Russell 2000 Reconstitution Effects Revisited
The costs to investors of passive investing and the relative merits of transparent index reconstitution rules are important investment management topics and subjects of perennial interest to researchers and investors alike. This is particularly true regarding the Russell 2000 Index, the preeminent benchmark index for the US small capitalization equity market.1 This paper updates prior research on the impact of index reconstitution on the performance of the Russell 2000 Index and reviews related work.
Russell last investigated the impacts of reconstitution on index performance in Cariño and Pritamani (2007). That study examined the period 2000 to 2006 and found a marked decrease in index impacts over the period 2003 to 2006 as compared to the period 2000 to 2002. Considering index performance from reconstitution until August 31, the average impact on Russell 2000 Index performance over the period 2000 to 2002 was estimated to be -3.03%. The estimated impact was reduced to -0.94% when the returns of stocks added to and deleted from this index were controlled for size, sector and momentum effects. Turning to the period 2003 to 2006, the estimated impact was +0.07% before factor correction and -0.16% after correction, a reduction of more than 80% in factor-corrected impacts.
Cariño and Pritamani (2007) observed that index methodology and exchange trading improvements may have helped reduce reconstitution impacts on index performance.2 However, no tests of the statistical significance of the difference in index impacts was conducted, so it was not clear that the reductions in estimated impacts were indeed the result of any systematic effect. This study uses the same analytical approach and adds nine years of reconstitution data to cover the period 2000 to 2015. We also use a bootstrapping procedure to conduct statistical significance tests on changes in underlying security prices.
Our results for the period 2000 to 2006 are close to those of Cariño and Pritamani. For example, we find average impacts before factor correction from reconstitution day to August 31 over the period 2000 to 2002 to be -3.06%, compared to -3.03%. We find average impacts from reconstitution day to August 31 over the period 2003 to 2015 to be -0.05%. When impact is corrected for factor effects, the magnitude increases to -0.18%, surprisingly close to the Cariño and Pritamani estimate of -0.16% average impact over the period 2003 to 2006.
Statistical significance testing shows that the average differences in index impacts between the periods 2000 to 2002 and 2003 to 2015 are almost all highly statistically significant, regardless of the length of the post-reconstitution window and the level of factor correction employed. These results support the Cariño and Pritamani hypothesis that index and exchange improvements may have reduced the impact of reconstitution trading on index performance.3 These results are also consistent with the findings of Petajisto (2011) and Chang, Hong and Liskovich (2015).
We further test for the statistical significance of estimated impacts themselves. These results are particularly interesting for the period 2003 to 2015. Index impacts for five- and ten-day event windows are all highly statistically significant regardless of the level of factor correction. The reconstitution day to July 31 and August 31 windows, however, reveal a different pattern. Raw index impacts without any factor correction are not found to be statistically different from zero. Adding size correction leads to the July event window impacts becoming statistically significant. However, adding both size and sector correction to the August event window impacts is not sufficient for obtaining statistical significance of estimated impacts. August event window index impacts are found to be statistically significant only when size, sector and momentum corrections are used together.
This pattern of results appears to validate the Cariño and Pritamani (2007) study design. First, the results are consistent with the distinction between short-term price-pressure effects that partially reverse, resulting in smaller long-term effects, in that all short-term effects are highly statistically significant but longer-term effects are not.
Second, the ability of increasing levels of factor correction to increase the statistical significance of longer-term effects speaks to the effectiveness of the characteristics-based factor-correction method used.
—Download the complete paper—
1 See, for example, “A big issue for the small-cap Russell 2000,” Chris Dieterich, Barron’s, Feb. 28, 2015
2 Index methodology improvements included the quarterly listing of IPO issues and the introduction of provisional reconstituted indexes in advance of the official indexes. Also, introduction of the NASDAQ “closing cross” trading mechanism for NASDAQ-traded securities greatly reduced risk of trading at the close of reconstitution. All of these innovations were introduced in 2004.
3 Relevant to the hypothesis that index methodology improvements may have reduced reconstitution impacts, Russell also introduced banding in its indexing methodology, starting in 2007.