Informed Trading and Price Discovery before Corporate Events
Shmuel Baruch University of Utah shmuel.baruch@business.utah.edu, Marios Panayides University of Pittsburgh mpanayides@katz.pitt.edu Kumar Venkataraman * Southern Methodist Universitykumar@mail.cox.smu.edu
Abstract
We show that stock prices incorporate less “news” before negative events than positive events. For a sample of Euronext-Paris stocks, we document a buy-sell asymmetry in order submission strategies: informed traders use more price aggressive (market) buy orders before positive events and less price aggressive (limit) sell orders before negative events. Motivated by this pattern, we build on Kaniel and Liu (2006) to model the execution risk that informed agents impose on each other and relate the asymmetry to costly short selling. When investor base is narrow, security borrowing is difficult, and the magnitude of the event is small, we show that the buy-sell asymmetry is pronounced and that the price discovery before negative events is lower. Overall the results support that informed traders influence the process of price formation in financial markets, as predicted by microstructure theory.
First Draft: March 2013 This Draft: March 2015
Keywords: informed traders; order submission strategies; short sales; buy-sell asymmetry.
Earlier versions of this paper were titled “Informed trading before corporate events: Theory and Evidence.” For their comments, we thank Amber Anand, Leonce Bargeron, Kerry Back, Indraneel Chakraborty, Jeff Coles, Kevin Crotty, Dave Denis, Diane Denis, Amy Edwards, Robert Engle, Tom George, Hans Heidle, Stacey Jacobsen, Pankaj Jain, Ron Kaniel, Praveen Kumar, Sebastian Michenaud, Bruce Lehman, Ken Lehn, Swami Kalpathy, Bige Kahraman (EFA discussant), Albert Menkveld, Ling Peng, Gideon Saar, Bob Schwartz, Wing Wah Tham, Erik Thiessen (Erasmus discussant), Shawn Thomas, Rex Thompson, Laura Tuttle, Mathijs Van Dijk, Vish Viswanathan, Sunil Wahal, Ingrid Werner (Stern discussant), Chad Zutter, and seminar participants at Arizona State University, University of Houston, University of Pittsburgh, University of Toronto, Rice University, Southern Methodist University, Lehigh University, Baruch College CUNY, the 2013 Erasmus Liquidity Conference, the 2013 Stern Microstructure Conference, the 2013 Northern Finance Association Conference, the 2013 European Finance Association Conference and the U.S. Securities and Exchange Commission (SEC). We thank Laurent Fournier and Carole Huguet of Euronext-Paris for providing data. We are grateful in particular to Hank Bessembinder for helpful discussions during the development of the project.
* Corresponding author: Kumar Venkataraman, Southern Methodist University, 6212 Bishop Blvd., Dallas, Texas 75275-0333, Phone: 214 768 7005; Fax: 214 768 4099; email: kumar@mail.cox.smu.edu.
Informed Trading and Price Discovery before Corporate Events
Abstract
We show that stock prices incorporate less “news” before negative events than positive events. For a sample of Euronext-Paris stocks, we document a buy-sell asymmetry in order submission strategies: informed traders use more price aggressive (market) buy orders before positive events and less price aggressive (limit) sell orders before negative events. Motivated by this pattern, we build on Kaniel and Liu (2006) to model the execution risk that informed agents impose on each other and relate the asymmetry to costly short selling. When investor base is narrow, security borrowing is difficult, and the magnitude of the event is small, we show that the buy-sell asymmetry is pronounced and that the price discovery before negative events is lower. Overall the results support that informed traders influence the process of price formation in financial markets, as predicted by microstructure theory.
Keywords: informed traders; market versus limit orders; short sale constraints; buy-sell asymmetry. JEL classification: G11, G12, G14, G18.
1. Introduction
Informed traders play a central role in price formation in financial markets. In microstructure models (see Kyle (1985), Glosten and Milgrom (1985)), informed traders receive a private signal about the true value of the security. Their private information is incorporated in security price when market makers respond to the informed order flow by adjusting the bid-ask quotes. Consistent with theoretical
models, there is widespread evidence of both insider trading and significant stock price movements before
corporate announcements, such as mergers and acquisitions and earnings releases.1
In this study we address the following questions regarding informed trading and price discovery. First, how do informed traders build positions, and what explains their strategies? Second, how do their decisions to use aggressive or passive strategies affect the process through which private information is incorporated into security prices? And third, how do their strategies affect the speed of learning before corporate events? Understanding informed traders’ behavior is central both to validating the well-developed theoretical literature that relates informed trading to price formation, and to designing better surveillance systems to monitor and detect insider activity.
Despite the importance of understanding how informed agents build positions, there is relatively little empirical evidence explaining their strategies, mainly because the available data sources are not sufficiently detailed. Publicly available data sources, such as NYSE’s Trade and Quote (TAQ) database, report all the transactions in a market but do not identify the specific trades of informed agents. Some data on transactions are available from regulatory Form 4 filed by corporate insiders with the U.S. Securities and Exchange Commission (SEC) (see Seyhun (1986)), or from the SEC’s case files of defendants formally charged with insider trading (see Meulbroek (1992)); however, these datasets do not contain
order level information which is necessary to characterizethe aggressiveness of the traders’ strategies.
We examine a dataset provided by the Euronext-Paris exchange that contains detailed information on all orders submitted for all stocks; however the Euronext data do not directly identify the orders of
1 See Keown and Pinkerton (1981), Meulbroek (1992), Bodnaruk, Massa and Simonov (2009), Jegadeesh and Tang (2010), and Griffin, Shu and Topaloglu (2012), among others.
informed agents. To do so we employ a research design based on Chae (2005), Graham, Koski and Loewenstein (2006) and Sarkar and Schwartz (2009), and examine a sample of corporate announcements
related to M&As, SEOs, repurchases, dividend initiations and dividend terminations in 2003.2 While
these events convey new information, the timing of these events is not known to the public in advance. The prior literature shows that uninformed traders lower participation to reduce adverse selection risk
before scheduled corporate events, such as earnings announcements (see Lee, Mucklow, and Ready
(1993)), but uninformed traders do not anticipate unscheduled events where timing information is not
publicly available in advance of announcement. In our model, informed agents receive a private signal
about the unscheduled event and build positions in the direction of signal.3 We identify informed trading
as the increase in buying activity before positive events and the increase in selling activity before negative events relative to a control period for the same firm. By holding each firm as its own control, we reduce the influence of firm attributes on the cross-sectional variation in order submission strategies.
We present a model, which builds on Kaniel and Liu (2006), of the informed trader’s choice of market versus limit orders. Informed agents face a tradeoff between transacting with certainty at a current market price by placing a market order versus risking non-execution in an attempt to get a better price by placing a limit order. In addition to paying the bid-ask spread, market orders tip off the market maker about the private signal and increase market impact cost. When other informed agents receive the same signal and use market orders, the execution risk of a limit order strategy is particularly high. Our model predicts that informed agents use limit orders when there is sufficient uncertainty about the presence of other informed agents, and use market orders if they are certain that other informed agents are present.
2 We examine the Euronext-Paris data in 2003 because more recent order-level data obtained from NYSE-Euronext have important limitations, which we describe in Section 2.1.
3 Informed agents include corporate insiders such as board members, directors and employees as well as well-connected market participants such as bankers, analysts and hedge fund managers. A WSJ article dated 06/06/2013 that describes abnormal trading activity before Smithfield’s acquisition announcement says “When multiple bidders vie for a company, it isn’t unusual for hundreds of people to know about the possible deal before it surfaces – including employees of banks, law firms, and other outside advisors, not to mention the people inside the companies themselves.”
These predictions follow Kaniel and Liu’s (2006) theoretical work showing that if the mass of informed traders is sufficiently low then informed traders might use limit orders.
Informed agents face less competition when the nature of private information conveys a decrease
in stock price. This is because informed agents are less likely to sell stocks with unfavorable information
if they do not already own the stock.4 Consistent with this prediction, we report a novel empirical
regularity of asymmetry in the informed agents’ order submission strategies before positive and negative
events. Specifically we observe an increase in aggressively priced buy orders before positive events but a
decrease in aggressively priced sell orders before negative events (henceforth “buy-sell asymmetry”).5 As
in Diamond and Verrecchia (1987), our model relies on short sale constraints to explain the buy-sell asymmetry observed in the data. However, in our model, the asymmetry emerges not only because some informed sellers decide to abstain, but also because informed sellers become liquidity providers.
Our model yields cross-sectional predictions in buy-sell asymmetry based on competition among informed sellers. Informed sellers can be one of two types; the first type already owns the stock while the second type does not. The probability that the informed seller is of the first type increases with broadness of investor base. When investor base is narrow, when the cost of borrowing shares is large, and when event is small such that potential gains cannot justify the borrowing costs, a limit order equilibrium emerges in which the first type of agent uses limit orders and the second type abstains from trade. On the other hand, when investor base is broad, borrowing costs are small, or when event is large, the second type of agent borrows the shares, and both types trade. Because of the execution risk that informed traders
impose on each other, both types use market orders. When the private information conveys an increase in
stock price, our model predicts that informed buyers always anticipate competition and use market orders.
4 Corporate insiders face more constraints when they trade on bad news than on good news (see Marin and Olivier (2008)). For example, insiders in many markets are prohibited from selling short their own stock (e.g., Section 16 of the U.S. Securities Exchange Act of 1934), or sell stock holdings that are part of a compensation contract below a certain threshold. Informed sellers who do not own the stock incur the cost of borrowing shares. When borrowing costs are high, informed sellers anticipate less competition from other agents before an information event.
5 A related literature on block trading examines buy-sell price impact asymmetry; that is, the well-known result that security purchases convey more information than security sales (see Kraus and Stoll (1972), Keim and Madhavan (1995)). Saar (2001) motivates the idea that informed sellers may trade for liquidity reasons while informed buyers do not, which is a source of price impact asymmetry.
Motivated by prior work, our empirical measures of short sale constraints include stock index membership, availability of exchange-listed stock options, and eligibility for Euronext’s Deferred Settlement Service. Index stocks have broad investor base, active participation by institutions and lower borrowing cost (see D’Avolio (2002), Nagel (2005)). Stock options allow informed sellers to establish equivalent positions at lower cost in short constrained stocks (see Battalio and Schultz (2011), Hu (2013)). Stocks eligible for Euronext’s Deferred Settlement Service (called “SRD”) facility are easier to borrow for establishing short positions (see Foucault, Sraer and Thesmar (2011)).
We find strong empirical support for the model’s predictions. When informed traders anticipate competition (i.e., index constituent stocks, stocks with listed options, SRD-eligible stocks and large news announcements), price aggressiveness increases for both buy orders before positive events and sell orders before negative events. In contrast, when short selling is costly (i.e., non-index stocks, stocks without listed options, and SRD-ineligible stocks) or when news is small, informed buyers and informed sellers use a different mix of market and limit orders. Specifically, price aggressiveness of buy orders increases
before positive events but price aggressiveness of sell orders decreases before negative events.6
Importantly, our results support that informed traders’ strategies affect the process through which private information is incorporated into security prices. Following Biais, Hillion and Spatt (1999), we estimate the efficiency of price discovery before corporate events using “unbiasedness regressions”, where we regress the close-to-close return in the interval Days [-5,+1] on the return in interval Days [-5,X] where X represents half-hour snapshots between Days [-5,-1]. Barclay and Hendershott (2003) interpret the slope of the unbiasedness regressions as the signal:noise ratio. In our context, as the slope moves closer to “one”, the interpretation is that the pre-event security price reflects the post-event security value with increasing precision. For both positive events and negative events, at the beginning of the pre-event window the slope of unbiasedness regression is not statistically different from zero. However, as the
6 Short constrained stocks tend to be smaller, more volatile and less liquid. In Section 4, we show that other stock attributes (e.g., liquidity) do not lead to asymmetry in competition among informed agents before positive and negative events and thus do not contribute to buy-sell asymmetry. The key testable prediction of our model is the buy-sell asymmetry for short constrained stocks but not for stocks without short constraints.
event gets closer (i.e., Day [-1]), the slope for positive events is not significantly different from one while for negative events, the slope is lower (between zero and one) and significantly different from one. These results indicate that the stock price incorporates more “news” before positive events than negative events. Results based on subsamples strongly support that informed traders’ strategies affects the speed of learning and informational efficiency of prices. In scenarios when informed traders decrease the price aggressiveness of orders (i.e., short sale constrained stocks with negative events), the slope of the unbiasedness regression towards the end of the pre-event window is between zero and one, and statistically different from one. In scenarios when informed traders increase the price aggressiveness of orders (i.e., negative events in firms with no short sale constraints and all positive events), the slope of the unbiasedness regression is positive and not statistically different from one. These results provide direct empirical support for our model, and in general, the well-established theoretical literature predicting that informed traders’ strategies influence the process of price formation in financial markets.
In addition to improving our general understanding of the interaction between informed traders and price discovery, our study points to an unintended consequence of the widespread ban on short selling by regulators around the world in response to the 2007-09 financial crisis. Beber and Pagano (2013) find that the short sale ban lowers the information efficiency of prices, particularly surrounding events with negative information. Because the short sale ban reduces competition among sellers, our model predicts that informed sellers who already own the stock use less aggressively priced orders, which impedes the flow of negative private information into prices. Thus our study offers a specific mechanism by which short selling impedes the price discovery process, as shown by Boehmer and Wu (2013).
In the Diamond and Verrecchia (1987) and also Saar (2001) framework, informed traders trade upon arrival; in other words, they do not employ limit orders. In fact, most theoretical work posits that
informed agents exclusively use market orders to exploit their information advantage.7 Novel exceptions
are Kumar and Seppi (1994), Chakravarty and Holden (1995), Kaniel and Liu (2006), Goettler, Parlour
7 See Kyle (1985) and Glosten and Milgrom (1985) for studies of strategic trading in a dealer setting and Glosten (1994), Rock (1996), Seppi (1997) and Back and Baruch (2013) for strategic trading in a limit order book setting.
and Rajan (2009), Boulatov and George (2013) and Rosu (2014). In these models, informed traders do find it optimal, under certain conditions, to be less aggressive and submit limit orders. However, none of these limit order papers relate informed trader strategies to the direction of the private signal, and with the exception of Kaniel and Liu (2006), none provides empirical evidence that informed traders may use limit
orders.8 Our results support theoretical models that link informed strategies to competition and further to
the information efficiency of the stock price (Holden and Subrahmanyam (1992)). This study is the first, as far as we are aware, to show that short sale constraints impede the price formation process by affecting the competition among informed traders before a corporate event.
The rest of the paper is organized as follows. Section 2 describes the sample, the data sources and summary statistics. Section 3 shows asymmetry in informed trader strategy before positive and negative events. In Section 4, we present a model that explains the results and provides new testable predictions. In Section 5, we test whether competition among informed agents influences informed trader strategies. Section 6 presents an analysis of order exposure while Section 7 examines price discovery based on unbiasedness regressions. Section 8 concludes the study.
2. Data and Methodology
2.1. Sample and data
We examine the Euronext-Paris, Base de Donnees de Marche (BDM) database for the year 2003. The BDM database contains detailed information on the characteristics of all orders submitted for all stocks listed on Euronext-Paris. This includes the stock symbol; the date and time of order submission; whether the order is a buy or a sell; the total size of the order (in shares); the displayed size (in shares); an order type indicator for identifying market or limit orders; a limit price in the case of a limit order; and instructions on when the order will expire.
8 Consistent with their model, Kaniel and Liu (2006) show that limit orders have price impact. Recent experimental and empirical work suggests that informed traders use limit orders. See Barclay, Hendershott, and McCormick (2003); Bloomfield, O’Hara, and Saar (2005, 2013); Anand, Chakravarty and Martell (2005), Hautsch and Huang (2012), among others.
We examine the 2003 sample period because more recent order-level data purchased from the Euronext market have important inaccuracies. In particular, orders that never get executed, or orders with a hidden component that are partly executed, do not get reported to the database. The omission affects the accuracy of reconstructed limit order book, the analysis of order submission strategies, and the construction of control variables used in some specifications. Further, trading in Euronext stocks is highly consolidated (over 90%) during our sample period, where the vast majority of orders were submitted and executed on Euronext-Paris. With the explosion in alternative trading venues, European equity markets in recent years are highly fragmented. In a fragmented market, the informed agent’s choice of the trading venue to execute an order needs to be modeled, which introduces a layer of complexity. For these reasons, the market structure in 2003 provides a simple laboratory to test theoretical models on informed trading.
The Euronext database does not provide any information on trader identity. We therefore examine trading activity before unscheduled corporate events. The timing of information release for scheduled events, such as earning releases, is publicly available in advance of announcement. Chae (2005), Graham, Koski and Loewenstein (2006) and Sarkar and Schwartz (2009) show that, although traders do not know
in advance the information contained in scheduled events, those traders with some discretion on timing of
trades alter behavior prior to these events. For example, Lee, Mucklow and Ready (1993) find that market makers widen bid-ask spreads and lower inside depth before earnings releases. In contrast, informed agents alone are aware of the pending news release in the case of unscheduled, or “surprise” events, where the timing of information release is not known in advance. Consistent with this idea, Sarkar and Schwartz (2009) show that trading activity before unscheduled events is characterized by one-sided
markets. Following the prior literature, we attribute the change in trading activity observed before
unscheduled events to the actions of informed agents.
We identify unscheduled events using the Global SDC database compiled by Thomson Financial
Securities Data and the AMADEUS database provided by Bureau van Dijk. We focus on five types of
unscheduled events: M&As, SEOs, repurchases, dividend initiations and dividend terminations. We use Bloomberg and Factiva search engines and identify the date of the first news story about the event. We
eliminate Euronext-Paris stocks that switch from continuous trading to call auctions (or vice-versa) or were de-listed from Euronext during Days [-30,+10] surrounding the event (Day [0]). The final sample consists of 101 unscheduled corporate events for 95 unique stocks.
In microstructure models, informed traders build positions in the direction of private signal before the information is widely available. We therefore classify events as positive and negative news based on the announcement (Days [0,+1]) cumulative abnormal returns (CAR), where CAC40 Index return serves as the benchmark. In Table 1, Panel A, the results suggest that the positive event sample has larger market capitalization, higher stock price, lower return volatility and smaller bid-ask spread than the negative event sample. The median percentage bid-ask spread is economically large (approximately 0.80%) suggesting that the cost of building a position using market orders is non-trivial.
In Table 1, Panel B, for each unscheduled event, we compare order flow on Days [-5,-1] with the order flow on control period Days [-30,-10] for the same firm. Informed order flow is measured as the change in order flow for buy orders before positive events and the change in order flow for sell orders
before negative events.9 Relative to control period, the daily number of buy (sell) orders and the average
order size increase before positive (negative) events, which supports that informed traders are active before unscheduled events. We also report the daily ratio of marketable orders to limit orders (henceforth, market:limit ratio) and the percentage of limit orders with a hidden size. We find that the percentage of limit orders with a hidden size is similar on event and control days. A higher (lower) market:limit ratio on event days relative to control days suggests that informed traders use more (less) aggressively priced orders before events. The market:limit ratio increases before positive events but decreases before negative events indicating that informed agents use a different mix of market and limit orders based on the direction of the signal.
9 For a sample of French M&A announcements, Aktas et al. (2007) report that trading volume spikes in the five days preceding a merger announcement and that trading volume and bid-ask spreads increase for both acquirers and targets in the Days [-65,-6] before the announcement. Informed trading during the control period would lower the statistical power of our design to find support for the model. We attempt to minimize the impact of leakage to the general public by deleting all events with pre-event “rumors” reported in press articles.
In Table 2, we report the event CAR for the positive and negative events. The 58 positive events have a mean (median) Days [0,+1] CAR of 4.84% (2.74%) and the 43 negative events have a mean (median) Days [0,+1] CAR of -4.53% (-2.57%). The largest CARs are observed for M&As followed by SEOs. We observe both positive and negative events for all the announcement types with the exception of
dividend termination.10 We find that the stock price moves in the direction of news before the event
(Panel B) and that the stock price incorporates more information before positive events than negative events (Panel C). Specifically, for all event types, the magnitude of Days [-5,+1] CAR in Panel B is larger than Days [0,+1] CAR in Panel A. In Panel C, for positive events, the Days [0,+1] CAR accounts for 44.83% of the Days [-5,+1] CAR. In other words, the stock price reflects 55.17% of the “news” before positive events. In comparison, the stock price reflects 26.48% of the “news” before negative events. What explains the higher price discovery before positive events? Can it be informed trader strategies? We explore these questions using multivariate regressions to model informed trading and unbiasedness regressions to model price discovery before positive and negative events.
2.2 Methodology
The multivariate analyses of order submission strategies is estimated on an event-by-event basis using all orders submitted on event days (Days [-5,+1]) and control period (Days [-30,-10]) for the firm. The change in order flow surrounding the event, which we attribute to informed traders, is estimated by
including daily indicator variables for each event day. For example, the DayMinus5 coefficient captures
the change in order characteristics observed on Day [-5] relative to control days for the same firm. Using same-firm order flow on control days as a benchmark reduces the impact of firm characteristics on order submission strategies. Because informed agents can be active on any of the five days before the event, we calculate a cumulative informed trading measure that aggregates individual day indicator coefficients. Specifically, the cumulative measure captures the combined influence of informed trading over the five
10 Asquith and Mullings (1986) and Graham et al. (2006) find negative announcement returns for 28% and 36%, respectively, of their dividend initiation sample. The former study notes that “for these firms investors are anticipating the initiation of dividends and were disappointed by the amount of the initial dividend.” Consistent with the idea, the Factiva dividend announcement of one of our firms with negative returns says “However, this is a special dividend. It is not expected to pay a regular dividend.”
days before the event without any econometric constraint on each of the Days [-5,-1]. In interpreting the results, we focus on both individual day indicator coefficients and the cumulative measure, and the associated t-statistics. In the interest of parsimony, we present results that are aggregated across events. Details of our aggregated method are found in Appendix A.
3. Multivariate Analysis of Price Aggressiveness and Price Impact before corporate events
3.1. Price Aggressiveness
We begin our investigation by examining the preference of informed agents towards an order’s price aggressiveness attribute in a multivariate regression. The regression specification accounts for the impact of market conditions on an order that arrives at time of submission ‘t’ for event ‘i’, as follows:
PriceAggressiveit = γ0 + γ1DayMinus5 + γ2DayMinus4 + γ3DayMinus3 + γ4DayMinus2 + γ5DayMinus1+ γ6Day0&Plus1 + γ7DayPlus2 + γ8OrderExposureit + γ9PriceAggressiveit-1 + γ10HiddenOppSideit + γ11DisplayedSizeit-1 + γ12OrderSizeit
+ γ13Spreadit + γ14DepthSameit + γ15DepthOppit +γ16Volatilityit + γ17WaitTimeit +
γ18TradeFreqHourit + γ19BookOrderImbalanceit + γ20TradeSizeit-1 +
γ21MktVolatilityit-1 +γ22Ind.Volatilityit-1 (1)
Following Biais, Hillion and Spatt (1995) and BPV (2009), PriceAggressive is an ordinal variable
that takes the value of 1 for the most aggressive order and 7 for the least aggressive order.11 The market
conditions capture (1) the state of the limit order book, including bid-ask spread, displayed depth at the best quote, cumulative order book imbalance, standing limit orders at the same price as the incoming order, and revelation of hidden orders at the inside quote by the last trade; (2) trading conditions for the stock, such as recent volatility, trading frequency, and waiting time between recent order arrivals; (3) order attributes, such as size and exposure; and (4) control variables such as recent industry volatility, market volatility, and time-of-the-day effects. Industry and market volatility helps account for commonality in economic fundamentals, as in Chordia, Roll and Subrahmanyam (2000). A detailed description of the variables is provided in Appendix B. To render results more comparable across stocks,
11 Following Liu and Agresti (2005) and Gelman & Hill (2007), we select a linear specification over a non-linear specification (ordered probit) because the dependent variable represents a large number of price aggressiveness (seven) categories. When fitting a proportional odds model, there is little gain in efficiency when using more than 4 levels of the category variable over an OLS. Further, a linear regression specification allows an easy estimation of the economic significant of day-indicator variables and the cumulative measure.
we normalize the depth and spread variables by dividing the actual observation by the median for that stock on control days, while order size and trade size are normalized by dividing the actual observations by the stock’s average daily trading volume on control days.
Columns (1) and (2) of Table 3 reports coefficients of the price aggressiveness regression along
with corresponding t-statistics, estimated on an event-by-event basis and then aggregated across firms
using the approach described in Appendix A. The coefficients on the control variables are consistent with those reported in the prior studies. Specifically, traders submit less aggressively priced orders (i.e., choose limit orders over market orders) when (a) the inside bid-ask spread is wide, (b) same side book depth is thin, which signals less competition (c) opposite side book depth is deep, or the last trade reveals the presence of hidden orders, both of which signal active counterparties, (d) volatility is high, consistent with a volatility capture strategy (see Handa and Schwartz (1996)), (e) book imbalance signals less competition on same side relative to opposite side of the book, and (f) the limit price of the previous order (a proxy for omitted market conditions) is less aggressive. The impact of market volatility and industry volatility are not significant; however, own-stock volatility has a significant influence.
We focus on the coefficient estimates of indicator variables, DayMinus5 to DayMinus1. Note that
the least aggressive order is categorized as “7” and the most aggressive order is categorized as “1”. Thus a
negative (positive) DayMinus3 coefficient points to an increase (decrease) in price aggressiveness on Day
[-3] relative to same-firm control days. Since the research design attributes abnormal activity on Day [-3]
to informed agents, a negative DayMinus3 coefficient implies that informed agents use more aggressively
priced orders to build positions before the event.
For buy orders submitted before positive events (column (1)), all the five coefficients
corresponding to Days [-5,-1] are negative, and DayMinus3 and DayMinus1 coefficients have t-statistics
below -2.0. Overall the cumulative effect coefficient is -0.60 (see Panel B, t-statistic=-2.96) implying that
informed buyers submit more aggressively priced buy orders before positive events. For sell orders
submitted before negative events (column (2)), three of the five coefficients corresponding to Days [-5,-1] are positive and the DayMinus5 coefficient is statistically significant (t-statistic=2.76). The cumulative
effect coefficient of 0.17 (t-statistic=-1.70) suggests that informed sellers submit less aggressively priced sell orders before negative events. The difference in cumulative effects before positive and negative event is highly significant (t-statistic=-23.57). The asymmetry in informed trader strategies before positive and
negative events, consistent with univariate results reported in Table 1, is a novel empirical regularity.12
In terms of economic significance, the cumulative effect coefficient in linear specification is interpreted as the change in price aggressiveness on Days [-5,-1] relative to control days [-30,-10] after
accounting for other determinants of price aggressiveness. Focusing on all positive events (column (1)),
the cumulative coefficient -0.6045 suggests that, relative to average price aggressiveness of 5.18 observed during control days, price aggressiveness increases on Days [-5,-1] by 11.67%. For negative events a similar analysis suggests that the aggregate coefficient for Days [-5,-1] represents a decrease in price aggressiveness of 3.27%. We also examine the change in distribution of the seven categories of price aggressiveness on event days relative to control days. Controlling for market conditions and order characteristics, we observe an 8.73% increase in frequency of aggressive orders (categories 1 to 4) on Days [-5,-1] for positive news versus a 2.90% decrease in frequency of similar aggressive orders on Days [-5,-1] for negative news.
In unreported analysis, we regress the change in price aggressiveness on Days [-5,-1] estimated using equation (8) for each event on the event’s CAR ratio, Days [0,+1]/Days [-5,+1], which measures the extent to which “news” is incorporated in stock price before announcement. The regression coefficient is 0.449 (t-statistic=2.13) which suggests that events with more aggressively priced orders are associated with a larger stock price drift in the direction of news before the event.
3.2. An analysis of price impact surrounding positive and negative events
The results on buy-sell asymmetry points to a possible mechanism by which block purchases convey more information than block sales - more aggressive strategies used by informed buyers relative to informed sellers cause a larger drift in security price for block buys than block sells. To investigate this
12 Results are similar if events are classified as positive and negative based on Days [-5,+1] CAR. Further, note that there is no clear pattern of aggressiveness between Day [-5] and Day [-1] suggesting that informed traders are not certain about the exact timing of the news release.
asymmetry, we estimate the implementation shortfall cost of an order using the framework proposed by Perold (1988) and implemented by Harris and Hasbrouck (1996) and Griffiths, Smith, Turnbull, and White (2000). Each order is associated with two components of implementation shortfall: (a) effective spread cost is the appropriately signed difference between the fill price and the quote mid-point at the time of order submission, and (b) the opportunity cost is the appropriately signed difference between the closing price on order expiration or cancellation date and quote midpoint at the time of order submission. For a limit order that goes unfilled, the effective spread cost is zero. For an order that is fully executed, the opportunity cost is zero. For orders that are not fully executed, the opportunity cost is positive if the stock price moves higher (lower) for buy (sell) orders after order is placed. The implementation shortfall cost for an order is the weighted sum of effective spread and opportunity cost, where weights are the proportion of order size that is filled and unfilled, respectively.
Table 3 presents coefficient estimates of the implementation shortfall regression on the prevailing market conditions and order attributes. We find that price aggressiveness is positively associated with effective spread cost, consistent with Harris and Hasbrouck (1996), and that hidden orders are associated with smaller opportunity costs, consistent with BPV (2009), who conclude that hidden orders are primarily used by uninformed agents to control exposure risk. Focusing on columns (5) and (6) in Panel B, the cumulative effect effective spread costs on Days [-5,-1] relative to control days is not statistically significant. However, in the case of opportunity cost, the cumulative effect coefficient is positive for buy orders before positive events (column (7), t-statistic=2.17) but negative for sell orders before negative events (column (7), t-statistic=-1.77), and the difference in cumulative effects before positive and negative events is highly significant. Thus the results suggest that the stock price tends to move in the direction of the order before positive events but not before negative events.
4. A Model on Execution Risk, Short Sale Constraints and Informed Trader Strategy
The goal of the model presented in this section is to explain the initial findings on buy-sell asymmetry and to develop additional predictions that go beyond our initial findings.
4.1. Motivation
It is common in the theoretical literature to posit that informed traders solely use market orders. There is however a small literature that points out that, given the choice, informed traders may employ
limit orders.13 To the best of our knowledge, there is nothing in the literature that directly relates informed
trader aggressiveness to the direction of the private information (i.e. positive versus negative). Kaniel and
Liu (2006) use a probabilistic trading model14 and show that if: (a) the mass of uninformed traders who
use limit orders is sufficiently large (so informed traders can camouflage their limit orders), (b) the mass of informed traders is sufficiently small, and (c) the likelihood that the information is long-lived is sufficiently high, then informed traders find it optimal to use limit orders.
We believe that an amalgam of the economics that underlies Kaniel and Liu (2006) together with asymmetry between buying and selling activity introduced by costly short selling, as in Diamond and Verrecchia (1987), is a good foundation to build our framework. In Kaniel and Liu (2006), the lower the probability that the next trader is informed, the more likely the current (informed) trade submits a limit order. In an environment with costly short selling, this is a plausible scenario. Since the economics is quite intuitive, we chose to construct the simplest model possible. If the economics is robust, as we think it is, a parsimonious model should suffice to generate valid predictions to bring to the data.
4.2. The Model of Informed Trading without Short Selling
We consider a static model with two identically informed traders who are large enough to
influence each other’s execution speed, execution risk, and trading costs.15 The informed traders learn the
realization of a signal 𝑣𝑣, and make the choice of using market orders or limit orders to build their position
before private information becomes widely known. The informed traders can also choose to abstain. An
13 Chakravarty and Holden (1995), Kaniel and Liu (2006), Goettler, Parlour and Rajan (2009), Boulatov and George (2013) and Rosu (2014)
14 Goettler, Parlour, and Rajan (2009) and Rosu (2014) also employ probabilistic trading models. Most notable difference is that competitive market makers are absent in their models, whereas in Kaniel and Liu (2006) market makers yield to public limit orders.
15 We are not the first to use a static model to gain insight on what is essentially a dynamic problem. For example, arm races and bank runs are typically studied using static models.
informational event is negative if the informed traders learn that the stock is overvalued and positive if the stock is undervalued.
If the event is positive, then both Trader One and Trader Two can and will trade (i.e. buy). Trader
One’s payoff is given by the following payoff matrix, where 𝑣𝑣 is the expected value of the signal. The
payoff to Trader Two is symmetric. Trader One’s payoff (Positive event)
Trader One
Market Orders Limit Orders Abstain
Trader Two Market Orders (𝑣𝑣-𝑣𝑣)2 b (𝑣𝑣-𝑣𝑣)2a 0
Limit Orders (𝑣𝑣-𝑣𝑣)2d (𝑣𝑣-𝑣𝑣)2 x 0
Abstain (𝑣𝑣-𝑣𝑣)2e (𝑣𝑣-𝑣𝑣)2y 0
Each informed seller can be one of two types. Type Zero has the stock in his/her portfolio and can
therefore sell the stock at will, while type C has to pay C to short the stock. To motivate the model’s
intuition, we consider in this section a scenario where short selling is banned, i.e. C=∞.16 We denote by p
the probability that a trader is of type Zero, which is a proxy for the broadness of the stock’s investor base
when informed traders are randomly selected from the universe of investors. If the event is negative,
Trader One’s payoff depends on his/her type, 𝜃𝜃 ∈ 0, 𝐶𝐶 , is given by
Trader One’s payoff (Negative event)
Type𝜽𝜽 ∈ 𝟎𝟎, 𝑪𝑪
Trader One
Market Orders Limit Orders Abstain
Trader Two Market Orders (𝑣𝑣-𝑣𝑣)2 b-𝜃𝜃 (𝑣𝑣-𝑣𝑣)2a-𝜃𝜃 0
Limit Orders (𝑣𝑣-𝑣𝑣)2d-𝜃𝜃 (𝑣𝑣-𝑣𝑣)2 x-𝜃𝜃 0
Abstain (𝑣𝑣-𝑣𝑣)2e-𝜃𝜃 (𝑣𝑣-𝑣𝑣)2y-𝜃𝜃 0
16 For roughly one-third of their sample of NYSE and NASDAQ stocks, Diether and Werner (2011) show that a limited supply of loanable shares reduces the ability of short sellers to trade.
The payoff matrix for type Zero is similar to the positive event. Type C, on the other hand, incurs
the cost C=∞ when selling. Thus, type C abstains. Although the parameters a,b,d,e,x,y, and 𝑣𝑣 − 𝑣𝑣 , in
the payoff matrices, and probability p vary from event to event, we posit that all information events are
standard in the following sense:
Assumption (Standard Event): The following holds: (i) 0<a<b and (ii) x>d implies y>e. 17
The first assumption states that Trader Two’s market orders impose execution risk on Trader One’s limit orders. That is, when Trader Two uses market orders, the price impact pushes market prices away from Trader One’s limit price. To prevent limit orders from becoming stale, Trader One has to keep
offering better prices, implying that using limit orders to build the position is suboptimal (a<b). The
second assumptions states that if Trader One prefers limit orders when Trader Two uses limit orders, i.e. if x>d, then it should also be that Trader One prefers limit orders when Trader Two abstains; i.e. x>d only if y>e. In particular, note that the model does not assume that x>d.
Our two equilibria are defined as follows:
Definition 1 (Limit Order vs. Market Order Equilibrium): A limit-order equilibrium is any equilibrium
that does not involve market orders and a market order equilibrium in any equilibrium that does not involve limit orders.18
Theorem 1, which provides the intuition for the model, represents a special case of Theorem 2 when short selling is banned. Theorem 2 is stated in the next section and proved in Appendix C.
Theorem 1 (Limit Order vs. Market Order Equilibrium, short selling is banned):
A. A positive event always has a market-order equilibrium.
B. Whenever there is an equilibrium that involves limit orders in a positive event, then there is also a limit-order equilibrium in the corresponding negative event.
C. Assume y<e: whether event is positive or negative, informed traders only use market orders.
17 The assumption x>d implies y>e is only used in part C of Theorems 1 and 2.
18 To clarify the definition, if the equilibrium has the property that type Zero uses limit orders while type C
D. Consider a negative event. Assume y>e, and assume further that p is sufficiently small (i.e. close to zero), then there is unique limit-order equilibrium.
E. Consider a negative event. Assume y>e, and assume further that p is sufficiently large (i.e. close to one), then there is a market-order equilibrium.
Parts A and B of the theorem attribute our empirical findings in Section 3 of buy-sell asymmetry to the execution risk that informed traders impose on each other. Part A explains why we see buy market-orders before positive events while part B proves that there are more (at least in a weak sense) limit order equilibria in the universe of negative events.
Part D generates our first prediction that adds to our initial empirical finding: the asymmetry
between positive and negative events is more pronounced when investor base is narrow (i.e., p is low).
Our novel prediction depends on assumption, y>e, that a monopolist informed seller prefers limit orders
to market orders. The assumption is motivated by the results in Kaniel and Liu’s (2006) that if the mass of informed trader is sufficiently low then informed traders might use limit orders. This assumption, though conceivable, is not standard. For example, if the stock is sufficiently liquid such that bid-ask spread is
small, or information is short lived such that immediacy is paramount, the assumption y>e is likely to be
violated; i.e. even a monopolist informed trader who faces no competition prefers market orders to limit
orders (i.e., y<e) implying that informed traders facing competition also prefer market orders (i.e., x<d).
Importantly, Part C of the Theorem shows us that a violation of assumption y>e does not imply a
reverse buy-sell asymmetry, and in particular, a violation of the assumption y>e does not offer predictions
that vary with the broadness of investor base. In other words, the presence of negative events in our data
that violate assumption y>e reduces the power of our statistical tests because, regardless of broadness of
investor base (i.e., p), these events do not contribute to asymmetry between positive and negative events.
Empirical evidence in support of buy-sell asymmetry implies that there are sufficient negative events in our sample that fit the framework of Kaniel and Liu’s (2006).
Finally, it is important to note the possibility of multiplicity of equilibria. For example, Part E of the theorem does not guarantee uniqueness of the market-order equilibrium when event is negative and
investor base is broad.19 However, the presence of events with multiplicity of equilibria should be viewed
as noise, making it harder for us to establish statistical significance, but does not bias our results.
4.3. Costly Short Selling
In this section, we extend the model by allowing C to be finite. In other words, the stock is
available to short, though at some cost. Hence, the decision of a trader of type C to abstain is endogenous.
Since this extension contains our original model as a special case (C=∞), Theorem 2 will be used to
formulate our empirical predictions.
Theorem 2 (Limit Order vs. Market Order Equilibrium, costly short selling)
A. A positive event always has a market-order equilibrium.
B. Whenever there is an equilibrium that involves limit orders in a positive event, then there is also a limit-order equilibrium in the corresponding negative event.
C. Assume y<e: whether the event is positive or negative, informed traders only use market orders. D. Consider a negative event. Assume (i) y>e, (ii) C is sufficiently large or the event is sufficiently
small (i.e. 𝑣𝑣 − 𝑣𝑣 is small), and (iii) p is sufficiently small (i.e. close to zero), then there is unique limit-order equilibrium.
E. Consider a negative event. Assume (i) y>e, and (ii) at least one of the following three conditions holds: p is sufficiently large (i.e. close to one), C is sufficiently small, or the event is sufficiently large (i.e. 𝑣𝑣 − 𝑣𝑣 is large), then there is a market-order equilibrium.
4.4. Testable predictions
Part D of Theorem 2 predicts that the buy-sell asymmetry depends on both firm characteristics (i.e., short sale constraints) and event attributes (i.e., magnitude of the event). Figure 1, which is based on Parts A, B, and D of Theorem 2, leads to the following hypotheses:
Hypothesis I (Buy-sell asymmetry): The aggressiveness of buy orders increases before positive
news and the aggressiveness of sell orders decreases before negative news.
Hypothesis II:The buy-sell asymmetry is pronounced when the event is small.
19 To guarantee uniqueness one needs x<d.
Hypothesis III: The buy-sell asymmetry is pronounced when investor base is narrow, or the stock is difficult to short.
Figure 1: Competition and Order Aggressiveness
Part E of Theorem 2 shows that the buy-sell symmetry is recovered when the event is large, investor
base is broad, or borrowing cost is small. Our empirical results, presented in the next section, support that
buy-sell symmetry is observed in these cases; however we cannot rule out that the results stem from part C
of our theorem. Recall that part C assumes that a monopolist informed trader prefers market orders (i.e.
y>e is rejected) which is plausible for liquid stocks. It is conceivable that a stock with broad investor base
and low borrowing cost is also a liquid stock. Similarly, stocks with low volatility are typically liquid, and it is conceivable such stocks are associated with small informational events. That said, we are unaware of any economic link, besides the one based on short sales built in this paper, that explains why the extent of competition among informed agents is different before positive and negative events. Thus, Hypotheses II and III are novel even though short sale proxies that we employ are correlated with other firm attributes.
Finally our model assumes that informed agents using market orders tip off market makers and cause the security price to drift in the direction of private signal. Thus the extent to which private information is
Positive News Negative News
Informed Want to Buy Informed Want to Sell
Otherwise
Intense Competition Mild Competition
Informed traders Informed traders
use use
Market Orders Limit Orders
Broad Investor Base Cheap to Short or Event is Large
reflected in the security price depends on the aggressiveness of informed traders’ strategies.
Hypothesis IV (price efficiency asymmetry): The stock price incorporates more “news” before
positive events than negative events, particularly when event is small, investor base is narrow, and borrowing costs are high.
5. Empirical Tests of Informed Trader’s strategy on price aggressiveness
To test model predictions, we examine the cross-sectional patterns in informed traders’ strategies by classifying corporate events along two dimensions (b) the broadness of investor base and the ease of short selling, and (b) magnitude of announcement period returns.
5.1. Index constituent stocks and informed trader strategy
Stock membership in an index influences the broadness of investor base. Index firms tend to be larger, actively traded and attract interest from analysts and buy-side institutions (Nagel (2005)). Further index stocks are easier to borrow because index funds are active lenders of securities. For these reasons, our model predicts that informed agents in index stocks employ aggressive strategies both before positive
and negative events.20 For non-index stocks, the model predicts a buy-sell asymmetry - informed buyers
use market orders before positive events while informed sellers use limit orders before negative events. In columns (1) - (4) of Panel A.1 of Table 4, we report an univariate analysis of daily market:limit ratio on event and control days. We classify stocks that belong to SBF120 index as those with broad
investor base.21 For index stocks, we observe an increase in market:limit ratio both before positive and
negative events. For non-index stocks, the market:limit ratio increases before positive events but decreases before negative events. Consistent with univariate results, the cumulative effects coefficient from price aggressiveness regressions (see Panel A.2.) for index stocks are negative both before positive (column (1), t-statistic=-2.31)) and negative (column (2), t-statistic=-1.74)) events. For non-index stocks, the cumulative measure is negative before positive events (column (3), t-statistic=-2.14)) but positive
20 For index stocks the market-order equilibrium is consistent with a broad investor base (i.e., Part E of Theorem 2) but also a scenario where the stock is so liquid (i.e., Part C of Theorem y<e due to narrow bid-ask spread) that the benefit of using limit orders is small.
21 The SBF120 index represents a broader cross-section of stocks than the CAC40 Index (see Bessembinder and Venkataraman (2004)).
before negative events (column (4), t-statistic=2.17)). The difference in the cumulative measures before positive and negative events is highly significant (t-statistic=-16.16) for non-index stocks but not for index stocks. Thus the results support a buy-sell symmetry when investor base is broad but a buy-sell asymmetry when investor base is narrow. The latter result is consistent with the limit-order equilibrium that is predicted for stocks with narrow investor base by Part C of Theorem 2.
5.2. Deferred Settlement Service and informed trader strategy
For certain eligible stocks, Euronext-Paris offers a unique mechanism, the Deferred Settlement
Service (“Service de Règlement Différé”, henceforth SRD), that offers investors the convenience to take long and short positions with deferred settlement of the trade until the end of the month (see Foucault, Sraer and Thesmar (2011)). Stocks that are eligible for SRD-facility are chosen by the exchange. An
investor who wishes to sell short an SRD-eligible stock must flag the order as deferred execution when
submitting to the broker. On executing the short sale, the broker effectively acts as a lender of the stock
until the end of the month for an additional fee.22 The SRD facility offers informed sellers a convenient
way to sell stocks that they do not own as long as they cover the short position by the end of the month. Anecdotal evidence indicates that short selling a stock that is not eligible for Euronext’s SRD facility is cumbersome because short sellers need to locate shares they want to sell in advance of executing a short sale. Thus it is easier for informed sellers to take position in eligible stocks relative to SRD-ineligible stocks.
Table 4, columns (5)-(8), report the results for SRD-eligible and SRD-ineligible stocks. We obtain the list of SRD-eligible stocks during our sample period from Euronext-Paris. For SRD-eligible stocks, the univariate statistics indicate an increase in market:limit ratio before both positive and negative events while for SRD-ineligible stocks, the market:limit ratio increases before positive events but decrease before negative events. The regression results support the patterns observed in the univariate analysis. For SRD-eligible firms, the cumulative measure is negative both for buy orders before positive
22 For a typical retail order in June 2001, Foucault, Sraer and Thesmar (2011) report that online brokers charge an additional fee of 0.20% of the order (with a minimum amount of 6.2 euros) for orders with deferred execution. In comparison, the median announcement return for negative events reported in Table 2 is 2.57%.
events (column (5)) and sell orders before negative events (column (6)). For SRD-ineligible stocks, the measure is negative for buy orders before positive events (column (7)) but positive for sell orders before negative events. The difference in cumulative measures is statistically significant for SRD-ineligible stocks (t-statistic=-13.31) but insignificant for SRD-eligible stocks (t-statistic=1.67). Thus we observe a buy-sell asymmetry for stocks that are difficult to short but not for stocks that are easier to short.
5.3. Options listing and informed trader strategy
Exchange-listed options offer informed sellers a viable outlet to build a short position. Battalio and Schultz (2011) show that, when stocks are short constrained, informed sellers build positions using option contracts, which creates a sell imbalance in options market. Options market makers, who absorb
the imbalance and hold long positions, offset the inventory risk by short selling the stock. Hu (2013)
shows that the hedging activities of options market makers are related to the information content of equity market imbalance. Options market makers enjoy special exemptions from locate requirements on inventory-hedging trades; thus short sale constraints do not significantly affect their ability to implement short positions. Via this mechanism, an options listing lowers the short sale constraints. Within our framework, one implicit assumption is that options market makers use market orders to hedge their inventories in the stock market.
Results in Table 4, Panel B, broadly support the model prediction of buy-sell asymmetry. For stocks with listed options, the market:limit ratio increases both before positive and negative events while for stocks without listed options, the market:limit ratio increases before positive events but decreases before negative events. Similarly, for stocks with listed options, the cumulative measure is negative both for buy orders before positive events and sell orders before negative events. For stocks without listed options, the measure is negative for buy orders before positive events but positive for sell orders before negative events (column (4)), and the difference in cumulative measures before positive and negative events is statistically significant.
In an unreported analysis, we examine trading activity in the options market for stocks with listed options. We find no evidence of a statistically significant increase in the number of daily options trades
before positive events. Before negative events, the number of daily options trades increases on Days [-5,-1] relative to control period. On Day [-[-5,-1], the options activity is nearly twice as large and the increase is statistically significant relative to control days. More options activity observed before negative events relative to positive events in short constrained stocks supports that informed traders use synthetic short position in the options market as a substitute for short sales in the underlying stock.
5.4. Announcement returns and informed trader strategy
Informed sellers with more incentives to locate difficult-to-borrow shares when the negative news is large. When the negative news is small, the benefits from short selling might not out-weight the borrowing cost. For this reason informed sellers who do not already own the stock might abstain. We
classify an event as large when the absolute value of announcement day return exceeds 5%.23 Consistent
with Hypothesis II, results in Panel B suggest informed sellers use limit orders before small negative events. The market:limit ratio decreases on Days [-5,-1] relative to the control days for small negative events (column (8)) whereas in all other scenarios, we observe an increase in the market:limit ratio.
Focusing on positive events, the cumulative measure on Days [-5,-1] in the price aggressiveness regression is negative for both small (coefficient=-0.35 with t-statistic=-2.09) and large (coefficient=-1.39 with t-statistic=-2.40) events. The economic significance of these results is large. For large positive events, we estimate an increase of 26.80% in buy order aggressiveness on event days relative to control days, which reflects a 42.06% increase in the frequency of aggressive buy orders (categories 1-4). The corresponding statistics are 6.68% and 4.91%, respectively, for small positive events. For small negative events, the positive cumulative measure (coefficient=0.31 with t-statistic=1.95) translates into a decrease of 6.05% in sell order price aggressiveness and a 6.05% decrease in frequency of aggressive sell orders. The difference in cumulative measures between small positive and small negative events is highly
23 As a robustness check, we classify events as large (above median) or small (below median) after dividing the absolute value of announcement return by the stock’s average bid ask spread measured on control days. The normalized measure captures the trade-off between using market orders and paying the spread versus using limit orders and receiving the spread. The coefficient signs and statistical significance are similar to those reported in Table 4.
significant (t-statistic= -15.44). We note that the cumulative measure for large negative events is positive but not statistically significant (t-statistic=0.45).
Our results are consistent with theoretical papers that model a dynamic limit order book with strategic traders. For example, Goettler, Parlour, and Rajan (2009) predict that informed traders use limit orders when volatility is low and Rosu (2014) predicts that informed traders use limit orders when realized signals are close to their expected value. In support of these predictions, and consistent with our model, informed traders use limit orders before small negative events. However, consistent with our model, and inconsistent with Goettler, Parlour, and Rajan (2009) and Rosu (2014), we show that informed traders use market orders before small positive events. That said, when the event is small, we find that the orders they submit are less aggressive than when the event is large.
5.5. Related discussions
Table 4 reports the number of events for each subsample. We note that many index stocks have exchange-listed options and are eligible for the SRD facility. In an unreported analysis, we find that the overlap leaves sufficient room for independence across these cuts of the data. For example, focusing on index stocks, 10 among 24 positive event stocks and five among 16 negative event stocks do not trade with listed options. Similarly, focusing on stocks without listed options, 13 among the 44 positive events and nine among 28 negative events are eligible for SRD facility.
An important model assumption is that informed agents using market orders impose execution risk on other informed agents using limit orders. To validate this assumption, we estimate the Days [-5,-1] cumulative measure from opportunity cost regressions for each sub-sample in Table 4. Recall that the opportunity cost is positive if the stock price moves higher (lower) for buy (sell) orders after order submission. In all sub-samples (with the exception of large negative events), the cumulative measure from the price aggressiveness regression and the cumulative measure from the opportunity cost regression have the opposite sign. This is a striking finding that supports the following interpretation. When informed traders anticipate more competition, they use aggressive orders (i.e., price aggressiveness measure is negative), which tips off the market and causes the stock price to move in the direction of the order (i.e.,
opportunity cost measure is positive). When informed traders anticipate less competition, they use passive strategies (i.e., price aggressiveness measure is positive) which are harder to detect by the market maker (i.e., opportunity cost measure is negative).
In an unreported analysis, we examine both the number of trades and the number of shares traded on Days [-5,-1] relative to control days. In sub-samples identified as “short constrained” in columns (4) and (8) of Table 4, we find that the increase in trading activity observed on Days [-5,-1] relative to control days is lower than other sub-samples. The lower trading activity supports that informed sellers who do not own the “short constrained” stock abstain from trading when benefits do not exceed the borrowing costs. 6. Empirical Tests of Informed Trader’s Strategy on Order Exposure
Do informed agents build positions using hidden orders? The theoretical predictions on informed trader’s usage of hidden orders are ambiguous. Harris (1996) argues that exposing size will attract interest from “reactive” traders who monitor markets and respond to interests posted by other traders. However exposing large orders might cause other traders to withdraw interest, or implement front running strategies, if they infer the presence of informed agents. In a stylized framework where informed agents are restricted to using limit orders, Moinas (2010) predicts that informed agents select an order exposure strategy that increases execution probability (see also Buti and Rindi (2013)). Boulatov and George (2013) show that informed agents expose order size when they face competition from other agents.
Only a handful of academic studies provide empirical evidence on the issue. In a laboratory setting, Bloomfield, O’Hara and Saar (2013) show that both uninformed and informed agents use hidden orders and that informed agents make higher profits in an opaque market when their private information is valuable. Using data from Euronext-Paris, BPV (2009) conclude that hidden orders are used primarily by uninformed agents to control exposure risk (see also Aitken, Berkman and Mak (2001), De Winne and D’Hondt (2007), and Kumar, Thirumalai and Yadav (2010)). One limitation is that extant research examines all orders submitted to the market, and in such a general setting, it is difficult to detect informed
order flow. We believe the event-study design that we implement isolates informed order flow with greater precision and provide useful evidence on order exposure.
In Table 5, follow
