Tick Size, Bid-Ask Spreads, and Market Structure

10/29/2001–coded–JFQA #36:4 Huang and Stoll Page 503 JOURNAL OF FINANCIAL AND QUANTITATIVE ANALYSIS VOL. 36, NO. 4, DECEMBER 2001 COPYRIGHT 2001, SCHOOL OF BUSINESS ADMINISTRATION, UNIVERSITY OF WASHINGTON, SEATTLE, WA 98195 Tick Size, Bid-Ask Spreads, and Market Structure Roger D. Huang and Hans R. Stoll� Abstract We propose a link between market structure and the resulting market characteristics— tick size, bid-ask spreads, quote clustering, and market depth. We analyze transactions data of stocks traded on the London Stock Exchange, a dealer market, and also traded as ADRs on the New York Stock Exchange, an auction market. We conclude that market characteristics are endogenous to the market structure. The London dealer market does not have a mandated tick size, and it exhibits higher spreads, higher quote clustering, and higher market depth than the NYSE auction market. Clustering of trade prices is similar in London and New York. I. Introduction In a financial market, the minimum tick size is the minimum allowable price variation. Minimum tick rules can apply to quotes, to trades, and to trade reports. On the New York Stock Exchange (NYSE), the minimum tick for quotes, for trades, and for trade reports was $1/8 until June 24, 1997, when the tick size was reduced to $1/16 (teenies). On the London Stock Exchange (LSE), there is no minimum tick size for quotes, trades, or trade reports. On the Nasdaq Stock Mar- ket, there was a minimum tick size of $1/8 for quotes until June 2, 1997. However, trades could take place at any price increment. Trade reports were rounded to the next eighth. In the Chicago Mercantile Exchange’s S&P 500 futures contract, the tick size is 0.10 index points or $25 per contract. Formal tick size rules do not exist in the LSE or in OTC bond markets and currency markets. The literature on market microstructure is replete with studies of attributes that affect or reflect market liquidity such as tick size, bid-ask spreads, quote �Huang, roger.huang.31@nd.edu, University of Notre Dame, Mendoza College of Business, Notre Dame, IN 46556; Stoll, hans.stoll@owen.vanderbilt.edu, Vanderbilt University, Owen Graduate School of Management, Nashville, TN 37203. We thank Mark Flannery, Sylvain Friederich, Jon Garfinkel, Frank Hatheway, Bob Jennings, Pete Kyle, Tom McInish, Andy Naranjo, Jay Ritter, Erik Sirri, Paul Schultz (the referee), and Paul Malatesta (the editor) for their comments and suggestions. We have also benefited from seminars at the London School of Economics, the Securities and Ex- change Commission, the University of Florida, the 1999 PACAP/FMA Conference in Singapore, and the Federal Reserve Bank of Atlanta Conference on Financial Markets 2000: E-Finance. This research was supported by the Dean’s Fund for Research and by the Financial Markets Research Center at the Owen Graduate School of Management, Vanderbilt University. 503 10/29/2001–coded–JFQA #36:4 Huang and Stoll Page 504 504 Journal of Financial and Quantitative Analysis clustering, and market depth. While we begin with tick size, the objective of this paper is to tie these various characteristics of markets into a more general view that reflects the underlying market structure. We examine the source and the impact of a minimum tick rule by considering stocks traded in different market structures. Specifically, we examine a set of stocks traded in London, where there is no mandated tick size, and also traded on the NYSE, where there is a mandated tick size. We conclude that market structure is the exogenous factor responsible for the presence of tick size rules and other market microstructure attributes. Related to the question of tick size is the empirical phenomenon of cluster- ing. Christie and Schultz (1994) used clustering on even eighths as evidence of implicit collusion on Nasdaq. We propose a measure of clustering that is akin to the Herfindahl Index, and we analyze clustering of quotes and of trade prices and the relation of clustering to spreads, tick size and market structure. In particular, we examine the hypotheses that price clustering is due to i) ease of negotiation, ii) implicit collusion, and iii) market structure. We conclude that market structure is responsible for the observed clustering. We also find evidence that suggests higher spreads in the LSE are partially offset by higher mandated depth behind the quotes. We trace the differential depth requirements to the differential market structures. Our general approach differs from many recent studies that focus on the effect of particular market features. For example, a flurry of recent studies, some prompted by the planned decimalization of the U.S. market, examines the pre- and post-effects of a reduction in tick size.1 On July 18, 1994, the Stock Exchange of Singapore reduced the minimum tick size for stocks trading at or above $25 from $0.50 to $0.10. Lau and McInish (1995) examine the effects of the reduction on both bid-ask spreads and market depth. The American Stock Exchange (AMEX) has reduced its tick size in stages. Crack (1995) and Ahn, Cao, and Choe (1996) examine the 1992 switch from 1/8th to 1/16th for stocks below $5. Ronen and Weaver (2000) examine the extension of the rule to all stocks on the AMEX on May 7, 1997. The Toronto Stock Exchange (TSE) moved from a fractional to a decimal trading system on April 15, 1996. Bacidore (1997), Porter and Weaver (1997), and Ahn, Cao, and Choe (1998) study the impact of the TSE’s switch on bid-ask spreads. The NYSE adopted the teenies on June 24, 1997. Bollen and Whaley (1998) and Goldstein and Kavajecz (1998) examine this change. Nasdaq changed the minimum quote increment from 1/8th to 1/16th for bid prices greater than $10 on June 2, 1997.2 Smith (1998) examines the change that occurred in the midst of a series of changes to implement the Order Handling Rules (OHR). 3 All these studies conclude that the adoption of a smaller tick size lowers spreads. The evidence on the market depth is less uniform but, by and large, suggests that it may be adversely affected. The most important difference in our study is its focus on the role of market structure in determining bid-ask spreads and tick size rules. In contrast, the earlier studies often hold market structure constant by examining a change in tick rule on 1See SEC (1994) Market 2000 report. 2For bid prices below $10, the tick size is 1/32nd. 3See Barclay et al. (1997) for an analysis of the impact of OHR on the first 100 stocks to be phased into compliance with the rule. 10/29/2001–coded–JFQA #36:4 Huang and Stoll Page 505 Huang and Stoll 505 the same exchange.4 We examine whether both tick size rules and spreads are en- dogenous in a broader model of exchange structures. In addition, we consider the extent to which other features of markets—the degree of quote and price cluster- ing and the depth of market are associated with market structure. Previous studies, more narrowly focused on changes in existing tick size rules, may be affected by changes in market-wide and firm-specific information before and after the adop- tion of a new tick size. Our study is robust with respect to these changes since we examine a given set of stocks traded at the same time in different markets. The remainder of the paper is organized as follows. Section II develops the analytical framework. Section III describes our data set, which consists of U.K. stocks that are also traded on the NYSE as American Depository Receipts (ADRs). The next four sections present the empirical evidence on spreads (Sec- tion IV), clustering (Section V), clustering and spreads (Section VI), and depth, tick size, and spreads (Section VII). Section VIII contains the conclusion. II. Analytic Framework The premise underlying our analytical framework is that the distinction be- tween auction and dealer markets is important. The key feature differentiating the two market structures is the treatment of public limit orders. In an auction mar- ket, limit orders are displayed and may trade against incoming market orders. In a pure dealer market, limit orders are held by each dealer, are not displayed, and can only be traded against the dealer’s quote. The distinction has implications for the level of spreads, for the existence of a tick rule, for the degree of clustering, for the quoted depth, and perhaps for other characteristics of markets. A. Spreads It is well established that the dollar spread varies cross-sectionally according to stock characteristics such as the stock price, volume of trading, volatility, and amount of informational trading. Under the null hypothesis that market structure has no effect on spreads, the relation between spread and stock characteristics would be the same in dealer and auction markets. But there is evidence for the alternative hypothesis that dealer market spreads are higher than auction market spreads even for the same stocks simply because of the effect of different market structures.5 The principal reason for lower spreads in auction markets is that limit orders narrow spreads in comparison to dealer spreads. In a dealer market, dealers determine the spread. In an auction market, limit orders determine the spread. B. Tick Rules The existence and importance of tick rules is also a function of market struc- ture. A dealer market, like that in London, does not mandate time priority (across 4An exception is the study by Ahn, Cao, and Choe (1998) who examine TSE stocks that are cross-listed on the NYSE, AMEX, and Nasdaq. 5See, for example, Huang and Stoll (1996). 10/29/2001–coded–JFQA #36:4 Huang and Stoll Page 506 506 Journal of Financial and Quantitative Analysis dealers). A tick rule, as Harris (1991) has pointed out, is essential if time pri- ority is to have meaning. Time priority has little meaning if the person who is first to quote the best bid can lose that position to someone who quotes only a penny more. Conversely, if there is not time priority, a tick rule is not necessary. Consequently, it is not surprising that dealer markets, where there is no time pri- ority, have evolved with little attention to a tick rule. Even on Nasdaq, the 1/8 tick was a convention for quoting bids and asks, not a rule, and the convention did not apply to transactions. Transactions could be negotiated at price increments of $1/256 or in decimal format with up to eight digits to the right of the decimal. In an auction market, where limit orders have standing, a tick rule is needed to make time priority meaningful. Without a tick rule, customers could easily step ahead of dealers or conversely dealers and floor brokers could easily step ahead of customers.6 In summary, a tick rule is endogenous. It arises in auction markets to facilitate orderly trading and give time priority meaning. A tick rule, while the outcome of an auction market, can have an independent effect on spreads, as demonstrated by Harris (1994). Tick size increases spreads for stocks with spreads that would otherwise be less than the tick size. For exam- ple, suppose an $8 stock would normally have an 8-cent spread. If the minimum tick is 12.5 cents, the spread can be no less than 12.5 cents. The tick size raises spreads in an auction market, particularly for low price stocks. Yet it remains possible that spreads in a dealer market exceed those in an auction market. C. Clustering Another market feature that may be affected by market structure is the degree to which prices cluster. Clustering is the tendency for prices to fall on a subset of available prices. Clustering is defined with respect to a price grid. For example, if prices are quoted in eighths, clustering exists if all eight price positions are not used equally. Clustering can be measured by the fraction of prices at even eighths as in Christie and Schultz (1994). We define a measure of clustering for stock i, Ci, that is similar to the Herfindahl Index, Ci � � k �Fik � F�ik� 2 � where Fik is the observed frequency of price k and F �ik is the theoretical frequency under the assumption of a uniform distribution. For example, if the available prices are eighths, the theoretical frequency with which a price falls on each eighth is 1/8. If the actual frequency is also 1/8, C � 0.0. If only even eighths are used, C � 1/8. A feature of this measure is that a doubling of available prices accom- panied by a doubling of used prices will result in a smaller clustering measure. For example, if a decline in the tick size from 1/8 to 1/16 results in the use only of even sixteenths, the clustering measure is C � 1/16. The clustering measure is half as large, which is appropriate given that twice as many prices are used, as was the case when only even eighths were used. 6The NYSE does not follow a strict time priority rule. To minimize the breaking up of large orders, the time priority rule applies only to the first limit order. The remaining limit orders follow a size priority rule; namely, limit orders that match the size of the market order at the best price are given priority over other limit orders that might have been placed earlier. 10/29/2001–coded–JFQA #36:4 Huang and Stoll Page 507 Huang and Stoll 507 Price clustering may arise for at least three reasons. First, prices cluster to simplify negotiation. Traders cannot use an infinite set of numbers. If there is no tick size or if the tick size is small, clustering would occur simply as a matter of trading convenience. Under this view, we expect clustering within a market to increase with stock price and with the spread, and we do not expect it to differ between auction and dealer markets. Independent of market structure, the higher the stock price and/or the greater the spread, the less the importance of a small price increment and the larger the price increment traders are likely to choose. For example, in negotiating for a house, the price increment will not be $0.25. Second, clustering is a function of market structure because it is related to the number of traders with standing. In a dealer market, only recognized dealers have standing and display quotes. Dealers are often required to trade in size, and they must cover a variety of costs. These obligations and costs can lead to wider quoted spreads and greater clustering. In an auction market, limit orders have standing. Public investors do not incur some of the costs of a dealer, and they have an incentive to place limit orders that beat dealer quotes. The presence of many limit orders tends to narrow spreads and reduce clustering because more prices are likely to be used. In a dealer market, clustering will be most evident in quotes and will tend to be negotiated away in trades. In an auction market, limit orders allow the public to pre-negotiate prices inside dealer quotes by placing limit orders. Consequently, we expect less quote clustering in an auction market (where every price is more likely to be used) than in a dealer market (where fewer participants reduce the chance that every price is used). Third, Christie and Schultz (1994) and Dutta and Madhavan (1997) argue that clustering of prices on even eighths reflects coordination by dealers in Nasdaq to raise spreads above competitive levels. Christie and Schultz find nearly total avoidance of odd eighths for some but not all Nasdaq stocks, and they conclude that coordination among dealers raises spreads. Both the market structure and implicit collusion imply greater clustering in dealer than in auction markets. We distinguish the two by examining the nature and degree of clustering in quotes in comparison to trade prices. The coordination view implies substantial clustering in both quotes and trade prices. If dealers are to profit from wide spreads asso- ciated with quote clustering, they must trade at the quoted prices; hence, trade prices must also cluster. In summary, we investigate three hypotheses in regard to clustering. The ease of negotiation hypothesis implies increased clustering with increased price, but does not imply that clustering should be different in different market structures. The market structure hypothesis of clustering implies that quote clustering in a dealer market exceeds quote clustering in an auction market for the same reason that spreads in a dealer market exceed spreads in an auction market. However, the minimum tick size is a complicating factor. Whereas clustering could go to zero, the minimum 1/8th tick in an auction market puts a lower bound of 1/8th on the spread in an auction market. To distinguish implicit collusion and market structure, we look at the amount and pattern of quote clustering in comparison to the amount of trade price clustering. Completely effective dealer coordination implies quote clustering and trade price clustering of the same level, for it is the 10/29/2001–coded–JFQA #36:4 Huang and Stoll Page 508 508 Journal of Financial and Quantitative Analysis transactions that yield any profits. If trade prices cluster significantly less than quotes, we would reject the hypothesis of systematic implicit collusion. 7 D. Depth Define depth to be the quantity bid offered at the inside quote. Depth will be related to other market features such as spreads, tick size, and degree of clustering. In particular, we expect depth to be less in an auction market because limit orders narrow the spread. Depth is reduced because the spreads narrowing limit orders are small. They are small because large limit orders do not want to take the risk of being “picked off” by informed traders, and because many limit orders may be placed by individual investors seeking to better the quote. Depth will tend to be larger in pure dealer markets because dealers implicitly operate at larger spreads and larger tick size. III. Data We begin with a sample of 31 FTSE 100 firms that are traded in the U.S. in 1995. Five firms are deleted from the list for trading less than 200 times during January or December of 1995. An additional firm was lost for switching ex- changes and another one for merging during the year. Of the remaining 27, there are 19 listed on the NYSE, four on the AMEX, and one on Nasdaq. Our final sample is the set of 19 British stocks traded as ADRs on the NYSE.8 By examining the same stocks under different market structures, we hold constant stock characteristics and are able to investigate the role of market struc- ture. NYSE quote and transaction data are from TAQ. We restrict the data set to quote and trade prices on the NYSE and exclude quotes and prices from the re- gional exchanges and Nasdaq. The Transaction Data Service of the LSE supplied the U.K. transactions data. The data is error-checked with the typical filters. 9 Days when either the NYSE or LSE is closed are excluded. The sample list and some descriptive statistics are provided in Table 1. It shows the ADRs’ number of market makers on the LSE along with their ADR ratios. The ADR ratio is the number of ADRs that correspond to one U.K. share. For example, an ADR ratio of 1/4 indicates one ADR is collateralized by four U.K. shares. The ADR ratio ranges from 1/2–1/10, reflecting the lower price of shares in the U.K. as compared with the U.S. Under the law of one price, a U.K. share adjusted for the ADR ratio has the same value as the ADR. For example, British Airways U.K. price of £4.27 would be £42.70 after adjusting for the 1/10 ADR ratio. This pound price is equivalent to the dollar price of $67.35 7Of course, we cannot rule out that there are some traders or times in which dealers are able to extract monopoly rents. 8It would also be of interest to examine U.S. firms that are traded on the LSE. We exclude this sam- ple for structural and data reasons. The system for U.S. stocks on the LSE is the SEAQ-International (SEAQ-I). However, unlike SEAQ, SEAQ-I is more a brokerage system than a dealership market and quotes posted on the SEAQ-I are no