Price Clustering Asymmetries in Limit Order Flows

• DOI: http://doi.org/10.1111/fima.12136
• Date: 01 December 2016
• Author: Travis Box,Todd Griffith
• Published date: 01 December 2016

Price Clustering Asymmetries in Limit

Order Flows

Travis Box and Todd Griffith∗

We explore the relation between limit order price clustering and price efficiency. We find that

executed sell limit orders cluster more frequently on round increments than buy limit orders

and that this asymmetry in clustering is consistent with the well-documented asymmetry in price

response to marketable orders. In addition, we find that the degree of clustering is positively

related to volatility and that asymmetry in clustering depends on whether stock prices arerising

or falling—sell limit orders cluster morefrequently as prices are rising, although buy limit orders

cluster more as prices are declining. Our results indicate that predictable patterns in limit order

pricing contribute to short-run deviations from price efficiency.

This paper examines how the clustering of limit orders on round pricing increments affects

the execution prices of marketable orders and, thereby, contributes to temporary deviations from

price efficiency. We define price efficiency to be the degree to which security prices reflect all

available information, both timely and accurately.1Prior research shows that upon the arrival of

material firm-specif ic or macroeconomic information, liquidity-demanding traders improveprice

efficiency by quickly submitting marketable orders in the direction of permanent price changes

and in the opposite direction of temporary pricing errors (Kyle, 1985; Brogaard,Hendershott, and

Riordan, 2014; Chaboud et al., 2014). However, when those marketable orders execute against

less-precise clustered limit orders, stock prices are slower to reflect permanent price changes.

Prior literature offers three explanations for why limit orders might execute more frequently on

discrete price sets: First, traders might simply wish to reduce cognitiveprocessing costs (Wyckoff,

1963). Second, investors may desire to increase order execution probability or lower negotiation

costs (Harris, 1991; Cooney, Van Ness, and Van Ness, 2003). Third, liquidity providers might

be uncertain about the underlying value of a security (Ball, Torous, and Tschoegl, 1985). In

any of these scenarios, the clustered order price represents a lower precision estimate of the

security’s fundamental value. Thus, the predictable clustering patterns in limit orders might delay

the incorporation of new information into asset prices.

Research shows that markets react differentlyto buy and sell orders, and that the price impact of

purchases exceeds that of sales (Kraus and Stoll, 1972; Holthausen, Leftwich, and Mayers, 1987,

1990; Gemmill, 1996). Saar (2001) provides a theoretical justification for this impact differential

based on information effects. If informed traders are active market participants, and their strategy

results in buy orders conveying more information than sell orders on average, then equilibrium

prices are expected to adjust more for buys than for sells. Similarly,Chan and Lakonishok (1993)

argue that institutional investors do not typically hold the market portfolio, and the decision

of which stock to sell does not necessarily convey negative information. The choice to buy a

∗Travis Box is an assistant professor in the Department of Finance at the University of Mississippi in University,

MS. ToddGriff ith is a Ph.D.student in the Department of Finance at the University of Mississippi in University, MS.

1Wedef ine price efficiency in accordance with prior literature such as Fama(1970), Brogaard, Hendershott, and Riordan

(2014), and Saffi and Sigurdsson (2010).

Financial Management •Winter 2016 •pages 1041 – 1066

1042 Financial Management rWinter 2016

single stock out of the numerous possibilities available, however, is likely to convey favorable

firm-specif ic news.

By using the comprehensive NASDAQ Total View-ITCH limit order data set, which provides

unique trade direction identifiers, we are able to study the clustering frequency of executed

buy and sell limit orders separately. In aggregate, executed sell limit orders cluster on round

increments more often than executed buy limit orders. Consistent with extant literature, we

find that marketable buy orders have higher price impacts than sell orders. In addition, we

demonstrate that marketable orders transacting at clustered prices have higher price impacts than

trades occurring at nonclustered prices. Because the intersection of a limit sell (buy) order and

a marketable buy (sell) order results in a trade, the clustering in limit sell orders is equivalent to

clustering in marketable buy orders. Therefore, the aggregate asymmetry in clustering frequency

between executed buy and sell limit orders reported in this study provides another explanation

for this previously documented disparity in market order price impact.

Our paper also explores the link between executed limit order clustering and return volatility

by analyzing how extreme stock price movements affect the likelihood that shares transact on

round pricing increments. We find a positive and monotonic relation betweenthe frequency with

which limit orders execute on round increments and return volatility. As price shocks become

more extreme, liquidity supplying limit orders execute onr ound numbers more often. In addition,

the asymmetry in clustering between buys and sells depends on whether stock prices are rising

or falling. When prices are rising, the stock return is positive, sell limit orders placed on round

increments execute more frequentlythan b uy limit orders. Conversely,when prices are falling, the

stock return is negative, buy limit orders priced on round numbers execute more often than sell

limit orders. In our sample, the distribution of extreme price movements is not symmetric, with

major price advances occurring more frequently than declines. Because price shocks are more

often positive than negative, asymmetries in clustering frequencies between executed buy and

sell limit orders could be a mechanical consequence of an asymmetric stock return distribution.

To illustrate our main findings, Figure 1 plots executed limit orders for Tesla Motors (TSLA)

over twodistinct 45-minute trading inter valsin 2012. For each panel, the darker lines denote limit

orders executed on round increments, while the lighter lines represent orders executedat all other

prices. Panel A describes a series of sell limit orders executing on August 7, 2012, where Tesla’s

stock price increased rapidly from $29.50 to just below $31.00 per share. The magnitude of the

adjustment implies that the market became much more optimistic during a short period. The long

dark lines at round prices (such as $30.00, $30.10, $30.90, etc.) indicate that large pockets of

clustered orders delay the convergence of Tesla’s shares to their new equilibrium price. If this

extreme price change resulted from positive information entering the market, the speed at which

prices reflected this information was hindered by clustered sell limit orders slowing the transition

to the new equilibrium value. Panel B of Figure 1 depicts a similar event where Tesla’s prices

declined rapidly.Here, large quantities of clustered buy limit orders intersect with marketable sell

orders, prolonging the stock’s price decline.

To the extent that clustering represents a lower precision estimate of a security’s reservation

value, we provide strong evidence that price clustering in executed limit orders contributes to

transitory deviations from price efficiency. For instance, when prices are rising (falling), fewer

precise sell (buy) limit orders execute on round increments more frequently, delaying information

from being timely and accurately impounded into security values. This is particularly true when

price movements are volatile, because executed limit order clustering is increasing in return

volatility.

The rest of the paper is organized as follows. Section I discusses the data used for our analysis.

Sections II and III report the results from our empirical investigation along with motivational