Stock Exchange Governance and Market Quality

Chandrasekhar Krishnamurti Division of Banking and Finance

Nanyang Business School Nanyang Technological University

S3-B1-B-76 Nanyang Avenue Singapore 639798

Tel: (65) 6790-5702 Fax: (65) 6791-3697

E-mail: ackrishna@ntu.edu.sg

John M. Sequeira Department of Finance and Accounting

NUS Business School National University of Singapore 1 Business Link, BIZ1 Building,

Singapore 117592 Tel: (65) 6874 3033 Fax: (65) 6779 2083

E-mail: bizjms@nus.edu.sg

and

Fu Fangjian William E. Simon School of Business

University of Rochester Rochester, NY 14627 Tel: (716) 2754318

Email: fufa@simon.rochester.edu

Revised: July 2002

Corresponding Author:

Chandrasekhar Krishnamurti Division of Banking and Finance Nanyang Business School Nanyang Technological University S3-B1-B-76 Nanyang Avenue Singapore 639798 Tel: (65) 6790-5702 Fax: (65) 6791-3697 E-mail: ackrishna@ntu.edu.sg

2

Stock Exchange Governance and Market Quality

Abstract

Does the organization structure of a stock exchange matter? We address this interesting issue utilizing the unique setting prevailing in India. India has two major stock markets, namely, the Bombay Stock Exchange (BSE) and the National Stock Exchange (NSE). These two exchanges adopt similar trading systems, trade essentially identical stocks, and follow the same trading hours. However, both exchanges have different organizational structures: BSE is a mutualized form of organization whereas NSE is structured as a demutualized exchange.

Domowitz and Steil (1999) propose that demutualized exchanges have better incentives to provide innovative services that increase the value of the exchange. Their proposition implies that demutualized exchanges are superior to mutualized exchanges in governance. Another inference that can be drawn from Domowitz and Steil’s work is that a demutualized exchange, due to its superior governance structure, should provide a better quality market. Using the Hasbrouck (1993) measure of market quality, we demonstrate empirically that NSE provides a better quality market compared to BSE. We attribute this divergence in market quality to the differences in the governance of the respective stock exchanges. Furthermore, we show that NSE provides a better quality market even after controlling for firm specific factors. Our study has important policy implications for other emerging countries. We show that entrenched monopolies may be successfully persuaded to modify their behavior by engendering competition in the market place, utilizing technology as a catalyst. Another contribution of our study is that we show that latest technological advances in networking, communications, and information processing, may be utilized to break down barriers to entry in the trading services industry.

3

1. Introduction

In sharp contrast to the plethora of studies that have examined the impact of corporate

governance provisions adopted by manufacturing and service firms,1 there has been very

little research attention devoted to the governance of stock exchanges. A notable

exception is the recent work of Domowitz and Steil (1999), who examine the

interrelationships between stock exchange automation, governance, and the quality of

markets. Traditionally, stock exchanges have been organized as non-profit, mutual/

membership associations. A recent trend has been the conversion of mutualized

exchanges into publicly owned corporations, which are themselves listed and traded on a

stock exchange. Domowitz and Steil (1999) list several benefits of the demutualized

stock exchanges as compared to mutualized stock exchanges. The primary driver for

such benefits is the favorable governance structure associated with demutualized

exchanges. In their paper, Domowitz and Steil (1999) argue that members of mutualized

stock exchanges have incentives to oppose innovations even if these increase the value of

the exchange. Since traditional stock exchanges are mostly regional monopolies, they

could, in the extreme case, even oppose enhancements to quality of services that they

provide if such improvements are thought to diminish the welfare of the respective

exchange members.

One important implication of the Domowitz and Steil (1999) argument is that

demutualized stock exchanges should provide a better quality market than mutualized

ones. For expositional convenience, we will refer to this as the “Domowitz and Steil

Proposition.” In fact, the primary focus of our paper is to examine the Domowitz and

Steil Proposition empirically using data from two competing stock exchanges in India

1 Gompers, Ishii, and Metrick (2001), in a recent study, find a striking relationship between corporate governance and stock returns. In their paper, they also conclude that firm value is highly correlated with a “Governance Index” that they develop.

4

which differ in their governance structure – namely, the Bombay Stock Exchange and the

National Stock Exchange. Our efforts to make direct comparisons of market quality in

the two stock exchanges are greatly facilitated by the simultaneous trading of at least

forty major stocks on both the exchanges, with both utilizing similar trading systems.

While prior studies have compared the quality of stock markets with different trading

systems, reliable comparisons are, in fact, difficult to achieve.2

Our study, therefore, makes unique contributions to the literature in several ways.

First, we compare the quality of two markets, which are similar in most respects except

their governance structure. Previous studies to date have not, as yet, examined the impact

of governance structure on the quality of stock markets. Second, since our study is

conducted in an emerging market setting, it has policy implications for other emerging

markets3. Third, our paper describes how advances in trading technology have

implications for competition in the stock market trading industry. In particular, we show

the importance of information technology in breaking down barriers to entry. Finally, we

draw some policy implications for regulators who supervise “natural monopolies” such as

stock exchanges.

The rest of the paper is organized as follows. Section 2 provides the background

information on Indian stock markets outlining the principal differences between the two

stock markets analysed. Section 3 discusses the methodology used in this paper. Section

4 describes the data and empirical results are presented in Section 5. The final section

contains our conclusions.

2 Prior studies have compared quality of different markets by examining measures of liquidity and execution costs of comparable stocks. In such studies, one cannot be certain that firm-specific characteristics are not, in fact, driving the observed results. See, for instance, Affleck-Graves et.al (1994), Huang and Stoll (1996b), and Bessembinder and Kaufman (1997). 3 Parisi, Nail and Soto (1991) compare the Santiago Stock Exchange with the Electronic Stock Exchange to examine which exchange plays a leadership role. They are, however, unable to examine the effect of governance on market quality since observed differences may be attributable either to the variations in trading systems or differences in governance.

5

2. Bombay Stock Exchange and the National Stock Exchange – A Tale of Two

Stock Exchanges

2.1 Genesis of the National Stock Exchange

Until the early 1990’s, the Bombay Stock Exchange (BSE) was the premier stock

exchange in India. At that time, BSE was plagued with a variety of problems, principal

among them, the outdated trading and settlement procedures, and poor regard for investor

protection. The recalcitrant attitude of the brokers and the administration of BSE

frustrated the efforts of the regulator, namely, the Securities Exchange Board of India

(SEBI), in pushing through critical reforms to bring the premier exchange of India on par

with the best run exchanges in the world. The complacency of the administration of BSE

may be attributed to their overestimating the significance of barriers to entry in the

provision of stock broking and trading services. BSE also underestimated the

determination of the Government of India to reform the stock markets, who took the

unprecedented step in creating a new stock exchange with the help of government owned

and controlled financial institutions.

The recent technological advances in communication and computing technologies

aided the efforts of these institutions. A new stock exchange – the National Stock

Exchange (NSE), emerged as a consequence of the government’s efforts and initiatives.

The ensuing competitive market structure in Indian stock markets, as an outcome of the

reform process, is unique and without parallels. NSE dramatically altered the stock

market landscape of the country within a short period of its inception, using current best

practices in computing and network technology to provide state of the art trading services

to Indian investors. In short, NSE dramatically improved the quality of trading services

and thereby challenged the dominant position held by BSE until that time.

6

Indian investors were attracted to the high quality of trading services and systems

provided by NSE. In particular, NSE’s systems and procedures explicitly incorporated

investor protection. As a result, BSE rapidly lost its market share. In order to stem the

erosion of its market share further, BSE was forced to take steps to improve its services.

NSE, however, retained its first-mover advantages on account of its early lead in

adopting leading edge technologies and procedures.

Another significant difference between the two exchanges pertains to governance.

BSE follows the mutual form of organization whereby stockbrokers own and operate the

exchange. The president of the exchange is elected by the broker members and,

therefore, represents the interests of the brokers; the interests of investors are

consequently relegated to a secondary position. On the other hand, NSE follows the

corporate or demutualized form of organization, defined as the separation of ownership

of the exchange from membership. According to Domowitz and Steil (1999), the

incentives of a mutualized exchange differ significantly from a demutualized one.

Members of mutualized exchanges have incentives to oppose innovations, even if such

innovations enhance the value of the exchange. This possibility arises if innovations that

reduce the demand for their intermediation services result in a decline in their brokerage

profits that is not balanced by a corresponding increase in the value of the exchange.

This line of reasoning implies that NSE has incentives to adopt actions that increase the

quality of its market, while BSE is more likely to resist changes that increase quality, but

have a deleterious effect on their brokerage profits. Thus, we have reasons to believe that

the quality of NSE’s market will be higher than that of BSE, ceteris paribus. Hence,

market quality comparison between NSE and BSE forms the core motivation for our

paper.

7

2.2 Differences between NSE and BSE

In this section, we summarize the principal differences between NSE and BSE. Since

the overall objective of our paper is to make quality of market comparisons between the

two exchanges, we first start by outlining the principal differences between them. This

would allow us to appreciate the theoretical underpinnings and enable us to formulate

appropriate expectations regarding quality differences. We use the following dimensions

with which to make these comparisons: Ownership, Governance, Trading Systems,

Connectivity, Leadership, and Technology use.

2.2.1 Ownership

NSE has been incorporated as a tax paying company and is owned by a group of

large developmental financial institutions. On the other hand, BSE is organized as a

brokers’ association. The broker members who “own” seats on the exchange own and

operate BSE.

2.2.2 Governance

NSE is professionally managed by a fulltime Managing Director who reports to a

Board of Directors, and is, as such, managed by professionals who do not directly or

indirectly trade on the exchange; ownership and management of the exchange are,

therefore, completely separated. Until NSE emerged, most of the stock exchanges in

India were owned, controlled and managed by brokers who held seats on the various

exchanges. This mutualized form of organization was prevalent in Indian stock

exchanges, and BSE was no exception. In the case of BSE, one of the broker members is

elected to administer the exchange and is designated as the president of BSE, who also

owns a seat on the exchange, like the other broker members. Unlike BSE, NSE is the

first demutualized exchange in India.

8

2.2.3 Trading System

NSE uses the latest innovations in computing and network technologies to bring the

best available trading system to its customers, and endeavours to constantly improve the

quality of its trading system in its bid to attract and retain its customers. A fully

automated screen based trading system, referred to as NEAT and developed by NSE,

enables members from all over the country to trade with one another on a real-time basis

with ease and efficiency. Three of the more prominent features of NEAT are discussed

below. First, NEAT is a completely automated system for order matching that is

completely order driven and provides complete anonymity to its trading members.

Second, NEAT operates on a strict price time priority such that all buy orders received on

the system are sorted with the best-priced order getting the first priority for matching with

an incoming market sell order. For instance, within orders which have the same price,

time priority is enforced, that is, orders placed first are executed first. This matching

process is computerised, keeping the system transparent, objective and fair. Each order

remains in the system until it is matched with an incoming market order or until it is

cancelled, whichever occurs earlier. Third, is the tremendous flexibility afforded by

NEAT that provides users with several time related orders such as Good-Till-Cancelled,

Good-Till-Day, Immediate-or-Cancel. Moreover, traders are able to take advantage of

price-related and volume-related orders that may be built into an order. Fully accessible

information on total order depth in a security, last traded price, quantity traded during the

day, amongst others, allows traders to make informed decisions regarding price and order

quantity.

BSE has, however, been a laggard in the use of technology. Floor trading was still

the prevalent mode of transacting when NSE commenced operations in November 1994.

Since then, a rapid erosion of market share forced BSE to establish its own computerized

9

trading system. This move was a strategic response to a comparatively superior trading

system established by NSE. By June 1995, BSE inaugurated its own electronic trading

system known as BOLT. BOLT adopted many of the features that were built into NEAT.

BSE is a laggard compared to NSE; NSE stands out clearly as the originator of various

new initiatives. In fact, whenever NSE unveils a new initiative, BSE usually responds,

rather reluctantly after a time lag, with a view to protect its market share rather than a

genuine effort to increase customer satisfaction.

2.2.4 Connectivity

NSE established a national network of terminals in over 300 towns and cities which

included smaller towns and cities not currently served by any stock exchange. A reliable

network technology is employed which uses satellite communication to link all member

terminals to its main computer located in Bombay. This equal access to traders all over

the country sharply contrasts BSE, which serves only the city of Bombay. BSE was

initially operated as a regional monopoly with a charter allowing its trading members

exclusive rights and access to its trading floor in Bombay. Shortly after NSE commenced

its trading operations, its runaway success seemed to threaten the very livelihood of BSE

brokers. The cost of a seat on BSE fell from about Rs. 40 million to about Rs. 10 million

within a year of the opening of NSE. To counter this, BSE quickly set up BOLT, which

started operating from 1995. The connectivity of BOLT was still, however, restricted to

the city of Bombay, while NEAT’s influence continued to over the entire country. As the

first automated exchange in India, NSE had a freer hand in determining its reach. With

no physical trading floor limitation, it could logically argue that its reach is able to cover

the entire country. BSE and all the other regional exchanges in India are regional

monopolies, and according to their charter, are explicitly prohibited from setting up

offices or providing their services in locations outside their respective jurisdictions. This

10

explicit prohibition served to limit the connectivity and reach of BSE; for instance, a BSE

member could not set up an office and provide trading services in the city of Delhi, which

has its own stock exchange. NSE is unaffected by these limitations, has enjoyed much

network externalities, at a cost to its most powerful rival, BSE.

2.2.5 Leadership

From the outset, NSE behaved like a leader – it set the trend for other regional

exchanges including BSE to follow. This leadership status bestowed NSE with the first-

mover advantages that BSE sadly lacked. After ceding its position as the premier

exchange in the country, BSE has had little choice but to trail behind NSE. NSE has

managed to maintain its position by being the innovator since its inception; its moves are

carefully thought out and flawlessly executed. Moreover, NSE is motivated by its

ambition to meet or exceed current international standards pertaining to the provision of

trading services whereas BSE has had to consider appropriate responses to major NSE

initiatives. Frequently, BSE’s ability to react has been stalled by the actions of its own

members whose private incentives conflict with those of the exchange as a whole.

2.2.6 Technology

NSE owes its existence to advances in information technology, which it has

effectively employed to break down barriers to entry in the stock market trading business.

In continuing to apply the latest technology, NSE views itself as the front-runner

exchange. Having pioneered the use of automated order matching and increasing its

reach to cover the entire country via satellites, NSE has, in effect, cleverly used

technology to break the monopoly of BSE. For instance, on NSE, the time lag between

placing an order and getting confirmation from anywhere around the country, takes less

than two seconds. NSE’s superior technology enabled it to consolidate a previously

highly fragmented order book, and this consolidation has resulted in substantial increases

11

in liquidity. BSE began the decade of the 1990’s with a deep suspicion of technology; its

members seeing technology as a major hindrance to maintaining its rent seeking

activities. In fact, BSE’s members were not in favour of a transparent system. When BSE

chose rather reluctantly to embrace automation, its primary motive was to recapture some

order flow that it had lost to NSE. In choosing to implement this technology, BSE was

torn between the goals of value maximization of the exchange and the rent seeking

activities of its broker members. As a consequence, BSE has perpetually lagged behind

NSE in adopting new and innovative technology.

2.3 Automation, Governance and Quality of Markets

Although we identify six major differences between NSE and BSE, the ownership

structure is considered to be the main driving force distinguishing both markets. In fact,

we argue that it is the difference in ownership that determines the characteristics of each

exchange. Everything that NSE has achieved since inception, BSE potentially could have

accomplished in a superior manner. As the premier exchange in India, BSE had first-

mover advantage and the resources to pursue excellence and to maintain its dominant

status. That BSE chose not to pursue NSE’s strategy of creating a high quality automated

nationwide market for trading securities, needs to be emphasized. Had BSE adopted

NSE’s approach, it would have pre-empted the creation of NSE. BSE’s adherence to its

status quo in trading technology becomes apparent once its ownership structure is

detailed. Essentially, BSE is organized as a mutualized exchange or a broker’s

association, where ownership is not separated from exchange membership. As exchange

members receive their profits from intermediating non-member transactions, these

members have vested financial interests in resisting innovations that may actually

increase the value of the exchange, especially if these innovations reduce their potential

profits. The brokers of BSE resisted automation and the attendant improvements in

12

market quality as they perceived that automation would lead to market transparency. In a

transparent system, these brokers would be precluded from certain rent-seeking activities

which are deemed illegal. Unlike the automated system, a manual trading system would

not easily detect the illegal activities of these brokers. It is with these considerations that

the floor trading system continued in spite of alleged inefficiencies. This culture of

pleasing the brokers at the cost of investors is very strong in BSE.

According to Domowitz and Steil, “trading market automation permits

demutualisation”, meaning that the corporate structure of organization of a stock

exchange is feasible when computerized trading replaces floor trading. We illustrate in

this paper a situation where the presumed causality runs in the reverse direction. NSE’s

corporate organizational structure allowed it to pursue trading automation unhindered by

potential conflicts of interest with its broker-members. Based on a governance structure

comprising professional managers who report to the Board of Directors, NSE’s primary

objective is consequently, value maximization. Had NSE adopted the mutual form of

organization, it would have failed to incorporate trading automation since the interests of

the broker members would have taken precedence to goals of value maximization for the

exchange. Several prominent financial writers around this period, in fact, questioned the

wisdom of the government of India in “starting a second stock exchange in Bombay”.

That NSE did not become the “second exchange” in Bombay is mainly attributable to its

governance as compared to BSE. NSE’s professional managers had absolute freedom to

decide on a trading technology that was most appropriate. In order to survive, NSE

needed to break down the barriers to entry created and maintained by existing stock

exchanges. NSE was quick to dismiss floor trading as a preferred option, seeing this as

merely mimicking the other exchanges who subjected themselves to the inherent

limitations associated with floor trading. In its drive to create a better securities market,

13

NSE had two basic objectives: to improve the quality of the market and reduce the cost of

transacting. NSE viewed these objectives as necessary to enable them to capture market

share and become the dominant player in the business of trading securities. NSE’s choice

of trade automation and nationwide connectivity through satellites served to break

barriers of entry in the stock broking business. In addition, these initiatives created

positive externalities.

Having established the superiority of NSE’s governance as compared to BSE, we

next turn to market quality comparisons in order to substantiate the Domowitz and Steil

(1999) proposition.

3. Empirical Methods

3.1 Prior Literature

A unidimensional quantification of market quality is absent in extant literature.

Prior research has focused on liquidity, informational efficiency, and volatility

characteristics of markets, as the criteria for market quality comparisons. Another view

of market quality is based on transaction costs. In stock markets, transaction costs may

be classified either as explicit or implicit costs (execution costs). The execution cost is

the difference between the actual transaction price and the benchmark price that is

considered to be in some sense, efficient. Unlike commission costs which are explicit,

execution costs are not directly observed and are not easily measured. Hence, most prior

studies have defined different measures of execution costs4 and compared these estimates

across markets (e.g., Roll 1984, Berkowitz, Logue and Noser 1988, Stoll 1989, Huang

and Stoll 1994, Chan and Lakonishok 1993, Hasbrouck 1993).

4 Measurement of this difference between the actual transaction price and the benchmark price is generally aimed at estimating this figure for both the buyer and seller engaged in a particular transaction. This difference is often calculated as an average transaction cost measure in comparative market analysis, and is used to determine and evaluate market microstructure studies when transaction costs are minimized [Huang and Stoll (1996b) and Bessembinder and Kaufman (1997)]. Such a procedure is applied in this paper.

14

Several measures of transaction (or execution) cost have been used, principal

among them, are quoted bid-ask spreads and effective bid-ask spreads. In a pure

dealership market, transactions take place only at the quoted bid or the ask price.

Appropriate measures of transaction cost should, therefore, be based on such

quoted/posted bid-ask spreads (see for example, Demsetz 1968, Branch and Freed 1977,

Benston and Hagerman 1974, Huang and Stoll 1996, Barclay et al. 1999). In many

security markets, however, trades frequently take place inside the spread.5 When this

occurs, the quoted spread will tend to overstate the investors’ expected trading costs,

making the effective spread, which is simply the average difference between the price at

which a dealer sells at one point in time and buys at an earlier point in time, a better

measure for trading costs (e.g., Roll 1984, Stoll 1985).

Using two samples of market orders, one based on orders submitted by retail

brokers and another submitted electronically to the NYSE, Petersen and Fialkowski

(1994) estimate the spread generated using both orders and report a significant difference

between the posted spread and the effective spread paid by investors. Applications of this

effective spread method to measure trading costs have been used in numerous studies

(e.g., Barclay 1997, 1999, Bessembinder 1997, Christie and Huang 1994, Christie et al.

1994, Huang and Stoll 1996b).

An alternative approach to measure transaction costs is to calculate the difference

between the actual transaction price and the efficient price. Broadly defined, the efficient

price, which is considered as the true value of a stock, is an unbiased price estimate that

can be achieved in any relevant trading period by any randomly selected trader.

Beebower (1989), for example, uses daily high-low midpoint prices and closing prices as 5 Blume and Goldstein(1992), for example, find that between 12 to 31 percent of the trades occur inside the spread. Once transactions occur inside the quoted spread, it is often the case that the quoted spread will tend to overstate the investors’ expected trading costs and, as such, is not an appropriate measure of transaction costs.

15

efficient prices, and the effective spread to compute transaction costs. This measure may,

however, place an excessive weight on trades that are not representative of most trades

over a particular trading period and is, therefore, likely to be a biased representation of

actual transaction prices. To address this, Berkowitz et al. (1988) propose a measure of

transaction costs based on the volume-weighted average price over the trading day.

Although their measure yields results that appear less biased as compared to other

measures that only use single prices, two main problems may arise: (1) the measure may

weight an aberrant trade very heavily, especially when a very thinly traded stock with

small market value experiences a relatively large transaction; (2) when gaming takes

places, that is, traders adjusting their trading behavior to affect the “efficient price” when

they become aware that they are being evaluated on this basis. Another approach

introduced in Hasbrouck and Schwartz (1988) is an overall measure of average execution

costs for all trades of a certain stock, in a particular market, over a certain period of time.

This measure relies on the notion that execution costs increase the volatility of short-term

price movements relative to the volatility of long-term price movements. Although the

measure is an average value for all trades in a certain market, it is strongly influenced by

the preponderance of small trades since it was not designed to measure how execution

costs are affected by trading size. A newer approach developed by Hasbrouck (1993)

measures the transaction costs in stock markets based on the methodology that a non-

stationary time series can be divided into a random-walk component and a residual

stationary component. When applied to stock transaction prices, the random-walk

component is identified as the efficient price with the stationary component (termed the

pricing error) representing the difference between the efficient price and the actual

transaction price.

16

The dispersion of the pricing error that results from this division, measures how

closely actual transaction prices follow a random walk and, therefore, constitutes an

appropriate measure for transaction costs.

In our paper, we apply the Hasbrouck (1993) method to compute the variance of

pricing errors, which is used as the principal metric for market quality comparison.

Having restricted access to the order book, we were not able to obtain data on quoted bid

and ask prices, and hence, unable to compute quoted or effective spreads. We partially

overcome this handicap by using the methodology of Roll(1984). Roll’s technique

involves extracting spreads from transaction prices. Comparisons using these extracted

spreads are used to infer about the relative quality of NSE and BSE. We apply the

multivariate regression approach of Hasbrouck and Schwartz (1988) to identify the

source of the observed differences in market quality. This approach uses the intuition

that firm-specific trading characteristics may have an effect on overall market quality.

In our paper, we use two different methods of transactions cost measurement, namely,

the Hasbrouck (1993) approach to estimate the dispersion of the pricing error, and the

Roll (1984) approach, to compute the implicit effective bid-ask spread. An extension of

the Hasbrouck (1993) method that includes other factors that may affect transaction costs

is also discussed6.

3.2 Hasbrouck’s (1993) approach

Hasbrouck’s (1993) approach to measure transaction costs is based on the fact

that any time series that exhibits homogeneous non-stationarity may be decomposed into

two separate components, namely, a stationary series and a pure random walk. The

stationary component is defined as the predictable momentum in the series at each point

of time while the random-walk component is simply taken as the mid-point of the

6 This method relies on the methodology introduced in Hasbrouck and Schwartz (1988).

17

predictive distribution for the future path of the original series. When applied to stock

transaction prices, the random-walk component is taken to be the efficient price. The

residual stationary component resulting from the decomposition (also known as the

pricing error) represents the deviation of transaction prices from the efficient price. Since

the dispersion of the pricing error tracks how closely actual transaction prices follow the

efficient price, it is, therefore, viewed as an appropriate measure of transaction costs.

Hasbrouck proposes the standard deviation of the pricing error, sσ , as a summary

measure of market quality that measures how closely the transaction price tracks the

efficient price. This measure of market quality can then be applied to different markets to

make appropriate inferences about market quality. The role of sσ as a proxy for market

quality depends on the assumption that as transaction costs and other trading barriers are

reduced, actual transaction prices should track the efficient prices more closely. The

details regarding the estimation procedure are provided in Appendix.

3.3 Roll’s (1984) approach

In our paper, we also use Roll’s (1984) approach to measure the effective bid-ask

spread which is generally considered as the transaction cost of that stock. The Roll (1984)

approach is a simple but efficient method that allows the effective bid-ask spread to be

inferred directly from a time series of transaction prices only, based on the following

assumptions:

(1) the stock is traded in an informationally efficient market;

(2) the probability distribution of observed price changes is stationary.

To compute the effective bid-ask spread, Roll (1984) shows that the existence of

trading costs in a perfect market leads to negative serial dependence in successive

observed transactions price changes in the otherwise random walk behavior of the

18

efficient price of a stock. On this basis, Roll (1984) derives the covariance between

successive price changes as follows:

4/),( 21 sPPCov tt −=∆∆ + , (12)

where 1−−=∆ ttt PPP and ttt PPP −=∆ ++ 11 ; the ),( 1+∆∆ tt PPCov term represents the first-

order serial covariance of transaction price changes with a value equal to the negative

square of one-half of the bid-ask spread; and P∆ is the underlying stock return with a

variance of 2/2s and an autocorrelation coefficient of –1/2.

In this model, s is not the quoted bid-ask spread. Since successive price changes

are recorded from actual transactions, s is, in fact, the effective bid-ask spread, a dollar-

weighted average spread faced by investors who trade at the observed prices. This, as

such, makes s an appropriate measure of trading cost. Rewriting equation (12) allows us

to define the effective bid-ask spread in relation to the negative covariance in price

changes as follows:

CovSpread −=∧

2 , (13)

where ∧

Spread now represents the estimated effective bid-ask spread. We use the

percentage returns of prices, namely, 11 / −−− ttt PPP , to compute )P,P(Cov 1tt +∆∆ . The

average stock transaction price varies across firms, making this measure more appealing

as a statistic for comparing effective spreads of different stocks. The resultant measure is

jS∧

, the estimate of the percentage bid-ask spread for stock j which is given as follows:

jj CovS∧∧

−= 200 . (14)

19

jCov∧

is the estimated serial covariance of returns7 of stock j which is scaled by 200 to

convert all measurements into percentage form. Both daily and weekly returns are

computed to obtain two different estimates of serial covariances for each stock.

4. Data

The dataset used in this study comprises 40 pairs of common stocks that are

simultaneously listed and traded on both BSE and NSE. All the selected stocks are

Indian index stocks that are observed to have large numbers of observations for each

trading day and hence higher liquidity as compared to other similar stocks. We

specifically selected these stocks to minimize stock-specific effects; for example, stocks

with larger market capitalization generally have lower trading costs (see Hasbrouck

1993). Previous studies, which include Huang and Stoll (1996) and Barclay (1997) that

attempt to compare the transaction costs between the NYSE and NASDAQ, employ

stocks that are only matched in a limited manner, resulting in findings that tend to be

strongly affected by stock-specific factors. For our paper, the forty pairs of stocks that are

traded on both BSE and NSE are identical (see Table 1 for a complete listing of the 40

stocks).

High frequency intraday transaction data are used in the first part of the study based

on Hasbrouck’s (1993) approach to compute transaction costs. All transaction records

for the 40 sample stocks cover the period 1 January 1997 to 31 July 1997 are obtained

from NSE and BSE. This dataset comprises a total of 144 trading days with the total

number of transaction exceeding 15 million. In the second part of the study that applies

Roll’s (1984) approach, we use daily and weekly prices data on both markets over the

7 Roll (1984) estimated the implicit bid-ask spread and examined its relation to firm size on the NYSE and AMEX over a period of 20 years. He found that the implicit bid-ask spread was negatively related to the market capitalization of stocks, a result that provides validates his method in estimating the implicit bid-ask spread as a measure of trading costs.

20

period 1 January 1997 to 31 March 2000 is used. For our study, we obtain information

on market value, average weekly /daily price, average trading size, and average trading

number per day, for each pair of stocks traded on both exchanges to determine factors

relevant in transaction cost measurement.

In constructing the time series of returns and trades, we ignore natural time

conventions. Instead, we view the data is viewed as an untimed sequence of observations,

with the time subscript t incremented each time a new transaction occurs. Overnight

returns are not used to avoid the overnight effects on stock prices. Lee and Ready’s

(1991) approach is used to infer the direction/sign of a trade by comparing its price to the

price of the preceding trade(s), and is further based on classifying each trade into four

categories: an uptick, a downtick, a zero-uptick, and a zero-downtick. A trade is

considered an uptick (downtick) if the price is higher (lower) than the price of the

previous trade. When the price is at the same level as the previous trade (that is a zero

tick), and when the last price change was an uptick, then the trade is considered a zero-

uptick. On the other hand, if the last price change was a downtick, then the trade is

considered a zero-downtick. Essentially, if two successive prices are the same, the latter

one is assumed to follow the sign of the preceding one. A trade is also classified as a buy

(sell) transaction if it occurs on an uptick (downtick) or a zero-uptick (zero-downtick).

Such an approach allows us to classify trade directions since quote data are not available

and is widely used by many studies8.

5. Empirical Results

In this section, we present the empirical results of transaction costs comparisons

for “multi-trading” stocks that are listed simultaneously on NSE and BSE. We compare

both markets using two different methods to compute the transaction costs: the standard

8 On such example is found in Holthausen, Leftwich and Mayers (1987).

21

deviation of the pricing error using Hasbrouck’s (1993) approach, and the daily price

covariance using Roll’s (1984) approach.

5.1 Standard deviation of the pricing error using Hasbrouck’s (1993) approach

Hasbrouck’s (1993) approach involves the estimation of the standard deviation of the

pricing errors which is obtained sequentially. In the first stage, equation (3) is estimated

using a VAR length of three lags, consistent with current best practice9. The trade

innovations tt vv ,2,1 , from the estimated VAR model are transformed into their vector

moving average representations (VMA) using equation (4). Estimates from the first

equation in the VMA are used to compute the 'α s and 'β s. The estimated pricing error

given by equation (8) and equation (9) (with all 0=tη ) are used to obtain the Beveridge

and Nelson (1981) adjusted estimate of tS , from which, the variable of interest, namely,

the standard deviation of the pricing error, sσ , is obtained using equation (11). Standard

deviations of the pricing errors for the paired sample stocks from both markets are

computed and are compared to determine which exchange is more cost-efficient.

Table 2 lists the computed values of Hasbrouck’s (1993) measure for all forty multi-

trading stocks on both markets and the ratio of this measure in NSE relative to its value

for BSE. The smallest standard deviation of the pricing error is 0.0152 and 0.0934 for

NSE and BSE, respectively; the largest values are 0.8639 and 1.4901 for NSE and BSE,

respectively.

Based on Hasbrouck’s (1993) approach, the standard deviation of pricing error

represents the percent value of transaction costs for a particular stock. Clearly, both the

minimum and maximum values of Hasbrouck’s measure are greater on BSE. Moreover,

the mean standard deviation of the pricing error is 0.27 and 0.64 on NSE and BSE,

9 Serial dependencies beyond three lags tends to be negligible and have little impact on the estimates in the model.

22

respectively, suggesting average transaction costs for the paired sample stocks are 0.27

percent of stock prices on NSE, as compared to 0.64 percent of the stock prices on BSE.

The mean ratio (NSE/BSE) of 0.4034 confirms that the average transaction cost of these

multi-trading stocks on NSE is lower than BSE. Hasbrouck’s (1993) results of 0.33

percent of stock prices on the NYSE are closer to NSE’s figure of 0.27% while BSE

reports values of 0.64%.

A Wilcoxon signed ranks test is also used to compare the mean standard deviation of

the pricing error for these “multi-trading” stocks to determine whether there is significant

difference in Hasbrouck’s (1993) measure between the two markets. The test statistic is

highly significant at the 1% level with a value of –5.417, strongly supporting our

hypothesis that the average standard deviation for the pricing error for “multi-trading”

stocks on NSE is significantly lower for identical set of stocks traded on BSE. Since the

standard deviation of the pricing errors is used as a measure for transaction costs, all our

results based on Hasbrouck’s (1993) approach suggest that the average trading costs on

NSE are significantly lower than BSE.

From the list of forty, only two stocks (or 5% of the total) register higher values for

Hasbrouck’s (1993) measure on NSE with the remaining 38 stocks (95%) exhibiting

lower values on NSE; no stocks are found with equal values for Hasbrouck’s measure.

The standard deviation of the pricing errors are indeed smaller on NSE for most stocks,

providing further confirmation that the trading cost on NSE is lower than BSE even at the

individual stock level.

5.2 Daily price covariance using Roll’s (1984) approach

We conduct robustness check using Roll’s (1984) approach. We compute two

types of effective bid-ask spreads using the daily price data: a rupee-spread and a

percentage-spread. The rupee effective bid-ask spread enables us to compare the bid-ask

23

spreads of paired sample stocks on both markets. As the rupee-spread is positively related

to the specific stock’s price, it is therefore, unsuitable for comparison across different

stocks. This measure is primarily used to compare identical stocks traded on two different

markets. Percentage bid-ask spreads, on the other hand, which use the stock price level

as the denominator, allows spreads of stocks, which vary in the level of stock price level,

to be compared.

Table 3 presents the rupee-based and percentage-based covariances of the daily

prices for 39 out of the 40 multi-trading stocks10. Roll (1984) states that the covariance of

stock prices in an efficient market are expected to be negative since market makers adjust

their quotes around the efficient price quickly. We find, however, that most of the

covariances computed for the 39 stocks are, in fact, positive; for rupee-based covariances,

a total of 10 (or 12.8%) are negative; only 7 (or 8.9%) are negative for percentage-based

covariances.

Our empirical result is, however, not surprising and is consistent with Roll (1984) and

Harris (1990), who also obtain positive covariances using daily prices. Roll (1984) and

Harris (1990) argue that since the stock market is far from being fully efficient, the speed

of price adjustment tends to be slower, and usually takes between several hours to weeks

to adjust to the right level. This consequently results in positive correlation in daily price

rather than the theoretical proposition of negative serial correlation. Amihud and

Mendelson (1987) and Hasbrouck and Ho (1987) also obtain similar results. Moreover,

Harris (1990) finds that when longer time intervals are used (for example, weekly prices),

the number of stocks having positive covariances tend to fall as more negative

correlations are expected, given that the price adjustment will take place. Based on these

10 Data pertaining to Ponds India Limited was not readily available.

24

factors, we decided to abandon spread estimation using Roll’s methodology with daily.

We estimate spreads using weekly data based on the findings of Harris (1990).

5.3 Estimation of effective spreads for weekly prices using Roll’s (1984) approach

Effective bid-ask spreads computed using the rupee-based covariance of weekly

transaction prices are given in Table 4. Unlike the results obtained using daily data, 32

stocks on NSE and 30 stocks on BSE exhibit negative covariances when weekly prices

are used. Over weekly intervals, the Indian stock price prices tend to move closer to their

efficient price levels.

Based on the assumption of an informationally efficient market, effective spreads that

are inferred from any observation interval should be equal. Our results using daily and

weekly prices differ. We find that covariances estimated from the daily prices exhibit

more positive values than those estimated from weekly prices. Roll (1984) attributes this

deviation from the theory to informational inefficiency of the respective market. For our

study, this clearly indicates that both NSE and BSE are not perfectly information-

efficient markets.

Table 4 also presents the rupee-based bid-ask spread, which we use as a measure of

transaction costs for the multi-trading stocks on NSE and BSE. Out of 39 stocks, we have

a total of 29 usable pairs of estimations (accounting for 74% of the total). These are

indicated in bold. Of these 29 pairs, 20 stocks have lower bid-ask spreads on NSE than

BSE. As the bid-ask spread is regarded as a proxy for transaction costs, this suggests that

transaction costs on NSE are lower than BSE.

Use of the rupee-based bid-ask spreads is inappropriate in a comparison of trading

costs across different stocks as these spreads are directly price related. A better measure

of trading costs in this case is the percentage-based bid-ask spread, results of which are

presented in Table 5. A total of 29 stocks (or 74.4%) on NSE and 30 stocks (or 76.9%) on

25

BSE exhibit negative covariances. The ratio of the bid-ask spread for stocks on NSE to

the bid-ask spread for the identical stocks on BSE are shown in the last column. Of the

29 usable pairs, 20 stocks have ratios less than unity indicating that spreads are lower in

NSE than BSE. The mean effective bid-ask spread is 3.469 and 3.916 percent of the stock

price on NSE and BSE, respectively, while the mean bid-ask spread ratio is 0.8925,

suggesting the bid-ask spread on NSE is about 11% lower as compared to BSE. This

difference in the effective spreads is statistically significant11, implying that transaction

costs on BSE are, on average, significantly higher than NSE. Also, 69% of the stocks are

found to have a higher bid-ask spread on BSE.

We have thus shown empirically that NSE is a better quality market than BSE, a

finding that is consistent with the Domowitz and Steil (1999) proposition. Can we

attribute the better quality governance in NSE as compared to BSE for the observed

differences in the quality of markets? Outlined below are possible structural explanations

that explain our findings.

5.4 Why is the quality of the market better on NSE?

We find substantial empirical evidence in our paper that the transaction costs on NSE

is lower than BSE. Important differences in market structures between the two exchanges

may possibly account for the lower cost structure on NSE, which is discussed below.

BSE commenced operations in 1875 and is the oldest stock market in India, which for

many years, has been the major Indian stock market. Unlike NSE, BSE was beset with

problems related to an outdated trading and settlement mechanism, lack of transparency

in transactions, delays in physical delivery and registration of transfer, poor disclosure,

price manipulation, insider trading, poor liquidity and excessive speculation. NSE, was

11 The Wilcoxon test statistics of -2.693 and standard t-test statistic of -2.88 in both markets are statistically significant.

26

established in 1994 by policy makers with the intent of improving the trading systems,

practices and procedures in Indian stock markets. Market reforms were carried out using

a combination of new technology, new processes and enforcement of new regulations,

with NSE pioneering the introduction of a national network of computerized trading, a

clearinghouse, a completely automated screen based trading system, and special facilities

for institutional trading in India. NSE has transformed and improved the Indian stock

markets by increasing market transparency, providing more efficient settlement and

delivery, and implementing collateralization based on risk of positions undertaken.

In Table 6, we summarize the main structural differences between the two exchanges

that enable NSE to operate under an umbrella of lower transaction costs, and which has

allowed NSE to dominate BSE. The factors shown in italics are related to

ownership/governance. Following Domowitz and Steil (1999), we argue that a

demutualized ownership structure has enabled NSE to score higher on these governance

variables: use of technology, internal control systems, transparency, regulation and

investor protection. Small and medium investors would be attracted to NSE due to the

perceived better quality of governance and protection. In addition, the geographical

reach, which is not a governance factor, is likely to augment trading activity on NSE, and

therefore have a beneficial effect on execution costs.

While some of these factors such as technology use and investor protection serve to

enhance the market quality by attracting liquidity traders, it is not clear whether having a

more transparent market with better internal control systems, transparency and regulation

enhance the quality of market as measured by execution costs (transaction costs). A

market with less regulations and weak internal control systems would be very attractive

to traders who seek to exploit inside information. Such a market would also be attractive

to institutional traders if the exchange is negligent in enforcing regulations. In fact BSE

27

would be more attractive to insiders and institutional traders on account of the perceived

laxity of that exchange. Such a market can theoretically provide lower execution costs if

some traders place limit orders of large sizes.

Based on the comparison of the two exchanges, we expect NSE to have a higher

degree of trading activity (number of trades) and BSE to have a larger trade size. Both

these factors are expected to result to result in an improvement in market quality via a

reduction in execution costs. Some of the governance factors for which we have no

proxies: technology use and investor protection, may have an influence on market

quality. To disaggregate the source of market quality differences we conduct a

multivariate study as explained below.

5.4.1 Multivariate analysis using Hasbrouck and Schwartz’s (1988) method

Hasbrouck and Schwartz (1988) compare execution costs across different markets by

using a multivariate approach. We use a variation of their model to examine the effect of

governance and other factors on trading costs. We use a multivariate regression model to

control for firm-specific factors that have a bearing on execution costs. Our explanatory

variables include the average price (AP), the average size per trade (AST), the average

number of trades per day (ATD), and the market value (MV) for a particular stock over

the sample period. The AST variable is used to account for the occurrence of block

trades of a particular stock while the ATD is used as a proxy for the liquidity of the stock

on a particular exchange. We argued in the previous section that ATD would be higher

on NSE as compared to BSE while the AST of BSE is expected to be larger. Price and

market capitalization are used as control variables following the work of Hasbrouck and

Schwartz (1988).

We regress transaction costs against a list of variables, which include the stocks’

price, liquidity, market value, according to the following equation:

28

iiiiiii eDaMVaATDaASTaAPaaC ++++++= 543210 , (15)

where iC represents the transaction costs for stock i , defined as the standard deviation of

the pricing errors; iAP , is the average price for stock i ; iAST , is the average size per

trade for stock i ; iATD , is the average number of trades per day used as the proxy for

liquidity; iMV , is the market capitalization for stock i ; and iD , is a dummy variable to

distinguish between the two markets, set equal to one for NSE stocks and zero for BSE

stocks. The dummy variable is included to capture potential governance factors that

distinguish the two markets.

Table 7 provides descriptive statistics pertaining to AP, AST and ATD on both

exchanges. The average stock price of 546.46 rupees is higher on BSE as compared to

467.11 rupees on NSE. Stocks traded on NSE have a smaller mean AST of 259 shares as

compared to 363 shares on BSE, suggesting larger trades on BSE as compared to NSE.

This is mainly due to the characteristics of BSE associated with frequent insider trading

and institutional block trading. Wilcoxon and t-test statistics for the difference in AST

between these two markets are significant at the 1% level.

The mean value of 2299 daily trades on NSE is larger than the 1196 recorded for

BSE and the difference is statistically significant using both the Wilcoxon and the

standard t-test. This indicates that the trading activity on NSE is higher compared to BSE.

As the average trade per day is often used as a proxy for the liquidity of stocks, this is

indicative that NSE is relatively more liquid than BSE. The increased trading activity is

perhaps a natural consequence of the nation-wide network established by NSE as

compared to a city-wide network employed by BSE.

29

A total of fifteen regressions are estimated based on all possible combinations of

the independent variables in our study12. Selected results of these regressions are

presented in the Table 8. Values of the adjusted R-square range from 0.278 to 0.403.

The dummy variable, iD , is negative and statistically significant in all regressions,

suggesting that transaction costs on NSE are lower than BSE even after controlling for

trading activity, trade size and the control variables.

Estimates of the average price (AP) are positively related to transaction costs in

all regressions and significant in regressions 4 and 5. The statistical significance wanes as

other trading activity and market value variables are included.13 A negative relation is

observed between the average size per trade (AST) and transaction costs, implying that

larger trade size lowers transaction costs. This negative relation is significant in

regressions 3 and 6. The number of trades per day (ATD) is significant and negatively

related to transaction costs in regressions 4 and 7. The significance drops when the AST

variable is included (correlation of 0.23). As expected, stocks which trade actively tend

to be associated with lower transaction costs. Also, since the ATD is a proxy for

liquidity, it suggests that higher liquidity does, in fact, result in lower transaction costs.

The market capitalization (MV) is negatively related to transaction costs indicating that

stocks with larger market value generally exhibit lower transaction costs, a result

consistent with that obtained in Roll (1984) and Hasbrouck (1993). In all cases, the

dummy variable is negative and statistically significant. We cautiously interpret this

finding to indicate that the residual impact of superior governance in the NSE results in a

lower transaction cost as compared to BSE. Interestingly, BSE benefits from its ability to

12 This procedure is valid when the number of regressors is small since the total number of regressions is a function of [2exp(k) – 1], where k represents the number of regressors (see Maddala, 1977). 13 The correlation between AP and MV is 0.39, while AP has a negative correlation of –0.28 with the AST variable.

30

attract insiders/institutional traders as evidenced by the negative impact of the AST

variable. Our descriptive statistics showed that BSE has significantly higher trade size as

compared to NSE.

Our results are consistent with Krishnamurti and Lim (1999), who similarly find that

NSE, with lower execution costs, lower price volatility, and higher liquidity, is a better-

quality market than BSE. It is, therefore, not surprising to observe that NSE has taken

over much of the transaction volume from BSE since its inception. This is clearly

demonstrated in Figure 2 which tracks the trading volume of both markets over the period

April 1996 until January 1998; and which shows NSE having a consistently higher

trading volume as compared to BSE since its inception.

6. Conclusion

The primary role of a stock exchange is to provide trading services to its ultimate

clients. Prior studies have concentrated on comparing the quality of markets with

different trading systems. Little attention has been paid to the critical issue of the quality

of markets and stock exchange governance. The reason for this neglect is due to the fact

that until recently, most major exchanges have been organized as mutual associations. A

recent trend has been in the demutualization of stock exchanges, ostensibly, to better deal

with emerging competition from electronic trading networks. In a recent paper,

Domowitz and Steil (1999) provide convincing arguments in favor of demutualized

exchanges. They reason that members of a mutualized exchange have incentives to

oppose innovation that increase the quality of its services, and hence the value of the

exchange, while stakeholders of a demutualized exchange, are likely to favor any

measure that enhances the value of the firm.

31

In our paper, we examine the Domowitz and Steil proposition empirically using data

from a big emerging market, namely, India, which has two major exchanges with

distinguishably different organizational structures. The National Stock Exchange adopts a

demutualized form of organization unlike the Bombay Stock Exchange which continues

to pursue the mutual structure. Consistent with the Domowitz and Steil proposition, we

find that National Stock Exchange has a better quality market as compared to Bombay

Stock Exchange based on the market quality measure derived by Hasbrouck (1993). Our

basic conclusions are robust to an alternate measure of market quality. Multivariate

regression results indicate that the superior governance of NSE is at least partially

responsible for its better market quality.

Our study has implications for emerging markets where regulators do not have

sufficient experience or clout in dealing with recalcitrant monopolists. We have shown

that competition has been more effective than regulatory dictates in transforming the

errant behaviour of the members on the Bombay Stock Exchange; findings that surely

have important policy implications for other emerging nations. There exists evidence that

economic incentives are likely to be more potent in transforming the behavior of rent

seekers rather than directives from policy regulators.

The role played by technology in transforming Indian stock markets is

particularly evident in our findings. Applying the latest technology, the National Stock

Exchange was able to penetrate the barriers to entry in the stock-broking industry, which

had the consequent benefit of reducing the cost of providing these services. Being able to

pass on the lower trading costs to the investors, the National Stock Exchange has posed a

serious threat to the very existence of the oldest exchange in India, namely the Bombay

Stock Exchange. That technology is responsible for breaking barriers to entry and for

dramatically improving in the quality of markets in Indian stock exchanges, is

32

unequivocal; the Bombay Stock Exchange must re-examine its governance structure if it

wishes to remain relevant in a global environment where only the best will survive.

33

REFERENCES

Affleck-Graves, J., Hegde, S. P., and Miller, R. E., 1994, Trading mechanisms and the

components of the bid-ask spread, Journal of Finance 49, 1471-1488. Amihud, Y., and Mendelson, H., 1987, Trading mechanisms and stock returns: An

empirical investigation, Journal of Finance 42, 533-553. Barclay, M. J., 1997, Bid-ask spreads and the avoidance of odd-eighth quotes on

NASDAQ: An examination of exchange listings, Journal of Financial Economics 45, 35-60.

Barclay, M. J., Christie, W. G., Harris, J. H., Kandel, E., and Schultz, P. H., 1999, Effects

of market reform on the trading costs and depths of NASDAQ stocks, Journal of Finance 54, 1-34.

Beebower, G., 1989, Evaluating transaction costs, in The Complete Guide to Securities

Transactions: Enhancing Investment Performance and Controlling Cost, W.H. Wagner (ed.), Wiley, New York.

Benston, G. J., and Hagerman, R. L., 1974, Determinants of bid-asked spreads in over-

the-counter market, Journal of Financial Economics 1, 353-364. Berkowitz, S. A., Logue, D. E., and Noser Jr. E. A., 1988, The total cost of transactions

on the NYSE, Journal of Finance 43, 97-112. Bessembinder, H., 1997, The degree of price resolution and equity trading costs, Journal

of Financial Economics 45, 9-34. Bessembinder, H., and 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.

Beveridge, S., and Nelson, C. R., 1981, A new approach to decomposition of economic

time series into permanent and transitory components with particular attention to measurement of the ‘Business cycle’, Journal of Monetary Economics 7, 151-174.

Branch, B., and Freed, W., 1977, Bid-ask spreads on the AMEX and the bid board,

Journal of Finance 32, 159-163. Blume, M., and Goldstein, M., 1992, Displayed and effective spreads by market,

Working Paper, Wharton Business School, University of Pennsylvania. Chan, L. K.C., and Lakonishok, J., 1993, Institutional trades and intraday stock price

behavior, Journal of Financial Economics 33, 173-199. Christie, W. G., and Huang, R. D., 1994, Market structures and liquidity: A transactions

data study of exchange listings, Journal of Financial Intermediation 3, 300-326.

34

Christie, W. G., and Schultz, P., 1994, Why do NASDAQ market makers avoid odd-

eighth quotes?, Journal of Finance 49, 1813-1840. Demsetz, H., 1968, The cost of transacting, Quarterly Journal of Economics 82, 32-53. Domowitz, I., and Steil, B., 1999, Automation, Trading Costs, and the Structure of the

Trading Services Industry, Brookings-Wharton Papers on Financial Services. Glosten, L., 1987, Components of bid-ask spread and the statistical properties of

transaction prices, Journal of Finance 42, 1293-1307. Gompers, P.A., Ishii, J.L., and Metrick, A., 2001, Corporate Governance and Equity

Prices, Working Paper, Wharton Business School, University of Pennsylvania. Harris, L., 1990, Statistical properties of the Roll serial covariance bid/ask spread

estimator, Journal of Finance 45, 579-590. Hasbrouck, J., 1991b, The summary informativeness of stock trades: An econometric

analysis, Review of Financial Studies 4, 571-595. Hasbrouck, J., 1993, Assessing the quality of a security market: A new approach to

transaction-cost measurement, Review of Financial Studies 6, 191-212. Hasbrouck, J., and Ho, T, 1987, Order arrival, quote behavior and the return generating

process, Journal of Finance 42, 1035-1048. Hasbrouck, J., and Schwartz, R. A., 1988, Liquidity and execution costs in equity

markets, Journal of Portfolio Management 14, 10-16. Holthausen, R. W., Leftwich, R. W., and Mayers, D., 1987, The effect of large block

transactions on security prices: A cross-sectional analysis, Journal of Financial Economics 19, 237-267.

Huang, R. D., and Stoll, H. R., 1996a, Competitive trading of NYSE listed stocks:

Measurement and interpretation of trading costs, Financial Markets, Institutions and Instruments 5, No. 2, 1-55.

Huang, R. D., and Stoll, H. R., 1996b, Dealer versus auction markets: A paired

comparison of execution costs on NASDAQ and the NYSE, Journal of Financial Economics 41, 313-357.

Krishnamurti, C., and Lim, E., 1999 Competition, liquidity and volatility: A comparative study of Bombay Stock Exchange and National Stock Exchange, working paper, Nanyang Technological University, Singapore.

Lee, C. M., and Ready, M. J., 1991, Inferring trade direction from intraday data, Journal

of Finance 46, 733-746.

35

Maddala, G.S., 1977, Econometrics, McGraw-Hill Book Company. Parisi, F., Nail, L., and Soto, C., 2001, Evidence of a Leadership Role in the Chilean

Stock Exchanges, University of Chile Working Paper. Petersen, M. A., and Fialkowski, D., 1994, Posted versus effective spreads: Good prices

or bad quotes?, Journal of Financial Economics 35, 269-292. Roll, R., 1984, A simple implicit measure of the effective bid-ask spread in an efficient

market, Journal of Finance 39, 1127-1139. Stoll, H. R., 1985, The stock exchange specialist system: An economic analysis,

Monograph Series in Finance and Economics, Salomon Brothers Center, New York University.

Stoll, H. R., 1989, Inferring the components of the bid-ask spread theory and empirical

tests, Journal of Finance 44, 115-134.

36

Appendix: Hasbrouck’s (1993) approach

According to Hasbrouck (1993), the observed transaction price at time t, tP , may

be decomposed into two components as follows:

ttt SMP += , (1)

where tM is the efficient price, representing the value of a certain stock that includes all

public information available at time t and all the private information that may be inferred

from previous stock transactions and tS , the pricing error, representing the deviation of

the actual transaction price from this efficient price.14

Two important assumptions that Hasbrouck (1993) makes are:

(1) The efficient price, tM , follows a random walk:

ttt WMM += −1 , (2)

where tW are uncorrelated increments, 0)( =tWE , 22 )( wtWE σ= , and 0)( =τWWE t for

τ≠t ;

(2) The pricing error, tS , is a zero-mean covariance-stationary stochastic process.

tS , the pricing error, is central to Hasbrouck’s (1993) approach capturing various

microstructure effects such as price discreteness, inventory control, the non-information

based component of the bid-ask spread (order processing costs) and the transient

component of the price response to a block trade that cannot be explicitly modeled. From

a transaction cost perspective, the pricing error, tS , is defined as the cost to the buyer

while – tS , is the respective cost to the seller, maintaining the zero-sum principle as

discussed earlier.

14 Here t is assumed to index stock transactions.

37

To assess the market design and regulation, the standard deviation of the pricing

error, sσ , is proposed as a summary measure of market quality that measures how closely

the transaction price tracks the efficient price. The role of sσ as a proxy for market

quality depends on the assumption that as transaction costs and other trading barriers are

reduced, actual transaction prices should track the efficient prices more closely.

Derivation of sσ requires the estimate of parameters from a vector autoregressive

(VAR) model that explicitly captures the dynamic structure of stock transactions15. In the

bivariate VAR model that we have adopted below, current trades and transaction prices

are dependent on information derived from the previous transactions:

,,222112211

,122112211

tttttt

tttttt

vXdXdRcRcXvXbXbRaRaR

++++++=++++++=

−−−−

−−−−

LL

LL (3)

where tX represents the trade variable, in this case, the trade volume at time t, which is

signed to be positive if the agent who initiates the trade is buying, and negative if the

agent is selling; tt vv ,2,1 , represent trade innovations, which are zero-mean, serially

uncorrelated disturbances; and tR is the standard returns variable computed by taking the

first difference of the natural logarithm of observed transaction prices, that is,

1−−= ttt PPR .

Hasbrouck then transforms the VAR model in equation (3) to a vector moving

average (VMA) model to express the exogenous variables in terms of current and lagged

disturbances as follows:

LL

LL

+++++++=

+++++++=

−−−−

−−−−

2,2*21,2

*1,2

*02,2

*21,1

*1,1

*0

2,2*21,2

*1,2

*02,1

*21,1

*1,1

*0

ttttttt

ttttttt

vdvdvdvcvcvcX

vbvbvbvavavaR. (4)

15 One major advantage of a VAR model is that it is a powerful framework that is general enough to capture unspecified lagged dependencies. From a market microstructure perspective, this is essential, as market imperfections result in lagged effects that include inventory control mechanisms, lagged price adjustment, and price discreteness.

38

From equations (1) and (2), the pricing error is given by ttt WS ηα += , where tη is a

disturbance term that is uncorrelated with tW . Accordingly, it follows that:

11 −− −+=−= tttttt SSWPPR . (5)

Since tW and tS are serially uncorrelated, tR possesses nonzero autocovariances at the

first lag only. This implies that tR may be represented as a first-order moving average

process, namely:

1−−= ttt aR εε . (6)

Two parameters, namely, { 2, εσa }, fully characterize the mean and autocovariances of the

return process. Applying the identification restriction in Beveridge and Nelson (1981)

refereed to as the BN restriction that tη =0, and using equations (5) and (6), we obtain:

.11 −− −+=− ttttt WWW ααεε (7)

Equation (7) implies that aa −= 1α , tt aW ε)1( −= , tt WS α= , 222 )1( εσσ aw −= , and

222εσσ as = . When transaction prices are available, the pricing error for a particular

transaction may be estimated using equation (8) as follows:

)(),( 22

11, LL +++−=−= −−−

∧

ttttttt RaaRRaaRRS ε (8)

where ∧

tS represents the estimated pricing error. As tR in equation (4) is used to

determine the value of the pricing error, it follows that:

,111,21,201,11,10 LLL ++++++++= −−− ttttttt vvvvS ηγηββαα (9)

where, tη is a disturbance term that is orthogonal to all components of tv . Thus, using

the BN restriction, the value of the pricing error may be computed directly if estimates of

the 'α s and 'β s in equation (9) are available. Use of the BN restriction implies that the

pricing error is information correlated [see Glosten (1987)].

39

Using Hasbrouck’s (1991b) approach and the BN restriction, values of 'α s and

'β s are computed as follows:

∑∞

+=

−=1

* ,jk

kj aα ∑∞

+=

−=1

*

jkkj bβ , (10)

where the *ka ’s and *

kb ’s are estimates obtained from equation (4). Accordingly, the

variance of the pricing error, 2sσ , is given by:

[ ]

=∑

∞

= j

jjj

js vCov

βα

βασ )(0

2 , (11)

which is an identification-invariant best linear estimate of tS , conditional on current and

lagged tv . The standard deviation of the pricing error is obtained by taking the square

root of the variance. This measure tracks how closely the transaction price is to the

efficient price and allows a meaningful measure of transaction costs of stocks to be

obtained.

40

Table 1 List of the 40 paired sample stocks traded on both NSE and BSE

Company Name Asea Brown Boveri Limited Associated Cement Andhra Valley Power Supply Co. limited Arvind Mills Limited Ashok Leyland Limited Asian Paint India Limited Bharat Heavy Electricals Limited BSES Castrol India Limited Cochin Refineries Limited Colgate Palmolive India Limited East India Hotels Limited Grasim Industries Limited Gujarat Ambuja CMT Housing Development Finance Limited Hindustan Lever Limited Hindustan Petroleum Industrial Credit and Investment Corporation of India Limited Industrial Finance Corporation of India Limited Indian Hotel Company Limited Indo Gulf Fertilizers and Chemicals Corp. Limited Indian Rayon and Industries Limited ITC Larsen & Toubro Mahindra & Mahindra Limited Mangalore Refinery and Petrochemicals Limited Mahanagar Tel Nigam Limited Nestle India Limited Oriental Bank of Commerce Ponds India limited Ranbaxy Laboratories Limited Reliance Industries Limited State Bank of India Tata Chemicals Limited Tata Power Co. Tata Tea Tata Engr. & Loco. Thermax Limited Tata iron & Steel Co. TVS-Suzuki

41

Table 2 Hasbrouck’s (1993) measure for Paired Sample Stocks in NSE and BSE Stocks

BSE(%)

NSE(%)

Ratio

(BSE/NSE)

Andhra Valley Pwr. 0.4370 0.1716 0.3927 Arvind Mills 0.5458 0.1248 0.2287 Asea Brown Boveri 1.3727 0.5550 0.4043 Ashok Leyland 0.9676 0.1819 0.1880 Asian Paints 0.9762 0.3578 0.3665 Associated Cement 0.8645 0.4310 0.4986 Bharat Heavy Els. 0.3483 0.2159 0.6199 Bses 0.3980 0.0972 0.2442 Castrol India 0.6287 0.2379 0.3784 Cochin Refineries 0.2894 0.1852 0.6399 Colgate-Palmolive 0.5111 0.1529 0.2992 Eih 1.4901 0.6690 0.4490 Grasim Inds. 1.3783 0.3778 0.2741 Gujarat Ambuja Cmt 0.4223 0.1429 0.3384 Hdfc Bank 0.8448 0.8639 1.0226 Hindustan Lever 0.9546 0.4183 0.4382 Hindustan Ptl Corp. 0.4831 0.3289 0.6808 Icici 0.3896 0.0750 0.1925 I F C I Ltd. 0.0934 0.0152 0.1627 Indian Hotels Co. 1.1899 0.7472 0.6280 Indian Rayon&Inds. 1.0697 0.3229 0.3019 Indogulf Fert. 0.1174 0.0771 0.6567 Itc 0.4136 0.1086 0.2628 Larsen & Toubro 0.3786 0.0904 0.2388 Mahanagar Tel Nigam 0.2705 0.1101 0.4070 Mahindra & Mahindra 1.0382 0.3837 0.3699 Mrpl 0.0955 0.0350 0.3665 Nestle India 0.8387 0.2679 0.3194 Oriental Bk.Of Com. 0.1226 0.0513 0.4184 Ponds 0.7593 0.8611 1.1341 Ranbaxy Labs. 1.1797 0.4189 0.3551 Tata Chemicals 0.6699 0.1664 0.2484 Tata Engr.& Loco. 0.4220 0.1184 0.2806 Tata Iron & Steel 0.4020 0.0625 0.1555 Tata Power Co. 0.2988 0.0792 0.2651 Tata Tea Co. 0.5675 0.1666 0.2936 Thermax 0.5792 0.3842 0.6633 Tvs-Suzuki 0.9522 0.6084 0.6389 Reliance Inds. 0.4919 0.0751 0.1527 State Bank Of Ida. 0.3804 0.0617 0.1622 Mean 0.6400 0.2700 0.4034

42

Table 3

Covariance Using Daily Price Data )1(

Rupee-based cov )2( Percentage-based cov )3( Stocks NSE BSE NSE (%) BSE(%) Andraval -0.0172* 0.2865 -0.00057* 0.00661 Arvindmi 0.1702 0.2158 0.00294 0.00868 Abb 15.744 14.114 0.00854 0.00674 Ashokley 0.0067 1.5248 0.00375 0.02960 Asianpa 0.9659 2.4130 0.00114 0.00327 Acc 1.0723 1.3080 0.00153 0.00878 Bhel 3.8321 8.3449 0.00385 0.00791 Bses 4.5356 5.2174 0.00546 0.00720 Castrol 0.9956 2.0589 0.00157 0.00265 Cochinre 1.5430 -0.0748* 0.00512 0.00182 Colgate 1.9709 2.6615 0.00143 0.00265 Eihotel 1.9523 2.5452 0.00091 0.00234 Grasim 13.484 14.265 0.01580 0.01860 Gujambce 0.4143 -1.6079* 0.00140 0.00166 Hdfc 0.7911 0.7598 0.00438 0.00611 Hindleve 217.00 226.09 0.00338 0.00481 Hindpetr 2.3589 2.5068 0.00830 0.00836 Icici 0.9455 0.0659 0.00764 0.01270 Ifci 2.7988 -0.0097* -0.00153* -0.00161* Indhotel 14.794 17.377 0.00945 0.00962 Indrayon 6.1814 3.0869 0.02680 0.02300 Indogulf 0.0551 0.2910 0.00402 0.01280 Itc -11.410* -6.5783* 0.00059 0.00162 L&T 1.8837 1.3965 -0.00038* -0.00051* Mtnl 3.4143 3.4463 0.00488 0.00500 Mm 18.382 15.232 0.01490 0.01240 Mrpl 0.0363 0.0110 0.00952 0.00504 Nestle -2.9822* -3.9896* 0.00064 0.00028 Orientba 0.1831 0.6784 0.00631 0.00533 Ranbaxy 44.801 44.691 0.01120 0.00915 Tatachem 0.7023 0.6313 0.00381 0.00435 Telco 0.9063 4.4395 0.00611 0.01100 Tisco 1.0041 0.9195 0.00482 0.00456 Tatapow -0.1244* 0.3158 -0.00091* 0.00225 Tatatea 10.411 8.4523 0.00671 0.00716 Thermax 1.6432 -0.0003* 0.00252 -0.00072* Tvssuzuk 13.048 15.666 0.00343 0.00406 Reliance 0.5314 1.3080 0.00000 0.00009 Sbin 0.7088 2.3413 0.00004 0.00006 Notes: (1) One stock (Ponds India Limited) is not recorded in DataStream, resulting in a sample of 39- paired multi-trading stocks are studied by this method; (2) Covariance of )( 1−− tt pp ; covariance of

)( 1−− tt pp /1−tp ; * Negative value of price covariance.

43

Table 4 Rupee Covariance and Effective Bid-Ask Spreads using weekly data

NSE

BSE

Stock

Covariance

b-a spread

Covariance

b-a spread

NSE/BSE(1)

Andraval -2.7499 3.3166 -3.0020 3.4652 0.9571 Arvindmi -3.2711 3.6172 -4.3000 4.1584 0.8699 Abb 91.281* -19.108 )2( 50.121* -14.202 1.3455 Ashokley -0.0817 0.5734 -0.1650 0.8124 0.7058 Asianpa -29.744 9.5754 -38.350 12.423 0.7708 Acc -1.0241 2.0308 -4.0930 4.0586 0.5004 Bhel 40.575* -12.778 44.068* -13.317 0.9595 Bses -47.178 13.778 -41.326 12.897 1.0684 Castrol -48.230 13.931 -57.328 15.188 0.9172 Cochinre 0.9826* -1.9890 -3.7942 3.9074 -0.5090 Colgate -30.863 11.144 -36.931 12.191 0.9142 Eihotel -4.5132 4.2615 9.9939* -6.3416 0.6720 Grasim -15.243 7.8317 -19.312 8.8152 0.8884 Gujambce -3.1384 4.0949 -3.4541 3.7283 1.0983 Hdfc 11.252* -6.7288 8.2844* -5.7737 1.1654 Hindleve -2885.7 107.76 -2918.1 108.36 0.9944 Hindpetr -24.052 9.8379 -28.913 10.787 0.9121 Icici -2.1154 2.9175 -3.6010 3.8068 0.7664 Ifci 0.0105* -0.2059 15.5971* -7.9332 0.0260 Indohotel -82.629 18.234 -71.494 16.961 1.0750 Indrayon -35.608 -11.970 -41.362 12.901 -0.9278 Indogulf -0.7107 1.6912 -0.6120 1.5697 1.0774 Itc -165.75 25.826 -167.58 25.968 0.9945 L&T 25.422* -10.114 29.953* -10.984 0.9208 Mtnl -5.0707 4.5171 -3.6333 3.8236 1.1813 Mn -64.203 16.073 -64.816 16.150 0.9953 Mrpl -0.1781 0.8462 -0.2621 1.0276 0.8235 Nestle -15.839 7.9835 -30.258 11.034 0.7235 Orientba -1.9511 2.8021 -1.8365 2.7181 1.0309 Ranbaxy -184.47 27.245 -155.30 24.998 1.0899 Tatachem -13.447 7.3561 -8.1446 5.7247 1.2850 Telco -6.4241 5.0845 -17.058 8.2849 0.6137 Tisco -10.740 6.5736 -15.453 7.8857 0.8336 Tatapow -4.3230 4.1800 -7.1000 5.3453 0.7820 Tatatea -35.607 11.970 -35.591 11.967 1.0002 Thermax 15.067* -7.7864 35.691* -12.632 0.6164 Tvssuzuk -1.3650 2.3435 6.1205* -4.9627 -0.4722 Reliance -6.6987 5.7163 4.5519* -4.2670 -1.3397 Sbin -18.269 8.5485 -24.851 9.9702 0.8574 Notes: (1) the bid-ask spread in NSE/ the bid-ask spread in BSE; (2) The variable used are

jj Covs −= 2 , the estimated bid-ask spread, where jCov is the serial-covariance of returns on stock j

during observation period.; the sign of the covariance was preserved after taking the square root. (Roll, 1984); and * denotes entries with positive covariance of which there are 7 such items in the NSE and on BSE.

44

Table 5 Percentage Covariance and Bid-Ask Spreads using weekly data

NSE

BSE

Stock

Covariance

b-a spread

Covariance

b-a spread

NSE/BSE(1)

Andraval -0.00014 2.391 -0.00023 3.033 0.78833 Arvindmi -0.00043 4.138 -0.00089 5.963 0.69388 Abb 0.00041* -2.615 0.00017* -4.045 0.64660 Ashokley 0.00026* -3.237 0.00035* -3.731 0.86760 Asianpa -0.00019 2.757 -0.00022 2.933 0.93999 Acc 0.00000* -0.167 -0.00040 3.975 0.04206 Bhel 0.00060* -4.899 0.00070* -5.307 0.92312 Bses -0.00083 5.755 -0.00079 5.632 1.02184 Castrol -0.00059 4.866 -0.00049 4.432 1.09792 Cochinre -0.00007 1.683 -0.00016 2.553 0.65922 Colgate -0.00033 3.617 -0.00038 3.883 0.93150 Eihotel -0.00000 0.238 -0.00023 3.046 0.07797 Grasim 0.00037* -3.857 0.00045* -4.238 0.91010 Gujambce -0.00028 3.353 -0.00025 3.188 1.05192 Hdfc 0.00011* -2.117 0.00004* -1.318 1.60622 Hindleve -0.00066 5.146 -0.00070 5.299 0.97113 Hindpetr -0.00075 5.488 -0.00091 6.023 0.91117 Icici -0.00006 1.561 -0.00006 1.521 1.02630 Ifci -0.00023 3.003 -0.00029 3.429 0.87577 Indohotel -0.00041 4.035 -0.00032 3.589 1.12427 Indrayon -0.00073 5.430 -0.00108 6.582 0.82498 Indogulf -0.00030 3.458 -0.00064 5.502 0.62850 Itc -0.00066 5.134 -0.00067 5.188 0.98959 L&T 0.00006* -1.589 0.00019* 2.750 0.57782 Mtnl -0.00004 1.223 -0.00004 1.266 0.96604 MM -0.00006 1.521 -0.00004 1.316 1.15578 Mrpl -0.00047 4.336 -0.00095 6.161 0.70378 Nestle -0.00043 4.142 -0.00030 3.476 1.19160 Orientba -0.00032 3.567 -0.00038 3.919 0.91018 Ranbaxy -0.00030 3.481 -0.00032 3.555 0.97918 Tatachem -0.00049 4.436 -0.00037 3.847 1.15311 Telco 0.00021* -2.905 0.00028* 3.365 -0.8633 Tisco -0.00005 1.445 -0.00012 2.154 0.67085 Tatapow -0.00054 4.626 -0.00064 5.044 0.91713 Tatatea -0.00050 4.477 -0.00043 4.152 1.07828 Thermax 0.00022* -2.953 0.00084* 5.786 0.51037 Tvssuzuk 0.00013* -2.263 0.00018* 2.661 0.85043 Reliance -0.00024 3.079 -0.00034 3.693 0.83374 Sbin -0.00012 2.209 -0.00024 3.118 0.70847

Notes: (1) The bid-ask spread in NSE/the bid-ask spread in BSE; (2) the bid-ask spreads are presented as a percentage of the stock prices; (3) The variable used is

jj Covs −= 2 , the estimated bid-asks

spread, where jCov is the serial-covariance of returns on stock j during observation period.; the sign of the covariance was preserved after taking the square root (Roll, 1984); * items with positive covariance for which there are 10 such items in NSE and 9 on BSE

45

Table 6 A Comparison of the Bombay Stock Exchange and the National Stock Exchange

Factors BSE

NSE

Geographic Reach

Mainly limited to Bombay City

Across the Whole of India

Ownership Structure Brokers’ Association Organized as a company

Use of Technology Laggard and follower Pioneer in computerized trading

Internal System Control Insider Trading and market manipulation are rampant

Stickler for market integrity

Transparency Relatively secretive Follows an open policy

Regulation Investor Protection

Resists regulatory change Low standards

Self-regulated; better risk management systems High standards

46

Table 7 Descriptive Statistics of Independent Variables in Multivariate Regressions

BSE NSE Number of sample stocks 40 40 Average Price ( AP) )1( Minimum )2( 20.09 23.46 Maximum )3( 4357.03 3705.17 Mean 546.46 467.11 Average Size per Trade (AST) )4( Minimum 29 22 Maximum 903 545 Mean 363 259 Wilcoxon test )5( -4.396* (0.000) t-test -4.215* (0.000) Average Number of Daily Trades ATD )6( Minimum 12 46 Maximum 7802 19451 Mean 1196 2299 Wilcoxon test )7( 4.241*(0.000) t-test 2.143*(0.038) Notes:

)1( AP is the average price over the sample period. )2( The stock “HDFC Bank” has the highest average price over the sample period on both the

NSE and BSE. )3( The stock “MRPL” has the lowest average price over the sample period on both NSE

and BSE. )4( AST is the average size per trade over the sample period. )5( From the Wilcoxon ranks test for AST, we obtain 7 positive ranks, 1 tie and 32 negative

ranks; t-statistics are based on positive ranks. )6( ATD is the average number of trades per day computed as the total number of trades/total

number of days over the sample period. )7( From the Wilcoxon signed ranks test for ATD, we obtain 36 positive ranks and 4 negative

ranks; t-statistics value are based on positive ranks. * significance at the 5% level (p-values in parenthesis)

47

Table 8 Regressions with Transaction Cost as Dependent Variable

Equation Constant AP AST ATD MV Dummy R2

1 0.800 6.548E-5 -4.573E-4 -1.373E-5 -1.376E-7 -0.404 0.402 (1.650) (-1.947) (-1.149) (-0.218) (-5.493*) 2 0.773 6.995E-5 -4.303E-4 -1.190E-5 -0.397 0.403 (1.810) (-1.917) (-1.220) (-5.592*) 3 0.871 -6.055E-4 -8.965E-6 -0.424 0.377 (-2.946*) (-0.919) (-6.021*) 4 0.608 1.019E-4 -1.934E-5 -0.341 0.373 (2.873*) (-2.125*) (-5.178*) 5 0.585 1.024E-4 -0.363 0.336

(2.823*) (-5.443*)

6 0.886 -6.748E-4 -0.441 0.370 (-3.533*) (-6.501*) 7 0.664 -1.951E-5 -0.349 0.305

(-2.049*) (-5.069*) 8 0.686 -6.742E-7 -0.374 0.278 (-1.085) (-5.188*) Notes: The dummy variable is assigned a value of 1 for all NSE stocks and 0 for BSE stocks; t-statistics are given in parentheses;* denotes significance at the 5 percent level.

48

49

Figure 1 Trading Volume on BSE and NSE

0

5000

10000

15000

20000

25000

Volu

me

(rup

ee, m

illio

ns)

Apr-96

May-96

Jun-96

Jul-9

6Aug-96Sep

-96Oct-

96Nov-9

6Dec

-96Ja

n-97Feb

-97Mar-

97Apr-9

7May

-97Ju

n-97Ju

l-97

Aug-97Sep

-97Oct-

97Nov-9

7Dec

-97Ja

n-98

NSE BSE