Does an electronic stock exchange need an upstairs market?

Transcription

1 Does an electronic stock exchange need an upstairs market? Hendrik Bessembinder * and Kumar Venkataraman** First Draft: April 2000 Current Draft: April 2001 * Department of Finance, Goizueta Business School, Emory University, 1300 Clifton Road, Atlanta, GA ** Department of Finance, Edwin L. Cox School of Business, Southern Methodist University, P.O.Box , Dallas, TX , Phone: , kumar@mail.cox.smu.edu We thank Seung Ahn,, Chris Barry, Bill Christie, Jeffrey Coles, Naveen Daniel, Herbert Kaufman, Rex Thompson, and seminar participants at the 2000 FMA Annual Meetings, Arizona State University, Texas Christian University, and Southern Methodist University for valuable comments and discussion. We are grateful to Patricia Ranunkel of Bank Indosuez (Paris) and Marianne Demarchi of the Paris Bourse for information on the upstairs market in Paris.

2 Abstract This paper investigates the costs and benefits of using an upstairs market, where trades are facilitated through search and negotiation, to execute large equity transactions. Though many theoretical models have addressed the role of the upstairs market, empirical research in this area has been limited, mainly due to lack of data. The Base de Donnees de Marche (BDM) database from the Paris Bourse identifies upstairs-facilitated trades and trades in the downstairs (electronic limit order) market. Using data on 92,170 block transactions in a broad cross-section of firms from the Paris Bourse, the paper (a) tests several theoretical predictions on upstairs trading, and (b) investigates the effect of market structure on trading costs. Results indicate that the upstairs market at the Paris Bourse is an important source of liquidity for large transactions. Results support the hypothesis that the role of an upstairs broker is to lower the risk of adverse selection in the upstairs market by certifying a block order as being uninformed.

3 1. Introduction Traders, in particular those wishing to execute large orders, are interested in controlling the risk of order exposure to reduce execution costs. Placing large limit orders to attract counterparties also provides free trading options to the market, and risks being picked off by better-informed traders. Large traders also risk being front-run by other traders, which would increase the market impact of their orders. If traders place large market orders, they potentially face high transactions cost due to the wide effective bid-ask spreads, particularly for orders that exceed quote sizes. This reflects the markets perception that large orders are more likely to reflect informed trading. 1 As a consequence, large traders may prefer to break up large orders into a series of small orders. The small orders could be directed to a centralized market (e.g., NYSE, NASDAQ, Paris Bourse) as limit orders or be executed against public limit orders or market makers' quotes. 2 An alternative used by large traders is to employ the services of a financial intermediary, who specializes in locating counterparties for large transactions and negotiating the trade price. The mechanism of trade facilitation through search and negotiation is frequently referred as the upstairs market, and the financial intermediary as an upstairs (or block) broker. Though an informal upstairs market exists in most financial centers around the world, empirical research in this area has been limited, mainly due to the lack of data. This paper extends our understanding of the role of an upstairs market, focusing in particular on the Paris Bourse, where the upstairs 1 Adverse selection models in microstructure (e.g., Glosten and Milgrom (1985)) suggest that market makers charge a bid-ask spread to compensate for information asymmetry. In a competitive equilibrium, the gain on trades with the uninformed investors just offsets the loss on trades with the informed investors. Easley and O'Hara (1987) demonstrate that an informed investor would prefer to trade larger quantities at any given price to maximize profits. Their model implies that the market makers' spreads would widen for larger trade sizes. 2 In the literature, the continuous intraday markets (such as the NYSE floor) and the batch auction markets (such as the exchange openings) are also referred as the downstairs markets. (See Grossman (1992), Madhavan and Cheng (1997)). Also, a transaction involving more than 10,000 shares is typically referred to as a block trade.

4 2 market competes with an electronic limit order market. We argue that the upstairs market will play a more significant role when it complements an electronic trading system as in Paris, than when the downstairs market includes a trading floor, as on the NYSE, since the trading floor may be able to replicate some of the advantages of an upstairs market. Theoretical models on the upstairs market focus on the informational role of the block broker and can be broadly classified in two categories. The first (e.g., Seppi (1990)) views the block broker as a repository of information on the trading motives of block initiators. In particular, the broker is able to distinguish between informed and uninformed traders. This allows the broker to screen out informed traders from the upstairs market and lower adverse selection costs for large liquidity traders. Evidence from the NYSE (e.g., Chakravarty (2001)) indicates that the specialists and floor brokers are able to deduce the identity of trade initiators, thereby lowering the risk of adverse selection. 3 This is similar to the certification role of the upstairs broker in Seppi (1990). The second category of models (e.g., Grossman (1992)) views the broker as a repository of information on hidden or unexpressed trading interests of large investors. 4 This information increases the effective liquidity during the search process in the upstairs market and provides lower execution costs. Empirical evidence suggests that the NYSE specialist, by virtue of her position at the center of a trading post, has information on unexpressed trading interests on the 3 Benveniste, Marcus and Wilhelm (1992) argue that the long-standing professional relationships between the floor traders and specialists result in information exchange, which can mitigate adverse selection costs. 4 Grossman (1992) suggests that the contingency orders left with brokers are very simple and, hence, very risky to the investor. Therefore, most investors preferences are hidden and not continuously expressed in the downstairs market. Due to long-term trading relationships with institutional investors, the upstairs broker is aware of their current holdings and hidden trading interests. This information is valuable during the search process. 2

5 3 floor of the exchange. This is similar to the role of the upstairs broker in the Grossman (1992). 5 Therefore, the downstairs market in a floor-based market structure may be able to duplicate the information and liquidity advantages of the upstairs market. In contrast, trading in the downstairs market in an electronic trading system is automated and anonymous. An electronic trading system more closely resembles the downstairs market considered by Seppi (1990) and Grossman (1992), suggesting that the upstairs market will play a more significant role in an electronic than in a floor-based trading system. Smith, Turnbull and White (2001) also study the upstairs market in an electronic trading system, the Toronto Stock Exchange (TSE). However, the market structure at the Paris Bourse differs from the TSE on some important dimensions. For example, all firms in the TSE have a designated market maker. In contrast, the liquid stocks at the Paris Bourse do not have a designated market maker, thus making Paris a better laboratory to test predictions of theoretical models. More specifically, this paper extends the existing literature on upstairs markets in the following ways. First, the paper presents evidence on the permanent and temporary price effects of block trades across various types of firms, with different liquidity characteristics. 6 Madhavan and Cheng (1997) focus on the DJIA index firms, while the Smith, Turnbull, and White (2001) analyze price effects of all firms listed on the TSE, without differentiating on firm s liquidity. As theoretical models (e.g., Grossman) predict that the benefits of upstairs intermediation differ 5 In addition, Venkataraman (2001) suggests that the trading rules in a floor-based market structure allow large traders to selectively participate in block trades and better control the risk of order exposure. Hence, large traders are more likely to express their demands in the downstairs market in a floor-based market structure. 6 As first defined in Kraus and Stoll (1972), the permanent component represents the change in the market's perception of the security's value due to the block transaction. The temporary component represents the transitory price movement necessary to provide the liquidity to absorb the block. Some empirical studies in microstructure, such as Huang and Stoll (1996) and Bessembinder and Kaufman (1997), have defined the permanent and temporary components of trades as price impact and realized spreads respectively. 3

6 4 across stocks of varying characteristics, we extend the literature by investigating this issue. Second, the paper investigates the value of relationships and reputation in the trading process by comparing execution costs in the upstairs and downstairs market. Trading in the downstairs market in Paris is anonymous, while the search process in the upstairs market involves reputation effects among market participants. This is in contrast to a floor-based system where reputation effects among participants could affect execution costs in both markets. However, theory suggests that traders will not select the market to execute their order in random. For example, Seppi (1990) suggests that liquidity-motivated traders may prefer to execute their orders in the upstairs market, while informed traders will be unable to, and will therefore use the downstairs market. In this paper, we implement econometric techniques to control for this selection bias to present evidence on the difference in execution costs of a typical order. While the Madhavan and Cheng (1997) paper presents similar evidence in a floor-based trading system, our paper extends the literature by presenting the first selectivity-adjusted comparisons in an automated trading system. Third, we argue that the market structure at the Paris Bourse is more suitable for testing the predictions of theoretical models on the role of the upstairs markets, and we provide direct tests of these models. The theoretical models compare the benefits of a negotiated/signaling mechanism in the upstairs market with a pure auction mechanism in the downstairs market. 7 As discussed above, the downstairs market in a floor-based trading system is not a pure auction mechanism, while the downstairs market at the Paris Bourse is an electronic limit order book that is similar to the market structure in theoretical models. 7 For example, Seppi (1990) describes the difference between the two mechanisms ".. From our perspective, the most significant difference between these two market-making mechanisms is that the dealers negotiate block trades with investors whose identities are known (in the upstairs market). This is in sharp contrast to the comparatively anonymous execution of orders by specialists (in the downstairs market)". 4

7 5 Fourth, we exploit the existence of variations in crossing rules on the Paris Bourse to present evidence on the relevance of these rules. Upstairs trades in most Paris Bourse stocks must be executed at prices at or within the best bid-offer (BBO) quotes in the downstairs market at the time of the trade. However, for a subset of liquid stocks (called eligible stocks), the Paris Bourse allows block trades to be executed at prices away from the BBO at the time of the trade. The possibility of allowing worse execution may open the upstairs market to a broader set of traders. An investigation of the effect of different crossing rules may have interesting implications for off-market trading rules in the United State, especially after the introduction of decimalization, which tightens bid-ask spreads but reduces the relevance of the inside quotes for large traders. We analyze price effects of 92,170 block trades in a broad cross-section of 225 firms. Results indicate that the upstairs market at the Paris Bourse is an important source of liquidity for large transactions. Almost 67% of the cumulative French franc intraday block trading volume is executed in the upstairs market. Test based on a probit model reveals that the likelihood of routing a trade to the upstairs market is higher: (a) during periods of wider spreads and larger order imbalance in the downstairs market, (b) for larger orders, (c) in stocks with higher prices, (d) in stocks that are eligible for the special rules of block trading in the upstairs market, and (e) for a seller-initiated trade. Overall, results suggest that the role of an upstairs broker is to lower the risk of adverse selection in the upstairs market by certifying a block order as being uninformed. The benefits of lower adverse selection are split between the block initiator and counterparties, and both participants are better off by transacting upstairs. The rest of the paper is organized as follows. Section 2 describes the market structure and the crossing rules at the Paris Bourse. Section 3 describes the testable predictions of the 5

8 6 theoretical models of upstairs trading, while Section 4 describes the sample and the distribution of block trades at the Paris Bourse. In Section 5, we present the empirical analysis of the price effects in the upstairs and downstairs market. Section 6 presents the analysis of the block initiator's choice process of the trading venue and the unconditional price effects in the two markets after controlling for selection bias in the data. Section 7 summarizes the results and discusses policy implications for electronic stock exchanges. 2. Market Structure and Crossing Rules at the Paris Bourse The Paris Bourse is an order-driven, electronic, continuous market. In a typical upstairs transaction, an institutional investor (block trade initiator) submits a large order to a member firm (upstairs broker) of the Paris Bourse. Normally, the block initiator has a long-standing relationship with the block broker. Depending on the size of the order and the depth in the limit order book, the broker can (a) send the order to the downstairs market to execute against standing limit orders, (b) act as a dealer (i.e., principal) and execute the block against her own inventory in the upstairs market, or (c) act as a broker (i.e., agent) and search for counterparties in the upstairs market. On most occasions, the block initiator does not indicate the specific market to execute the block order and leaves it to the discretion of the block broker. In France, a block broker acts as a dealer on rare occasions when the customer insists on an immediate execution (see Demarchi and Thomas (1996)). As the block broker deals with numerous institutional investors on a daily basis, she has some information on the current holdings of these parties and their latent trading interests in certain stocks. Based on this information, the block broker contacts the potential counterparties and negotiates the transaction price. During the search process, the identity of the block initiator 6

9 7 is not revealed, though the counterparties are informed about the total size of the block. All upstairs transactions are reported immediately to the Paris Bourse. The Paris Bourse publishes a majority of the upstairs transactions with no delay. Block trades in which a member firm acts a dealer may be reported with delay to enable the member firm to reverse its position. As soon as the upstairs trade is published by the trading system, market participants learn all the details of the upstairs transaction (except whether the member firm acted as a dealer or a broker). The rules on the Paris Bourse for crossing the upstairs trades are less restrictive than the NYSE or the TSE. At the NYSE, the upstairs trade is exposed to the limit order book and the trading crowd for price improvement on both the buy and the sell side of the cross. At the TSE, upstairs trades need to be executed at or within the best bid-offer (BBO) quotes in the downstairs market at the time the order is received. For a majority of stocks listed on the Paris Bourse, upstairs trades need to be executed at or within the BBO quotes in the downstairs market at the time the trade is executed. However, for a subset of liquid stocks (called eligible stocks), the block trading facility at the Paris Bourse allows block trades to be executed at prices away from the BBO at the time of the trade. 8 The exchange rules require that the block trade price must be within the weighted average quotes at the time of the trade. 9 8 For eligible stocks, a trade is classified as a block trade if the trade size exceeds the normal block size (NBS) for that stock. The Paris Bourse defines a NBS for each eligible stock on the basis of its daily trading volume and depth in the limit order book. 9 The weighted average Bid (Ask) gives the weighted average price of executing a market sell (buy) order of order size equal to NBS against the limit order book. Hence, it takes the depth at and away from the inside quotes into account. Please note that the limit order issuers at the Paris Bourse can specify that a portion of their order size be hidden. However, the weighted average quotes do not account for this hidden depth in the limit order book. 7

10 8 3. Theoretical Models on the Role of Upstairs Markets A. Benefits and Costs of Upstairs Trading Theoretical papers on the role of upstairs markets model the benefits and costs of upstairs intermediation in many ways. Grossman (1992) suggests that the informational role of the block broker results in higher effective liquidity in the upstairs market. This higher liquidity reduces the temporary price effect of the block and results in lower execution costs to the block initiator. In the Seppi (1990) setting, a large liquidity trader can credibly signal to the block broker that he is uninformed. This helps avoid the information externality of the downstairs market, and significantly lowers the execution cost. 10 Though the benefits of trading in the upstairs market could be significant, the search process in the upstairs market is costly. If a block initiator is informed, then the no bagging commitment may be too expensive and she may prefer to trade through market orders in the downstairs market (Seppi (1990)). In the Keim and Madhavan (1996) setting, the cost of upstairs facilitation is the brokerage commission, which is an increasing function of the number of counterparties located. In the Burdett and O'Hara (1987) setting, the cost of upstairs trading is the leakage in the downstairs market due to front-running strategies by some upstairs participants. 11 In the Grossman (1992) setting, the cost of upstairs trading is the extra volatility (price uncertainty) of trading in a decentralized market where prices are not clearly visible. To summarize, a trade initiator selects the trading mechanism through which to execute his block on the basis of expected execution costs in each mechanism. 10 For example, the signal could be in the form a 'no bagging' commitment i.e. the block initiator commits to not trade in the stock until the counterparties have traded the acquired position off their inventory. This commitment is relatively inexpensive to a liquidity trader, but quite costly to an informed trader. 11 Since block trades have predictable price effects, a trader can make quick gains (on average) by trading ahead of a block trade i.e., purchase the asset before a block buy or sell the asset before a block sell. This strategy is frequently referred to as 'front-running'. 8

11 9 B. Testable predictions on block trading The theoretical models in the literature offer a rich set of testable prediction: 12 Hypothesis I: Burdett and O'Hara (1987) offer an explanation for the Kraus and Stoll (1972) result that blocks are sold, not bought in the US. They conjecture that the asymmetry in the number of trades may be due to an asymmetry in the motivations of buyers and sellers. The seller of a large block may be trading on superior information, but may also be trading for liquidity reasons. In a block buy, however, such liquidity trading may be harder to justify. Hence they argue that informational content of a block buy would be larger than a block sell. Their conjecture results in the following testable predictions: (a) block sells will be observed more frequently than a block buy; (b) a block buy will have a lower probability of being routed to the upstairs market; and (c) the information content of a block buy will be larger than the information content of a block sell. Hypothesis II: For a given firm, larger trades are more likely to be executed in the upstairs market. Adverse selection models, such as Easley and O'Hara (1987), predict that the cost of information externality increases with order size. Hence, a large liquidity trader has a larger incentive than a small liquidity trader to use the services of an upstairs broker (Seppi, 1990) and avoid the adverse selection cost. Similarly, the potential liquidity benefits of upstairs facilitation (Grossman, 1992) could outweigh the search costs for large orders. To test this hypothesis, I calculate the proportion of trades (and trading volume) executed upstairs for each stock as a function of trade size categories. 12 While Burdett and O'Hara (1987) focus on the permanent price effect in the upstairs market, Keim and Madhavan (1996) model permanent and temporary price effects in the upstairs market. Grossman (1992) and Seppi (1990) model the block initiator's cost and benefit of executing trades in upstairs and downstairs markets. Grossman (1992) analyzes the temporary price effects, while Seppi (1990) analyzes the permanent price effects in both markets. 9

12 10 Hypothesis III: For a given trade size, firms with less liquidity in the downstairs market will have a higher level of upstairs participation. This hypothesis follows from Grossman (1992) who shows that the degree of upstairs facilitation will be higher if the perceived benefits of upstairs trading (in the form of execution costs) are higher. Hypothesis IV: Burdett and O'Hara (1987) suggest that the counterparties are not aware of the total size of the block and hence, underestimate the informational content of the block trade, i.e., they lose money on average on these trades. 13 These predictions are consistent with the findings of the 1971 SEC study on institutional investing, which found that the counterparties incurred trading losses on their transactions. In contrast, Keim and Madhavan (1996) predict that the counterparties are aware of the total block size and the informational content of the trade, i.e., they do not lose money on average on these trades. While the data does not provide specific information on the extent to which block size is revealed during the facilitation process, the issue of accurate pricing can be addressed by analyzing the compensation received by the upstairs counterparties i.e., the liquidity component of the block trade in the upstairs market. Hypothesis V: Seppi (1990) predicts that conditional on a block trade being routed to the upstairs market, the permanent price effect will be smaller than that of a trade sent to the downstairs market. The no bagging commitment in the upstairs market is costly to an informed investor, who may prefer to trade using anonymous orders in the downstairs market. Since orders to the upstairs market have less probability of being initiated by an informed trader and market participants are aware of the screening mechanism, the information content of an upstairs block 13 A puzzling issue is why counterparties would continue to trade despite making trading losses. Burdett and O'Hara (1987) counter this argument with the following (a) the large trade size results in lower transactions cost in the upstairs market that compensates for the mis-pricing, (b) an investor who desires a position in the stock for portfolio reasons would prefer to be a liquidity supplier rather than a liquidity demander, and (c) investors may be risk averse and prefer a known block price to an unknown price in the future. 10

13 11 trade is expected to be lower. Hypothesis VI: Grossman (1992) predicts that the (unconditional) temporary price effect of a block trade would be lower in the upstairs market relative to the (unconditional) price effects of a same trade in the downstairs market. The block broker's role as a repository of information about unexpressed demand increases the effective liquidity in the upstairs market and reduces the temporary price effect. Hypothesis VII: Proposition 1 in Keim and Madhavan (1996) states that the temporary price effect in the upstairs market is positive for buys and negative for sells. If the marginal search costs are increasing in order size, the absolute temporary effect is an increasing and strictly concave function of trade size. The concavity arises because the block broker, at the margin, may choose to search for more counterparties for the block instead of making a concession on the block price. As the search cost depends on the probability of locating counterparty, and the probability would differ across firms, this hypothesis is testable using time-series data on each firm. 14 I estimate the following regression model for each stock and report mean/median regression coefficients across stocks. Model 1: τ(q) = β 0 + β 1 * Q + β 2 * Q 2 + β 3 * I + ε; for each stock (6) where τ(q) = ln(p b /P 1 ) is the temporary impact measure in return form, and I is a dummy variable which equals 1 for a buyer-initiated trade, and 0 otherwise. P b refers to the block trade price while P 1 refers to the post-trade price of the asset in the downstairs market. The hypothesis predicts that β 1 > 0 and β 2 < 0. Hypothesis VIII: Proposition 4 of Keim and Madhavan (1996) predicts that the permanent price effect in the upstairs market is positive for buys and negative for sells. Also, the absolute 14 Due to data constraints, Keim and Madhavan (1996) test model predictions by pooling a cross-section of stocks. 11

14 12 permanent component in the upstairs market is an increasing and concave function of order size. To test this proposition, I specify the following regression Model 2: P(Q) = β 0 + β 1 * Q + β 2 * Q 2 + β 3 * I + ε ; for each stock (7) where P(Q) = ln(p 1 /P d ) is the permanent impact measure in return form. P 1 measures the posttrade value of the asset, while P d measures the value of the asset before the leakage of information of a block trade. The hypothesis predicts that β 1 > 0 and B 2 < 0. Hypothesis IX: Proposition 2 in Keim and Madhavan (1996) is as follows: For a given order size, the temporary price component is positively related to the cost of locating counterparties, the degree of risk aversion, and the variance of the risky asset's return. This proposition has testable implications across a cross section of firms. We might expect the cost of locating counterparties to be higher for firms with less liquidity. Also, Keim and Madhavan (1996) suggest that short sale constraints and the difficulty in locating traders with large holdings of a particular asset could result in asymmetric temporary effects for block buys and sells. For different trade size categories, I calculate the time-series average of the temporary price effect of buy and sell trades for each stock. Next, I estimate the following regression for a given trade size: Model 3: τ i = β 0 + β 1 * Liquidity i + β 2 * Volatility i + β 3 * I + ε i, (8) where t i is the average temporary price effect of block trades for the stock i in the trade size category, NBS is the normal block size, and volatility is the variance of daily stock returns using quote midpoints. The model predicts β 1 < 0, β 2 > 0 and β 3 > 0. Hypothesis X: Burdett and O'Hara (1987) and Keim and Madhavan (1996) predict that the search process in the upstairs market would result in leakage effects in the downstairs market before the block trade is announced, as some traders would indulge in front-running strategies to earn quick 12

15 13 profits. The leakage price effect is expected to be positive for block buys and negative for block sells. Further, the extent of leakage would increase with the number of counterparties contacted and time taken for facilitation. Hence, I estimate the following regression for a given trade size: Model 4: L i = β 0 + β 1 * Liquidity i + β 2 * Volatility i + β 3 * I + ε i, (9) where L i = ln(p 0 /P d ) is the average leakage measure of block trades in the stock i in the trade size category. P 0 is the price of the asset before a block trade while P d measures the price of the asset before the leakage of information of a potential block trade. The search may be easier for more liquid and less volatile firms, and for block sells relative to block buys i.e., (β 1 < 0), (β 2 > 0) and (β 3 > 0). 4. Sample selection and the distribution of the block trading volume 4.1. Sample selection As our objective is to investigate the significance of an upstairs market across a broad cross-section of firms, we select the component firms of the SBF-250 Index on April 1, 1997 as the initial sample. SBF250 represents all sectors of the French economy and includes all component firms of the CAC40 and SBF-120 indexes, which represent larger and more liquid firms at the Bourse. Trade and quote data on the sample are obtained from the Base de Donnees de Marche (BDM) database. To remain in the sample, a firm must (a) not trade in a batch auction market, as prices in the downstairs market will not be continuously available for calculating price effect (deletes 13 firms), (b) trade common equity with voting rights (deletes 5 firms), (c) have some trade and quote data during the sample period (deletes 2 firms), and (d) have normal trade 13

16 14 and quote behavior (deletes 5 firms). 15 After these deletions, the final sample size is 225 firms. The sample period is from April 1997 to March The sample is further divided into liquidity quintiles based on the average daily trading volume during the sample period. We focus on the price effects of large transactions that occurred during the regular market trading hours. While this approach considerably reduces the transactions analyzed, it has many advantages. First, theoretical models (e.g., Seppi (1990)) suggest that the role of the upstairs market is to provide liquidity for large orders. By focusing on large trades, we directly test the predictions of theoretical models. Second, in order to understand the factors that affect the choice between the two markets, we need to restrict our analysis to large transaction executed when both markets are open (i.e., normal trading hours). Third, it is easier to measure price effects of block trades when the downstairs market is open. Fourth, our approach follows Madhavan and Cheng (1997), who analyze block transactions in the NYSE during the regular trading hours. Hence, at a very preliminary level, this approach may allow us to understand the role of upstairs markets across exchanges with different market structures Definition of block trade Following the definition of a block trade in the NYSE, the empirical literature typically defines a transaction greater than 10,000 shares as a block trade. However, our intuition suggests that the block size should differ across firm characteristics such as trading volume, stock price, depth in the limit order book, etc. 16 We extend the block trading literature by defining a block 15 Some firms have large quoted depth on only one side of the market for many months, and subsequently delist. During this period, trades only occur on the deeper side of the market. It is possible that professional market makers may be providing price support for these stocks before delisting occurs. These firms will introduce bias in my analysis, as there are no quote revisions subsequent to a trade for many months. Names of firms available from the authors on request. 16 Angel (1997) shows that the distributions of stock prices are very diverse across world markets. For example, on April 1, 1997, the average stock price at the Paris Bourse and the NYSE are FF 800 (or U.S. $142) and $41 respectively. Also, the stock prices are more widely dispersed at the Paris Bourse than the NYSE. Since traders are 14

17 size for each firm. First, for firm i, we calculate the average market price, average daily downstairs trading volume and the average depth on the inside quotes in the limit order book for 15 month m. Next, we define block size as NBS i,m = MAX [NBS 1, NBS 2, NBS 3 ], where (a) NBS 1 = 7.5 * (Average depth on the inside quotes in the limit order book), (b) NBS 2 = 2.5 % of (Average daily downstairs trading volume), and (c) NBS 3 = FF 500,000 / Average Price, i.e., the block trade size is at least FF 500,000. The NBS for a quarter is based on the average value of NBS i,m for the previous quarter. 17 To minimize the effect of abnormal trading activity in a single month, we identify firms where the absolute change in NBS from one quarter to the next is greater than 100% (14 observations). If the change in NBS is consistent with a permanent change in trading activity, we do not make any corrections of the block size (3 occasions). If the change is due to a stock split, then we change the NBS on the day in which the split is effective (3 occasions). If the increase in NBS is due to abnormal trading behavior in a single month, then we retain the NBS from the previous quarter (8 occasions). 18 To qualify as a block transaction, the transaction size (in shares) must exceed the NBS of the security, and the transaction size (in FF) must exceed FF 500,000. Finally, we use filters to delete trades and quotes that have a high likelihood of reflecting errors. 19 more concerned about the dollar (or FF) size of the trade, it is reasonable to suggest that the block size (in shares) at the Paris Bourse should be smaller than the NYSE. 17 The Paris Bourse uses a similar algorithm to define the block size for the 74 firms that are eligible stocks for the special rules of block trading. As a robustness check, we compare our measure of NBS for the eligible stocks with block sizes provided by the Paris Bourse, and find a correlation of Hence, for the eligible stocks, we use the block size measures provided by the Paris Bourse. 18 For stock splits, my adjustments follow the procedure used by the Paris Bourse. An analysis of NBS for eligible stocks indicates that the Bourse ensures that the change in trading activity is permanent before announcing a change in block size. Also, during the sample period, the Bourse did not change the NBS of any eligible stock by greater than 100%. 19 Trades are omitted if (a) trade price is non-positive (b) involves a price change (since the prior trade) greater than absolute value of 25% (c) occurs on a day when change in overnight price is greater than 15% (d) occurs on the day of stock split. Quotes are deleted if (a) bid or ask is non-positive (b) bid-ask spread is negative (c) change in bid or ask price is greater than absolute value of 10% (d) bid or ask depth is non-positive. 15

18 Distribution of the block trading volume Table 1 presents the sample summary statistics and the distribution of the block trading volume during regular trading hours in the upstairs and downstairs market. All sample firms are classified into liquidity quintiles. The average stock price and market size of the sample on April 1, 1997, is FF 799 and FF 13,544 million respectively. There is no clear relationship between stock price and market liquidity. As expected, the average market size increases monotonically from FF 1,614 million for the least liquid quintile to FF 48,670 million for the most liquid quintile. During the sample period, there were 92,170 block trades. Of these, 31,088 (33.7%) were executed in the upstairs market. The size of an average block trade in the upstairs market is FF 11.5 million, while the size of an average block in the downstairs market is only FF 2.9 million. The substantial difference between mean and median trade sizes suggests that some trades in both markets are very large. The sub-sample analysis, by liquidity quintiles, provides similar results. As expected, the number of trades, the average trade size and trading volume show an upward trend with stock liquidity. 20 Results suggest that the upstairs market at the Paris Bourse is a significant source of liquidity for large transactions. Almost 67% of the cumulative block trading volume is executed in the upstairs market. In the TSE, Smith, Turnbull and White (2001) study all transactions (block and non-block) and report that 56% of the total volume is executed upstairs. In the NYSE, Hasbrouck, Sofianos and Soseebee (1993) report that 27% of the block volume in all NYSE-listed stocks was executed in the upstairs market, while Madhavan and Cheng (1997) find that 20% of the block volume in the DJIA index stocks was executed in the 20 One potential concern is whether our definition of NBS is biasing the results of trade distribution. However, the results are similar in quintile 4 and 5, where we use the block size provided by the Paris Bourse. 16

19 17 upstairs market. At a preliminary level, the results are consistent with our conjecture that the upstairs market may play a more significant role in an electronic stock exchange. Table 2 presents the distribution of buyer and seller-initiated block trades. Trades are classified as buyer and seller-initiated using the Lee and Ready algorithm. 21 Results suggest that blocks are bought and sold with the same frequency during the sample period in the Paris Bourse. This result differs from the empirical finding in the U.S. market (e.g., Kraus and Stoll (1972)) that blocks are sold with a higher frequency. Consistent with Burdett and O Hara (1987) prediction (Hypothesis I), we find that a higher proportion of seller-initiated trades (36%) are executed in the upstairs markets as compared to buyer-initiated trades (28%). The analysis by trade size indicates that upstairs trades, regardless of direction of order initiation, are three to four times larger than downstairs trades. Results in Table 3 strongly support the conjecture that larger traders are more likely to be executed in the upstairs market (Hypothesis II). Trades are classified as (a) small if (NBS trade size< 2*NBS), (b) medium if (2*NBS trades size< 5*NBS), and (c) large if (trade size 5*NBS). For small block trades, only 20% of the trades and cumulative trading volume is executed in the upstairs market. However, for large block trades, almost 80% of the trades, and 87% of the cumulative block trading volume is executed upstairs. 22 It is surprising to note that the upstairs participation rate, in all trade size categories, is highest for firms with the highest liquidity in the downstairs market. One potential explanation is that block initiators are 21 The classification is crosschecked using order-level data for the downstairs trades, and found to be error-free. This is not surprising as downstairs trades cannot be executed within the inside quotes i.e. cannot be price-improved. However, it is possible that some upstairs trades may be misclassified. To minimize errors, we delete 3,098 upstairs trades that occurred at quote-midpoints. Please note that the deletion of these low-cost trades will increase the average execution costs in the upstairs market. 22 A potential concern is the effect of order splitting strategies in both markets. Unfortunately, the data set does not allow us to identify such trading strategies. 17

20 18 concerned about the market impact of their orders in the downstairs market, as the block size of the higher quintile is significant. Hence, they route the larger orders to the upstairs broker to lower execution costs. Results in Table 4 support the conjecture that firms with less liquidity in the downstairs market have a higher level of upstairs participation (Hypothesis III). First, trades are classified into trade size (in FF) quintiles, and all trades in the same quintile are assumed to be of similar size. Next, within each trade size quintile, we calculate the upstairs participation rate for each firm and report median upstairs participation rates for firms in the same liquidity quintile. Results indicate that within a trade size quintile, upstairs participation rate increases for less liquid firms. For example, in trade size quintile 3, the upstairs participation rate increases from 19.9% for firms in the most liquid quintile to 57.1% for firms in the least liquid quintile. The results also suggest that it is important to control for firm liquidity when defining a block trade. 5. Price effects in the upstairs and downstairs markets 5.1. Empirical measures of price effects As first defined in Kraus and Stoll (1972), the temporary and permanent component of the price change measures the liquidity and informational effects of a block trade. Figure 1 provides a graphical representation of the expected price effects of a block buy order. The temporary component (τ(q)) represents the compensation to the liquidity providers (i.e., counterparties) and can be measured by the amount of price reversal after the block trade i.e. τ(q) = ln(p b ) - ln(p 1 ), where P b is the block trade price and P 1 is a measure of the post-trade price of the asset We proxy for P 1 using four different measures: (1) the mid-point of the first quote reported 30 minutes after the block trade, (2) first quote mid-point reported after 12:00 noon on the next trading day (i.e., t 1 ), (3) mid-point of the 18

21 The permanent component (P(Q)) has two components - post-trade impact and leakage effect. The post-trade impact (π(q)) represents the change in the market's perception of a 19 security's value after the announcement of the block trade, i.e., π(q) = ln(p 1 ) - ln(p 0 ), where P 0 is the pre-trade value of the security and is proxied by the quote midpoint reported before the announcement of the block trade, i.e., at t 0. The leakage effect (L(Q)) represents price movements in the downstairs market while the block is being facilitated (or shopped ) in the upstairs market. i.e., L(Q) = ln(p 0 ) - ln(p d ), where P d is the security value when the upstairs broker initiates the search process. 24 Finally, the total execution cost (T(Q)) represents the total cost of trading to a block initiator and is the sum of the liquidity and information effect, i.e., T(Q) = P(Q) + τ(q) = ln(p b ) - ln(p d ). All measures are expected to be positive for a block buy and negative for a block sell. The price movements are adjusted for market movements by subtracting the SBF120 index's market return from the stock's return, without any adjustment for the stock's market beta Price effects in the upstairs and downstairs market Table 5 and Table 6 present execution cost measures of seller and buyer-initiated block trades. Since these price effects are univariate measures, it may be important to remember that the average size of an upstairs trade is three to four times larger than the average size of a downstairs trade. For a seller-initiated trade (Table 5), the average execution cost (with a pretrade price movement of 3 days) is 59.4 basis points (bp) in the upstairs market and 73.7 bp in closing quotes on the next trading day (i.e., t 1 ) (4) mid-point of the closing quote on the 3 rd trading day after the block trade (i.e., t 3 ). As the results are similar, we report price effects using mid-point of the closing quotes on the next trading day. 24 We proxy for P d using three measures: (1) the mid-point of the quotes reported 30 minutes before the trade, (2) the mid-point of the closing quotes on the trading day before the block trade (t -1 ), and (3) the mid-point of the closing quotes on the 3 rd trading day before the block trade. In the tables, we present the results using (2) and (3). Demarchi and Thomas (1996) suggest that most block orders in France are facilitated within a day. 19

22 20 the downstairs market. The break-up of total trading costs among the permanent and temporary price effects is extremely interesting. Across all liquidity quintiles, the information content of an upstairs trade is significantly lower than the informational content of a downstairs trade. On average, a seller-initiated trade lowers prices by 11 bp in the upstairs market and 57 bp in the downstairs market. However, compensation to the counterparties (measured by the temporary price effect) is larger in the upstairs market (48.4 bp) than in the downstairs market (16.7 bp). In both markets, average trading costs are lower for stocks with higher liquidity. For a buyer-initiated trade (Table 6), the benefit of executing a trade in the upstairs market is significantly larger. The average execution costs (with a pre-trade price movement of 3 days) are 65.9 bp in the upstairs market and bp in the downstairs market. Execution costs in the upstairs market are lower by at least 50 bp across liquidity quintiles (except quintile 4). An analysis of the break-up of total trading cost clearly indicates that the cost advantage in the upstairs market originates from the lower adverse selection component of execution cost. The permanent price effect is significantly lower in the upstairs market (64.2 bp) relative to the downstairs market (156.8 bp), and this trend holds across liquidity quintiles. Interestingly, counterparties in the upstairs market receive insignificant compensation for their services, while counterparties in the downstairs market lose money on average (-37.6 bp), i.e., subsequent to the block buy in the downstairs market, the stock price moves higher than the price at which the block was executed. Collectively, results in Table 5 and Table 6 support the Burdett and O'Hara (1987) conjecture that informational content of a block buy would be larger than a block sell (as discussed in Hypothesis I). The temporary price effects provide evidence that the counterparties to block transactions in the upstairs market in Paris do not make trading losses during the sample 20

23 21 period. 25 It is difficult to say whether the counterparties have estimated the informational content of trade accurately, as counterparties could be earning lower than expected revenues. Nevertheless, it is encouraging to observe that counterparties do not make trading losses. Since prices are negotiated in the upstairs market and participation is voluntary, it may be difficult to explain trading losses as a long-term equilibrium. To summarize, a block trade initiator incurs lower trading costs in the upstairs market than in the downstairs market. This result holds across most liquidity quintiles, and the execution cost advantage of upstairs market is significantly larger for a buyer-initiated trade. An analysis of the components of execution cost provides strong support of the certification role of the upstairs broker (Seppi, 1990). While trades in both markets contain information, the adverse selection component of execution cost is significantly lower in the upstairs market relative to the downstairs market (consistent with Hypothesis V). This suggests that liquidity traders may prefer to use the services of an upstairs broker and obtain lower trading costs. Also, as counterparties receive larger compensation in the upstairs market, large liquidity providers may prefer to provide liquidity in the upstairs market and lower the risk of adverse selection. Table 7 contains summary statistics on execution costs in the two markets by trade size categories. Results indicate that execution costs are lower in the upstairs market in most trade size categories and the cost differential is larger for block buys than block sells. As discussed earlier, the permanent price effect is significantly lower in the upstairs market and is consistent with the certification role of the block broker. Also, the informational content of a block buy is significantly larger than a block sell. 25 This result differs from the 1971 SEC study on institutional investing that found that counterparties incurred trading losses on these types of transactions (as discussed in Hypothesis IV). 21

24 Regression analysis of the price effects Results of a regression of the temporary and permanent price effects in the upstairs market on transaction size are presented in Table 8. This analysis is restricted to a sub-sample of 132 firms that have at least 30 upstairs trades during the sample period. Regression coefficients are estimated for each firm and the mean and median values of regression coefficients across the sample of firms are reported. The trade size variable normalized by the NBS to control for intertemporal changes in liquidity during the sample period. The buy/sell dummy variable measures the asymmetric effects of block buys and sells. Results in Panel A provide weak evidence on a concave relationship between temporary price effect of upstairs trades and trade size (as discussed in Hypothesis VII). The coefficients are of the predicted sign i.e., positive on the trade size variable and negative on the squared trade size variable. However, the coefficient on the squared trade size variable is statistically insignificant. The coefficient on the buy/sell dummy variable suggests that the counterparties earn lower compensation for participating in buyer-initiated trades. Results in Panel B do not support a concave relationship between absolute permanent price effect of upstairs trades and trade size (as discussed in Hypothesis VIII). The coefficients on trade size and squared trade size are statistically insignificant, which suggests that information content of a trade does not increase with trade size. However, results are consistent with the Seppi (1990) model that order size may provide little additional information on the risk of adverse selection in the upstairs market, as the identity of the block initiator is known. Also, consistent with Burdett and O Hara (1987) (Hypothesis I), we find that the information content of a buyer-initiated trade in the upstairs market is significantly higher than the information content of a seller-initiated trade. 22