Trading Patterns and Excess Comovement of Stock Returns

Abstract

In April 2000, 30 stocks were replaced in the Nikkei 225 Index. The unusually broad index redefinition allowed for a study of the effects of index-linked trading on the excess comovement of stock returns. A large increase occurred in the correlation of trading volume of stocks added to the index with the volume of stocks that remained in the index, and opposite results occurred for the deletions. Daily index return betas of the additions rose by an average of 0.45; index return betas of the deleted stocks fell by an average of 0.63. Theoretical predictions for changes in autocorrelations and cross-serial correlations of returns of index additions and deletions were confirmed. The results are consistent with the idea that trading patterns are associated with short-run excess comovement of stock returns.

There is abundant evidence that security prices can move together either too little or too much to be justified by fundamentals. What could be causing this comovement if not fundamentals? Empirical studies have uncovered a variety of common factors in returns, such as size, value, and industry factors. In academic literature, debate is ongoing as to whether these factors are related to fundamental risk or, alternatively, reflect mispricing driven by investor demand. Providers of commercial risk models have stayed away from this debate and include a broad set of factors to explain common variation in asset returns.

We argue that one driver of comovement of returns is commonality in trading activity. We tested this hypothesis by using an unusual index redefinition of the Nikkei 225 Index in April 2000 in which 30 stocks were replaced. Upon inclusion in an index, a stock becomes exposed to the trading shocks experienced by other stocks in the index. Whenever index funds experience inflows or outflows, they trade index stocks as a basket. Also, index arbitrageurs delta-hedge their index derivative positions, which requires simultaneous trading in the basket of the underlying securities. Consistent with these observations, we documented a large and significant increase in the correlation of trading volume of the 30 stocks added to the Nikkei 225 with the trading volume of stocks that remained in the index, and we found the opposite results for the deleted companies.

We investigated whether the change in trading activity has consequences for returns. We found that after the Nikkei redefinition, the daily return betas of the additions with respect to the stocks that remained in the index rose by an average of 0.45 but the daily index return betas of the deletions fell by an average of 0.63. Thus, index membership alone explained a surprising amount of the comovement among stock returns.

We also made predictions about changes in autocorrelations and cross-serial correlations for added and deleted stocks following index redefinition. These predictions, which are not featured in existing research, were motivated by the idea that pricing effects from shocks to correlated investor demand should eventually subside. That is, although security returns of index stocks should comove excessively in the short run, at longer horizons, returns should revert to reflect fundamentals. We found strong support for these predictions. A particularly interesting result is that following index redefinition, daily return autocorrelations of additions decreased whereas return autocorrelations of deletions increased, which suggests that additions (deletions) become more (less) exposed to transitory index-trading shocks. Taken together, our results suggest that commonality in trading baskets induces significant excess comovement of stock returns.

Our findings have important implications for modeling risk in equity markets. Index membership is likely to be an important common factor even after accounting for industries and fundamental factors. This aspect is especially important in risk models geared toward daily returns—in that the effects of correlated index trading tend to subside as the return horizon increases. Our results also imply that short-term shocks to index demand add to the transaction costs of index investing.

We are grateful to Nicholas Barberis, Malcolm Baker, Ken Froot, Emir Kamenica, Tom Knox, Mike Rashes, Jorge Rodriguez, Jeremy Stein, Jeff Wurgler, Tuomo Vuolteenaho, seminar participants at Harvard, and especially, John Campbell and Andrei Shleifer for helpful discussions.

Notes

¹ The immediate change in beta around the event is a result of high returns to the additions driven by institutional rebalancing. While purchasing additions, institutions were also selling deletions and remainders, driving down their prices. Note in Figure 2 that the consequences of the redefinition lasted for 300 trading days, after which the redefinition left the window used to estimate beta.

² See also Beneish and Whaley (1996); Lynch and Mendenhall (1997); Kaul, Mehrotra, and Morck (2000); Wurgler and Zhuravskaya (2002); Denis, McConnell, Ovtchinnikov, and Yu (2003); Chen, Noronha, and Singal (2004).

³ For example, Barberis et al. (2003); Greenwood and Sosner (2003).

⁴ We attempted to track down additional inclusions and deletions after 1990 but found that most of the deletions were subsequently delisted. Additional information on this redefinition and all changes in the composition of the Nikkei indices can be found on the Nihon Keizai Net web page (http://www.nni.nikkei.co.jp/).

⁵ See http://www.nni.nikkei.co.jp/FR/SERV/help/npr0010.html.

⁶ Because the actual index included the deletions before the event and the additions after the event, our results on comovement are mechanically stronger for comovement with the actual index than with the remainder index. We also experimented with replacing the remainder return with a price-weighted remainder return, which we constructed by using only the remainder stocks but with weights given by their actual Nikkei 225 weights. These results were stronger than those reported in the article.

⁷ “Winsorizing” sets the most extreme values to the values at the 99th percentile of the distribution.

⁸ We left out these data so that our results would not be contaminated by short-term correlations of volume that might have arisen from rebalancing by institutional investors as they attempted to match the new index.

⁹ See, for example, Scholes and Williams (1977) and Chapter 4 in Lo and MacKinlay (1999).

¹⁰ Additional evidence that index membership reduces pricing efficiency is given in Greenwood (forthcoming 2007).