Page 1

ssrn.com/abstract=1710924

Short-Selling, Uptick Rule, and Market Quality: Evidence from

High-Frequency Data on Hong Kong Stock Exchange∗

Pengjie GaoJia HaoIvalina Kalcheva Tongshu Ma†

This draft: March 1, 2011

First draft: November 1, 2006

Comments welcome

Abstract

Much empirical research has been conducted concerning the effect of short-selling on market

quality and volatility. However, the evidence is inconclusive and still a matter of debate. Using

intraday data in a pure order-driven market we show that allowing for short-selling decreases

the adverse selection costs for less-visible firms, firms with less analyst coverage, larger adverse-

selection cost component of the bid-ask spread, low price per share, and high relative tick size

(given the same market capitalization). Allowing for short-selling also decreases (increases)

intraday volatility for less- (more-) visible stocks. In addition we document that with the uptick

rule in place (not in place) there is not statistically significant difference in liquidity (intraday

volatility) between stocks that are allowed for short-selling and those that are not.

Keywords: short-selling, market quality, liquidity, volatility, uptick rule

JEL classification codes: G1, G2

∗We are grateful to staff members of Research Department, and in particular Alexandra Young, at the Hong

Kong Security and Futures Commission (SFC) and Karen Lam of the Hong Kong Stock Exchange and Clearing

House Incorporated (HKSE) for numerous discussions about short-sale regulations in Hong Kong. We thank

George Gao, Karen Lam, David Xue, Alexandra Young, and Xueping Wu for helping us with some of the data used

in this study. Seminar participants from Northwestern University provided useful feedback. We benefited from

discussions with Torben Andersen, Hank Bessembinder, Kent Daniel, Ravi Jagannathan, Mitchell Peterson, Todd

Pulvino, Ernst Schaumburg, Annette Vissing-Jorgensen, Wallace Mok, Xueping Wu and Sudip Datta. Morgan

Stanley Market Microstructure Fund provides partial funding for this research project. Jon Ross from the David

Eccles Business School and Patricia Li´ ebana from the Kellogg School of Management provided indispensable

computational assistance in this project. The views herein are solely those of the authors, and not those of

SFC, HKSE, Morgan Stanley and Company, or any other person or entity. We are responsible for any remaining

omissions and errors. Earlier versions of this manuscript were titled “Does Removing the Short-sale Constraint

Improve Quality? Evidence from Hong Kong”

†The authors are from University of Notre Dame, Wayne State University, University of Arizona and

Binghamton University, respectively.The emails for the authors are pgao@nd.edu, jia.hao@wayne.edu,

kalcheva@arizona.edu, and tma@binghamton.edu, respectively. Corresponding author: Ivalina Kalcheva, Address:

Eller College of Management, McClelland Hall 315L, 1130 E. Helen St. P.O. Box 210108, Tucson, Arizona 85721-

0108, Tel: + 1-520-621-0747, Fax: + 1-520-621-1261, E-mail: kalcheva@arizona.edu.

Page 2

ssrn.com/abstract=1710924

Short-Selling, Uptick Rule, and Market Quality: Evidence from

High-Frequency Data on Hong Kong Stock Exchange

Abstract

Much empirical research has been conducted concerning the effect of short-selling on market

quality and volatility. However, the evidence is inconclusive and still a matter of debate. Using

intraday data in a pure order-driven market we show that allowing for short-selling decreases

the adverse selection costs for less-visible firms, firms with less analyst coverage, larger adverse-

selection cost component of the bid-ask spread, low price per share, and high relative tick size

(given the same market capitalization). Allowing for short-selling also decreases (increases)

intraday volatility for less- (more-) visible stocks. In addition we document that with the uptick

rule in place (not in place) there is not statistically significant difference in liquidity (intraday

volatility) between stocks that are allowed for short-selling and those that are not.

Page 3

I. Introduction

How do short-sale constraints matter for the functioning of the financial markets? What is the

role of the “uptick rule,” which generally prevents securities from being sold short at a declining

price? This has been a topic of debate among market practitioners, regulators, and financial

economists at least since the 1930s (Jones, 2008). Recently, the interest in short-selling and price

tests has increased dramatically in light of the 2007–2009 financial crisis, which caused regulators

across the globe to implement some sort of short-sale constraint (Beber and Pagano, 2010). Our

paper revisits this debate by using high-frequency data in a pure order-diven market and taking

advantage of the fact that on the Hong Kong Stock Exchange (HKSE hereinafter) short-selling

is allowed1by regulators for stocks that meet certain requirements, and more importantly, for

the stocks that are allowed to be sold short there was a period when the uptick rule was in

place and when it was not, creating a suitable natural experimental environment. The HKSE is

one of the largest and most liquid equity markets in the world where short-selling is practiced

(Bris, Goetzmann, and Zhu, 2007). The HKSE achieves a high level of market transparency

and operates as a pure order-driven market similar to the prototype public limit order book

envisioned in Glosten (1994). More details on the trading mechanism can be found in Appendix

On the HKSE before January 1994, short-selling was prohibited. After January 1994, only

some stocks were allowed for short-selling, while others were not. The list of designated securities

for short-selling was revised on a quarterly basis after 1996 — stocks which met the criteria of

“eligible stocks” were added into the short-sale list, while those no longer eligible were removed

from the list. In this case, we have a complete history of whether each individual stock is

allowed for short-selling on each day. The fact that stocks become eligible for short-selling on

different dates is an advantage as it implies a reduced likelihood that comparisons will be affected

by contemporaneous changes in market-wide factors affecting the inferences of the relationship

between the variables we study (Bessembinder, 2000). The details of the regulation on the

1According to the regulations of the SFC of Hong Kong, breach of short-sale constraint is a criminal offense

punishable by fines and imprisonment according to the security laws in Hong Kong. The level of fines and

imprisonment were revised (from a maximum of $10,000 and 6 months) to HK$100,000 and 2 years imprisonment

with the introduction of the Securities (Amendment) Ordinance 2000 on July 3, 2000.

Page 4

selection of stocks eligible for short-selling is provided in Appendix A. Generally, stocks eligible

for short-selling on the HKSE are to a great extent large and liquid stocks.

In the pilot program, the HKSE also instituted the uptick rule, which mandated that a

short-sale could not be made below the best current ask price. The uptick rule was abolished in

March 1996 and reinstated on September 7, 1998, due to changes in market conditions during

the East Asian financial crisis. Hence, from March 1996 to April 1998, short-sales were operated

without the uptick rule. At other times in our sample, short-sales were operated with the uptick

rule.

Our study employs high-frequency transaction data from a pure order-driven market to

examine the relationship between short-selling, the uptick rule, and market quality by looking

at an array of measures at the stock level: spreads, order imbalance, relative depth of the

limit order book, and intraday return volatility. Using intraday data allows us to decompose

the total spread into its order-processing and adverse-selection components which provides an

opportunity for a direct test of the Diamond and Verrecchia (1987) model as explained in the

next section. We adopt the quasi-experiment framework (using the terminology of Meyer, 1995)

and apply difference-in-difference (DID hereinafter) estimation as our main empirical tool. As

the regulation changes are not random and the choice of stocks to be allowed for short-selling is

not entirely exogenous, using control sample and control covariates in the empirical methodology

is critical. Such a methodology allows us to make more reliable inferences.

Using the DID approach, our empirical results suggest that short-sale constraints affect

market quality. The results suggest that the impact of short-sale constraints on market quality

measures is not homogeneous across stocks and depends on firm’s visibility. Less-visible firms are

defined as those with less analyst coverage, larger adverse-selection cost component of the bid-ask

spread, low price per share, and high-relative tick size (given the same market capitalization).

When the uptick rule is not in place, allowing for short-selling generally reduces transaction

costs. More importantly, most of the reduction is associated with the reduction of the adverse

selection component of the bid-ask spread and concentrates in less-visible stocks. In fact, with

the uptick rule in place, the difference in transaction costs between stocks that are allowed to

be sold short and those that are not, is statistically and economically insignificant. With the

Page 5

uptick rule in place, however, allowing for short-selling decreases intraday volatility for less-

visible stocks but increases the intraday volatility for more-visible firms. With the uptick rule

not in place, we do not find statistically significant difference in intraday volatility between

stocks that are allowed for short-selling and those that are not.

We take advantage of two distinctive institutional differences between HKSE and the U.S.

exchanges to gauge the visibility level of a firm. First, unlike U.S. stock exchanges, which have a

uniform tick size across most of the stocks, the tick size schedule on the HKSE is a step function

of the underlying security’s price. For example, for stocks traded between HK$0.01 and HK$0.25,

the tick size is HK$0.001; while for the stocks traded above HK$200.00 and below HK$1000.00,

the tick size is HK$1. The complete schedule of tick sizes is illustrated in Appendix C. The

schedule is valid from October 1996 to October 2005.2Second, the majority of the securities

on the HKSE are low-priced stocks, but it should be noted that they are not “penny stocks”

given the U.S. standards. For example, Angel (1997) notes that the median U.S. stock sells for

about $40, while a typical Hong Kong share is about $2. More importantly, on the HKSE the

low-priced stocks are spread rather evenly across large and small firms. According to the model

presented in Angel (1997), a highly visible firm will have a lower optimal relative tick size than

a less visible firm. Given the step tick function on the HKSE, lower-priced securities according

to Angel (1997) will be less-visible firms and higher-priced securities will be more visible.

Therefore, in our analysis we look at three different price groups — “low,” “medium,” and

“high” — as a proxy for a firm’s visibility in order to investigate whether the relationship

between short-selling, the uptick rule, and market quality is not homogeneous across stocks.

The reader should keep in mind that the “Low-Price” Group does not include “penny stocks”

given the U.S. standards. By sorting on price per share, our Low-Price Group includes firms

that are better characterized as firms with less analyst coverage (public information is more

2Effective from November 1, 2005, the HKSE implemented Phase 1 of the proposal to reduce minimum trading

spreads for shares with a price above HK$30, so the minimum tick size schedule is no longer valid. Specifically,

for stocks priced above $30 and below $50, the minimum tick size is $0.05; above $50 and below $100.00, the

minimum tick size is $0.100; above $100 and below $200, the minimum tick size is $0.100; above $200 and below

$500, the minimum tick size is $0.200; above $500 and below $1,000, the minimum tick size is $0.500; above

$1,000 and below $2,000, the minimum tick size is $1.000; above $2,000 and below $5,000, the minimum tick

size is $2.000; and above $5,000 and below $9,995, the minimum tick size is $5.000 (all units are in HK dollars).

www.hkex.com.hk/rule/exrule/sch-

2 eng.doc. During our sample period, there is no reduction of minimum tick size for the sample of traded stocks

we consider.

Page 6

limited), larger adverse-selection cost component of the bid-ask spread, low price per share, and

high-relative tick size (given the same market capitalization). Our evidence suggests that the

direction and the strength of the effect of short-sale constraints on market quality measures is

dependent upon a firm’s visibility.

We provide evidence on the effect of short-selling constraints in an order-driven market.

This is important because today more than half of the world’s stock exchanges are order-driven

(Ro¸ su (2010)), with no designated market makers (e.g. Euronext, Helsinki, Hong Kong, Tokyo,

Toronto), while in many hybrid markets designated market makers have to compete with a

limit order book (NYSE, Nasdaq, London). One could imagine that the effects of short-selling

constraints will be different across different market structures — e.g. order-driven vs. quote-

driven. One reason, for example, could be that in a quote-driven market a market maker could

still be allowed to short-sell under a short-sale ban environment (as in the recent U.S. shorting

ban) while this is not the case in a pure order-driven market. Despite the growth of order-driven

markets around the world, we are not aware of a theoretical study that compares and contrasts

the effects of short-selling constraints on stock-level liquidity, volatility, and price informational

efficiency across the two market structures, order-driven and quote-driven. This is even more

important in light of the recent spree of short-selling bans around the world in response to the

2007-2009 global financial crisis (Beber and Pagano, 2010).

The rest of the paper is organized as follows.The next section reviews the literature

and develops our hypotheses. Section III explains our empirical methodology in more detail,

including the data used and the sample selection process, and it introduces the proxies for market

quality. Section IV presents the results, and Section V concludes.

II. Literature Review and Hypothesis Development

The short-selling restrictions that regulators have imposed, in and outside U.S. for the most

part comprise but a single event, and even more importantly, most shorting bans are triggered

as a response to deteriorating market conditions. For example, Jones (2008) studies the short-

sale regulation changes in the 1930s that were a response by regulators to worsened market

conditions. Boehmer, Jones, and Zhang (2009) study the SEC emergency order as of September

Page 7

2008 to temporarily ban most short-sales in approximately 1,000 financial stocks for the short

period of time from 9/19/2008 to 10/8/2008, and Boehmer, Jones, and Zhang (2008) study the

repeal of the uptick rule by the SEC on July 6, 2007. Diether, Lee, and Werner (2009) and

Alexander and Peterson (2008) invoke regulation SHO3to also study short-sale price tests. In

contrast to these studies, the current study does not fall into the group of “before-vs.-after”

studies (to use the terminology in Bessembinder, 2000). One important feature of our sample

is that, at each point in time, there are stocks that are allowed for short-selling as well as those

that are not, and the list of stocks eligible for short-selling is updated quarterly. Hence, we can

also control for market-wide changes that are caused by events unrelated to short-sale regulation

changes. Further, for stocks included in the short-selling list there is a period of time when the

uptick rule is in place and a period when it is not in place. We perform firm-level cross-sectional

analysis by comparing stocks affected by the constraints4and stocks that are not at the same

time given different levels of firm visibility using high-frequency data in a pure order-driven

market.

Boehmer, Jones, and Zhang (2009) show that the financial stocks subject to the short-sale

ban suffered a severe degradation in market quality, as measured by spreads, price impacts, and

intraday volatility. Outside the U.S. the cross-country study by Charoenrook and Daouk (2008)

focuses on the market-wide aggregate impact of relaxing the short-sale constraints rather than

the market quality at the individual stock level.5They find that when short-selling is possible,

there is greater aggregate market liquidity. Beber and Pagano (2010) study short-selling bans

around the world in response to the 2007–2009 crisis, using daily data. Chang, Cheng, and

Yu (2007) use the HKSE’s short-sale regulation changes, as in the current study. Our study,

however, differs from theirs in terms of the research question, the empirical design, and the

3It is worth noting that Regulation SHO also contains additional regulations on requiring locating stocks and

more frequent disclosure of short interests. Therefore, strictly speaking, the empirical tests of Regulation SHO

are a joint-test of all of these regulation changes.

4We note that, on September 8, 1995, the SEHK launched its Traded Stock Options Market. In our sample,

we cannot identify which firms have traded options. However, the fact that the sample contains firms with traded

options strengthens our results. Figlewski and Webb (1993) and Danielsen and Sorescu (2001) have argued that

the introduction of traded options represents an economically important relaxation of short-sale constraint. Even

if the whole sample consists of firms with traded options, given that to a large extent our sample comprises of

large and liquid stocks, this will strengthen our result.

5Bris, Goetzmann, and Zhu (2007) is also a cross-country study that focuses on market-wide aggregate impact

of relaxation of short-sale constraints, but they study primarily the effects on price informational efficiency.

Page 8

frequency of data. They focus on short-sale constraints’ effect on stock valuation and volatility

using daily prices and returns data. In contrast, we use high-frequency transaction data. The

high-frequency data allow us to measure intraday spreads, order imbalance, relative depth of the

limit order book, and intraday return volatility. Moreover, we can decompose the intraday total

spread into its order-processing and adverse-selection components to directly test Diamond and

Verrecchia (1987) theoretical model as explained later in this section. Chen and Rhee (2010)

also use the HKSE data, but their focus is on price adjustment to new information.

Boehmer, Jones, and Zhang (2008) show that the repeal of the uptick rule causes market

liquidity to worsen slightly, although they find no evidence that repeal of the uptick rule

contributed to the bout of volatility experienced by U.S. stocks in late July and early August

2007. Alexander and Peterson (1999) examine the impact of the uptick rule on short-sell orders

sent to the NYSE. They find that the execution quality of short-sell orders is adversely affected

by the uptick rule, even when stocks are trading in advancing markets. Alexander and Peterson

(2002) consider how smaller tick size is related to short-sale order executions by looking at the

1997 reduction in tick size from 1/8th to 1/16th. They argue that the move to decimalization

in conjunction with the uptick rule lessens the impact of the uptick rule for short market-

orders. However, smaller tick size hurts the short at-the-quote limit orders when the spread is

larger than the minimum tick size. They also suggest that short-sellers may place market-orders

more frequently than at-the-quote limit orders, but also cancel their orders more frequently and

quickly. Jones (2008) studied the regulation changes in the 1930s U.S. equity market using daily

data. He found that requiring brokers to secure written authorization before lending shares to

a customer decreased liquidity, consistent with Diamond and Verrecchia (1987). However, the

uptick rule and “strict uptick rule” seemed to increase rather than decrease market liquidity

because these rules essentially force short-sellers to use limit orders rather than market orders,

thus providing liquidity. Diether, Lee, and Werner (2009) use the recent Regulation SHO to

evaluate how market liquidity is affected by the suspension of the uptick rule and the conversion

to price-test on the NYSE, as well as the suspension of the bid-price test and the conversion to

price-test on NASDAQ. They found that Regulation SHO reduces the depth at the offer side

relative to the bid side, reduces buy-order imbalance, and increases quoted spreads for the NYSE

Page 9

sample but not much for the NASDAQ sample. Alexander and Peterson (2008) also examine

the recent Regulation SHO on market liquidity and their results are consistent with those in

Diether, Lee, and Werner (2009). Diether, Lee, and Werner (2009) and Alexander and Peterson

(2008) find that the pilot program suspending the uptick rule widens spreads slightly.

On the theoretical front, to our knowledge, Diamond and Verrecchia (1987) is the only study

that explicitly models the effect of short-sale constraints on stock-level liquidity. The basic

structure of the Diamond and Verrecchia (1987) model is based on Glosten and Milgrom (1985):

market makers are risk-neutral, earn zero expected profits from each trade, and face no inventory

costs or constraints. The latter assumption means that they are willing to keep selling or buying

the asset at fair prices. Short-prohibition in their model excludes both informed traders and

uninformed traders from short-selling, and hence it leaves the probability that a sale comes from

an informed trader the same whether short-selling is prohibited or not. Thus, short-prohibition

does not make sell orders more informative. Short-prohibition does not change anything on the

buy side in their model as well.

Generally, Diamond and Verrecchia’s (1987) common-priors rational-expectations model

predicts that short-sale constraints impede the flow of private information. Short-sale prohibition

increases the average bid-ask spread, compared with the case in which short-selling is costless,

because the speed of adjustment of prices to values that imply a small bid-ask spread is reduced.

In other words, the short-sale prohibition reduces the speed at which the price converges to the

true liquidation value, and therefore it reduces the speed at which the bid-ask spread narrows

over time.

The size of the spread and the price of the stock are determined by supply and demand. The

increase in the bid-ask spread, when short-selling is prohibited, is due to a lower supply of stocks

for sale because of some investors wanting to sell but not owning the stock. This fact manifests

itself as a no-trade period in the model, when no transaction takes place during a given trading

interval. Short-selling constraints lead to an increased incidence of no-trade outcomes, which is

one of the main points of Diamond and Verrecchia (1987). The no-trade outcome is important

in that the frequency of no-trades is informative to the market maker in the sense that probably

informed traders have not entered the market yet. A no-trade period is a sign of the release of

Page 10

private information.

The Diamond and Verrecchia (1987) model is a rational-expectations model that predicts

the effects of short-selling restriction on liquidity in a quote-driven market. Interestingly, most

empirical studies that focus on price effects of short-selling constraints provide evidence mostly

consistent with differences-in-beliefs6models, as in Miller (1977).We are not aware of a

theoretical study that predicts the effects of liquidity in a differences-in-beliefs setting in an

order-driven market per se.

Given our analysis above and the order-driven environment in the HKSE, the effect of short-

sale constraints on liquidity and volatility is ultimately an empirical question. For example,

Diamond and Verrechia (1987) ignore the depth dimension of liquidity as they assume one unit

size for all trades; however, one could imagine that trade size (or depth of the limit order) would

matter. In any case, using the intuition presented in Diamond and Verrecchia (1987), our three

main hypotheses are as listed below.

H1:When a stock is initially allowed for short-sale, on average, the bid-ask spreads, and

especially the adverse selection component of the bid-ask spreads, will drop.

Note that in the world of Diamond and Verrecchia, the market makers incur no inventory

cost. Therefore, the bid-ask spread is really the asymmetric information component rather than

the overall bid-ask spread. In the empirical analysis, we will decompose the total spread into

adverse-selection and order-processing components.

H2: When there is no uptick rule, allowing a stock for short-sale will reduce bid-ask spreads,

and particularly the adverse-selection components of the bid-ask spreads, more than when the

uptick rule is present.

We speculate that firms with increased incidence of no-trade outcomes when short-selling is

prohibited will be the less-visible firms as measured by: firms with less analyst coverage, larger

adverse-selection cost component of the bid-ask spread, low price per share, and high-relative

tick size (given the same market capitalization). Even more importantly, these are stocks that

ultimately have the most to gain from short-selling in terms of liquidity.

H3: Among the stocks initially allowed for short-sale, on average the bid-ask spreads will

6For example, see Figlewski (1981); Boehme, Danielsen, and Sorescu (2006); Danielsen and Sorescu (2001);

and Jones and Lamont (2002), among others.

Page 11

drop more significantly for stocks with less public information (or more private information) —

i.e. less visible firms.

The last hypothesis is also related to the results reported by Cohen, Diether, and Malloy

(2007) and Beber and Pagano (2010). Cohen, Diether, and Malloy (2007) find that an increase

in shorting demand leads to negative abnormal returns of 2.98% in the following month,

and more significantly, their results are stronger in environments with less public information

flow, suggesting that the shorting market is an important mechanism for private information

revelation. Beber and Pagano (2010) find that short-selling bans are associated with decrease in

liquidity as measured by daily bid-ask spreads and that this effect is more pronounced for small-

cap and more volatile stocks. However their study falls into before-vs.-after studies in which the

short-selling bans are triggered by recent market changes. Our paper uses high-frequency data,

which give us the possibility to look at intraday spreads and more importantly to decompose

it into order-processing and adverse-selection components, which is the core issue behind the

Diamond and Verrecchia (1987) model. The high-frequency data also allow us to study order

imbalances, relative depths of the limit order book, and intraday volatility in an order-driven

market.

Another important issue is whether short-sales are connected to market crashes. Bernardo

and Welch (2004) suggest that fear of financial crises, rather than fear of a real liquidity shock, is

the true cause of financial crises. One implication of their model is that implementing constraints

that hinder some market participants from front-running other investors can effectively prevent

financial crises from occurring. This implication is consistent with the finding of Allen and Gale

(1991) that short-sales can potentially destabilize an economy. On the other hand, Hong and

Stein (2003) argue that if some investors are constrained from selling short, their accumulated

unrevealed negative information will not be manifested until the market begins to drop, which

further aggravates market declines and leads to a crash. The Bai, Chang, and Wang (2006)

rational-expectations equilibrium model suggests that in the absence of information asymmetry,

short-sale constraints reduce stock price volatility because they limit the range of fluctuations in

aggregate stock demand. But in the presence of information asymmetry, short-sale constraints

can cause the price volatility to increase as less-informed investors perceive higher risks and

Page 12

demand larger price adjustments in accommodating trades. Therefore, we expect that for less-

opaque, more-visible firms, once the short-sale prohibition is lifted the volatility will rise and

that for more-opaque, less-visible firms, the volatility will drop. Kraus and Rubin (2003) model

also predicts that volatility could either increase or decrease depending on the parameter values

of the model.

Thus, we also investigate the individual stock’s volatilities using the realized intraday

volatility measure developed in Andersen, Bollerslev, Diebold, and Labys (2003).Previous

studies on the effect of short-sale constraints have relied on daily data to estimate stock volatility

and skewness. We use a realized volatility measure calculated using transaction data, and the

realized volatility measure has been shown to be more precise than those calculated using daily

data (e.g. Andersen, Bollerslev, Diebold, and Labys, 2003).

In his study of the 1930s U.S. short-sale regulation changes, Jones (2008) finds that among

the three events for which short-sales were more tightly regulated, volatility did not change

significantly in two of the events, while in the third event there was weak evidence that volatility

actually became higher afterwards. Bris, Goetzmann, and Zhu (2007) test whether short-sale

constraints stabilize or destabilize financial markets by examining the skewness of market and

individual stock returns in 46 countries. They find strong evidence that in markets where short-

selling is either prohibited or not practiced, market returns display significantly less negative

skewness. However, at the individual stock level, short-sales restrictions appear to make no

difference. Charoenrook and Daouk (2008), in their cross-country study of 111 countries, find

that when short-selling is possible, aggregate stock returns are less volatile.They find no

evidence that short-sale restrictions affect either the level of skewness of returns or the probability

of a market crash. Chang, Cheng, and Yu (2007), using daily returns from HKSE, find that

when short-sales are allowed, individual stock returns exhibit higher volatility and less positive

skewness. Diether, Lee, and Werner (2009) find that relaxing short-sale constraints increased

market volatility for their NYSE sample but not the NASDAQ sample. Alexander and Peterson

(2008) find no effects on volatility when short-sale constraints are suspended either in NYSE

or in NASDAQ. Ho (1996) finds an increase in stock return volatility when short sales were

restricted during the Pan Electric Crisis in the Singapore market in 1985–1986.

Page 13

III. Methodology and Data

In this section, we first describe the overall methodology for conducting our tests, then we

describe our sample selection and provide summary statistics, and finally we conclude with the

construction of our measures of market quality.

A. Methodology: Difference-in-Difference Estimation

We first consider a simple time-series comparison of means or medians of the values of

interest in determining the impact of short-sale regulation changes on the following proxies of

market quality: proportional quoted spread, proportional effective spread, order-processing and

adverse-selection costs, and intraday return volatility. The construction of these measures will

be discussed later in this section. It is possible, however, that the time-series comparison may

simply capture the secular trend for the market as a whole for the measures. This is particularly

plausible if the pre-event and post-event windows that we choose are too long. Therefore, in

most of our analyses, we intentionally choose a short event window around short-sale regulation

changes. To further minimize such a possibility, we also construct a market capitalization, price

and dollar volume matched sample, and we examine whether there is any secular trend in this

control sample.7To mitigate the impact of time-series trends when examining the short-sale

regulation changes’ effects, we use the difference-in-difference regression approach (Meyer, 1995;

Bertrand, Dufflo, and Mullainathan, 2004; Angrist and Krueger, 2001). Specifically, we estimate

the following difference-in-difference regression:

yi= α0+ β1Shortablei+ β2NoRulei+ β3Periodi+ β4Groupi+ γXi+ ?i,

(1)

where yiis the individual stock’s average market quality measure (e.g. spread, intraday volatility,

etc.) during either the pre- or post-event window. Shortableiequals one if the stock is eligible

for short sale, NoRuleiequals one if the stock eligible for short sale is during the period without

the uptick rule, Periodiequals one if the observation lies in the post-event period, and Groupi

7We choose these variables because they have been shown to be related to cross-sectional difference in spreads

in other studies. For a survey, see Appendix A of United State Security and Exchange Commission’s report on

the comparison of order executions across equity market structure (2001). The document can be obtained from

www.sec.gov/pdf/ordrxmkt.pdf.

Page 14

equals one if the observation belongs to the treatment group (stocks that ever become eligible

during our sample). Xiis a vector of stock-specific control covariates to mitigate the imperfect

matching of treatment and control groups. In the regressions, we use the logarithm of individual

stocks’ market capitalization, inverse of the price, and logarithm of dollar volume. For both the

treatment group and control group, the pre- (post-) window values of the control covariates are

measured at the beginning of the month prior (the end of the month after) a stock’s becoming

eligible for short sale. The testable hypotheses boil down to testing: (1) whether the event of

stock becoming eligible for short-selling impacts individual stock market quality, or β1= 0; (2)

whether the event of “no uptick rule” impacts individual stock market quality, or β2= 0 ; and

(3) whether the removal of the short-sale constraint during the no-uptick-rule period impacts

individual stock market quality, or β1+ β2= 0.

Note that we have average pre-event and post-event measures for both the treatment sample

and the control sample. So if there are N stocks that ever became shortable, then the number

of observations in regression (1) is 4N. As a convention, in the tables in which we report the

number of observations, we only report the number of stocks eligible for short-sale rather than

the total number of observations in the regression. Bertrand, Dufflo, and Mullainathan (2004)

point out several methodological issues in the application of difference-in-difference estimation.

Their simulations illustrate that the treatment effects could be significantly overestimated when

the pre- and post-event windows contain repeatedly sampled observations without appropriate

adjustments. Their evidence also suggests that the t-values we compute for the treatment effect

are rather conservative.

We further acknowledge that the selection of stocks to the short-list is not an exogenous

event. According to the HKSE, stocks are selected based on market capitalization and turnover

ratios, and both criteria are associated with the market microstructure attributes that we are

examining here. Therefore, our empirical analysis is not carried out in a “natural experiment”

but a “quasi-natural experiment” setting (using the terminology of Meyer, 1995). We take into

account such selection bias by using a difference-in-difference approach and by using individual

the dummy variables Groupi and Periodi are significantly different from zero, the control is

particularly problematic. Fortunately, in many of our difference-in-difference estimations, none

of the coefficient estimates from Groupior Periodiindicates any serious problem.

B. Data

We rely on five data sources. The data on the stock returns, number of shares outstanding,

and dollar trading volume are from the Daily Stock Price and Returns File — Hong Kong, the

Pacific-Basin Capital Markets Databases (PACAP). The PACAP data cover all stocks listed on

the HKSE for the period January 1995 to December 2002. Unfortunately, PACAP stops covering

the HKSE stocks after December 2002, so we supplement this information for 2003 and 2004

from our second database, the Standard and Poor’s Global Advantage Issues database. Since the

PACAP and Standard and Poor’s Global Issues databases do not share a unique common link,

we hand matched them based on ticker symbol, SEDOL/CUSIP code, and company name. Our

third data source is the intraday transaction data from the HKSE. Because the HKSE started

to release the bid-ask records only in March 1996, our sample of complete intraday transaction

data starts from March 1996 and ends in December 2003.8

These data (hereinafter “HKTAQ”) are similar to the Trade and Quote database (TAQ)

released by the NYSE. These data contain two files: Trade Records and Bid-Ask Records.9

However, these data differ from the TAQ data provided by the NYSE in two aspects. First,

the Bid-Ask Records file provided by HKSE records not only the best bid and ask prices and

8Therefore, our sample excludes the first batch of 17 Hong Seng Index constituent stocks. Additionally, even

though in most of our empirical exercise, we require the [-30, +30]-day event window, we relax this particular

requirement for the stock introduced to the short-sale list on March 25, 1996 (since for this batch, the trade and

quote data only have 25 calendar days before the event). If we exclude all of this batch of observations, we lose

a large number of stocks eligible for short sale.

9The trade and bid-ask files provided by the HKSE have two identifiers: STKCODE and STK ID. STKCODE

is not unique and can be reused once the company is delisted from the HKSE, but STKCODE is unique. In the

cleaned HKTAQ database on the server, we retain STK ID. The I/B/E/S International Summary File contains

both the STKCODE (albeit under the different name called EXCHANGE TICKER) and SEDOL Code. The

DataStream database contains the Local Code, where the first two digits are for country/region and the other

digits are for local exchange code. That is, we may link HKTAQ, I/B/E/S, and DataStream together via the

STKCODE. An alternative source of accounting information is Standard and Poor’s Global Advantage. However,

there is no common link identification from Global Advantage to the other databases; even though we can locate

the variant of SEDOL code, it does not seem to match the other three databases. So we hand matched Global

Advantage with the other databases. To check the consistency of our matching, we also referenced the Master

file of the HK Trade and Quotes database, DataStream, and I/B/E/S. As the ticker symbols may be reused by

different issues in the HKSE, the name match is of critical importance to ensure the integrity of the matching