not preclude the ability for fundamental factors to exhibit explanatory power over future firm returns”. Nonetheless, the focus of this study is on the weak-form efficiency in the price data, not the semi-strong form efficiency related to fundamental factors such as earnings.

8 As pointed out by a referee, more institutional ownership and investor awareness for added stocks are “well established in the literature”. But, this is the first time these factors have been brought to bear on weak-form efficiency. Also, the results reported here come from longer windows (e.g., two years around), while many previous studies use much shorter windows, such as overnight or intraday (e.g., Kappou et al. 2010).

9 This section is based on the information posted on the S&P website (www.standardandpoors.com/indices).

10 Likewise, skipping the days between the announcement and actual inclusion leaves the results materially intact.

11 A four-year sample period around the event day gives similar results, while potentially susceptible to more contaminations from other events.

12 The sample size is even smaller (less than 30) in the case of other measures such as institutional ownership, analyst following, and investor awareness, rendering the results even less reliable.

13 Among the 878 added stocks, 615 (263) are listed on the NYSE (Nasdaq) prior to the rebalancing. Simple statistical tests show that those listed on the Nasdaq experience more efficiency gains, suggesting better efficiency of the specialist system adopted at the NYSE, consistent with Kappou et al. (2010). The test results are not reported to save space, since it is not the focus of the current study.

14 We also experiment with the alternative value weighting but find little difference.

15 The variance ratio test is another test of randomness or serial independence, which we do not adopt owing to its asymptotic nature. While an asymptotic test necessitates a long sample period for optimal performance, a typical event study uses a short sample period to minimize potential contaminations. For example, Lo and MacKinlay (1988) employ over 20 years of data to optimize the asymptotic test while an event study typically taps no more than two years around the event to pinpoint the effects. Furthermore, the variance ratio test involves arbitrary choice of increments.

16 This condition is well met in all sample periods used in this article. For example, in each year around the event in a two-year sample period, both numbers exceed 100 in all data series tested.

17 See Mood (1940) for the original distribution theory and Freund (1992) for more accessible details for the runs test.

18 While the abnormal runs are normally distributed, as discussed earlier, it is not clear if the market-adjusted abnormal runs also follow the normal distribution, casting doubt on the validity of inferences drawn from their means. Yet, results from the distribution-free nonparametric test based on medians are free from the issue.

19 While the efficiency hypothesis seems one-sided, a two-sided test is warranted in view of the potential contaminations explored later. Moreover, the sign test is chosen over the Wilcoxon signed rank test, which requires a symmetric distribution, while the (in)efficiency measure is skewed, as detailed later.

20 It is beyond the scope of this study to explore the sources of the skewness or other statistical properties of the data. Regardless, the risk of skewness is invalidation of parametric tests (t or z tests) based on means, which assume symmetry in the normal distribution. That is why in this article we draw inferences mainly from the nonparametric tests based on medians, as is the convention in the literature.

21 As a further robustness check, we also run the runs test on the average return for the sample stocks and find similar results, i.e. the average return becomes more random, with or without market adjustment. Yet, no formal test can be performed on two data points.

22 Between the two alternative measures used by S&P in its decision of inclusion, monthly trading volume and annual turnover of trading value, as reviewed earlier, the latter is clearly contaminated by the price effect, That might be why trading volume is a liquidity measure adopted by many previous studies.

23 Earlier studies (e.g. Beneish and Whaley 1996), and Liu (2006) calculate similar volume ratios. The results are robust to relocating the estimation window of at least 100 trading days within the prior trade year.

24 Hegde and McDermott (2003, Table 5, p. 432) report higher market-adjusted volume for additions to the S&P 500 in two relatively short windows: [−60, −10] vs. [+10, +60], which is arguably consistent with the results reported here, while not directly comparable in view of the different estimation windows.

25 Incidentally, we do not have access to intra-daily data for bid/ask spread, an alternative measure of liquidity used by, e.g. Beneish and Whaley (1996) and Hegde and McDermott (2003).

26 The results reported in Tables 4b do not change materially when raw returns are used in the estimation of the bid/ask spread.

27 Setting the estimation window at (−250, −1) for the first two event windows yields stronger results (i.e. post-evet spread even higher relatively, meaning even lower post-event liquidity), as evidenced by the decreasing spread as the window extends both before and after the event. The symmetrical pair-wise comparisons might be more meaningful in view of the auto-covariance nature of the implied spread.

28 We also experiment with two other alternative measures of volatility, i.e., unsigned daily return (see e.g. Jones and Seguin 1997) and standard deviation of daily return, and obtain similar results.

29 Another concern is that press coverage could be contaminated by the effects associated with the inclusion since a larger price swing should attract more attention.

30 Including changes in liquidity or volatility as additional explanatory variables has no material impacts on other variables and overall goodness-of-fit. This is not surprising given the lack of either effect, as reported before.