The Components of the Bid-Ask Spread in a Limit-Order Market: Evidence from the Tokyo Stock Exchange

The Components of the Bid-Ask Spread in a Limit-Order Market:
Evidence from the Tokyo Stock Exchange




Hee-Joon Ahn
College of Economics and Commerce
Sookmyung Women’s University
Seoul, Korea

Jun Cai
Department of Economics and Finance
City University of Hong Kong
Tat Chee Avenue
Kowloon, Hong Kong

Yasushi Hamao*
Department of Finance and Business Economics
Marshall School of Business
University of Southern California
Los Angeles, CA 90089-1427

Richard Y. K. Ho
Department of Economics and Finance
City University of Hong Kong
Tat Chee Avenue
Kowloon, Hong Kong


Forthcoming in Journal of Empirical Finance










We would like to thank Kee-Hong Bae, Ananth Madhavan, and the seminar participants at Korea
University for their helpful comments. We also thank the City University of Hong Kong for its generous
support. Financial supports from RGC Competitive Earmarked Research Grants 1998-2000 and 1999-
2001 (Cai) and the Faculty Research Fund at Marshall School of Business, University of Southern
California (Hamao) are gratefully acknowledged. Caroline Biebuyck provided the editorial assistance.
*Contact author, hamao@usc.edu, 213-740-0822

---

The Components of the Bid-Ask Spread in a Limit-Order Market:
Evidence from the Tokyo Stock Exchange




Abstract

This paper analyzes the components of the bid-ask spread in the limit-order book of the Tokyo Stock
Exchange (TSE). While the behavior of spread components in U.S. markets has been extensively studied,
little is known about the spread components in a pure limit-order market. We find that both the adverse
selection and order handling cost components of the TSE exhibit U-shape patterns independently, in
contrast to the findings of Madhavan, Richardson, and Roomans (1997) for U.S. stocks. On the TSE,
there does not exist an upstairs market that allows large trades to be prenegotiated or certified as on the
New York Stock Exchange (NYSE). This feature of the TSE provides a valuable opportunity to examine
the relationship between trade size and spread components. Our results show that the adverse selection
cost increases with trade size while order handling cost decreases with it.

Key words: Bid-ask spread, adverse selection cost, order-processing cost, trade size.

JEL classification: C14; G15

2

---

1. Introduction
Over the past years, limit-order trading has received growing attention as more exchanges implement
electronic public limit-order books and open up the market-making process. A number of studies have
examined various aspects of the limit-order market. In particular, Glosten (1994), Handa and Schwartz
(1996), Rock (1996), Seppi (1997), Viswanathan and Wang (1998), and Foucault (1999) offer a variety of
equilibrium models on limit-order trading.1 Biais, Hillion, and Spatt (1995) offer an empirical analysis of
the supply and demand of liquidity and interaction between the order book and order flow in the Paris
Bourse. Harris and Hasbrouck (1996) investigate the relative importance of market and limit orders. Ahn,
Bae, and Chan (2001) analyze the interaction between transitory volatility and order flow composition in
a limit-order market. Chung, Van Ness, and Van Ness (1999) and Kavajecz (1999) examine whether
quoted spreads reflect the trading interest of specialists or limit-order traders.2
The purpose of this paper is to examine the components of the bid-ask spread in a limit-order market.
Existing market microstructure theories on the components of the bid-ask spread are largely developed
within the framework of quote-driven single (multiple) dealer markets. In addition to the order-
processing costs, the bid-ask spread must cover the following two components: the inventory and
information costs in a dealer market.3 However, the bid-ask spread is not unique to the dealer markets.
Cohen, Maier, Schwartz, and Whitcomb (1981) establish the existence of the bid-ask spread in a limit-
order market when investors face transaction costs of assessing information, monitoring market, and
conveying orders to the market. Glosten (1994) shows that the limit-order market will have a positive

1 See Seppi (1997) for a more detailed summary of other equilibrium models of limit-order market.

2 Other empirical studies of limit-order market include Frino and McCorry (1995) and Hollifield, Miller, and Sandås
(1999).

3 Demsetz (1968) and Tinic (1972) identify the order-processing costs incurred by the providers of market liquidity.
Stoll (1978), Ho and Stoll (1983), and Amihud and Mendelson (1980), emphasize the inventory holding costs.
Copeland and Galai (1983), Easley and O’Hara (1987), and Glosten and Milgrom (1985) concentrate on the
information costs faced by liquidity suppliers when trading with informed traders.
1

---

bid-ask spread arising from the possibility of trading on private information. Nevertheless, empirical
evidence on the bid-ask component in a limit-order market has been extremely limited.4
We examine the bid-ask component in a limit-order book of the Tokyo Stock Exchange (TSE). On
the TSE, there are no designated market makers with an obligation to take positions in the market. Every
transaction is executed by a saitori who maintains each offer to buy or sell in an order book, which is
open to all exchange members on the floor. All liquidity is supplied by traders who submit limit or
market orders. In this sense, the TSE may be better described as a market where a multiple number of
dealers provide market-making at their own discretion (Takagi, 1993).
A number of earlier papers explored various aspects of the TSE. Amihud and Mendelson (1989,
1991, and 1993) examine liquidity provision and price discovery on the TSE. Hamao (1992) and Takagi
(1993) present an overview of the institutional features of the TSE trading. Lindsey and Schaede (1992)
compare the role of a saitori with that of the specialist of the NYSE. George and Hwang (1995) and Kim
and Rhee (1997) investigate the effectiveness of the TSE price limit rules. Two recent studies examine
the intertemporal behavior of the market microstructure on the TSE. Lehmann and Modest (1994) study
the size of the bid-ask spread and its cross-sectional and intraday stability, among other issues. They
report U-shape intraday patterns in bid-ask spreads, return volatility, and trading volume. Hamao and
Hasbrouck (1995) investigate the properties of intraday trades and quotes on the TSE for a number of
representative firms. They find evidence consistent with asymmetric information effects within the limit-
order book.
Our primary findings are as follows. First, we found that both the adverse selection and order-
processing cost components exhibit U-shape patterns. This contrasts with the finding by Madhavan,
Richardson, and Roomans (1997) that on the NYSE the adverse selection component declines and order-
process component increases during the day. The TSE evidence of an increase in adverse selection costs
around the end of the trading day suggests that transactions around this period convey private information

4 Brockman and Chung (1999) and Chan (2000) study the bid-ask components on the Stock Exchange of Hong
Kong. De Jong, Nijman, and Roell (1996) examine the bid-ask component on the Paris Bourse.
2

---

that would otherwise be released during the non-trading hours that follow the exchange close. Second,
we also find that adverse selection costs increase with trade size while order-processing costs decrease
with it. This result is compared with the evidence on the NYSE that medium trades contain more
information than large trades, as reported by Barclay and Warner (1994) and Huang and Stoll (1997). We
believe that this difference in the relation between trade size and information content of a trade between
the two exchanges comes from how the two exchanges treat large order flows. On the NYSE, large block
orders are sent to the upstairs, where transactions are made through a search-brokerage mechanism. On
the TSE, there is no separate venue for block trading. All trades, large or small, are consolidated into the
central electronic order book, where investor identity is not revealed as in upstairs trading on the NYSE.
Our evidence that large trades convey more information provides strong support for the theoretical
prediction given by Easley and O’Hara (1987) that informed investors prefer to trade large volume. The
majority of the world’s stock markets take the form of the limit order system. Most of the extant studies,
however, focus on quote driven markets or a hybrid system that adopts both limit order book and dealer
system. Our study implies that the process of how information is incorporated into stock price through
trading on a limit order market is different from the existing evidence reported based on the quote driven
or hybrid system.
The remainder of the paper is organized as follows. Section 2 describes the tick-by-tick transaction
data, the process of filtering the data, and summary statistics. Section 3 presents a simple characterization
of the bid-ask component model and estimation procedure. Section 4 presents the empirical results for
the cross-sectional difference and the intraday pattern of the bid-ask components. The relation between
trade size and bid-ask components is also examined. Section 5 concludes the paper.

2. Data sources, filtering, and summary statistics
2.1 Data sources
We obtain the real-time TSE trades and quotes data from Nikkei Economic Electronic Database
System (NEEDS) historical tick data. The database is time-stamped to the nearest minute and includes
3

---

the information on all quotes and trades in both price and quantity.5 It also has detailed flags indicating
the conditions of each trade and quote. These flags include the opening/closing trade indicators, buy/sell
indictors, and special and warning quote indictors, among others.6 The NEEDS database essentially
reflects all the trade and quote information broadcast to TSE members by the TSE. The database is the
most detailed and extensive among the known data sets on the Japanese stock market.
Our initial data set includes all transactions and quotes recorded for the 225 stocks that constitute the
Nikkei Price Average Index. The Nikkei 225 component stocks, selected from the first section of the
TSE, are highly liquid and widely known as a good representation of the Japanese stock market. The
sample period is from January 5, 2000 to March 31, 2000, for a total of 60 trading days. The first trading
day of the year, January 4, 2000, is excluded since trading was conducted only during the morning
session of the day.

2.2 Data filtering process
2.2.1 Tick size
We use only the transactions and quotes on the TSE in our analysis.7 On the TSE, the tick size is a
step function of stock prices. For example, during the sample period used in this study, stocks priced less
than or equal to ¥2,000 have a tick size of ¥ 1; stocks priced between ¥2,001 and ¥3,000 have a tick size
of ¥5; and stocks priced between ¥3,001 and ¥30,000 have a tick size of ¥10. While some stocks are
priced above ¥30,000 and traded on greater tick sizes, most stocks trade at prices below ¥30,000. In our
initial sample of Nikkei 225 constituent stocks, six are priced above ¥30,000 and therefore excluded from
the sample. Exclusion of these six firms is to avoid possible confounding effects on the spread due to
significantly larger tick sizes. We then partition the remaining 219 firms into three groups based on
individual tick sizes (i.e. ¥1, ¥5, and ¥10).

5 Even if time-stamped to the minute, the data are recorded to preserve the sequence of events within each minute.
6 On August 24, 1998, the TSE officially abolished warning quotes. However, the TSE order matching system is not
updated until the end of year 2000 and the system occasionally shows warning quotes for some stocks.
4

---

2.2.2 Classification of buy and sell transactions
To ensure a clean data set, we apply several filters to individual transactions and quotes for each stock
in the sample. Each transaction during the morning and afternoon sessions is classified as a buyer or a
seller-initiated trade by the price location flags available from the dataset. These flags indicate the
location of each trade at the bid, ask, between the bid and ask, or outside the bid-ask range. Almost all of
the trades for our sample firms are made either at the bid or ask. However, there is a small number of
trades that are made either within or outside the spread. For an average firm in our sample, of the total
19,682 transactions made during the regular sessions for the entire three-month period, 0.12 percent (24
trades) are made within the spread while less than 0.01 percent (2 trades) are made outside the spread.
Since the TSE is a pure order-driven market without responsible market makers, under normal conditions
of a double auction, a trade should always hit either the bid or ask. Therefore, we exclude all trades made
within or outside the spread.

2.2.3 Opening and closing transactions in the morning and afternoon sessions
The trading day on the TSE is divided into morning (9:00-11:00 a.m. local time) and afternoon
(12:30-3:00 p.m.) sessions. All observations recorded before 9:00 a.m. or after 3:00 p.m. are excluded
from the sample. Therefore, we do not include transactions made during the off-hours trading session,
which often deals with negotiated large block trades through a system called ToSTNet (Tokyo Stock
Exchange Trading Network System).8 To examine the intraday patterns, we divide each trading day into
nine 30-minute intervals, four in the morning session and five in the afternoon session.

7 There are five regional exchanges in Japan: Osaka, Nagoya, Hiroshima, Fukuoka, and Sapporo Stock Exchanges.
Some of the Nikkei constituent stocks also trade in regional exchanges. However, regional exchanges are
dominated by the TSE in terms trading volume. Trade-throughs are not allowed.
8 According to Tokyo Stock Exchange Fact Book (2000), off-hour trading was introduced for block orders on
November 14, 1997. ToSTNet-1 was established on June 29, 1998 to deal with off-hours trading for single issue
orders and basket cross orders. TosTNet-2 was established on August 7, 1998 to allow orders be executed at the
TSE’s closing prices.
5

---

On the TSE, the opening and closing transactions in each session are made under a batch clearing
process called itayose while the rest of the trades during the day are made under a continuous double
auction called zaraba. Since the model for our empirical test applies to normal trading conditions under a
continuous double auction, we only need transactions made under zaraba. Thus, we exclude all itayose
trades. With the opening and closing trade and quote indicators, the data set enables us to accurately
identify all trades made at itayose.


2.2.4 Price limit on the TSE
One important feature of the TSE is the price limit rule. The TSE uses price limits to trigger
indicative quote dissemination, to halt trade temporarily, and to allow continued price discovery through
quote adjustment in some circumstances.9 For a stock trading between ¥1,001 and ¥1,500, the tick size is
¥1, the maximum price variation between trades without trade being halted is ¥20, and the daily price
limit is ¥200. The daily price limits are quite large, ranging from 10 to 30 percent for most stocks priced
below ¥30,000. Consequently, these limits are rarely hit. For 156 of our 219 sample stocks, the price
never hits the daily price limit during the entire three-month sample period. In contrast, the maximum
price variation allowed between trades is on the order of 0.5 to 1.5 percent of the stock price, and thus the
limit is hit more frequently. For example, for 96 out of the 219 sample firms, the stock price reaches the
intraday price limit during the sample period. However, the average number of hits per day is low at 0.62
times (0.19 percent of the total number of transactions on an average day).
It is possible that these price-limit rules affect our estimation process based on order arrival sequences.
For example, the price limits can hinder the natural price discovery process by suppressing part of the
orders that are informative and, thus, would reduce the amount of information asymmetry that would
otherwise be present in the market. This is especially true for the case of daily price limits since trading is
allowed only within the daily limit during the entire day. Therefore, our estimates of spread components

9 It is a debatable issue whether these rules help price continuity. See Kim and Rhee (1997) for an empirical study
of the effects of daily price limit on TSE.
6

---

could be smaller than the amount of the information asymmetry that would appear in the absence of the
price limits.10

2.2.5 Special quotes
The TSE requires the saitori to post a special quote when a major order imbalance between buy and
sell orders arises. Once posted, the special quote remains until a counter order arrives and a new
equilibrium is established, at which point the quote is withdrawn. In the absence of the arrival of any
counter order, the quote can be renewed at an interval of every five minutes within the variation limit set
forth by the exchange. The display mechanism of a special quote has several distinct features. A special
quote is always accompanied by a null quote (zero price and zero quantity) on the other side. The
quantity displayed with a special quote is the net buying and selling order quantities rather than the
quantity available at the quote. In addition, the price location flag for the trade following a special quote
does not indicate the usual ‘transaction at ask’ or ‘transaction at bid’ but the arrival and execution of a
buy order or a sell order. Since a trade or trades matched with special quotes cannot be seen as part of a
successive normal order flow, we exclude any trades following special quotes during the regular trading
hours.11

2.2.6 Partial execution of large market orders

Sometimes an incoming market order is too large to be absorbed by the current bid or ask quotes. In
this case, the saitori partially fills the order up to the size of the current quote. Then the remaining
portion is converted into a limit order at the current quote.12 For example, suppose the current ask is ¥500
for 1,000 shares. If a market buy order of 1,200 shares arrives, it can only be partially executed against
the current ask up to 1,000 shares. The remaining 200 shares will be converted into a limit buy order (an

10 As an alternative approach we apply a filter that excludes the entire 30-minute interval if a trade hits either the
daily or intraday price limits. However, our results are not sensitive to this filtering method.
11 Issuances of special quotes could be related with severe information asymmetry in the market. In a separate
analysis, we include the transactions executed against special quotes. However, the estimation results are virtually
identical to those estimated without special quotes.
7

---

ask quote of 200 shares at ¥500). Two outcomes can follow: (1) an incoming market sell order or limit
sell order priced at ¥500 or below hits this ask quote or, (2) when there is no immediate arrival of either
type of orders, the 200 share buy order walks up the book and hits the limit sell order at the next best
price.13
Particular attention on buy-sell classification is required when a trade is executed under the first
situation (i.e. an incoming order hits the limit order originated from partial execution of a market order).
The trade, in the above example, is classified as a sell transaction in the original dataset. However, since
this 200 share trade originates from the market buy order of 1,200 shares, it is more reasonable to treat it
as a buy transaction instead of a sell transaction. A large market order (and often a series of large market
orders on the same side) indicates the arrival of new information about the security. A series of
transactions initiated on the same side will be more consistent with the information arrival. Therefore, we
reclassified all trades that result from partial execution of large market orders. We also repeat all
following analyses using the original buy-sell classification provided by the data set. Our reclassification
rule, however, generally results in better, if marginal, convergence in the estimation of spread
components. We report the results based on the reclassification. 14


2.2.7 Final sample
It also happens that the tick size of some stocks changes in our sample. In this case, we exclude the
half-hour section if the tick size changes (for example, from ¥1 to ¥5 or vice versa) during the interval.
To ensure enough observations within each half-hour interval, we impose the condition that there are at
least five valid transactions during the interval. Then, we drop the entire trading day if it has fewer than

12 The saitori used to display a warning quote briefly to invite liquidity in this situation before the TSE’s rule change
on warning quotes in 1998. Since the rule change, the quote is treated as part of general quotes.
13 The saitori uses discretion on the exposure duration of the limit order converted from a partially executed market
order.
14 The way to obtain immediacy of a large order in the TSE’s institutional setting might be not to submit a market
order but to submit an opposing limit order with a price that guarantees an immediate and full trade execution.
Indeed, a limit sell (buy) order submitted in this fashion should be regarded as a market buy (sell) order in disguise
of a limit order. Hence, it would be more relevant for our purpose to treat these limit orders as opposing market
8

---