Do “thinly-traded” stocks benefit from specialist intervention?

M. Nimalendran∗ Bank of America Professor Warrington College of Business University of Florida Gainesville, FL 32605 [email protected]

Giovanni Petrella Assistant Professor of Banking Catholic University Largo Gemelli, 1 20123 Milano (Italy) [email protected]

This Draft: July 15, 2002

∗

Corresponding author.

Do “thinly-traded” stocks benefit from specialist intervention?

Abstract This paper addresses the issue of the optimal trading system for less actively traded stocks. Several studies have examined the quality and performance of alternative market structures for actively traded stocks. However, very few empirical studies compare the performance of different market structures for less actively traded (i.e., “thinly-traded”) stocks. In 1997 the Italian Stock Exchange (ISE) implemented the Thin Stocks Project to improve market quality of “thinly-traded” stocks. Under this project, stocks defined as “thinly-traded” by the ISE were given the option to trade either on a pure order driven market with limit order book (POD), or a hybrid order driven market with specialist and limit order book (HOD). In this paper we compare the relative performance of the two trading systems for “thinly-traded” stocks using several metrics of market quality. We find that the hybrid order driven system offers superior performance along several quality metrics. In particular, the hybrid order driven system offers lower execution costs, greater depth, significant increase in the depth-to-spread ratio, and a decrease in adverse selection cost. Very illiquid stocks benefit more from adopting a hybrid system than moderately illiquid ones. The results are robust to different measures of execution costs, sampling periods, and modeling approaches. Our findings support the analysis by Grossman and Miller (1988) that a specialist can enhance liquidity of “thinly-traded” stocks. Keywords: Thin Markets; Specialist; Market Quality; Trading Costs. JEL Classification: G14; G18; D23; L22.

Do “thinly-traded” stocks benefit from specialist intervention? 1.

Introduction The structure and design of exchanges is an important issue that has received

considerable attention in market microstructure studies.1 The research in this field has mainly focused on the implications of market structure for metrics of market quality such as explicit and implicit trading costs, liquidity, price discovery, and informational efficiency. One of the most important developments in financial markets over the past decade is the proliferation of new markets and automated trading systems. Many of these automated markets use an order driven system, with a high degree of transparency where current and away public limit orders are continuously displayed. Even though there has been a tremendous growth in order driven systems, there is little empirical research into the relative performance of order driven and quote driven systems, especially for less frequently traded (i.e., “thinly-traded”) stocks. Recent reorganization of the Italian Stock Exchange (ISE) offers a unique institutional setting to study the relative performance of pure order driven and hybrid systems for trading illiquid stocks. In May 1997, the ISE implemented the Thin Stocks Project.2 Under this project, stocks defined as “thinly-traded” by ISE were given the option to trade under two alternative regimes: a pure order driven system with a limit order book (POD), or a hybrid order driven system including a specialist and a limit order book (HOD). In this paper we examine the quality of the two market structures for trading illiquid stocks and test several hypotheses. Examining the effects of trading architecture on market quality for less actively traded stocks is important for several reasons. First, several empirical studies document a statistically significant relation between thinness and poor market quality. Thinly traded stocks are usually associated with poor market quality indicators -- such as large spreads, high transitory volatility, and low price efficiency (Pagano, 1997), and high adverse 1

O’ Hara (1995), Madhavan (2000), as well as Biais et al. (2002) provide excellent summaries of the theoretical, empirical, and experimental contributions in this area. 2 Since 1994 several major European exchanges have adopted different trading rules for different types of securities (by market capitalization and/or its liquidity) or for different types of orders (small, medium, large). These systems have been introduced with the objective of improving the quality of the

1

selection costs (Easley, Kiefer, O’Hara, and Paperman, 1996). Second, negative effects associated with “thinness” may not only be detrimental for the market behavior of firms that are already listed, but they may also deter new companies from listing for fear of falling into a vicious cycle of low trading activity leading to low attractiveness for investors which in turn lead to even lower trading activity.3 Thus, improving the market quality for these firms could also help the listing of new companies. The third motive for this study is the widespread presence of “thinly-traded” stocks on the main stock exchanges around the world. For example, stocks with a turnover ratio lower than 25% of the overall market turnover account for almost 26% of the total number of stocks listed on the major European exchanges (see Table 1).4 There are several important differences between previous empirical studies of market quality and the present study. First, previous empirical studies that have compared the quality of alternative market structures have generally focused on actively traded stocks (Neal, 1992; Petersen and Fialkowski, 1994; Huang and Stoll, 1996; and Keim and Madhavan, 1997), whereas in this study we focus on “thinly-traded” stocks. Second, the present study compares different trading structures for companies listed on the same market. This feature reduces structural differences across stocks in the sample, and highlights differential effects induced by different trading mechanisms. Most studies have a much more limited ability to control for differences in structural characteristics of the underlying securities. For example, the studies of Huang and Stoll (1996), and Bessembinder and Kaufman (1997) that compare execution costs on NASDAQ and NYSE may suffer from important differences in the listing requirements and risk market for the securities that are traded (for a survey of these changes, see Demarchi and Foucault (2000)). The ISE’s project falls into the category of improving liquidity for “thinly-traded” stocks. 3 In a game theoretical model of market participation Pagano (1989) shows that, depending on traders' beliefs, two very different equilibria arise: (i.) low trading activity will give rise to low liquidity equilibrium and self-fulfilling liquidity trap; and (ii.) high trading activity will tend to high liquidity equilibrium and liquidity virtuous cycle. Whether the market will settle on one or another equilibrium mainly depends on the expectations held by economic agents. Any new trader entry (exit) makes the market more (less) liquid. Such kind of participation externality leads to unstable multiple rational expectations equilibria. 4 As documented by Easley, Kiefer, O’Hara, and Paperman (1996), on the London Stock Exchange (the largest European market in terms of capitalization and traded shares) the bottom 50% of listed stocks by capitalization account for only 1.2% of overall market trading volume, and over 1,000 stocks average less than one trade a day. Statistics for ISE confirm the above trend. In 1995 the bottom 100 less traded stocks, which account for 30% of listed stocks, contributed only 0.1% to the total trading volume of the market (Italian Stock Exchange, 1996).

2

characteristics of the securities they examine on the two markets. Third, the sample data differs from existing studies. This study examines the Italian Stock Exchange (ISE), an electronic screen-based market whose dimension and importance is increasing in the context of organized exchanges. According to the Federation of European Stock Exchange statistics (FESE, 2002), ISE is presently the fifth (fourth) largest exchange in Europe and the third (third) largest exchange in the Euro area in terms of total market capitalization (monetary trading volume). Finally, this study provides specific insights on the operation of order driven markets. Interest in limit order trading has grown rapidly in recent years as it plays a vital role in the liquidity provision for both pure (e.g., Paris, Tokyo, Stockholm) and hybrid (e.g., NYSE) order driven structures. In this paper we examine a comprehensive set of quality measures, and focus on specific features displayed by spread behavior in an electronic order driven system. In particular, the difference between quoted and effective spread, that has already been studied in quote driven environment (Neal, 1992), assumes a completely different meaning in an order driven environment. The remainder of the paper is organized as follows. In section 2 we discuss the theoretical and empirical studies related to the comparative analysis of alternative trading systems. In section 3 we describe the microstructure of the Italian Stock Exchange (ISE), and summarize the reforms adopted by ISE to improve the quality of the market for less actively traded stocks. In section 4 we describe the hypotheses to be tested and the research design. In section 5 we describe the data and the sample selection criteria. In section 6 we present the quality measures and the methodologies used to investigate the relative quality of the two trading systems. In section 7 we present the empirical results, and conclusions are provided in section 8.

2.

Literature Review Theoretical work in market microstructure has focused on the implications of

market design and trading rules on traditional measures of market quality such as trading costs, liquidity, price efficiency, and on the welfare of market participants.

3

Viswanathan and Wang (2002) analyze the effect of different market structures on the welfare of traders: they conclude that a risk neutral trader would prefer to trade in a limit-order market, instead of any hybrid or dealership market. Pagano and Roëll (1996) compare the price formation process in auction and dealer markets in the presence of informed trading: they show that greater transparency generates on average lower trading costs for uninformed traders, and they also find that the implicit bid-ask spread in a transparent auction is tighter than in a less transparent dealer market. Other theoretical studies provide less clear-cut predictions in terms of trading costs for different market structures. Biais (1993) compares fragmented (dealer-based) and centralized (auction-based) markets and shows that the average bid-ask spread is the same in the two markets. Madhavan (1992) develops a model to compare price formation processes in quote driven and order driven markets in a world with adverse information and strategic trading. However, with reference to trading costs, Madhavan (1992) states that: “The determination of which system has lower operating costs must be resolved empirically” (p. 626). On the empirical ground, several studies have compared trading costs in dealer and auction markets using a matched sample of stocks listed on these markets. Neal (1992) compares the costs of trading options on a specialist-based system (AMEX) and a multiple market maker system (CBOE). Based on a sample of 26 AMEX and 15 CBOE options, Neal finds that the specialist system has lower transaction costs for “lowvolume” options. By contrast, the transaction cost differential disappears for “highvolume” options. Affleck-Graves et al. (1994) focus on the differential magnitude of the spread components for NYSE/AMEX versus NASDAQ stocks using a matched sample of stocks. After controlling for differences in stock price, dollar volume of trading, market capitalization, and standard deviation of daily return, they find that spreads are higher on the exchanges than on the dealer market. Porter and Weaver (1996) also investigate spreads for NYSE, AMEX, and NASDAQ firms using a matched sample. They find that the quoted spread on the NYSE is about 43 percent lower than on the NASDAQ. Huang and Stoll (1996) compare execution costs for a matched sample of large capitalization NYSE and NASDAQ firms. They report that average execution costs on NASDAQ

4

exceed those for NYSE firms by a factor of two to three. Finally, Bessembinder and Kaufman (1997) extend the previous study and analyze a matched sample of small, medium, and large capitalization firms listed on NASDAQ and NYSE. They find that execution costs are generally lower on the NYSE than on NASDAQ, and the differentials are larger for small and medium capitalization issues. Most of the theoretical and empirical studies have focused on the NYSE/AMEX and NASDAQ markets. These markets represent quote driven systems where a dealer takes the opposite side of a transaction. Compared with the NASDAQ, the main difference is that the NYSE/AMEX is a floor-based system with a specialist providing liquidity and price continuity. In addition, the NYSE specialist also maintains the limit order book.5 On the other hand most of the automated markets (for example ISE, Paris Bourse, Stockholm Stock Exchange, Toronto Stock Exchange) are continuous order driven markets with public display of all limit orders (i.e., with high transparency) and without reliance on dealers. In the past five years several European markets have also introduced different trading rules for different types of securities (market capitalization/liquidity) and for different types of orders (small, medium, large). In the case of the ISE, the Thin Stocks Project allows dealer intervention for very illiquid stocks.

3.

The Microstructure of the Italian Stock Exchange Trading on the Italian Stock Exchange (ISE) takes place through an electronic

screen-based system. The system supports two trading mechanisms: a call auction used to open trading, and a continuous auction operating throughout the trading day. An electronic limit order book supports the system in both phases and makes possible the realization of trades by automatically matching buy and sell (limit and market) orders. The book is open to all intermediaries either to observe the state of the book (price and quantity for all orders on both sides), or to insert orders into the system.

5

Chung, Van Ness, and Van Ness (1999) examine the interaction between limit order traders and specialist in providing liquidity for NYSE stocks. They find that competition between traders and specialists exerts a significant impact on the bid-ask spread.

5

Until May 1997, ISE operated with a pure order driven system, working without designated market makers to provide liquidity, for all listed stocks. On May 1997, ISE undertook a specific project – named Thin Stocks Project – to enhance secondary market activity and quality for “thinly-traded” stocks. Based on this project, a new organization of trading hours was adopted, and a hybrid order driven system with a specialist was implemented for trading less active stocks. To identify “thinly-traded” stocks ISE computes a Synthetic Liquidity Index (SLI), which is a numerical index used to rank all companies listed on the ISE in order of decreasing liquidity. This index is computed annually on the basis of several firmspecific variables that are (supposed to be) associated with market liquidity. ISE uses the following variables to calculate SLI: a. b. c. d.

firm market capitalization; average quoted bid-ask spread; daily average monetary trading volume; turnover indicator (number of shares traded relative to the number of floating shares); e. free float (percent of publicly available shares, i.e. not in charge of controlling shareholders); f. trade frequency (number of days with trading relative to the total number trading days). For each of the above variables, ISE assigns a rank for each stock in order of decreasing liquidity. Suppose there are N firms traded on the ISE, the ranks would be assigned as follows, where “highest” is the highest value of the variable and “lowest” means the lowest value of the variable:

Variable

Rank for the stock with: Highest value

Lowest value

Firm market capitalization

1

N

Average quoted bid-ask spread

N

1

Daily average trading volume

1

N

Turnover indicator

1

N

Free float

1

N

Trade frequency

1

N

6

Using the above rankings, the Synthetic Liquidity Index (SLIj), for firm j is calculated from the Ranki,j for the ith variable and jth firm as follows: SLI j =

1 6 ∑ Rank i, j 6 i =1

Based on the SLI value all the companies are then ranked, from the first to the Nth, in order of decreasing liquidity (or increasing SLI value), and assigned to one of the two ISE’s market segments - Active Stocks segment or Inactive Stocks segment – according to the company’s position in the SLI ranking. On May 1997, when the project started, 225 companies were listed on the Italian stock market. The ISE defined the Inactive Stocks segment as the last 100 stocks. The two segments differ in terms of trading hours and trading mechanisms (see Figure 1). Specifically, listed companies belonging to the Inactive Stocks segment are allowed to choose the trading system for their own stocks between a pure order driven system based on a limit order book, or a hybrid system with both a limit order book and a company-designated specialist. The first change introduced by the Thin Stocks Project was to modify the trading hours by extending the pre-opening phase,6 and shortening the continuous trading phase with respect to more actively traded stocks. Increasing the time for the opening phase aims at consolidating the order flow to determine ‘more accurate’ opening prices; whereas, shortening the continuous trading phase is intended to enhance market depth and breadth in order to reduce transitory price volatility.7 The second important modification of ISE’s market microstructure introduced by the Thin Stocks Project is the participation of a specialist designated by the firm. The specialist is obliged to continuously display bid and ask quotes, within the limits of a maximum allowed spread, and for the execution of a maximum number of lots per trading day. The maximum spread and the maximum lot size for each stock is determined 6

The opening auction at ISE unfolds in three phases: (i.) the pre-opening, in which approved intermediaries enter their orders and a theoretical opening price is determined; (ii.) the validation, that works to confirm the validity of the theoretical opening price; and (iii.) the opening, where the contracts are automatically concluded at the unique opening price, if all requested conditions are properly fulfilled (e.g., the opening price, compared with the previous day closing price, complies with the allowed maximum price variation). 7 This change is consistent with the seminal model proposed by Garbade and Silber (1979).

7

by the ISE.8 Specialist quotes are posted on the limit order book, and displayed on the trading terminals to all market participants. Therefore, ISE specialist, unlike NYSE specialist (Ready, 1999), is not monopolistic in managing and displaying the book. Rather, any intermediary has access to the electronic limit order book and can place orders that compete with the specialist’s quotes. The second obligation of the specialist concerns issuer information and financial analysis. Specialists commit themselves to enhance the quality of the information released by companies issuing “thinly-traded” stocks. In particular, the specialist is obliged to publish not less than two financial analyses on the stocks for which he acts as designated specialist. For providing market making and financial analysis services, specialists are compensated through a reduction in trading fee by ISE. Further, they benefit from bid-ask spread revenues and from lower adverse selection costs due to greater knowledge of the firm (because they establish continuous relationships with the company).9 Finally, they can also receive a side payment from the issuer company. If this is the case, the relationship between the specialist and the issuer is regulated by a specific agreement. This agreement can take two forms: the issuer company can provide the specialist some funds to comply with market making obligations or, alternatively, the issuer company can share specialist profits (and losses) due to his market making activity.10 Lastly, the ISE provides direct assistance to the “thinly-traded” stock issuer with respect to several activities related to the enhancement of information quality (e.g., by hosting investors meetings, by providing technological support to realize web pages, etc.).

8

On December 31, 1999, the average (median) maximum spread established by ISE for specialist supported thin stocks was 4.4% (4%), and the average (median) value of the maximum number of tradable lots per day was 54,600 USD (50,400 USD). 9 This point will be explicitly addressed later. 10 A positive involvement of the issuer company in the liquidity of its share has been recently proposed by Amihud and Mendelson (2000), who argue that value-maximizing small firms should be willing to pay directly for analysts coverage.

8

4.

Research Design and Testable Hypotheses

4.1.

Empirical Models The issue of which market structure provides the best quality can be examined

from different perspectives. First, one can separately document quality metrics for stocks traded in different markets. For example, the study by Lee (1993) examines trading costs in different market locations. A second approach is to contrast execution costs in different markets based on matched pairs of stocks. This can be referred to as cross-sectional approach and has been used, among others, by Huang and Stoll (1996) and Bessembinder and Kaufman (1997). The basic idea is to infer the effects of market structure on market quality metrics after controlling for differences in stocks’ characteristics that are expected to be related with quality measures. However, this approach can suffer from stocks’ mismatching that might affect empirical comparisons and lead to incorrect inferences. We therefore employ an event study (ES) approach that avoids any problem due to mismatching of firms. In this approach, the same stock is examined in two different trading environments (in our case, pure vs. hybrid order driven systems). This reduces the likelihood that important firm-specific variables will be omitted from the analysis.11 The switching date from the pure order driven (POD) to the hybrid order driven (HOD) is considered as the “event date,” and the quality indicators before and after the switching date are statistically compared. This approach is similar to studies that examine stocks that move from one exchange to another. For example, Christie and Huang (1994) examine stocks that move from NASDAQ to NYSE or AMEX. However, the stocks that move from NASDAQ to NYSE/AMEX are very large companies that are very liquid – the mean capitalization of firms moving to NYSE in Christie and Huang’s sample is about $ 400 million. In our study we examine very illiquid stocks.12

11

An alternative solution, used by Huang and Stoll (1996) and Bessembinder and Kaufman (1997), consists in regressing the difference in market quality indicators between benchmark and control stocks on the difference in the variables used to select the benchmark sample. The sample selection procedure is correct when the difference in the variables is not statistically significant in explaining the difference in the quality indicators. 12 The mean (median) market capitalization in our sample is about $ 157 (82) million.

9

The simple event study approach is not free from shortcomings either. The possibility that stock fundamentals and risk characteristics may have changed after the switching date could bias the results. To control for this possible bias, we use an event study approach combined with a matched sample of similar firms that do not switch trading systems (CES approach). This allows us to control for any changes in the stocks characteristics over time. The CES approach relies on the results of two event studies: one conducted on the stocks moving from the POD to the HOD system, and the second on a set of matched firms that do not change trading regimes (i.e., control sample). By comparing the changes in the quality measures for the stocks that switched trading structures with the variations observed for the control sample, we isolate the pure effect due to market structure.

4.2.

Hypotheses Grossman and Miller (1988) show that a specialist can enhance stock liquidity

when trading volume is low, whereas an auction structure is preferable when volume is relatively high. This implies that the cross-sectional differences in spreads between market structures should be a function of the trading volume. When trading volume is high, the probability of synchronous buy and sell orders arrival will be high, allowing public orders to be matched without any intermediary taking the opposite side. By contrast, when the order flow is relatively low and/or asynchronous, there is need for an intermediary to cross orders that are not matched. Execution costs in the first scenario (high volumes) are lower than in the second (low volumes) for two reasons. First, the presence of an intermediary in the low volume scenario needs to be compensated for providing immediacy, and this entails higher execution costs. Second, large volumes reduce execution costs due to fixed costs and economies of scale in trading activities. In this paper our focus is on the second scenario, i.e. on less actively traded stocks. In this case, the presence of a dealer can enhance market quality by playing a truly intermediation-based role in the market trading process. Based on Grossman and Miller’s analysis, our first hypothesis is stated below.

10

Hypothesis I: For “thinly-traded” stocks a hybrid order driven system with specialist participation (HOD) provides superior market quality (in terms of trading costs, market depth, and overall liquidity measures) compared to a pure order driven system (POD). Theoretical studies by Glosten and Milgrom (1985), Glosten (1989) and Benveniste, Marcus and Wilhelm (1992), support the hypothesis that specialist intervention can reduce asymmetric information problems in securities markets. Glosten and Milgrom show that, when the level of asymmetric information is high, competing market makers can rationally decide not to post bid and ask quotes. By contrast, a monopolistic specialist can sustain severe asymmetric information costs by adopting a long-run view and an intertemporal cross-subsidization strategy. A monopolistic specialist, unlike competing market makers who operate with zero expected profits, is able to balance two opposite effects: losses caused by transaction with informed traders with profits coming from trades with liquidity traders.13 Glosten (1989) extends the previous model by considering an environment in which trades are not restricted to unit amounts. In this case, the specialist quotes a price schedule that specifies the price per share as a function of the quantity to be traded. In particular, Glosten supports the adoption of a price schedule that should reflect the probability of informed trading according to the traded quantity. Therefore, Glosten also shows that a specialist is able to control severe adverse selection problems; but in his model the specialist can survive through cross-sectional subsidization between trades, and not over time as in Glosten and Milgrom (1985). Benveniste, Marcus and Wilhelm (1992) present a model that highlights specialist’s characteristics in terms of long-term professional relationships with brokers. They show that longstanding relationships between brokers and specialists can mitigate the effects of asymmetric information. In particular, the specialist -- by sanctioning traders exploiting private information – can improve the terms of the trade faced by uninformed traders. The main implication of their model is that benefits of a specialist-

11

based market will be greatest when the potential for privately informed trading is greatest. Our second hypothesis is stated below.

Hypothesis II: For “thinly traded stocks”, the adverse selection component of the spread is lower under the hybrid system with specialist intervention compared to a pure order driven system. Unlike quote driven markets, where transaction price can be directly negotiated with dealers, in electronic order driven systems the transaction price should in principle be equal to the posted bid (market sell order) or ask (market buy order) quotes. In an electronic order driven environment, there is no dealer with whom to negotiate quotes displayed on the book, but only a computerized system that automatically matches market orders coming to the market with limit orders waiting on the book. As a result of this procedure, trades should occur at posted prices. However, when the quantity at the best quote is exhausted, market orders hit “non-best” quotes. In such cases, the quoted spread can be lower than the average (and marginal) trading cost for some trades, and thus understate the investor’s actual trading costs.14 Our third hypothesis is stated below.

Hypothesis III: In an electronic order driven market, either in the pure or in the hybrid version (with specialist), the quoted spread is lower than or equal to the effective spread computed with actual transaction prices.

13

The first type of transactions provides the specialist with (signals about) proprietary information on the true value of the security he is trading. In the second type of transactions, the specialist exploits the information he learned by transacting with liquidity traders. 14 The difference between quoted and effective spreads assumes a completely different interpretation in an order driven market relative to a quote driven market. In the latter, effective spread is usually lower than quoted spread due to specialist stopped orders, public limit orders, and matching market orders (Petersen and Fialkowski, 1994). By contrast, in order driven environments effective spread may be greater than quoted spread.

12

5.

Sample Selection and Data Description We test our hypotheses using a sample of twenty firms listed on the Inactive

Stocks segment of the Italian Stock Exchange (ISE) that voluntarily moved from the POD system to the HOD system during the period 1997-1998. The event under study is the switching to a different trading system. Our pre and post event windows include five months before the switch and five months after the switch. Intraday trade and quote data for the post and pre event window (-75, +74 days) were obtained from ISE’s Research Department. For each stock included in the sample, the ISE provided two data files. One file contained data on limit orders and the other file contained transaction record data.15 The file on limit orders included information on best quoted bid price, best quoted ask price, and depth on both sides for all orders entered into the system time stamped to the nearest second. The file on transaction data included information on transaction time of execution to the nearest second, transaction price, and number of shares traded for all transactions concluded in the sample periods. The sample selection procedure consists of four steps.

Step 1. The twenty “thinly-traded” stocks belonging to the Inactive Stocks segment of ISE that moved to the HOD system are identified as the sample under investigation.

Step 2. Stocks that satisfy all of the following conditions together are selected: a) to belong to the Inactive Stocks segment as defined by the Italian Stock Exchange; b) to be traded under the pure order driven system throughout the entire period; and c) to share the industry code with at least one of the stocks identified in Step 1. Following the above procedure, we obtain a set of “thinly-traded” stocks that trade under the POD system, and share the same industry code with at least one HOD

15

Usual filtering procedures to check for recording errors are applied. Further, we do not adjust for delays in trade reporting (as suggested by Lee and Ready, 1991, and Hasbrouck, Sofianos, and Sosebee, 1993) because the trades and quotes are reported in real-time on the ISE.

13

traded stock. Each of these securities is a potential candidate to be paired with a HOD traded stock.

Step 3. POD traded stocks selected at Step 2 are excluded from the sample if

p POD − p HOD p POD p HOD 2 ⋅ POD ≥ 1 , and if ≥ 3 or ≥ 3, p + p HOD p HOD p POD where, p is the average closing price in the sample period, and the superscripts POD and

HOD stand respectively for pure and hybrid order driven traded stocks. This screening is to eliminate matching candidates for which price levels are extremely far apart.

Step 4. We match each HOD stock to a POD stocks according to a set of characteristics that may be related to market quality indicators by minimizing a loss function with respect to these characteristics. Several empirical studies (Porter and Weaver, 1996; Bessembinder and Kaufman, 1997; Bessembinder, 1999) compare trading costs displayed by different market structures on the basis of matched samples of NYSE and NASDAQ listed companies. However, in constructing matched pairs, these studies evaluate stock homogeneity uniquely in terms of market capitalization (Bessembinder and Kaufman, 1997; Bessembinder, 1999) or stock price (Porter and Weaver, 1996). These approaches may lead to mismatching along other variables that may be related to market quality. To minimize this problem, we match pairs across several factors (listed below) using a comprehensive measure we named Index of Homegeneity (IOH). For each hybrid order driven stock ( i ), we compute the following Index of Homogeneity ( IOH i , j ) relative to the pure order driven traded candidate ( j ): 5

IOH i , j = ∑ k =1

Fk j − Fki Fki

[1]

where Fk represents the value of the factor k. We identify the best control stocks by minimizing the loss function described by equation [1].16 The factors used for matching are the following: market capitalization, stock price, return volatility, book-to-market ratio, and firm’s leverage. 16

Loss functions can also be computed in quadratic form. For this study, the functional form specifically adopted does not affect the results of the sample selection procedure. Although the functional form of the loss function is not relevant, it is by contrast important to use relative measures of deviation to give equal weight to all factors.

14

To give an example of the procedure used to create the matched pairs, consider the HOD traded stock Aedes belonging to the real estate sub-industry of the financial industry class.17 Acqua Marcia, Cmi, Garboli, Immobiliare Metanopoli, and Risanamento di Napoli are POD traded stocks belonging to the Inactive market segment and classified by Italian Stock Exchange in the same sub-industry. Acqua Marcia and Garboli do not pass the price level screening (Step 3). Thus, Cmi, Immobiliare Metanopoli, and Risanamento Napoli are left as candidates for matching the HOD traded stock Aedes. Applying algorithm [1], Risanamento di Napoli shows the best (lowest) Index of

Homogeneity with Aedes. Therefore, the matched pair consists of stocks Aedes and Risanamento di Napoli. Table 2 presents the matched pairs of HOD and POD stocks, along with the subindustry classification, the name of the specialist, and the switching date. The results of the matching procedure leave four HOD stocks without control stocks, because of a weak homogeneity with the candidates control stocks (Bna Priv, Bna Rnc, Linificio Rnc, and Terme Acqui Rnc). Therefore, out of twenty HOD stocks, the matching procedure leads to sixteen pairs made up of sixteen HOD traded thin stocks, and sixteen POD traded thin stocks. Summary statistics presented in Table 3 show that the mean differences in market capitalization, share price, return volatility, book-to-market ratio, and market leverage are not significantly different from zero. Therefore, the matching procedure achieves the objective of selecting a control sample made up of stocks structurally similar to our sample stocks, albeit traded under a different mechanism.

6.

Empirical Analysis

6.1.

Definitions of Quality Measures To test Hypothesis I we need to define a set of market quality measures related to

trading costs, market depth, and overall market liquidity.

17

Industry classification used throughout this study refers to that used by the ISE.

15

A wide range of metrics related to bid-ask spread captures the trading costs side of market quality. One of the most used is the quoted percentage spread (QPS), that accounts for trading costs related to a round-trip transaction executed at posted quotes:

QPSi ,t = 200 ⋅

Ai ,t − Bi ,t Ai ,t + Bi ,t

.

[2]

where Ai ,t is the best (lowest) ask quote for the stock i at time t, and Bi ,t is the best (highest) bid quote for the stock i at time t. The previous measure evaluates trading costs under the assumption that trades are executed at the quoted (bid or ask) prices. Nevertheless, as explained in Section 4.2, this does not necessarily happen. Therefore, it can be useful to distinguish between the ex

ante spread (and liquidity) expressed by quotes, and the ex post spread (and liquidity) implicit in trade prices. Effective spreads – that are ex post spread measures – are computed with actual transaction prices, and thus reflect the real conditions at which the trade is executed. In formula, the effective percentage spread (EPS) is:

EPS i ,t = 200 ⋅

Pi ,t − M i ,t

[3]

M i ,t

where Pi ,t is the transaction price for the stock i at time t, and M i ,t is the midpoint of the bid and ask quotes for stock i at time t. If the bid-ask spread can be considered as the price of immediacy, market depth can be thought as the quantity side of the immediacy. Even when the spread is low, quoting a lower depth can degrade market quality. Thus, we also look at market depth behavior. The depth variable in our study is expressed in terms of the average monetary value of the shares offered at the ask and bid prices. In formula:  BidQi ,t + AskQi ,t DEPTH i ,t = M i ,t ⋅  2 

  

[4]

where BidQi ,t ( AskQi ,t ) is the quantity available at the bid (ask) quote for stock i at time t.

16

To consider jointly trading costs and market depth, we also compute the Market Quality Index (MQUAL). This measure is computed by taking the ratio of depth to absolute spread: MQUALi ,t =

DEPTH i ,t Ai ,t − Bi ,t

[5]

where previous definitions apply. The market quality index quantifies the trade-off between spread and depth (Bacidore, 1997). A market with high (low) depth and low (high) spread would have a higher (lower) market quality index. A decrease (increase) in the market quality index would indicate that depth is declining (increasing) more than spread.

6.2.

Simple Event Study Methodology

In the simple event study (ES) approach we compare the behavior of the same stock under two trading regimes. The main advantage of the ES methodology relative to the matched-pairs comparison is the lack of mismatching risk. In addition, to identify the pure effect attributable to different market structures, we also control for two other confounding influences. First, we use a portfolio (or pooled) approach that relies on the diversification across firms to minimize idiosyncratic effects of new information that may induce price and quote changes unrelated to stocks’ liquidity. Second, the dispersion of switching dates across time also minimizes any systematic influence of market-wide liquidity fluctuations. Given the general framework specified in Section 4.1, to test whether observed differences in quality metrics between POD and HOD systems are statistically significant, we estimate the following model: Qt = α 1 + β 1 ⋅ ( HOD) t + ε t

[6]

where Qt indicates the value of the quality measure for transaction t; HODt is a dummy variable that equals 1 if the observation (quote or trade) occurs under the HOD regime, and 0 if the observation occurs under the POD regime, ε t is a white noise random variable i.i.d. (0, σ ε2 ) .

17

A significant estimate of β1 would indicate that the quality measure is different under the HOD system compared to the POD system. Note that αˆ1 is the mean estimate of the quality measure for the POD system, and βˆ1 gives the difference between HOD and POD systems: positive (negative) values imply an improvement (worsening) when moving from POD to HOD.

6.3. Event Study with Control Sample Methodology In this approach we compare quality indicators for firms that move to the HOD system (switching firms, henceforth) with the same indicators, during the same event window for firms that do not change trading location (control firms). Thus, the combined event study (CES) approach tests the significance of the difference between quality measure differences across trading structures after controlling for any systematic changes. Let ∆ t be the difference in quality measures between switching and nonswitching (control) firms:

∆ t = Qt ( NS ) − Qt ( SW )

[7]

where NS refers to the control stock (non-switching), SW refers to switching firms. To test for a statistically significant differential effect due to market structure, we estimate the following model:

∆ t = α 2 + β 2 ⋅ ( HOD) t + ε t

[8]

A significant estimate of parameter β 2 ( βˆ 2 ) quantifies the effect due to new trading regime HOD. Parameter estimate αˆ 2 measures the difference between switching and non-switching stocks in the HOD regime.18

If the control sample is properly chosen, parameter estimate for α 2 should not be different from zero. However, even when αˆ 2 is different from zero, by assuming that this value is constant across the two 18

periods (pre- and post-switching) then βˆ2 still measures the differential effect induced by the adoption of the HOD regime.

18

To consider the relative (percentage) effect of the microstructure changes due to market structure, we also specify a logarithmic version of [8]. Define Φ t =

Qt ( NS ) , then Qt ( SW )

the model we estimate is: ln Φ t = α 3 + β 3 ⋅ HODt + ε t

[9]

Estimates of β 3 coefficient ( βˆ3 ) can be interpreted as the pre-post (continuous) rate of variation of the differential between switching and non-switching firms:

βˆ3 = ln Φ ( post ) − ln Φ ( pre) = ln

Φ ( post ) . Φ ( pre)

The estimate of α 3 ( αˆ 3 ) approximates the percentage difference between benchmark and control stocks in the pre-switching period.

6.4. Estimation of the Adverse Selection Component of the Spread

To test Hypothesis II we need to estimate the adverse selection component of the spread. We consider a modified version of George, Kaul, and Nimalendran (1991) model, proposed by Neal and Wheatley (1998), that allows the spread to vary over time:

LRDt = ∆pt − ∆mt =

π 2

⋅ (Dt ⋅ QPS t − Dt −1 ⋅ QPS t −1 ) =

π 2

⋅ ∆(Dt ⋅ QPS t )

[10]

where pt is the logarithm of the transaction price; mt is the logarithm of the midquote; Dt is the trade direction indicator variable, with Dt = 1 in case of buy order and Dt = −1 in case of sell order; QPSt is the quoted percentage spread; π is the order processing component of the spread. To evaluate the impact of the trading system (POD Vs. HOD), we introduce in equation [10] a multiplicative dummy:

LRDt = (π 1 + π 2 ⋅ HODt ) ⋅ ∆( Dt QPS t ) + ε t

π 1 ⋅ ∆( Dt QPS t ) + ε t  = (π 1 + π 2 ⋅ HODt ) ⋅ ∆( Dt QPS t ) + ε t

19

if

HODt = 0

if

HODt = 1

[11]

where HODt = 0 if the observation occurs under the POD regime and HODt = 1 if the observation occurs under the HOD regime. Given that (1 − π 1 ) measures the adverse selection component of the spread, − πˆ 2 represents an estimate of the variation in the adverse selection component induced by the trading system switching.

7.

Results

Figure 2 shows the behavior of the daily average quoted spread, computed as in equation [2], for all the switching stocks in the event window (-75, 74). The figure provides graphical evidence that a structural decline in trading costs coincides with the switching date. In addition to the decrease in the level of the spread there is also a reduction in spread variability subsequent to the adoption of the hybrid system. Figure 3 shows the daily average market depth, computed as in [4], for all the switching stocks. Again, from graphical inspection there is a consistent increase in market depth. Market quality index also improves, and this is obvious given that it is simply the depth-to-spread ratio. To test more formally the effects of the microstructure change, we estimate equation [6] for monetary turnover, quoted bid-ask spread, market depth, and market quality index. For each of these variables we estimate a regression model with a dummy variable equal to one for the period under the HOD system. This is the simple event study approach. The results are reported in Table 4. We find that under the HOD system, volume turnover increases from an average of 43 million of Itl Lira under the POD system to 86.5. This is a more than 100 percent increase in volume turnover, which is economically and statistically significant even considering the double counting effects induced by specialist intermediation. Next we also document that quoted percentage spreads decrease from 2.79 percent to 1.98 percent (decrease of 81 basis points). This is a significant decrease of 29 percent from the POD to HOD system. The market depth increases substantially from 14.41 million of Itl Lira to 21.33, or a significant 48 percent increase. Finally, the market quality index, which is the

20

ratio of depth-to-spread also, increases significantly under the HOD system. The quality index improves from 0.80 to 1.45, which is an increase of 81 percent. However, the overall market quality improvement could also be only apparent. Even coinciding with the system switching, it could in principle be affected by (i.) company-specific and/or (ii.) market-wide factors, and not just by the adoption of the hybrid system. To control for company-specific factors as source of possible bias, we use multivariate regression models with a HOD dummy variable in addition to a set of control variables: Qt = α 1 + β 1 ⋅ ( HOD) t + Control Variables + ε t

[6-bis]

The results are reported in Table 5. All the signs of the coefficients are consistent with those expected on the basis of previous studies, the fit of the regressions greatly improve, and – most importantly – the trading system switching dummy is still highly significant and of the same sign as before. Those results confirm our previous findings even controlling for the influence of company specific factors. To control for market-wide factors as source of possible bias, we consider both switching and non-switching stocks in a combined event study analysis. Results for the quoted percentage spread variable are reported in Table 6. Part A of the table presents OLS estimates of models [8] and [9], where we consider as dependent variable the simple and logarithmic differences between switching and control stocks spreads in the pre- and post-event period. The combined approach confirms our previous results. For the full sample 75% (81.5% using the log version) of the stocks experience a statistically significant reduction of the spread at 5% level. None of the firms moving to the HOD system experience an increase in trading costs. To reconcile parameters’ estimates with economic interpretation, part B of Table 6 presents sample averages for ∆ t and ln φ t . Comparison of part A and part B results is broadly consistent with the economic interpretation provided in Section 6.3. Up to this point, all “thinly-traded” stocks – as defined by the Italian Stock Exchange – seem to benefit from specialist intervention. Thus, an important subsequent question arising is whether the effects of specialist intervention are constant across

21

stocks. For example, Grossman and Miller (1988) argue that stocks with very low liquidity are likely to benefit the most under a specialist system. To test this hypothesis, we partition the sample into two groups - very illiquid stocks (Thin Subsample) and moderately illiquid stocks (Moderately Thin Subsample).19 The results seem interesting. The magnitude of the improvement is much larger for the “very thin” subsample. The very thin stocks switching to the HOD system experience a reduction of 77% for the quoted spread, while trading costs for less “thinlytraded” stocks only drops by 17%. These results are consistent with the hypothesis that specialist participation can improve market quality for “thinly-traded” stocks, and that very “thinly-traded” stocks would benefit the most from moving to a hybrid system. Overall, the results of the combined approach support the hypothesis that the HOD system offers lower executions costs compared to the POD only for very inactive stocks. Based on these exploratory results we conduct further analysis in order to check whether the impact of the microstructure change vary with stock trading activity or market capitalization. To answer this question, we first partition the sample by quintiles in terms of volume turnover. Next we compute, for each indicator of market quality and each stock, averages of the post-switching and pre-switching periods. Then we compute the ratio between post-switching and pre-switching measures for each indicator and each stock. Lastly, we average those ratios across stocks belonging to the same quintile. The post/pre ratio indicates the relative change in the market quality induced by the microstructure change. Figure 4 presents the behavior of the post/pre ratios computed for spread, depth, and market quality index. From visual inspection of Figure 4 it seems that the relative impact of the specialist intervention varies with quintiles: for very “thinlytraded” stocks (1st quintile) and high volume (5th quintile) stocks, the benefits – where existing – are minimal, whereas the benefits are the greatest for mid-positioned (i.e., 3rd quintile) “thinly-traded” stocks.20

19

We classify moderately illiquid stocks as those whose market capitalization, as of December 31, 1997 was greater than three (two) times the median (mean) of the sample. 20 The daily average monetary turnover for quintile 3 stocks is about $ 200,000, and the average switching date market capitalization is about $ 80 million.

22

We need, as before, to control for the behavior of the non-switching stocks. Thus, we compute the net change for each quality indicator as the difference between the change experienced by switching stocks and the change of control stocks: Q  Q  Net change =  post −  post   Q Q switching pre  pre  non - switching   stocks stocks

[12]

where Q post ( Q pre ) is the average value of the quality indicator in the post(pre)-switching period. Figure 5 plots the behavior of the net change across quintiles. The improvements experienced by switching stocks are still very relevant. Interestingly, also the bell-shaped pattern resists after controlling for the behavior of the non-switching stocks. To test the hypothesis that a specialist can reduce asymmetric information costs (Hypothesis II), we compare estimates of the adverse selection component of the spread under the two trading regimes. Table 7 presents OLS estimates of model [9]. The adverse selection component represents 54.1% of the spread for the full sample, 54.8% for the very illiquid stocks, and 53% for the moderately illiquid stocks. More importantly, the change induced by the adoption of the HOD system, captured by the multiplicative dummy is statistically significant for the full sample (-2.7%) and for the very thin subsample (-3.7%), and it is not significant for the moderately thin stocks. Thus, specialist intervention leads to a reduction of the adverse selection costs for very “thinlytraded” stocks. Heidle and Huang (2002) analyze a sample of 96 companies that switched exchange listing during 1996 and report results consistent with ours. They find that the probability of informed trading declines by 35% when the firm relocates from NASDAQ to NYSE, the bid-ask spread improvements observed when the firms change listing are related to the decline in the probability of informed trading, and illiquid firms benefit most from a more transparent market. To quantify the total effect induced by the adoption of the HOD system, we need to consider jointly the reduction in the adverse selection component of the spread and the reduction in the spread. The percentage reduction in the adverse selection component ( %∆AS ) is given by

23

%∆AS =

(1 − π 1 − π 2 ) ⋅ QPS HOD − (1 − π 1 ) ⋅ QPS POD (1 − π 1 ) ⋅ QPS POD

=

(1 − π 1 − π 2 ) ⋅ (1 − m)QPS POD − (1 − π 1 ) ⋅ QPS POD (1 − π 1 ) ⋅ QPS POD

=

− π 2 − m ⋅ (1 − π 1 − π 2 ) (1 − π 1 )

[13]

where m indicates the percentage change of the spread when moving from POD to HOD system. The change in adverse selection costs is reported in Table 8. We find that the moving to a HOD system leads to a 42.5 percent reduction in the adverse selection component of the spread for very “thinly-traded” stocks, while it drops by 26.3 percent for moderately thin stocks. These results are consistent with Hypothesis II that specialists reduce information asymmetry problems and improve market quality for “thinly-traded” stocks. Hypothesis III states that, in an electronic order driven market, quoted spread is lower or equal to the effective spread computed with actual transaction prices. The intuition is that the average trade price differs from the best quote when order quantity is greater than the quantity quoted on the limit order book for the best quote (i.e., the best depth). In this case, the excess quantity is traded at the next-best quote. In table 9 we report currently quoted and effective spreads statistics for HOD and POD “thinly-traded” stocks.21 For the full sample, the difference between QPS_T and EPS averages 12 basis points, which is 8% of the mean quoted spread. T-statistics for difference in mean reject the null hypothesis of equal means for 81% (75%) of the sample at 5% (1%) level. The correlation coefficient for (QPS_T, EPS) averages 0.8. It is interesting to note that Petersen and Fialkowski (1994) document a 0.1 correlation between quoted spread and effective spread. They also document that on the NYSE the effective spread is lower than the quoted spread due to specialists’ price improvements (Ready, 1999). By contrast, in electronic order driven markets, we find that quoted spread represents the lower bound of trading costs. 21

Currently quoted spread is a version of the quoted bid-ask spread that considers only quotes in effect at the time a transaction occurred. This is to make it comparable with the effective spread, that is computed using effective transaction prices.

24

The difference between quoted and effective spreads can also be interpreted as a proxy for the slope of demand and offer schedules. Larger (narrower) jumps and discontinuities of the transaction price imply more (less) deep demand and offer curves.22 The magnitude of the difference between quoted and effective spreads in an electronic order driven environment raises interesting questions about the market depth as a contributor to the overall stock liquidity. Empirical evidence shows that this difference is larger for HOD than POD stocks: -8% (-9%) vs. -3% (-2%). Further, it is interesting that for the “very thin” subsample the pattern is more pronounced: the difference between quoted and effective spread is about 9% for HOD stocks and about 2% for POD stocks. This result suggests that (i.) a trade-off between spread and depth exists, (ii.) this tradeoff is more severe for HOD stocks, and (iii.) for very thin stocks smaller spreads are associated with lower depth. Designated specialists face a regulatory constraint that imposes a maximum allowable spread, and the specialist complies with the spread constraint, but quotes lower depth at the best quotations. This explanation is also consistent with strategic use of the quantity side of the schedule by liquidity suppliers (Lee, Mucklow, and Ready, 1993), and particularly by specialists (Kavajecz, 1999).

8.

Conclusions

In this paper we examine the impact of specialist intervention for trading illiquid (i.e., “thinly-traded”) stocks. Empirical evidence from the Italian Stock Exchange shows that a hybrid order driven system – with a specialist and a limit order book providing liquidity – offers better market quality than a pure order driven structure. In particular, we find significant improvements under the hybrid system compared to a pure order driven system across several quality metrics –trading activity, volume turnover, bid-ask spread, depth, depth-to-spread ratio, and adverse selection costs. We also find that “very thinly” traded stocks benefit more than less inactive stocks from the adoption of a hybrid trading system with specialist intervention.

22

Sandäs (2001) analyzes a pure limit order market, the Stockholm Stock Exchange, and attributes the slope of the price schedule to the adverse selection costs.

25

Results are robust to different variable specification (quoted, currently quoted, and effective spreads); different sampling periods; different comparative approaches (simple event study and combined approaches). Our results are also consistent with related empirical evidence. Madhavan and Sofianos (1998) analyze a sample of NYSE specialist and provide evidence that specialist services are most valued for “thinlytraded” stocks. Kavajecz (1999) compares the spread on a limit order book with the specialists quoted spread; he argues that specialists play an important role in narrowing the spread, especially for smaller and less frequently traded stocks. Chung, Van Ness, and Van Ness (1999) analyze the interaction between limit orders, specialist, and bid-ask spread; they find that specialists provide greater liquidity to low volume stocks. Heidle and Huang (1999) examine firms switching exchange listing during 1996 and analyze the probability of information-based trading across different exchanges; their findings are consistent with ours in terms of lower adverse selection when moving to a specialistbased market like NYSE. Furthermore, our results are also consistent with the behavior of the Italian Stock Exchange. The Italian market authorities have progressively reduced the number of stocks included in the Inactive Stocks segment of ISE that are eligible for the hybrid order driven system. On May 1997, when the Thin Stocks Project started, the Inactive Stocks segment was composed of the 100 less traded stocks of the ISE. In 1998, the segment contained only the last 80 thinly trade stocks, and this was further reduced to the last 50 stocks in 1999.23 Hence, less “thinly-traded” stocks have been progressively excluded from the Inactive Stocks segment, and from the possibility to opt for the hybrid order driven system. This behavior is consistent with our findings that specialist intervention is more valuable for very “thinly-traded” stocks. The third hypothesis we examine relates to the relationship between quoted and effective spreads and highlights a specific feature of electronic order driven markets. The difference between quoted and effective spreads assumes a completely different meaning 23

In April 2001, the Italian Stock Exchange launched a new market segment, named STAR, in substitution of the Inactive Stocks segment. The STAR market segment is built upon Thin Stocks Project principles. In fact, specialist intervention is required for each stock listed in the STAR segment. Additionally, companies listed in STAR must comply with restrictive requirements in terms of transparency, free float, and corporate governance (e.g., independent directors, internal control committees, etc.).

26

in an order driven markets relative to a quote driven ones. We find that in electronic order driven systems, quoted spread is significantly lower than effective spread. This evidence contrasts with quote driven (e.g., NASDAQ) and non-automated (e.g., NYSE) markets where dealers and specialists can provide price improvements relative to posted quotes. Further, the difference between quoted and effective spreads can be interpreted in a limit order book environment as proxy for the depth dimension of the overall stock liquidity. Empirical evidence reveals the presence of a trade-off between spread and depth, and that such trade-off is more severe for HOD traded stocks. This last finding may be attributed to the strategic behavior of the specialist in managing the quantity side of the book.

27

References Affleck-Graves, J., Hedge, S.P., Miller, R.E., 1994, Trading Mechanism and the Components of the Bid-Ask Spread, Journal of Finance 49, 1471-1488. Amihud, Y., Mendelson, H., 2000, The Liquidity Route to a Lower Cost of Capital, Journal of Applied Corporate Finance, winter, 8-25. Bacidore, J.M., 1997, The Impact of Decimalization on Market Quality: An Empirical Investigation of the Toronto Stock Exchange, Journal of Financial Intermediation 6, 92-120. Benveniste, L.M., Marcus, A.J., Wilhelm, W.J., 1992, What’s Special About the Specialist?, Journal of Financial Economics 32, 61-86. Bessembinder, H., 1999, Trade Execution Costs on Nasdaq and the NYSE: A PostReform Comparison, Journal of Financial and Quantitative Analysis 34, 387-407. Bessembinder, H., Kaufman, H.M., 1997, A Comparison of Trade Execution Costs for NYSE and NASDAQ-Listed Stocks, Journal of Financial and Quantitative Analysis 32, 287-310. Biais, B., 1993, Price formation and equilibrium liquidity in fragmented and centralized markets, Journal of Finance 48, 157-84. Biais, B., Glosten, L., Spatt, C., 2002, The Microstructure of Stock Markets, CEPR Discussion Paper # 3288. London, Centre for Economic Policy Research. Christie, W.G., Huang , R.H., 1994, Market Structures and Liquidity: A Transactions Data Study of Exchange Listings, Journal of Financial Intermediation 3, 300-326. Chung, K.H., Van Ness, B.F., Van Ness, R.A., 1999, Limit Orders and the Bid-Ask Spread, Journal of Financial Economics 53, 255-287. Easley, D., Kiefer, N.M., O’Hara, M., Paperman, J.B., 1996, Liquidity, Information, and Infrequently Traded Stocks, Journal of Finance 51, 1405-1436. FESE, 2002, European Stock Exchange Statistics, Bruxelles, june. Available at http://www.fese.be/ Demarchi, M., Foucault, T., 2000, Equity Trading Systems in Europe: A Survey of Recent Changes, Annales d’Economie et de Statistique 60, 73-115. Garbade, K., Silber, W., 1979, Structural Organization of Secondary Markets: Clearing Frequency, Dealer Activity and Liquidity Risk, Journal of Finance 34, 577-593. Garcia Coto, D., 1996, The Liquidity Tool, Federation of European Stock Exchange. George, T. J., Kaul, G., Nimalendran, M., 1991, “Estimation of the Bid-Ask Spread and its components: A new approach,” Review of Financial Studies 4, 623-656. Glosten, L., 1989, Insider Trading, Liquidity, and the Role of the Monopolist Specialist, Journal of Business 62, 211-236.

28

Glosten, L., Milgrom, P., 1985, Bid, Ask and Transaction Prices in a Specialist Market with Heterogeneously Informed Traders, Journal of Financial Economics 14, 71100. Grossman, S., Miller, M., 1988, Liquidity and Market Structure, Journal of Finance 43, 617-637. Hasbrouck, J., Sofianos, G., Sosebee, D., 1993, Orders, trades, reports and quotes at the New York Stock Exchange, New York Stock Exchange. Heidle, H., Huang, R.D., 2002, Information-Based Trading in Dealer and Auction Markets: An Analysis of Exchange Listings, Journal of Financial and Quantitative Analysis, forthcoming. Huang, R.D., Stoll, H.R., 1996, Dealer versus action markets: A paired comparison of execution costs on NASDAQ and the NYSE, Journal of Financial Economics 41, 313-357. Italian Stock Exchange, 1996, Alternative Solutions to Improve Market Liquidity for Thinly Traded Stocks, consultative paper, Milan (in Italian). Kavajecz, K.A., 1999, A Specialist’s Quoted Depth and the Limit Order Book, Journal of Finance 54, 747-771. Keim, D., Madhavan, A., 1997, Transaction Costs and Investment Style: An InterExchange Analysis of Institutional Equity Trades, Journal of Financial Economics 46, 265-292. Lee, C.M.C., Ready, M.J., 1991, Inferring Trade Direction from Intraday Data, Journal of Finance 46, 733-746. Lee, C.M.C., 1993, Market Integration and Price Execution for NYSE-Listed Securities, Journal of Finance 48, 1009-1038. Lee, C.M.C., Mucklow, B., Ready, M.J., 1993, Spreads, Depths, and the Impact of Earnings Information: An Intraday Analysis, Review of Financial Studies 6, 345374. Madhavan, A., 1992, Trading Mechanisms in Securities Markets, Journal of Finance 47, 607-641. Madhavan, A., 2000, Market Microstructure: A Survey, Journal of Financial Markets 3, 205-258. Madhavan, A., Sofianos, G., 1998, An empirical analysis of NYSE specialist trading, Journal of Financial Economics 48, 189-210. Neal, R., 1992, A Comparison of Transaction Costs between Competitive Market Maker and Specialist Market Structures, Journal of Business 65, 317-334. Neal, R., Wheatley, S.W., 1998, “Adverse selection and Bid-Ask Spreads: Evidence from Closed-End Funds,” Journal of Financial Markets 1, 121-149. O’Hara, M., 1995, Market Microstructure Theory (Cambridge, MA: Blackwell Publishing).

29

Pagano, M., 1989, Trading Volume and Asset Liquidity, Quarterly Journal of Economics 104, 255-274. Pagano, M., 1997, Market Size, the Informational Content of Stock Prices and Risk: A Multiasset Model and Some Evidence. In: Battigalli, P., Montesano, A., Panunzi, F. (edited by), Decisions, Games and Markets, (Boston, MA: Kluwer). Pagano, M., Roëll, A., 1996, Transparency and Liquidity: A comparison of Auction and Dealer Markets with Informed Trading, Journal of Finance 51, 579-611. Petersen, M.A., Fialkowski, D., 1994, Posted versus effective spreads: Good prices or bad quotes?, Journal of Financial Economics 35, 269-292. Porter, D.C., Weaver, D.G., 1996, Estimating Bid-Ask Spread Components: Specialist versus Multiple Market Maker Systems, Review of Quantitative Finance and Accounting 6, 167-180. Ready, M.J., 1999, The Specialist’s Discretion: Stopped Orders and Price Improvement, Review of Financial Studies 12, 1075-1112. Sandäs, P., 2001, Adverse Selection and Competitive Market Making: Empirical Evidence from a Limit Order Market, Review of Financial Studies 14, 705-734. Viswanathan, S., Wang, J.J.D., 2002, Market Architecture: Limit Order Books versus Dealership Markets, Journal of Financial Markets 5, 127-168.

30

Figure 1: Microstructure of the Italian Stock Exchange

Active Stocks

Trading Hours

Trading Mechanism a b

Inactive Stocks

Opening

Continuous

Opening

Continuous

8 am – 9:30 am

9:30 am – 5:30 pm

8 am – 12 pm

12 pm – 3:30 pm

Call Auction

PODa

Call Auction

PODa or HODb

Pure order driven with electronic limit order book (POD) Hybrid order driven market with limit order book and specialist (HOD)

31

Figure 2: Quoted Percentage Spread Relative to Switching Date from POD to HOD 6

5

4

3

2

1

0

Number of days relative to the switching date

HOD Regime

POD Regime

32

Figure 3: Market Depth Relative to Switching Date from POD to HOD 50 45 40 35 30 25 20 15 10 5 0

Number of days relative to the switching date

POD Regime

HOD Regime

33

0.80

4.00

0.70

3.50

0.60

3.00

0.50

2.50

0.40

2.00

0.30

1.50

0.20

1.00

0.10

0.50

0.00

0.00 Quintile 1

Quintile 2

QPS (left scale)

Quintile 3

DEPTH (right scale)

34

Quintile 4

Quintile 5

MQUAL (right scale)

Post/pre ratio

Post/pre ratio

Figure 4: Market Quality Variables by Quintile: Only Switching Stocks

Figure 5: Market Quality Variables by Quintile: Switching relative to Non-Switching Stocks 2.50

0.30 0.10

2.00 1.50

-0.30 -0.50

1.00

-0.70 0.50

-0.90 -1.10

0.00

-1.30 -1.50

-0.50 Quintile 1

QPS (left scale)

Quintile 2

Quintile 3

DEPTH (right scale)

35

Quintile 4

Quintile 5

MQUAL (right scale)

Net Change

Net Change

-0.10

Table 1 – Thinly-Traded Stocks on the Main European Stock Exchanges This table, drawn from Garcia Coto (1996), presents data on number, market capitalization, and trading volume of “thinly-traded” stocks listed on the main European stock exchanges. Stocks with a turnover ratio lower than the 25% of the overall average market turnover are considered “thinly traded”.

Number of stocks

Thinly-traded stocks with respect to the overall market

Total

Thinly-traded

Number (%)

Capitalization (%)

Trading volumes (%)

Amsterdam†

169

25

14.8

0.9

0.1

Athens†

118

20

16.9

13.0

1.9

238

50

21.0

12.8

2.9

456

254

55.7

26.5

2.1

Helsinki

66

10

15.2

2.2

0.3

Italy

315

108

34.3

17.6

3.3

Lisbon

89

18

20.2

8.3

1.2

2,158

293

13.6

4.4

0.7

156

53

34.0

1.2

1.7

111

19

17.1

2.8

0.3

270

76

28.1

9.2

1.2

251

102

40.6

8.2

0.8

25.9

8.9

1.4

Copenhagen Germany

†

London† †

Madrid Oslo

†

Paris† †

Switzerland

Equally-weighted average †

Estimates resulting from a market sample.

36

Table 2 – Sample This table presents stocks belonging to the Inactive Stocks segment of the ISE traded with the hybrid order driven (HOD) mechanism. For each HOD-traded stock the table indicates the sub-industry group (Source: Italian Stock Exchange Corp.), the designated specialist, the switching date from the POD to the HOD mechanism, and (if available) the matched POD-traded stock. Matching criteria are exposed in the body of the paper. The term ‘na’ stands for ‘not available’ matching stock.

Control Sample HOD-Traded Stock

Sub-Industry Class

Specialist

Operating From

Bna Ord Bna Priv Bna Rnc Finmeccanica Ord. Finmeccanica Risp. Gabetti Holding Ord Linificio Ord Linificio Rnc Trenno Ipi Terme Acqui Ord Terme Acqui Rnc Aedes Ord Aedes Risp Maffei Ord Vianini Lavori Ord Mittel Ord Vittoria Ass. Ord Bonifiche Ferraresi Nai - Nav. Alta Italia

Banking Banking Banking Electronics Electronics Real Estate Textiles Textiles Financial Misc. Real Estate Financial Misc. Financial Misc. Real Estate Real Estate Mineral & Oil Construction Financial Holding Insurance Industrial Misc. Transportation

Bca Roma Bca Roma Bca Roma Aletti Sim Aletti Sim Intersim Euromobiliare Euromobiliare Aletti Sim Intersim Finnat Euramerica Finnat Euramerica Intercassa Sim Aletti Sim Comit Montepaschi Caboto Caboto Aletti Sim Mediosim

May 19, 1997 May 19, 1997 May 19, 1997 May 19, 1997 May 19, 1997 May 19, 1997 May 19, 1997 May 19, 1997 May 19, 1997 June 9, 1997 June 23, 1997 June 23, 1997 July 7, 1997 July 7, 1997 September 8, 1997 November 25, 1997 December 9, 1997 December 9, 1997 January 2, 1998 January 2, 1998

37

Pair # 1

2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

POD-Traded Stock Bca Toscana na na Magneti Marelli Ord Magneti Marelli Rnc CMI Centenari e Zinelli na La Gaiana Monrif Rotondi na Risanamento Napoli Ord Risanamento Napoli Rnc Smurfit Sisa Boero Bartolomeo Riva Fin Allianz Subalpina Garboli AM Ausiliare

Table 3 – Sample Descriptive Statistics This table presents the distributional characteristics of HOD-traded stocks, matched POD-traded stocks, and the full sample. Market Capitalization stands for the factor firm size, is expressed in billion of Italian Lira, and is equal to the number of shares outstanding as of the end of 1997 times the closing price at Dec. 31, 1997. Price is the closing price at Dec. 31, 1997, expressed in Italian Lira. The volatility measure (Volatility) is the standard deviation of daily log returns. Bookto-Market is the ratio between the book value of common equity in billion of Itl. Lira and the market value of common equity in billion of Itl. Lira. Leverage is the market-based measure of leverage and is computed as the ratio of book assets to market value of equity. T-values of the difference in means (T-test) are reported.

Market Cap (T_MKTCAP) Price (CLO_PRC) Volatility (LVOL_L) Book-to-Market (B_M) Leverage (A_EQTY)

‡

HOD POD T-test Full Matched Sample HOD POD T-test Full Matched Sample HOD POD T-test Full Matched Sample HOD POD T-test Full Matched Sample HOD POD T-test Full Matched Sample

Mean

Std Error of The Mean

Min

Max

25th %ile

Median

75th %ile

314.10 277.86 0.217 ‡ 295.98 3,594.00 5,744.12 1.139 ‡ 4,669.07 0.079 0.105 1.092 ‡ 0.092 0.953 1.029 0.518 ‡ 0.991 5.23 7.33 0.522 ‡ 6.28

124.72 111.09

19.45 16.35

1424.30 1516.64

62.10 47.16

164.00 94.31

235.52 274.93

82.02 941.86 1,636.20

16.35 260.00 173.50

1516.64 12,910.00 24,500.00

51.00 1,117.55 1,699.00

123.89 2,249.50 3,525.00

245.31 4,418.50 6,600.00

948.48 0.014 0.019

173.50 0.026 0.032

24,500.00 0.195 0.274

1,377.5 0.031 0.038

2,695.00 0.045 0.088

5,245.00 0.123 0.165

0.011 0.094 0.113

0.026 0.46 0.24

0.274 1.85 1.72

0.037 0.75 0.67

0.088 0.91 1.12

0.146 1.09 1.27

0.073 1.600 3.702

0.24 1.05 1.08

1.85 22.23 53.51

0.68 1.50 1.82

1.01 2.43 2.41

1.22 8.31 4.18

1.989

1.05

53.51

1.66

2.42

6.27

indicates that the difference is not significant at 10% level.

38

Table 4 – Regression Models for Market Quality Variables with Dummy for HOD regime This table presents OLS estimates of model [6] for the full sample. HOD is a dummy variable that equals 1 if the observation occurs under the HOD regime. LCONVOL indicates the daily average volume turnover in millions of Itl. Lira. QPS indicates the daily average quoted percentage spread. DEPTH indicates the monetary value of the depth in millions of Itl. Lira. MQUAL stands for the depth-spread ratio.

Dependent Variable: LCONVOL

QPS

DEPTH

MQUAL

Indipendent Variables: Intercept

Par T-stat

3.76** 72.43

2.79** 45.44

14.41** 31.77

0.80** 11.71

Trading System Par Switching (HOD) T-stat

0.70** 9.91

-0.81** -9.72

6.92** 11.24

0.65** 7.00

F-test Prob > F Adj R^2

98.21 0.0001 0.05

94.41 0.0001 0.05

* (**) indicates significance at 5% (1%) level.

39

126.36 0.0001 0.07

48.99 0.0001 0.03

Table 5 – Regression Models with Control Variables This table presents OLS estimates of model [6-bis]. PERRET (PERRET1) indicates the (lagged) daily percentage return. CONTRANS indicates the daily number of trades. LOGVOL indicates the standard deviation of the intraday log-returns. INVLPRC indicates the inverse daily closing share price. LCONAVGV is the daily average volume turnover per trade in millions of Itl. Lira. For other variables definition please refer to the previous table.

Dependent Variable: LCONVOL

QPS

DEPTH

MQUAL

Indipendent Variables: Intercept

Par T-stat

1.10** 10.33

4.25** 22.13

-10.45** -6.00

-2.33** -9.17

Trading System HOD=1

Par T-stat

0.08* 2.27

-0.46** -6.59

4.42** 7.61

0.49** 5.23

Bid-Ask Spread (QPS)

Par T-stat

-0.17** -14.22

Return (t) (PERRET)

Par T-stat

0.002* 2.25

Return (t-1) (PERRET1)

Par T-stat

# of Trades (CONTRANS)

Par T-stat

0.03** 41.00

0.009** 4.50

Volatility (LOGVOL)

Par T-stat

0.90* 2.36

1.196** 3.23

Price inverse (INVLPRC)

Par T-stat

0.229** 4.65

Volume turnover (LCONVOL)

Par T-stat

-0.66** -14.31

Trade Size (LCONAVGV)

Par T-stat F-test Prob > F Adj R^2

-0.40 -1.95

0.0004 0.84

1.35** 32.74 739.57 0.00 0.77

0.34** 3.22 100.80 0.00 0.28

* (**) indicates significance at 5% (1%) level.

40

0.11** 6.76 -0.50 -0.16

0.002 0.92 -0.256 -0.52 1.082** 16.61

0.46 1.16

0.42** 6.82

10.17** 12.07

0.27* 1.98

134.83 0.00 0.34

91.65 0.00 0.26

Table 6 – Combined Event Study Part A of this table presents OLS estimates of of models [8] and [9]. The variable under observation is QPS (Quoted Percentage Spread). Regressions are estimated by pairs of stocks (switching, non-switching) on daily observations. T-stat for full sample and subsamples are computed using the Neal e Wheatley (1998) procedure. Part B presents sample averages for Delta and ln(Phi) variables. Delta is the absolute spread difference between stocks in the same pair (SW, NS). ln(Phi) is the natural logarithm of the ratio of the non-switching and switching stock spreads. Each measure is computed for the periods before (Pre) and after (Post) the change in trading mechanism.

Model [8] Full Sample

Very Thin Subsample

Model [9] Moderately Thin Subsample

Full Sample

Very Thin Subsample

Moderately Thin Subsample

Part A Alfa 2 T-stat

-0.75 ** -6.11

-0.54 -0.17

Beta 2 T-stat

2.97 ** 12.69

3.5 ** 12.42

-1.69 ** -13.75

-0.12 ** -4.15

0.02 ** 3.58

-0.7 ** -17.03

0.66 ** 3.44

0.66 ** 14.84

0.77 ** 15.21

0.17 * 2.63

Part B Delta pre-switching Delta post-switching

0.08 1.46

0.41 2.00

-1.36 -0.87

ln(phi) pre-switching ln(phi) post-switching

0.02 0.5

* (**) indicates significance at 5% (1%) level.

41

0.09 0.63

-0.87 -0.55

Table 7 – Estimation of the Adverse Selection Component of the Spread This table presents OLS estimates of model [9]. Regressions are presented for the full sample, the very thin subsample, the moderately thin subsample. The adverse selection component of the spread is expressed as a percentage of the quoted spread. Regression specifications (1), (3), and (5) include an intercept; regression specifications (2), (4), (6) exclude the intercept. For specifications (2), (4), and (6) the R2 has been redefined.

Full Sample (1)

(2)

0.459 ** 207.51

0.459 ** 207.52

0.452 ** 148.69

0.452 ** 148.68

0.470 ** 150.25

0.470 ** 150.25

HODt ⋅ ∆( Dt ⋅ QPSt )

Par T-stat

0.027 ** 9.09

0.027 ** 9.09

0.037 ** 9.59

0.037 ** 9.59

0.01 1.13

0.01 1.13

F-test Prob > F Adj R^2

53032 0.00 0.78

53033 0.00 0.78

33603 0.00 0.79

33605 0.00 0.79

18761 0.00 0.76

18761 0.00 0.76

0.541

0.541

0.548

0.548

0.53

0.53

-0.027

-0.027

-0.037

-0.037

--

--

* (**) indicates significance at 5% (1%) level.

42

-0.001 -1.04

(6)

Par T-stat

−πˆ 2

--

(5)

∆ ( Dt ⋅ QPS t )

AS Change

0.001 0.45

(4)

Par T-stat

(1 − πˆ1 )

--

(3)

Moderately Thin Subsample

Intercept

AS Component

0.001 0.21

Very Thin Subsample

--

Table 8 – The Adverse Selection Component of the Spread This table presents the quoted percentage spread (QPS), the adverse selection component expressed as a percentage of the spread (% of QPS), the adverse selection component quantified in percentage terms (Absolute Value = QPS⋅ % of QPS), the percentage change induced by the adoption of the HOD system (%∆AS). All figures are reported for the full sample, the very thin subsample, and the moderately thin subsample.

Full Sample

Very Thin Subsample

Moderately Thin Subsample

POD

HOD

POD

HOD

POD

HOD

QPS (%)

2.79

1.98

3.81

2.35

2.28

1.68

% of QPS

0.541

0.514

0.548

0.511

0.53

0.53

Absolute Value (%)

1.51

1.02

2.09

1.20

1.21

0.89

%∆AS

-0.326

-0.425

43

-0.263

Table 9 – Quoted Versus Effective Spreads This table compares currently quoted spread (QPS_T) and effective spread (EPS) for HOD and POD “thinly-traded” stocks. The table reports the average currently quoted spread (QPS_T), the average effective spread (EPS), the correlation coefficient between QPS_T and EPS (Corr), the absolute spreads difference (Difference) along with t-test for the null hypothesis of zero difference. T-stat for full sample and subsamples are computed using Neal e Wheatley (1998) procedure.

HOD Stocks Full Sample QPS_T EPS Difference T-stat Corr (QPS_T, EPS)

Very Thin Subsample

1.49 1.61 -0.12 ** -22.3 0.80

1.50 1.64 -0.14 ** -21.1 0.80

POD Stocks Moderately Thin Subsample 1.46 1.51 -0.05 ** -7.6 0.82

* (**) indicates significance at 5% (1%) level.

44

Full Sample 2.91 2.99 -0.08 ** -19.2 0.89

Very Thin Subsample 3.40 3.48 -0.08 ** -13.1 0.90

Moderately Thin Subsample 0.82 0.89 -0.07 ** -16.9 0.84