Informed Trading in Parallel Auction and Dealer Markets: The Case of the London Stock Exchange Pankaj K. Jain University of Memphis Christine Jiang University of Memphis Thomas H. McInish University of Memphis and Nareerat Taechapiroontong Mahidol University, Thailand June 2004 JEL classification: G10; G14; G15 Key words: Anonymity; Informed trading; Price impact; Multi-market trading; London Stock Exchange 1
Informed Trading in Parallel Auction and Dealer Markets: The Case of the London Stock Exchange Abstract In this paper, we examine trading on the London Stock Exchange where the same stocks trade side-by-side on the computer-based Stock Exchange Trading System (SETS) and the Dealer market. Both markets are successful. SETS has more trades overall, but the Dealer market has larger trades and dominates the execution of very large sized trades. Dealers are able to screen out informed orders as evident from the fact the permanent price impact of trades is much larger on SETS than on the Dealer market. However, the temporary price impact of trades is significantly larger on the Dealer market than on SETS, indicating that liquidity can be purchased on the Dealer market, but at a significant cost. 2
1. Introduction We study trading on the London Stock Exchange where the same stocks trades on the computer-based Security Execution Trading System (SETS) and the Dealer market. The LSE differs from other major markets in a number of additional ways. On the New York Stock Exchange (NYSE), a specialist has an obligation to maintain narrow spreads and stabilize prices. Upstairs brokers must expose negotiated trades to the downstairs floor and to the order book. In Paris, only trades over a given size threshold can be executed off-book, and only at or within the weighted average book price. In contrast, dealers on the Dealer market have no affirmative obligation to offer quotes and there is no requirement for interaction between the two markets. Nevertheless, the dealer markets play an important role in matching large trades, a function similar in nature to the upstairs market in New York. Thus they nicely complement the SETS, which is like the SOES system on Nasdaq, in terms of automated execution and lack of pre-trade negotiations or interactions between counterparties. The hybrid structure of the LSE has not yet been examined in the context of informed trading. although other existing studies on the LSE that use comparable data analyze the interaction of order flow between the two markets (see Ellul, 2001, and Friederich and Payne, 2001,) and the dynamics of market open and close (see Ellul, Shin, and Tonks, 2002). We provide a number of findings. We find that both the SETS and the Dealer markets are successful with the SETS market executing more trades, but the Dealer market executing larger orders overall and dominating the execution of very large orders. Dealers are able to screen out informed orders by posting wider quotes so that the permanent price impact of trades is much larger on SETS than on the Dealer market. The 3
permanent price impact of trades is not only significantly lower on the Dealer market than on SETS, but is actually negative on the Dealer market. This provides strong evidence that dealers on the Dealer market are able to effectively identify informed trades as predicted by the theoretical models of Easley Kiefer and O hara (1996), Seppi (1990), Pagano and Roell (1992), Benveniste, Marcus, and Wilhelm (1992). However, the temporary price impact of trades is significantly larger on the Dealer market than on SETS, indicating that liquidity for large trade sizes can be purchased on the Dealer market, but at a significant cost. 2. Trading systems and venues 2.1. Competition between trading systems Madhavan (1995) proposes a model providing a rationale for the existence of fragmented markets, focusing on the impact of disclosing trading information to market participants. He shows that informed traders and large traders who place multiple trades obtain lower expected trading costs in fragmented markets where their trades are not disclosed. On the other hand, dealers benefit from nondisclosure by decreasing price competition. Therefore, fragmented markets will not necessarily integrate into one market. Contrary to this view, Chowdhry and Nanda (1991) provide a theoretical model in which if more than one market for a security exists, one market will emerge as the dominant market, a winner takes most phenomenon. This prediction occurs as liquidity traders seek thick markets with the lowest execution costs and informed traders maximize their profits by hiding trades in the most liquid markets. Glosten (1994) suggests that the open electronic limit order book such as SETS is inevitable because it provides as much 4
liquidity as possible in extreme situations and does not invite competition from third market dealers, while other trading institutions do. Based on the discussion in this section we test the Market Domination Hypothesis, which states that trading in a given asset concentrates on a dominant market. Alternately, the Market Co-existence Hypothesis states that trading in a given stock can co-exist in different trading venues, possibly due to specialization of each venue. 2.2. Co-existing upstairs and downstairs markets Another branch of literature on market fragmentation allows for coexistence of parallel markets and the discussion centers around trading between upstairs and downstairs markets. Easley and O Hara (1987) develop the information-based trading model to explain why large (block) trades are made at less favorable prices than small trades. In addition, these block trades have persistent price effects. In particular, transaction prices are lower after block sales and higher after block buys, with only a partial price reversion to their previous levels. Seppi (1990), extending the work of Easley and O Hara (1987), develops a framework in which dealers (such as upstairs block traders in New York or the Dealers in London) are able to differentiate uninformed traders from informed traders based on reputation signals or other implicit commitments. This non-anonymous feature in upstairs markets enables the dealers to screen out informed traders from the upstairs market, and, as a result, lowers adverse selection costs for large liquidity traders. He argues that the lack of anonymity in off-exchange block trading enables investors and the 5
dealers to make no bagging the street commitments and face penalties on any subsequent trades if they fail to divulge information. Grossman (1992) also focuses on the information role of upstairs (fragmented) and downstairs (centralized) markets. He suggests that many large traders do not want to expose their orders to the public since such large trades may adversely impact the market price, may invite front running by other traders, and may introduce a free option problem (the risk of being picked off if market conditions change). A large order sent to the upstairs market is less exposed than one sent to the downstairs market and may be matched with other unexpressed liquidity. As a consequence, upstairs dealers serve as a repository of information on large investors hidden trading interest. Another model focusing on asymmetric information is proposed by Easley, Kiefer, and O Hara s (1996), hereafter, EKO. The authors show that the practice of cream skimming by dealers or trading locales supports the existence of off-exchange trading or market fragmentation. Dealers in off-exchange locales mitigate losses from trading with informed traders by purchasing retail order flow 1 or seeking only uninformed trades, and, as a consequence, diverting the remaining informed trades to the primary market. Thus, cream skimming permits uninformed traders in off-exchange markets to benefit from lower costs. Keim and Madhavan (1996) develop a theoretical model of the upstairs market where order size, beliefs, and prices are determined endogenously. They show that information sharing and risk sharing among traders in the upstairs market can reduce 1 Purchased order flow refers to the practice of dealers or trading locales paying broker for retails order flow. In general, this purchased order is guaranteed to be executed at the best prevailing price. 6
price impact. They also suggest that the benchmark price to calculate the permanent price impact should include the period prior to trade as a result of the information leakage. Empirical support for these predictions is provided by Madhavan and Cheng (1997) and Bessembinder and Venkataraman (2004). 2.3. Co-existing auction and dealer markets Economides and Schwartz (1995) and Schwartz and Steil (1996) surveyed institutional traders in North America, Europe, and Australia, and found that institutional investors prefer anonymous automated execution systems that provide low disclosure of identity of the trader submitting the order. Pagano and Roell (1992) provide a discussion of the relative benefits of auction and dealership markets with respect to the degree of transparency. Similar to the upstairs markets literature, they argue, negotiated dealership markets offer opportunities for screening of informed traders. This implies that dealers will trade with the uninformed while informed traders will trade in the order book or pay a bigger premium for trading large quantities with dealers. Pagano and Roell (1996) propose that increasing the transparency of the trading system can decrease trading costs. Similarly, Forster and George (1992) show that disclosure of the direction and size of liquidity trades in advance of trading can reduce the expected transaction costs of liquidity motivated traders, providing a motivation for so-called sunshine trading. Benveniste, Marcus, and Wilhelm (1992) conjecture that in non-anonymous specialist market structure the floor brokers have to repeatedly interact with the specialist who is able to ex-post identify the brokers with information-based trades. The specialist also has the means to sanction those brokers who fail to reveal information-based trades by not 7
providing better trade prices, refusing to fill orders above the quoted depth, or unwillingness to help work a large order. The predictions of the theoretical models of Seppi (1990), Grossman (1992), Easley, Kiefer, and O Hara (1996), Pagano and Roell (1992, 1996), and Benveniste, Marcus, and Wilhelm (1992) have been empirically verified by De Jong, Nijman, and Roell (1996), Smith, Turnbull, and White (2001), Fong, Madhavan and Swan (2001), and Booth, Lin, Martikainen, and Tse (2002) in the context of upstairs trading on the Paris Bourse, the Toronto Stock Exchange, the Australian Stock Exchange, and the Helsinki Stock Exchange, respectively. However, the unique market structure of the London Stock Exchange has not been studied in this context. The studies on these other exchanges find that the permanent price impact and adverse selection costs in negotiated upstairs markets are very low. They explain that off-exchange trading is not anonymous, and that asymmetric information plays less of a role in that market because of its ability to screen out the information motivated trades. In contrast, the downstairs market is an anonymous electronic order book, and, therefore, vulnerable to adverse selection problems. Nevertheless, the downstairs markets provide lower total execution cost for small trades. The existence of off-exchange and upstairs market provides a market with more efficient trading in terms of liquidity and do not harm the anonymous downstairs market. We examine whether the Dealers have relinquished their traditional role of primary quotebased markets and instead stepped into the shoes of becoming an upstairs market. Franke and Hess (2000) propose that the information differential between an anonymous screen-based trading system and a non-anonymous floor trading system should increase the attractiveness of the latter in the times of high information intensity. 8
Consistent with their hypothesis, they show that the order book market s market share is decreasing in trading volume and price volatility. We study these relationships in our cross sectional regression. Two recent empirical studies are closely related to ours. Heidle and Huang (2002) investigate whether auction markets (NYSE, AMEX) or dealer markets (NASDAQ) are better able to identify informed traders. Gramming, Schiereck, and Theissen (2001) examine the relation of degree of trader anonymity and the probability of informed trading on the two parallel markets at the Frankfurt Stock Exchange. Both these studies are based on the concept that the non-anonymous environment permits market makers to draw inference about the motives behind trades. They find that the probability of informed trading is lower on non-anonymous floor-based trading markets and directly related to variables that proxy for the degree of anonymity such as spread and adverse selection components. The decrease in spread is greater for firms with higher probability of informed trading prior to transferring from a dealer to an auction market. They conclude that the differences in the market structure result in the differences in risk of informed trading as informed traders prefer pre-trade anonymity. SETS market offers such an environment. In contrast, dealers on the Dealer market can engage in pre-trade negotiation. Such interactions with counterparties enable the dealers to better understand their motives and thus screen out informed traders. These observations on coexisting upstairs and downstairs markets as well as auction and dealer markets lead us to test the Information Screening Hypothesis, which states that with the co-existence of anonymous SETS and dealer markets, the dealer market is able to screen out informed trades so that these trades are routed to the 9
automated market. We test this hypothesis by looking the post trade permanent price impact and expect the impact to be higher for the SETS market. 2.4. Liquidity in auction and dealer markets As addressed in Seppi (1990) and Grossman (1992), off-exchange dealer markets involve a process of searching and matching of order flows. Since temporary price concessions are needed to induce dealers to accommodate orders due to inventory holding risk, and SETS is fully automated, additional liquidity cannot be negotiated by the offer of price concessions. Hence, temporary price impacts should be larger for an off-exchange dealer market than for an anonymous market. We call this the Liquidity Hypothesis which states that the temporary price impact on an anonymous auction market is less than that of a dealer market. In addition, for small sizes, dealers may not offer significant price improvement because of less intensive competition. For large orders, dealers have the ability to extract additional information about the motivation behind the order flow and benefit from such reputation effects, and significant price improvements are often achieved as a result. Thus, temporary price impact is a decreasing function of trade size for orders on the Dealers market. Bernhardt et al. (2004) also develop a model along these lines where larger orders have lower transaction costs on LSE. 3. Institutional background, data, and methodology 3.1. The London Stock Exchange The London Stock Exchange (LSE), which is one of world s leading stock exchanges, has experienced significant transformation to maintain and compete for order 10
flow and to improve price discovery. Before October 1997, the LSE was a pure quotedriven dealer market (SEAQ) that was relatively nontransparent about the order flow. There were no reports of collusion, but order flow was concentrated among five large market makers, and, consequently, there was dissatisfaction among traders. Moreover, retails investors complained that they were subsidizing large traders. These problems caused order flows to migrate to other European markets. In 1997, the LSE began to implement a phased introduction of a more transparent order-driven auction market called the Stock Exchange Electronic Trading System (SETS) to replace the SEAQ market for most liquid stocks. At first SETS traded stocks in the FTSE 100 index, but over time the stocks covered increased and in 2003 roughly 217 stocks from the FTSE 250 index are covered. Thin stocks that have never been components of these two indices are traded only on an old quote-driven market (SEAQ) and are not included in our study. 2 Dealers on the LSE can compete voluntarily for trades on SETS stocks on an offexchange dealer market, but are no longer obliged to post firm bid and ask prices as they did earlier and their quotes are no longer available to investors through publicly available price-display mechanism. Trades on the dealer market are not constrained by limit order prices on SETS or required to be partially executed against the limit order book as required by other hybrid markets such as the NYSE, Toronto Stock Exchange, Paris Bourse, or Helsinki Stock Exchange. Investors can choose their trading venues depending on their motivation. Investors, who require prompt and anonymous transactions, may prefer to execute market orders against the book in SETS. Passive customers may choose 2 A few stocks that have been deleted from these indexes continue to be traded on SETS. 11
to place limit orders on the book. Large traders, who do not want their trades to create extensive impact on prices in an order book market, may prefer to trade off-book on the Dealer market. Figure 1 illustrates how an order for SETS stocks is routed on the LSE. All orders must be submitted to member firms (dealers), previously called market makers, to handle the orders. Customers can instruct the member firm to execute the order immediately on SETS at the best available price on the limit order book or to place the order in the limit order queue on SETS. Alternately, the customer can instruct the broker to execute the trade immediately against the dealer s inventory (principal cross) or to cross the trade against other customers orders (agency cross). Customer can also split their order between the two markets. Dealers also trade with each other by placing non-anonymous trades with each other in the public market or by using one of four anonymous and private brokered trading systems as discussed in detail by Reiss and Werner (2004). Table 1 compares the characteristics of the SETS and Dealer market structures in 2000. Access to both SETS and the Dealer market are permitted only to brokers/dealers who are registered members of the LSE. Being a member enables dealers to connect directly to the exchange market. In addition, eligible members are exempted from stamp duty of 0.5% of share purchase value. Nevertheless, members must strictly follow rules and regulations for trading and reporting. Violations are subject to considerable fines. The limit order book market of the LSE is very transparent with respect to order flow and trade execution. Member firms can see all outstanding limit orders on the exchange screen. All trades on SETS are immediately reported for publication. However, the identity of traders is not displayed ex-ante on SETS. Traders only know the ID of the 12
broker through whom the order has been placed. In this respect, the ultimate counterparty is anonymous on SETS. On the contrary, dealers know the identity of traders exante but the quotes and depths on the Dealer market are not available to the public. With respect to counter-party identification the dealer market is non-anonymous. Trades are conducted via telephone and must be reported within three minutes of execution except for block trades greater than 8 NMS 3 that involve a Work Principal Agreement. These must be reported after the entire order is completed. In most cases, dealers trading systems report trades automatically. Trading in the dealership market does not rely solely on bilateral negotiation, but also uses the retail service providers (RSPs) system available from three broker-dealer firms. RSPs provide terminals for execution of retail orders without negotiation. The execution is also guaranteed to be within the book spread. Thus, this service directly competes with the limit order market. There is no minimum order size on SETS or in the dealership market. The standard settlement period for SETS trades in 2000 was T+5. Settlement periods on the dealership market show considerable variability. There are a few studies analyzing the hybrid market at the London Stock Exchange after the introduction of the limit order market (called SETS) in 1997. The institutional details are described in further details in those studies. Board and Wells (2000) provide a comparison of the LSE with Tradepoint. Naik and Yadav (1999) study the effect of the reform and find that trading cost for public investors is lower than before transformation. Friederich and Payne (2001) examine order flow interaction between a 3 NMS is normal market size measured from the average institutional trade size in a stock as computed and regularly updated by the Exchange. 13
limit order book and a dealer market focusing mainly on price volatility and the liquidity role of dealers. They conclude that dealers stabilize the market by supporting liquidity when trade size is above average or depth is low. Ellul (2001) investigates patterns in volatility and examines trader choice using the selection model introduced in Madhavan and Cheng (1997). He reports that dealers subsidize markets. 3.2. Data selection and processing This study uses data provided by the London Stock Exchange for stocks that are components of either the FTSE100 or FTSE250 indices in 2000, designated as SET1 or SET2. These stocks are traded on both the SETS and the Dealer markets. Normal trading hours are 8:00 am to 16:30 pm and we exclude trades and quotes outside these hours. All trades are in British Pounds (GBP). These data comprise a number of files. For each trade, the Trade Reports File has the firm symbol, date, time, price, number of shares, whether the trade is buyer- or sellerinitiated, which market was used for the trade (SETS or Dealer), type of order (market, limit), special designations (such as fill or kill), and the settlement date. We note that trade direction on the dealer market on the LSE is from the point of view of the dealers, so a dealer buy (sell) is assigned as a sell (buy) by us as we take the viewpoint of a trader. We exclude trades with settlement dates greater than SETS s standard settlement date of T+5, trades with a price or volume of zero, and trades with size greater than 8 NMS and trades designated WT (which are 8 NMS and are subject to a Work Principal Agreement), UT (occurring during opening and closing call period), RO (resulting from an option exercise), SW (resulting from a stock swap), CT (contra trades), and 14
PN (work principal portfolio notification). We also exclude trades for which (p t p t- 1)/p t-1 > 0.5 where p t is the trade price at time t, as this condition might result from potential data entry errors. All quote data are from SETS, and the Dealer market does not provide quote data. The Best Prices File includes the time and price (but not the depth) of all quote updates that are better than an existing bid or ask on SETS. We exclude quotes with either the ask, bid, ask size or bid size less than or equal to zero, and for which (a t a t-1 )/a t-1 > 0.5 or (b t b t-1 )/b t-1 > 0.5, where a t is the ask quote and b t is the bid quote. For all orders submitted to SETS, the Order History File contains details about the date and time when the order is entered, deleted, cancelled, or executed, along with its order type, quantity, and limit price. We use these details to obtain aggregate depth at each best limit price. Due to mergers, new listings, and delistings, stocks leave and join the index during the year and to ensure a sufficient sample period, we use only stocks that are members of either index for at least eighty days during the year 2000. The final sample comprises 177 firms after deleting 16 stocks not meeting the requirements enumerated above. The average market capitalization for these firms, obtained from the Compustat global file, is 7.5 billion pounds and the average stock price is 6.59 pounds. 3.3. Measurement of price impact To measure whether there is a discrepancy in the effect of trades on prices between the SETS and the Dealer markets, we use the method suggested by Keim and Madhavan (1996) and Booth, Lin, Martikainen, and Tse (2002). The advantage of this 15
method is the use of trade prices rather than quotes, which is helpful because we do not have quotes for the Dealer market. The total price impact can be decomposed into the permanent price impact and the temporary price impact. The permanent price impact reflects changes in the belief about a security s value due to new information conveyed by the trades. The temporary price impact measures liquidity effects from transitory price reversals. The total price impact reflects the extent of price concession or the difference between the trade price and the previous equilibrium price required to absorb the trade. We assume that a trade occurs at time t with price PT. The equilibrium price observed at time t-b before trade at time t is PB and the equilibrium price observed at time t+a after trade at time t is PA. The sequence of trades is b < t < a. We measure price impact as Permanent price impact (%) = BS*ln (PA/PB)*100 (1) Temporary price impact (%) = BS*ln (PT/PA)*100 (2) Total price impact (%) = BS*ln (PT/PB)*100 (3) where BS equals plus (minus) one for buyer (seller) initiated trades. We also study differences in price impact by trade size. We categorize trades by percentiles of all trades for each firm. This ensures a fair representation of all stocks in each trade size category instead of all trades from a stock getting lumped together in one category due to price differences in stocks. Note that the trade size cut-off varies across firms in this methodology. To identify the equilibrium prices before and after a trade, we plot the price movement around large GBP trades (top 5% in terms of GBP) in Figure 2. We calculate cumulative returns as follows. Each trade is labeled as trade 0, in turn. The previous twenty trades (regardless of trade location, size, buy/sell) executed prior to trade 16
0 (-1, -2, -3, ) and 20 trades executed after trades 0 (+1,+2,+3, ) are obtained. Then trade-to-trade returns calculated as the difference in log prices are estimated from each trade from trade 20 to trade +20. These returns are averaged and cumulated. Figure 2 shows that there are large price movements prior to trade 0 for trades on SETS for both seller-initiated trades and buyer initiated trades. This indicates that there is information leakage before trade 0. Conversely, price movements before trades on the Dealer market started immediately prior to trade 0. Prices after trade 0 seem to stay high for later trades for SETS, but after dealer trades prices reverse back. We also extend our analysis to price movements for 30 trades and 10 trades. The results show similar patterns. Following Booth, Lin, Martikainen, and Tse (2002), we choose equilibrium price PB before trade t at t-12 and equilibrium price after trade t at t+3 where PB is P -12 and PA is P +3. 4 Price movements in other trade value groups show similar patterns. 4. Empirical results Table 2 shows that the average daily number of trades per stock is higher on SETS than on the Dealer market, but the size of each trade in terms of both number of shares and monetary value is higher on the Dealer market. The size advantage of the Dealer market is evident even though these data exclude trades that are greater than 8 NMS, which take place predominantly on the Dealer market. Both the SETS and Dealer markets have substantial order flow, supporting the Market Co-Existence Hypothesis and providing evidence against the Market Domination Hypothesis. 4 Booth, Lin, Martikainen, and Tse (2002) find insignificant price movement 5 trades before and 3 trades after trade t. 17
Table 2 also shows that both the quoted and effective spreads are higher on SETS when orders are submitted to the dealer market. The higher effective spreads on SETS at the time when orders are sent to the Dealer market show that many traders are actually paying higher than average prices to trade on the Dealer market. We view this as the purchase of (high cost) liquidity when this liquidity is not available on SETS. Table 3 reports and compares the permanent, temporary, and total price impact of trades between the two parallel trading markets. Consistent with the Information Screening Hypothesis, the results show that there is a significantly smaller permanent price impact on the Dealer market compared with the SETS market both overall and for each size category. Further, consistent with the liquidity hypothesis, the overall temporary price impact is significantly lower on SETS than on the Dealer market. The declining trend in temporary price impact on large orders in Dealer market is consistent with the the possibility of larger trades causing permanent rather than temporary price effects as well as the findings of Bernhardt et al. (2004) who advance a price discount hypothesis as well. Next, we investigate whether these differences in permanent and temporary price impacts across trading systems survive after controlling for firm-specific characteristics. We estimate the following cross-sectional regression: Y = b 0 + b 1 SETS + b 2 Cap + b 3 Price + b 4 Volatility + b 5 Freq + b 6 Size + ε (4) where Y is the permanent price impact and temporary price impact, in turn, SETS is an indicator dummy variable that is assigned a value 1 for SETS and 0 for dealer market, Cap is the natural log of the market capitalization (millions of GBP), Price is the natural log of the price. Volatility is the natural log of the hourly return volatility (%) from the 18
SETS or Dealer markets, Freq is the natural log of the daily number of trade (000s) from the SETS or Dealer markets, and, Size is the natural log of share trade size from the SETS or Dealer market. Table 4 presents the regression results. Examining the regression with the permanent price impact as the dependent variable, the adjusted R-square exceeds 70%. SETS dummy has a statistically significant positive coefficient indicating that permanent price impact is greater for SETS trades where as dealers are able to screen out such losses. Price, volatility, and trade size are significantly positively related to permanent price impact and trading frequency is significantly negatively related to permanent price impact. Examining the regressions with the temporary price impact as the dependent variable, the adjusted R-square is 78%. SETS dummy has a statistically significant negative coefficient indicating that the liquidity is cheaper on SETS and costly on the dealer markets. Temporary price impact is lower for firms with a higher price, more frequent trading, and larger sized trades. But increased volatility increased the temporary price impact of trades. Regression results in Table 4 are consistent with those presented in Table 3 for SETS versus dealer comparisons as well as trade size comparisons. Coefficients for the control variables are also consistent with previous microstructure literature such as Stoll (2000). Qualitatively similar results in terms of adjusted R-squares, and direction, magnitude, and statistical significance of coefficients are obtained when we estimate 19
equation (4) separately for the SETS and Dealer markets and, therefore, those results are not reported for brevity. 5. Conclusion We examine trading on the London Stock Exchange where the same stocks are traded on a computer-based trading system called SETS and on a Dealer market. Dealers have no obligations to post quotes or to support the market in any way. We test the Market Domination Hypothesis, which states that if two markets trade the same security one market will dominate trading, and the Market Co-Existence Hypothesis, which states that two markets trading the same security can both prosper. We find that the SETS market has more trades overall, but the trades are larger on the Dealer market and the Dealer market dominates very large sized trades. Hence, the evidence supports the Market Co-Existence Hypothesis rather than the Market Domination Hypothesis. We find that Dealer markets are able to screen out informed trades. Consistent with this hypothesis we find significantly greater permanent price impact on SETS market compared with the Dealer market for both overall sample and for each size category. Further, we find that the temporary price impact on SETS market is less than on a dealer market indicating that liquidity can be purchased on the Dealer market, but at a significant cost. A multivariate investigation shows that our findings of higher permanent and lower temporary price impacts on SETS hold after we control for firm characteristics. 20
References Benveniste, L., Marcus, A., Wilhelm, W., 1992. What's special about the specialist? Journal of Financial Economics, 32, 61-86. Bernhardt, D., Dvoracek, V., Hughson, E., Werner, I., 2004, Why do large orders receive discounts on the London Stock Exchange? Working Paper, Ohio State University. Bessembinder, H., Venkataraman K., 2004. Does an electronic stock exchange need an upstairs market? Forthcoming in Journal of Financial Economics. Board, J., Wells S., 2000. Liquidity and Best Execution in the UK: A Comparison of SETS and Tradepoint. Working paper. Financial Markets Group. Booth, G., Lin, J., Martikainen, T., Tse, Y., 2001. Trading and pricing in upstairs and downstairs stock markets. Review of Financial Studies 15, 1111-1135. Chowdhry, B., Nanda, V.,1991. Multi-market trading and market liquidity. Review of Financial Studies 4, 483-511. De Jong, F., Nijman, Th., Roell, A., 1996. Price effects of trading and components of bid-ask spread on the Paris Bourse. Journal of Empirical Finance 3, 193-213. Easley, D., Kiefer, N., O Hara, M., 1996. Cream-skimming or profit sharing? the curious role of purchased order flow. Journal of Finance 51, 811-833. Easley, D., Kiefer, N., O Hara, M., Paperman, J., 1996. Liquidity, information, and infrequently traded stocks. Journal of Finance 51, 1405-1436. Easley, D., O Hara, M., 1987. Price, trade size, and information in securities markets. Journal of Financial Economics 19, 69-90. Easley, D., O Hara, M., Saar, G., 2001. How stock splits affect trading: a microstructure approach. Journal of Financial and Quantitative Analysis 36, 25-51. Economides, N., Schwartz, R., 1995. Equity trading practices and market structure: assessing asset managers' demand for immediacy. Working paper EC-95-08. New York University Stern School of Business. Ellul, A., 2001. As you like it: an investigation of trading behaviour and price volatility on auction and dealership market architectures. Working paper. Financial Markets Group. Ellul, A., Shin, H., Tonks, I., 2002. Toward deep and liquid markets: lesson from the open and close at the London Stock Exchange. Working paper. Indiana University. Fong, K., Madhavan, A., Swan, P., 2001. Why do market fragment? a panel-data analysis of offmarket trading. Working paper. University of Sydney. 21
Friederich, S., Payne R., 2000. Dealer liquidity in an auction market: evidence from the London Stock Exchange. Working paper. London School of Economics. Glosten, L.R. (1994) "Is the electronic open limit order book inevitable?", Journal of Finance, 49, 1127-1161 Gramming, J., Schiereck, D., Theissen, E., 2001. Knowing me, know you: trader anonymity and informed trading in parallel markets. Journal of Financial Markets 4, 385-412. Grossman, S., 1992. The information role of upstairs and downstairs markets. Journal of Business 65, 509-529. Heidle, H., Huang R., 2002. Information-based trading in dealer and auction markets: an analysis of exchange listings. Journal of Financial and Quantitative Analysis 37, 391 424. Keim, B., Madhavan, A., 1996. The upstairs market for large-block transactions: analysis and measurement of price effects. Review of Financial Studies 9, 1-36. London Stock Exchange, Market Enhancements: Guide to Release 3.1. November 2000. Madhavan, A., 1995. Consolidation, fragmentation, and the disclosure of trading information. Review of Financial Studies 8, 579-603. Madhavan, A., Cheng, M., 1997. In search of liquidity: an analysis of upstairs and downstairs trades. Review of Financial Studies10, 175-204. Naik, N., Yadav P., 1999. The effects of market reform on trading costs of public investors: evidence from the London Stock Exchange. Working paper. London Business School. Pagano, M., Roell A., 1992. Auction and dealership markets. What is the difference? European Economic Review, 36, 613-623. Pagano, M., Roell A., 1996. Transparency and liquidity: a comparison of auction and dealer markets with informed trading. Journal of Finance, 51, 579-611. Reiss, P. C., Werner I, 2004, Anonymity, Adverse Selection, and the Sorting of Interdealer Trades. Forthcoming in The Review of Financial Studies. Schwartz, R., Steil, B., 1996. Equity trading III: institutional investor trading practices and preferences, in. B. Steil et al. (eds). The European Equity Markets, London. Seppi, D., 1990. Equilibrium block trading and asymmetric information. Journal of Finance 45, 73-94. Smith, B., Turnbull, A., White, R., 2001. Upstairs markets for principal and agency trades: analysis of adverse information and price effects. Journal of Finance, 56, 1723-1746. Stoll, Hans R., 2000, Friction, Journal of Finance 55, 1479-1514. 22
Table 1 Comparison of SETS and Dealer market structure in 2000 This table compares the characteristics of the SETS and Dealer markets on the London Stock Exchange in 2000. SETS Dealer Trading mechanism Liquidity provided by Order-driven electronic limit order book market Public limit orders and voluntarily dealers Quote-driven multiple dealer telephone market Dealers Access Members only Members only Trader Anonymity Pre-trade, but not post-trade Non-anonymous Pre-trade transparency Post-trade transparency All outstanding limit order book prices and sizes are available to member firms. A member firm can observe the entire limit order book and the ID number of the broker placing the limit order. Immediate reporting of trades. Identity of the counterparties is fully revealed when transaction is confirmed. No pre-trade information is available to public. Quotes are provided based on bilateral inquiry. Reporting of trades is delayed by up to 3 minutes and is incomplete for Work Principal Agreements Minimum order size No minimum No minimum; Smaller orders are generally routed to retail service providers (RSPs) for immediate execution Settlement period T+5 No standard settlement 23
Table 2 Summary statistics This table presents summary statistics during 2000 for our sample of 149 firms. The overall, SETS, and Dealer results are presented in columns 2, 3, and 4, respectively. The first three rows present the daily number of trades, number of shares traded and GBP trading volume. Return volatility is the standard deviation of hourly returns in percent. Trade size is the average number of shares for each trade and GBP trade size is the monetary value of each trade. SETS relative quoted-spread (%) is computed as (askbid)/(ask+bid)/2*100 immediately prior to the trades on SETS as well as Dealer markets. Effective spread is the absolute value of the difference between the trade price and the midpoint of the spread at the time of the trade. Depth at best limit order quotes immediately prior to the trade is aggregated and given in both number of shares and in monetary value. The last column presents the t test of the mean difference between the SETS and Dealer markets. t-test of Mean Difference between SETS All SETS Dealer and Dealer Number of trades per day per stock 294 163 131 32 * Thousands of shares traded per stock per day 3,201 1,496 1,706-210 * GBP trading volume in thousands 19,520 9,266 10,256-990 * Return volatility (%) 1.60 1.20 1.58-0.38 * Trade size 10,696 8,536 14,539-6,003 * GBP trade size 50,560 40,347 68,901-28,554 * SETS Relative quoted-spread (%) 0.987 0.933 1.046-0.114 * Effective spread (%) 0.705 0.201 0.617-0.416 * SETS Depth 48,693 55,669 40,298 15,371 * GBP SETS depth (000s) 239,295 273,024 189,550 83,475 * *Significant at the 0.01 level. 24
Table 3. Price impact on SETS and Dealer markets For the 149 firms in our sample in 2000, this table presents the permanent, temporary, and total price impact of trades on SETS and the Dealer market both overall and by relative trade size category. The permanent price impact of trade at time t is computed as BS*ln (PA/PB)*100 where BS equals plus (minus) one for buyer (seller) initiated trade and if a trade occurs at time t, designate the price of that trade as PT, the price of the third subsequent trade as PA and the price of the twelfth previous trade as PB. The temporary price impact of trade at time t is computed as BS*ln (PT/PA)*100, and total price impact of trade at time t is calculated as BS*ln (PT/PB)*100. The last column presents the test of the mean difference between the SETS and Dealer markets. t-test of Mean Difference between SETS and Dealer GBP Trade value (percentile) All SETS Dealer Permanent price impact (%) All orders sizes 0.100 0.219-0.019 0.239 * <25% 0.003 0.122-0.055 0.177 * 25-50% 0.055 0.157-0.032 0.189 * 50-75% 0.141 0.224-0.002 0.227 * 75-90% 0.227 0.296 0.032 0.264 * 90-95% 0.204 0.311 0.025 0.285 * >95% 0.118 0.340 0.052 0.288 * Temporary price impact (%) All orders sizes 0.169 0.073 0.260-0.187 * <25% 0.277 0.123 0.351-0.228 * 25-50% 0.202 0.096 0.288-0.192 * 50-75% 0.134 0.067 0.234-0.167 * 75-90% 0.067 0.043 0.125-0.081 * 90-95% 0.054 0.037 0.080-0.043 * >95% 0.055 0.036 0.062-0.026 * Total price impact (%) All orders sizes 0.262 0.294 0.226 0.068 * <25% 0.260 0.244 0.267-0.023 * 25-50% 0.251 0.254 0.244 0.010 50-75% 0.274 0.294 0.226 0.068 * 75-90% 0.293 0.340 0.155 0.185 * 90-95% 0.258 0.349 0.105 0.244 * >95% 0.174 0.365 0.115 0.249 * *Significant at the 0.01 level. 25
Table 4 Cross-sectional regression analysis on price impact of trade This table presents the results of cross-sectional regressions of the price impact of trades for both the SETS and the Dealer markets. The sample contains 298 observations (149 firms X 2 markets. The regression equations are: Y = b0 + b1 SETS + b2 Cap + b3 Price + b4 Volatility + b5 Freq + b6 Size + ε where Y is permanent and temporary price impact, in turn. The permanent price impact of a trade at time t is computed as BS*ln (PA/PB)*100 where BS equals plus (minus) one for buyer (seller) initiated trade and if a trade occurs at time t, designate the price of that trade as PT, the price of the third subsequent trade as PA and the price of the twelfth previous trade as PB. The temporary price impact of a trade at time t is computed as BS*ln (PT/PA)*100. SETS is a dummy variable that equals 1 if the observation is aggregated from the SETS market. Cap is the natural log of the market capitalization (millions). Price is the natural log of the price. Volatility is the natural log of the hourly return volatility (%) from the SETS or Dealer markets. Freq is the natural log of the daily number of trade (000s) from the SETS or Dealer markets. Size is the natural log of share trade size from the SETS or Dealer markets. Permanent Price Impact Temporary Price Impact Independent variables Coeff. t-statistics Coeff. t-statistics Intercept -0.2224-1.74 1.3699* 12.19 SETS 0.2849* 21.62-0.1956* -16.92 Cap 0.0008 0.08-0.0024-0.26 Price 0.0274* 2.00-0.1060* -8.82 Vola 0.1031* 4.90 0.1176* 6.37 Freq -0.0399* -4.77-0.0180* -2.45 Size 0.0311* 2.28-0.0948* -7.93 Adj. R 2 0.7062 0.7836 F-value 119.97 175.58 26
Customer submits order to member firms with/without trading venues Member firm handle order in one of three ways according to customer s instructions Dealer market executes entire order against his own inventory (principal cross) or matches order with other customer s order (agency cross) Mix Partially executed in Dealer market and work the rest in limit order book. directly submits order to SETS limit order book market executes immediately as market orders or enters as a new limit order Member must report all trades from Dealer market within 3 minutes, except Work Principal Agreement orders. All orders executed in SETS are automatically reported. Figure 1. Order flow of SETS stocks on the London Stock Exchange. 27
0.0003 0.0002 0.0001 0-0.0001-0.0002-0.0003-20 -19-18 -17-16 -15-14 -13-12 -11-10 -9-8 -7-6 -5-4 -3-2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Trade relative to trade at time 0 SETS buy Dealer buy SETS sell Dealer sell Figure 2. Cumulative average returns around large GBP trades. We identify the 5% of trades that have the greatest GBP value. We label each of these trades, in turn, as trade 0. For each trade 0, we identify the twenty previous trades, trades -1 through -21, and the subsequent 21 trades, trades +1 through +21. We calculate the return for each trade from -20 to +20 as the difference in the log of the trade price minus the log of the previous trade price. These returns are averaged and cumulated beginning with trade -20. Mean values of cumulative average returns are plotted. 28