«January 2001 I would like to thank James Angel, Jeff Bacidore, Shane Corwin, Jason Greene, William Megginson, Jeff Netter, George Sofianos, Annette ...»
Stock Splits, Liquidity and Limit Orders
Marc L. Lipson
Department of Banking and Finance
Terry College of Business
University of Georgia
Athens, GA 30602-6253
I would like to thank James Angel, Jeff Bacidore, Shane Corwin, Jason Greene, William
Megginson, Jeff Netter, George Sofianos, Annette Poulsen, Gideon Saar and seminar
participants at Indiana University, Michigan State, and the NYSE JB seminar series for helpful
comments. I would also like to thank Jeff Benton and Katherine Ross for invaluable assistance and insight. All remaining errors are the property of the author. The comments and opinions expressed in this paper are the author‟s and do not necessarily reflect those of the directors, members or officers of the New York Stock Exchange, Inc. Contact Marc L. Lipson at the above address or as follows: email at firstname.lastname@example.org; FAX at (706) 542-9434; phone at (706) 542Please do not circulate or quote without permission.
Stock Splits, Liquidity and Limit Orders Abstract We use non-public NYSE system data to study the effect of stock splits on liquidity by documenting changes in the limit order book, execution costs, and trading activity. We find that depth available in the limit order book at various dollar distances from the mid-quote increases after a stock split, but the depth available at various percentage (split-adjusted) distances declines substantially. In general, we observe a slight increase in the use of limit orders rather than market orders, and fill rates are largely unchanged. Consistent with these results, the realized execution cost (in percent) of limit orders declines dramatically while the realized execution cost for market orders increases. Overall, despite a 10 basis point increase in the proportional effective half-spread, we find little evidence of a change in execution costs across all orders.
Finally, while trading activity generally declines following stock splits, we observe a substantial increase in the number of submitted orders, an increase in the proportion of trading volume that originates from individuals, and an increase in market buys by individuals.
Stock Splits, Liquidity, and Limit Orders
1. Introduction The motivation behind stock splits is a puzzle that continues to hold the interest of academics and practitioners alike. While ample evidence suggests that stock splits return prices to a “normal” trading range, the factors that determine this optimal trading range are not clearly understood.1 In fact, despite a positive market reaction to stock split announcements, market quality actually appears to decline after a split, with sharp increases in spreads and volatility. 2 This paper examines the effects of stock splits on several measures of market quality not previously studied. In particular, we examine changes in the limit order book, the execution costs of limit orders and market orders, and trading activity. Our results provide a more complete picture of the changes in market quality than what is provided by studies of trades and quotes alone.
Limit orders compete with market makers for order flow and also provide a pool of trading interest (the limit order book) that can absorb temporary order flow imbalances. For this reason, changes in limit order activity, both in terms of quantity and placement, may be closely related to changes in market quality.3 Furthermore, while a binding minimum tick size is likely to widen spreads and alter depth close to the mid-quote, it is unclear how depth further away See, for example, Lakonishok and Lev (1987) and Angel (1997) Merton (1987), Lamoureux and Poon (1987), Brennan and Hughes (1991), Maloney and Mulherin (1992), and Ohlson and Penman (1985), among others, examine changes in market quality around stock splits. Koski (1998) examines potential microstructure explanations for changes in volatility around stock splits.
McInish and Wood (1995), Harris and Hasbrouck (1996), Greene (1995) and Seppi (1997) discuss the relation between limit orders and market quality. The choice between limit and market orders is explored theoretically in Cohen, Maier, Schwartz and Whitcomb (1981) and Holden and Chakravarty (1995), while empirical evidence on the costs and pricing strategies of limit orders is provided in Biais, Hillion, and Spatt (1995) and Griffiths, Smith, Turnbull and White (1999).
from the quote will be affected. A unique contribution of this paper is to document changes in the limit order book, limit order volume, and limit order placements around stock splits.
Another contribution of this study is to examine execution costs for both market and limit orders around stock splits. Since public market orders often trade against public limit orders, an increased cost to one trader may be a savings to another. Thus, even though spreads increase following stock splits, the overall effect on execution costs is not clear. This is particularly true on the NYSE where specialists participate in only a small fraction of trades (see Sofianos and Werner (1997)).
Our analysis uses non-public system order data provided by the NYSE. These data allow us to distinguish between market and limit orders, to track cancellations and executions, and to identify orders originating directly from individuals. These data are also sufficient to recreate the limit order book following Kavajecz (1999), and measure trading costs as in Harris and Hasbrouck (1996). Our analysis examines 2-for1 or greater stock splits in NYSE listed companies during the years 1995 and 1996.
We find that the total depth available in the limit order book declines on the bid side but is little changed on the ask side. On the other hand, the distance from the mid-quote to the best price in the limit order book (limit-book spread) and to the prices at which total depth of 5,000 and 10,000 shares would be available in the limit order book (5,000 and 10,000 share spreads) are closer to the prevailing mid-quote in dollar terms. Similarly, there is an increase in the total depth available in the limit order book (cumulative depth) at various dollar distances from the prevailing mid-quote. To understand how these changes will impact liquidity, however, we need to examine proportional distances rather than dollar distances. This is particularly true if we are interested in potential effects on execution costs and volatility since these are measured proportionally.4 The limit-book spread and 5,000 and 10,000 share spreads are substantially larger as a proportion of the mid-quote following a stock split. For example, the 5,000 share spread increases on average from 1.57% to 2.17% on the bid side and from 1.03% to 1.99% on the ask side after a stock split. Similarly, at various proportionally identical (split-adjusted) distances from the prevailing mid-quote, the total number of shares available in the limit order book declines sharply. For example, cumulative depth up to $1/8 (split-adjusted) declines by about 2,600 shares on the bid side and by about 3,500 shares on the ask side following a split. These changes represent declines of over 30% from pre-split levels. We show that changes in the placement of limit orders, rather than changes in limit order executions, drive these changes. For example, we find that limit orders are placed, on average, 33 basis points away from the prevailing mid-quote before the split and 60 basis points after, while there is no significant change in execution rates. These results suggest a partial explanation for the changes in spreads and volatility commonly observed after stock splits may be changes in the limit order book.
We find that the execution cost for executed limit orders declines after a split while the cost for market orders increases. If we look at all executed orders, the mean change in the weighted average execution cost is indistinguishable from zero, while the median change is a decrease of 5 basis points. If we include a conservatively high penalty for non-execution of a limit order, the mean change in the weighted average execution costs across all orders is still Interpreting the results on dollar cutoffs is also difficult since there are two confounding effects. Traders may optimally employ dollar (tick) based trading strategies, in which case the increased depth at dollar distances suggests a greater willingness to place limit orders. On the other hand, if trading interests are based on proportional execution costs, then the increased depth at fixed dollar distances (proportionally greater distances) may simply reflect a movement along a demand schedule. We discuss interpretations of dollar distance results in the conclusion, paying particular attention to how our results relate to the ongoing debate on the effects of tick size on trading behavior.
indistinguishable from zero, though in this case the median change is an increase of about 5 basis points.5 These results paint a very different picture of trading costs than what is seen in spreads.
For example, the proportional effective half-spread (the average difference between the execution price and the mid-quote at the time of execution) increases by 10 basis points. Our results suggest, therefore, that the increase in trading costs associated with stock splits may not be as great as generally thought.
Finally, though we observe a decline in executed daily share volume after a stock split, we find a substantial increase in the number of orders (decrease in average order size), an increase in the proportion of order flow originating from individuals, and an increase in the number and volume of buys by individuals. These results provide direct evidence that stock splits are associated with an increase in trading activity by individuals.6 The rest of this paper is organized as follows. Section 2 discusses the sample and presents summary statistics. Section 3 presents the analysis of the limit order book, section 4 discusses limit order placements, and section 5 discusses execution costs. Section 6 discusses our results in light of a number of stock split debates.
2. Sample and Summary Statistics The sample includes all 2-for-one or greater stock splits by NYSE firms listed in the Center for Research in Security Prices (CRSP) data set during 1995-1996. For each stock split, we obtain intraday Trade and Quote (TAQ) data and NYSE system (superDOT) order data for the period starting 60 days prior to the stock split and ending sixty days after the stock split. To The costs associated with unfilled orders (implementation shortfall) were described in Perold (1988) and are discussed in Harris and Hasbrouck (1996) and Griffiths, Smith, Turnbull and White (1999), among others.
Studies of clientele effects include Merton (1987), Lamoureux and Poon (1987), Brennan and Hughes (1991), Maloney and Mulherin (1992), and Schultz (1998).
ensure valid estimates of the measures we examine, a split is included only if there are at least five days with system trades in the pre-split and post-split sample periods.
The number of splits, average split sizes, and firm characteristics are described in Table
1. This table lists 2-for-one and greater than 2-for-one stock splits separately. The sample of firms with 2-for-one stock splits is comparable to the sample of greater than 2-for-one stock splits with the exception of stock price. Not surprisingly, the stock price prior to the split is substantially greater for the larger splits. Given the small number of larger stock splits, we do not study these samples separately in the remainder of the paper. The mean pre-split stock price in our sample is $68.50, which is about twice the average stock price on the NYSE. These results are consistent with observations that stock splits return prices to “normal” trading ranges. For this reason, we follow Schultz (1998) and present all share and stock price values on a post-split basis throughout the paper.
2.1 Market Quality and Trading Activity Summary statistics on market quality and trading activity are presented in Table 2. Here and throughout the paper, statistical tests compare the pre and post-split time periods by examining the distribution across firms of the change in mean values using a t-test or Wilcoxon signed rank test. Specifically, for each firm we calculate the mean daily values in the pre- and post-split sample periods separately. We then examine the distribution of the firm-by-firm changes in the mean values. Thus, we assume independence across firms and across the pre-split and post-split time periods. In general, statistical results for median changes are qualitatively identical to those for means.7 In Table 2 we report both means and medians, though in the We also examined the distribution of pair-wise changes in medians, compared the distribution of pre-split and post-split event-time means and medians, and compared the distribution of pre-split and post-split daily firm values.
Statistical inferences are similar using these alternative methods.
remainder of the paper we mostly report means in the tables and discuss any differences between mean and median results in the text.
Panel A of Table 2 shows market quality measures. We begin with an analysis of quoted and effective half-spreads. The quoted half-spread is equal to one half the difference between the ask and bid prices, while the effective half-spread is equal to the difference between the execution price of an executed trade and the mid-quote at the time of execution. We present half-spreads to be consistent with our analysis in sections 3 and 5. As in Conroy, Harris and Benet (1990), we find that dollar quoted half-spreads and effective half-spreads decrease while proportional quoted and effective half-spreads (dollar spread divided by mid-quote) increase.
To examine changes in volatility, we look at two volatility measures: the standard deviation of daily (close to close) returns and the average of the squared daytime excess returns (open-to-close return less the mean open-to-close return during the pre or post-split time periods, as appropriate). We use mid-quotes for these measures in order to eliminate transient effects from bid-ask error. Consistent with Koski (1998), we find evidence of an increase in volatility.
We observe a significant increase in the mean volatility using the first measure, and the median volatility using both measures.