A Trans-Niagara Tale Of Informed Traders

8/7/2019 A Trans-Niagara Tale Of Informed Traders

1/50Electronic copy available at: http://ssrn.com/abstract=1361539

A Trans-Niagara Tale ofInformed Traders

This version: March 15, 2009

Abstract

The existing theory of international asset pricing cannot sufficiently explain whythere exist sizable cross-listing premia in the pairs of Canadian shares traded onthe exchanges separated by the Niagara Falls. This research provides a sufficient

condition under which cross-border dominance in private information leads topositive premia. The probability of informed trading (pin) proves to be a pithyand robust measure of asymmetric information priced in stocks fragmented acrossthe border, along the timeline, and beyond the event horizon. Relative to the NewYork Stock Exchange (nyse), the Toronto Stock Exchange (tsx) is heavier withinformed trades and accounts for more in information sharethis is an explicitevidence of informed traders contribution to cross-border price discovery. Theinformational imbalance leads to positive premia, whose parity-convergence isfostered by discretionary liquidity tradersrelating the dynamics of synchronouscross-listing premia to information asymmetry is novel. Lastly, the pin on a tsx-listed share, on average, rises upon cross-listing on the nysethis exacerbatingadverse selection not only explains negative event study returns on the tsx butalso unifies and extends previous claims in the literature.

Keywords: Cross-listing, Probability of Informed Trading, Information Share, ConvergenceSpeed, Bid-ask Spread

jel Classifications: G15, G14, D82



1 Introduction

Canada and the U.S. are the most integrated economies in the world and they share

comparable accounting standards and institutions. Even so, the existing theory of

international asset pricing cannot sufficiently explain why there exist sizable positive

cross-listing premia1 in Canadian stocks afloat on U.S. exchanges. This research pro-

vides a sufficient condition under which cross-border dominance in private information

gives rise to positive premia. The empirical findings herein effectively relate the cause,

the dynamics, and the regime switch of premia to information asymmetry. Overall,

the probability of informed trading (pin) proves to be a robust and pithy apparatus of

addressing how information is priced in stocks fragmented2 across the border, along

the timeline, and beyond the event horizon.

Over the last several decades, firms have raced to be listed on multiple exchanges. Ac-

cording to the World Federation of Exchanges, as of 2005, the global market capital-

ization of cross-listed stocks floated by 2,636 foreign companies amounted to U.S.$5.76

trillion, an increase of 16.3% from 2004. At the same time, in the U.S. alone, almost

2,000 cross-listings3 were recorded. By September 2005, investments in the Ameri-

can Depositary Receipts (adrs) reached U.S.$657 billion, an increase of 36% over the

preceding twelve months.4 The popularity of international cross-listings has prompted

many publications in this area, a majority of which focus on the benefits of cross-border

listings. See Karolyi (2006) for an excellent survey.i

1The relative premia of cross-listed stocks traded on U.S. exchanges against their respective homemarket shares, adjusted by the exchange rate.

2Fragmentation refers to the dispersal of trading in a security to multiple sites.3Including Levels i & ii Depositary Receipts (drs), Level i over-the-Counter (otc) drs, Rule 144a

private placement drs, ordinary shares, and Global Registered Shares (grss). See Bank of New Yorks(2006) The Depositary Receipt Markets.

4Yet, the Sarbanes-Oxley (sox) Act of 2002 has decelerated cross-listings in the U.S. according toDoidge, Karolyi, and Stulz (2007).

1



Cross-listings are a cross-border version of fragmentation. Consequently, the same

questions asked of domestically fragmented stocks can also be directed to cross-listed

shares.5 If a stock lists on an overseas exchange, where does price information originate

and where does price discoverythe dynamic process towards equilibrium pricetake

place? What is the dynamic relationship between the two, or more, prices reflecting

the same fundamental value? Do these virtually identical stocks in distinct markets

carry the same information content?

Hasbrouck (1995) confirms that the New York Stock Exchange (nyse), the flag-

ship exchange in the U.S., dominates other regional exchanges (such as Cincinnati and

Pacific) in terms of its contribution to price discoverythe order purchase agreements

may seek to divert small retail trades to regional locations but leave the larger and po-

tentially more information-based trades to the nyse. When a non-U.S. stock interlists

on the nyse, the host exchange may no longer be the overwhelming source of new

information being impounded to the cross-listed pair. Trades on the non-U.S. home

exchange can be more influential if more information (either private or public) on the

cross-listed share originates from the foreign exchange.

In this paper, I analyze pairs of Canadian cross-listed shares on the nyse and their

original shares listed on the Toronto Stock Exchange (tsx). The motivation of my

research is as follows: since Canadian shares and their cross-lists in the U.S. are intrin-

sically identical, and the markets and the economies they trade within are integrated

(so are the costs of capital), the noticeable cross-listing premia of the cross-listed pairs

are puzzling in the context of conventional asset pricing perspectives.

Information asymmetry is expected to resolve the missing link and the proxy I

use is the pin. The pin on a stock measures the proportion of informed transactions

5Previous studies on intra-border fragmentation include Hasbrouck (1995) and Easley, Kiefer,and OHara (1996).

2


4/50

among all trades in the stock. Following Easley, Kiefer, OHara, and Paperman (1996),

I estimate the pin for both the tsx- and the nyse-listed twins. Informed traders are

believed to make significant contributions towards price discovery which is a crucial

function of stock exchanges.

Legally and technically distinct shares listed by the same company on separate

exchanges are like derivative securities with same underlying assets. While their prices

are not always equal even after adjusting for the exchange rate, they cannot differ too

much in the long run either.6 This is Hasbroucks (1995) argument on the dynamics

of domestically fragmented stock prices, and I am extending the very idea beyond the

border. My initial hypothesis is that, a markets relative leadership in cross-border price

discoveryreflected in a higher information share7must, on average, accompany a

higher pin.

The cross-listing premium of a pair must diminish to the fraction tolerated by

transaction costs by arbitrage. Following Gagnon and Karolyi (2004), I estimate the

convergence speed of premia to the parity. A higher pin reflects a better idea on the

pricing and fundamentals of the companys listings. It naturally leads to my second

conjecture: the higher the pin on a cross-listed pair, the faster the convergence speed.

Foerster and Karolyi (1998) report that the post-cross-listing bid-ask spreads in

Canada narrow. Eun and Sabherwal (2003), Fleming, Ostdiek, and Whaley (1996),

and Jones and Seguin (1997) argue that less noise trades occur in the market with

lower trading costs. From this end, I reason that the pin on an existing share on the

tsx is expected to rise upon cross-listing on the nyse.ii

Following Eun and Sabherwal (2003), I choose Canadian stocks for several reasons.

6They are cointegrated according to Engle and Granger (1987), and Engle and Yoo (1987).7Hasbroucks (1995) information share is a relative contribution made by a stock exchange to price

discovery of shares fragmented on multiple exchanges. See Appendix A4 for a detailed derivation ofthe information share.

3


5/50

First, Canadian equities represent the largest group of stocks cross-listed in the U.S.

from a single country. Thus, reasonable inferences can be drawn from cross-sectional

analyses of the pin and other factors in the home and U.S. markets, while holding the

nationality of shares constant. Second, many of them trade actively on both the nyse

and the tsx. This is essential for conducting intraday analyses as markets have to be

liquid in order to achieve efficiency.

Third, since the trading time of the tsx coincides with that of the nyse (9:30am4:00pm,

est), Canadian securities offer a distinct benefit over cross-listed securities from Eu-

rope and Asia for which there is a little or no overlap trading time between their home

markets and the U.S. markets. Since the biases borne by trading-time differences are

minimized, analyses based on information asymmetry and market friction are more

reliable. Finally, Canadian stocks are listed in the U.S. as ordinary shares due to

compatible accounting standards, whereas most other cross-listed shares are adrs is-

sued by U.S. custodian banks. The so-called home bias discounts are therefore less

pronounced for Canadian cross-lists.

The main findings of my study are as follows: firstly, relative to the nyse, the tsx is

heavier with informed trades and accounts for more in information sharethis is an

explicit evidence of informed traders contribution to cross-border price discovery. If a

stock listed on an exchange shows a higher pin than its replica on the other cross-border

exchange, it reflects a larger proportion of informed traders who have better under-

standing of the cross-listed firm. The exchange with more informed traders is likely to

generate more information that stokes price discovery across the borderreflected in

a higher information share.

The first finding justifies the existence of positive cross-listing premia in the pairs

since a stock with a higher degree of adverse selection requires a higher return in

4


6/50

equilibrium.8 Secondly, the parity-convergence of premia is fostered by discretionary

liquidity tradersrelating the dynamics of synchronous cross-listing premia to informa-

tion asymmetry is novel in the literature. Contrary to my hypothesis, lower-pin pairs

converge are quicker in convergence to parity. This is because practitioners engaging

in statistical arbitrages keep themselves away from informed traders and make advan-

tage of the non-discretionary liquidity trader pool. Thus, a low pin on a pair with

a quickly vanishing premium is represented by discretionary liquidity traders. Pairs-

trades can be done in the absence of private information on the issuers of diverged

stocks while requiring timely execution and unwinding of positions .

Finally, the pin on a tsx-listed stock, on average, rises upon cross-listing on the

nysein other words, the information asymmetry surrounding the issuer on its home

exchange intensifies as it cross-lists overseas. This exacerbating degree of adverse se-

lection explains negative event study returns on the tsx. The last finding is a notable

contribution to the literature by focusing on relative cross-listing effects on the home

market information environment. Previous articles mention of reduced noise trade risk

in absolute magnitude.

The three key results effectively address how information asymmetry is priced in

stocks fragmented across the border, along the timeline, and beyond the event horizon.

Overall, the pin turns out to be a robust and pithy measure of asymmetric information.

The remainder of this paper is organized as follows. Section Two raises key hypotheses

with reasoning relating to existing articles in the literature. Section Three describes

the data. Section Four provides my main empirical results. I conclude in Section Five.

8In Appendix A5, based on an extended version of the noisy rational expectations model (Grossmanand Stiglitz (1980)), I provide a sufficient condition of home market liquidity dominance under whicha higher-pin tsx-listed stock must be priced lower than its nyse-listed replica in no-arbitrageequilibrium.

5


7/50

2 Theory and hypotheses

2.1 PIN, fragmentation, and information shares

Unlike articles which focus on the joint distribution of trades and prices,9 Easley, Kiefer

and OHara (1997a, 1997b) and Easley, Kiefer, OHara, and Paperman (1996) weigh on

parametric assumptions to estimate a relative measure of adverse selection using buy

and sell orders instead of price data. In their theoretic setting, there are (risk-averse)

competitive market makers, informed traders, and uninformed liquidity traders. In the

simplest format, liquidity traders are equally likely to buy or sell shares.

The four parameters of the maximum likelihood model are 1. the probability that

an information event occurs on a given day (); 2. the probability that an information

event is pessimistic (); and 3. the (Poisson) order arrival rates of informed and

uninformed traders ( and ). As a result, the probability of informed trading10 (pin)

measures the relative degree of private information-based trades among all trades.

Easley, Kiefer and OHara (1997b) argue as informed traders gain weight in the market,

adverse selection aggravates and the trading volume increases.iii

Fragmentation is the dispersal of trading in a security to multiple exchanges or markets.

As an early bridgework between fragmentation and informed trading, Chowdhry and

Nanda (1991) note that information lags between distinct trading locations yield tran-

sitory disparities in the prices of an identical security.iv Blume and Goldstein (1991)

and Lee (1993) state that price discoverythe convergence towards an equilibrium

price which is a crucial function of an exchangeon U.S. exchanges occurs primarilyon the nyse. Similar results are drawn by Harris, McInish, Shoesmith, and Wood

9Bagehot (1971), Grossman and Stiglitz (1980), Kyle (1985), and Glosten and Milgrom (1985).10pin +2 . See Appendix A3 for a detailed derivation of the pin.

6


8/50

(1995),v and by Gardner and Subrahmanyam (1994).11

When a nyse-listed stock trades not only on the nyse but also on the regional

exchanges, the fragmented security prices may not be identical but they also cannot

differ too much in the long run either. Since they share a common efficient component 12

this renders Simss (1980) original vector autoregressive (var) model unwieldy. That

is why Hasbrouck (1995) takes a formal error correction model (ecm)13 approach to

define information shares14 of exchanges. He finds that price discovery of fragmented

stocks appears to be concentrated on the nyse.

Easley, Kiefer, and OHara (1996) show that there is a significant difference in the

information content of orders executed in New York and in Cincinnati, and that thisdifference is consistent with the cream-skimming hypothesis, instead of the com-

petition hypothesis. The question pertaining to whether trades in different locations

within the U.S. carry the same level of information can also be directed to cross-border

fragmentation.

Branching out the fragmentation idea to the international finance literature, based on

U.S.-listed Canadian stocks, Eun and Sabherwal (2003) find that prices on the TorontoStock Exchange (tsx) and U.S. exchanges are cointegrated and mutually convergent,

following the ecm setting of Harris, McInish, Shoesmith, and Wood (1995). They

report that the U.S. share of price discovery ranges from 0.2 percent to 98.2 percent,

with an average of 38.1 percent.

11Extending the works of Hasbrouck (1991, 1995), Gardner and Subrahmanyam (1994) concludethat less informed trades are executed on the regional exchanges than on the nyse.

12Security prices are cointegrated if there exists a linear combination of the non-stationary prices

that can be toned stationary. A time series is strongly stationary if its probability distribution istime-invariant, and weakly stationary if up to its second momentsmean, variance, and covariance.13Engle and Granger (1987) and Engle and Yoo (1987).14Hasbroucks (1995) information share is a relative contribution made by a stock exchange to

price discovery of shares fragmented on multiple exchanges. Information shares are estimated by thevector ecm, provided that the multiple security prices are cointegrated. See Appendix A4 for detailedderivation.

7


9/50

Across the global equity markets, Bailey, Mao, and Sirodom (2006),vi and Chan,

Menkveld, and Yang (2008)vii describe intriguing multi-board trading structures in

Thailand and China, respectively, and explain how information asymmetry takes place

with the fragmented stocks therein. Also, foreigners are disadvantaged in Korea (Choe,

Kho, and Stulz (2005))viii while they wield superior information processing capability

in Thailand and Singapore (Bailey, Mao, and Sirodom (2007)).ix

The pin is a relative measure of private informed trades. If a stock listed on an

exchange has a higher pin than its replica on the other cross-border exchange, this

reflects a greater proportion of informed traders who have better understanding of the

cross-lister. Then, since informed traders are believed to contribute to price discovery,

it is also likely that the exchange with heavier intensity of informed trades tends to

generate more relevant information which fuels price discovery.

By definition, an exchange is said to lead the other exchanges if it accounts for

more price discovery (reflected in its higher information share). Hasbrouck (2007) notes

that a vector ecm analysis assigns quote changes to the influx of trades. Asymmetric

information is then reflected in a wide price change. In this sense, the vector ecm

analysis can be directionally equivalent to the pin estimation.x

Trades in cross-listed pairs are fragmented across the borderline that separates

the two most integrated economies and that appears to be a three-part hurdlehome

and host market indices, and the exchange rateto price discovery. My first hypoth-

esis attempts to verify the role of informed traders in determining cross-border price

discovery. Specifically,

relative to the cross-border exchange, the lead market (with a higher aver-

age information share) commands denser informed trades (with a higher

average PIN).

8


10/50

2.2 Convergence speed and informed traders

Easley, Hvidkjaer, and OHara (2002) note that, in equilibrium, a high-pin stock carries

an adverse-selection discount since it requires an additional return. Similarly, I reason

that a cross-listed pair yields either a positive or negative cross-listing premium15 if one

side carries denser private information.16 Unless that relative price spread is believed to

persist due to severe liquidity constraints, shortsale restrictions, and other deterrents,

an arbitrageur will long the discounted stock and short the other side with favorable

assumptions on the exchange rate.

The international finance literature has cumulated articles on arbitrage opportuni-

ties created by cross-listed shares. The early studiesMaldonado and Saunders (1983),

Kato, Linn, and Schallheim (1991), Park and Tavakkol (1994), Miller and Morey (1996),

and Karolyi and Stulz (1996)conclude that arbitrage profits for cross-listed shares

do not exist and thus they are priced at parity. Wahab, Lashgari, and Cohn (1992)

show that there are arbitrage opportunities in cross-listed pairs. Froot and Dabora

(1999) study pricing of a couple of dual-listed corporationsRoyal Dutch and Shell,

and Unilever N.V. and Unilever plcand find a sizable and significant price deviationfrom parity.17

Gagnon and Karolyi (2004) record significant price deviations in 581 adr-underlying

pairs under their studythey report discounts up to 90% and premia up to 70%. Chen,

Choi, and Kim (2008) reason that limits to arbitrage cannot falsify market efficiency,

due to practical infeasibility in some markets. Rather, they show thatwith intro-

15The relative premium of a cross-listed stock on a U.S. exchange against its home market basis

share, adjusted by the exchange rate.16In Appendix A5, based on an extended version of the noisy rational expectations model (Grossmanand Stiglitz (1980)), I provide a sufficient condition of home market liquidity dominance under whicha higher-pin tsx-listed stock must be priced lower than its nyse-listed replica in no-arbitrageequilibrium.

17See Kim, Szakmary, and Mathur (2000) for vector autoregressive (var) and seemingly unrelatedestimation (sure) methods that analyze adjustment in adr-implied prices.

9


11/50

duction of non-parametric econometrics in the cross-listed shares literatureas proxy

measures representing market efficiency improve, the convergence in adr-pairs of Asia-

Pacific firms speeds up over the sample period.

The speed at which cross-listing premium converges to the parity is measured by a

parameter proposed by Gagnon and Karolyi (2004). According to their empirical

model,18 each firms cross-listing premium can be explained by its first-lag term, and

its time-distributed risk exposure to the returns on the host and host market indices

and the foreign exchange rate. In the long run, the premium diminishes to the fraction

tolerated by the transaction costs by arbitrage.

For the synchronous cross-listing premium of a Canada-U.S. cross-listed pair,

its dynamics (convergence speed) is expected to depend on information asymmetry

controlled for market friction, liquidity constraint, and firm characteristics. Cross-

sectionally, a higher pin implies enhanced price discovery. In rational expectations

equilibrium, informed investors impound information to prices and catalyze price dis-

covery. Parity-convergence can, thus, be accelerated by the degree of average informed-

ness on the cross-lister. In this regard, my second conjecture states that

the higher the PIN on a cross-listed pair, the faster the parity-convergence

of cross-listing premia.19

18DRi(t) = i + i DRi(t 1) +1

j=1 USj R

U SM (t + j) +

1j=1

Cj R

CM(t + j) +1

j=1 F Xj RF X (t + j) + i(t).

Each firm(i)s cross-listing premium

DRi(t)

PUSi (t) PCi (t)

/PCi (t)

can be explained by 1. its

own lag (DRi(t 1)) associated with 2. the convergence speed parameter (i)the closer the abso-lute value to zero, the faster the convergence to the parity; and lag-distributed (yesterday ( j = 1),

today (j = 0), and tomorrow (j = +1)) returns on 3. the Dow Jones Industrial Average IndexRUSM (t + j), 4. the s&p tsx Composite Index RCM(t + j), and 5. the Canada-U.S. exchange ratereturn (RF X (t + j))a positive RF X implies a depreciation in the Canadian dollar. The forward-lagis due to information leakages and market impact, and is deemed standard in event models.

19By specification, a lower absolute value of parameter, below one, is equivalent to a higher conver-gence speed.

10


12/50

2.3 Cross-listing effects on PIN

Price discovery is enhanced with less noise trades. De Long, Shleifer, Summers, and

Waldman (1990) put, since noise traders do not reflect information on the fundamentals

their trades lead to diverging price from its intrinsic value, reducing price informative-

ness while increasing volatility (noise trader risk). In the literature, transaction costs

play an important role in the absolute magnitude of noise trades.

Foerster and Karolyi (1998) document that the post-cross-listing spreads in Canada

decrease. The augmented liquidity gives rise to tsx market makers competitive reac-

tion by setting bid-ask spreads smaller.xi The decrease in spreads on the tsx is heavily

weighed on the stocks whose trading volume contribution by the U.S. exchanges is rel-

atively large. Eun and Sabherwal (2003),xii Fleming, Ostdiek, and Whaley (1996),xiii

and Jones and Seguin (1997)xiv report that less noise trades occur in the market with

lower trading costs.

Reduced noise trader risk is a double-edged sword. It may be a boon to enhanced

price discovery, since less noisy fluctuation may contribute to setting a more precise and

stable process towards the fair price of a security. But, it also gives rise to questioning

whether less volatility entails a higher likelihood of informed trades? In other words,

does cross-listing exacerbate the home market information environment with more

perverse adverse selection? My last conjecture states that

after cross-listing on the NYSE, on average, the information asymmetry on

a TSX-listed stock intensifies (the PIN rises).

Provided that the pin on the tsx rises upon cross-listing while the spread moderates

on the tsx, the above hypothesis may sound unorthodox within the information-based

microstructure literature. Easley, Kiefer, OHara, and Paperman (1996) conclude that

11


13/50

the pin is positively correlated with bid-ask spread. Their reasoning is based on within-

market cross-sectional (cross-firm) analyses, whereas my attempted inference draws

from cross-border regime-switching events (cross-border listings). In sum, a cross-

listing has two effects. It brings on a higher pin on the original shareif my hypothesis

holdsand decreases the bid-ask spread according to Foerster and Karolyi (1998).

Cross-listings are a good source of additional liquidity to the existing home-listed

stocks and borrowing from the assertion of Easley, Kiefer and OHara (1997b) that

intensifying adverse selection captured by the pin and increasing trading volume are

positively correlated further leverages my hypothesis. The additional liquidity swarmed

back to the tsx market makers force them to set spreads narrower. See Admati and

Pfleiderer (1988) for a similar support.xv

3 Data and preliminary results

3.1 Data construction

As of March 28, 2007, 84 Canadian companies were cross-listed on the nyse. Through

the sample periodJanuary 1, 1998, through December 31, 2000I matched 55 tsx-

nyse pairs as listed in Table 1.20 The tick-by-tick trade and quote data21 for tsx-listed

Canadian stocks of the cross-listed pairs spans January 1998 through December 2000.

The Trade-And-Quote (taq) data through the same corresponding period for the cross-

listed shares on the nyse22 was accessed via Wharton Research Data Services (wrds).

20Following Eun and Sabherwal (2003), the augmented Dickey-Fuller (1981) unit root test is con-

ducted to each daily-price time series of the matched pairs with appropriate lag length, per Akaike(1974), to verify first-order integration (I(1)). Applying Johansens (1991) either trace test or eigen-value test yielded one co-integrating equation for each tsx-nyse co-listed pair. These results providejustification of constructing error correction models (ecms) to estimate the information shares of eachco-listed pairs exchanges.

21http://www.tsx.com/en/data/products services/historical/trading data/trades quotes.html22Canadian firms listed on nyse: http://www.nyse.com/about/listed/1.html?country=Canada&ListedCom

12


14/50

The equivalent period U.S.-Canada exchange rate intraday data was purchased from

Olson & Associates.

Unlike specialist-based, auction exchange nyse, electronic exchange tsx uses a

Central Limit Order Book (clob)xvi system, thus orders are required to be in the book

to have standing.23 I looked at decrements in the inside depth on one side of the quote

that correspond to uncommon trade sizes, like a trade of 1,300 shares. As a result,

matching trades with prevailing quotes of five-second lead following Lee and Ready

(1991)24 was deemed reasonable. The preliminary datasets for estimation of the pinxvii

follows Easley, Kiefer, OHara, and Paperman (1996).

I adopt Easley, Engle, OHara, and Wus (2008) log-likelihood function specification

for improved numerical stability in computing the pin.25 The arithmetic averages of

monthly pin estimates for all Canadian cross-listers on the tsx and the nyse are

plotted in Figure 1. It appears that the tsx, on average, dominates the nyse in terms

of the pin in annual estimates for the cross-listed pairs through the sample period. 26

The spikes in pin are seen in the post-decimalization period between November and

December 1999,xviii a finding consistent with Zhao and Chung (2006).

3.2 Preliminary results

The pin for tsx-and nyse-listed Canadian stocks are estimated following Easley,

Kiefer, OHara, and Paperman (1996) and Easley, Kiefer and OHara (1997a, 1997b).

Following Eun and Sabherwal (2003), the mid-points (= (ask+ bid)/2) of U.S.-Canada

23I owe this comment to Daniel Weaver of Rutgers Business School.24A trade is considered buyer-initiated if it is higher than the five-second earlier mid-quote, or

seller-initiated if lower.25A descriptive algorithm for estimation of the pin is explained in Appendix A3.26The annual estimates for the pin on the tsx are {0.242, 0.213, 0.206} in 1998, 1999, and 2000,

respectively, while the corresponding estimates for the nyse are {0.204, 0.212, 0.196}, over the sameperiod.

13


15/50

exchange rate bid and ask quotes are updated every minute. The bid and ask quotes

of the nyse-listed Canadian stocks are matched with their previous minutes exchange

rate quote mid-points. Based on the orthogonalized impulse responses of bid and ask

quotes from both the tsx and the nyse, the information shares27,xix of the tsx and

the nyse for each cross-listed pair are estimated.

In Table 1, the average across monthly estimates of information shares of each

pair over the entire sample periodJanuary 1, 1998 through December 31, 2000are

listed. The bid-ask spreads28 are adjusted by the mid-quote and thus measure the

relative discrepancy between bid and ask quotes free from exchange rates. About

twenty firms in the sample exhibit higher pin on the nyse than on the tsx through

the sample period. For some cross-listers, like Manulife Financial Corp. and Suncor

Energy Inc., there is no significant difference between the pin on the two exchanges.

Only nine firms in the sample show higher spreads on the tsx, and only two firms have

higher information shares on the nyse.

The arithmetic means of pin, spreads and information shares of the matched cross-

listers on the tsx and the nyse are as follows:29 1. pintsx = 0.240 > pinnyse = 0.211;

2. spreadtsx = 0.015 < spreadnyse = 0.022; and 3. istsx = 0.543 > isnyse =

0.457. In a rule-of-thumb summary, by scanning the results put out in Table 1, for a

Canadian cross-lister, on average, it appears that more price discovery takes place on

the tsx (the lead market) where the intensity of informed trades tends to be heavier

(a higher pin) and yet with lower spreads (competitive market making).

The impulse response function plots of bid and ask quotes for Abitibi Consolidate,

Inc.cross-listed on the nyse from the tsxare shown in Figure 2. Each of the four

27See Appendix A4.28spreadnyse

asknyse bidnyse(asknyse + bidnyse)/2

; and spreadtsx asktsx bidtsx

(asktsx + bidtsx)/2.

29For brevity, I do not present the monthly estimatesJanuary 1998 through December 2000ofthe pin and the spreads for the cross-listed pairs, however, they are available upon request.

14


16/50

consecutive charts specifies the source of innovation by two standard deviations. The

quotes on the nyse rarely affect the quotes on the tsx. To the contrary, positive

increases in ask and bid prices on the tsx are followed by changes in ask and bid prices

on the nyse, respectively. This chain reaction does not necessarily hold markedly for

all cross-listed stocks, and the degree to which an exchange responds to the other is

reflected in the relative magnitude of information share.

4 Results

4.1 PIN, fragmentation, and information shares

Based on monthly estimates, the statistical significance of the tsxs dominance over

the nyse in terms of the pin can be verified by the Wilcoxon signed-rank test.30 First,

I obtain the difference in the pin on the tsx and on the nyse for each firm, i, and for

each month, t,

d(i, t) pintsx(i, t) pinnyse(i, t),

then the Wilcoxon-test statistic is defined as

V0 {(i,t)}

1{d(i,t)>0} it,

where it is the rank of {|d(i, t)|}. In the first column of Table 2, the Wilcoxon-test

statistic, under the null hypothesis

H0 : pintsx = pinnyse,

30It is a non-parametric pair-wise comparison test proposed by Wilcoxon (1945). Table 2 summarizesthe test results for hypotheses regarding the pin, spreads, and information shares.

15


17/50

is very strongly rejected at a 1% right-tail significance level, which implies

H1 : pintsx > pinnyse.

Harris, McInish, and Wood (2002) report that the nyses contribution to price dis-

covery against its regional counterparts increases as its spreads against the regionals

decrease. In the cross-border context, the tsxs competitive market making vis-a-vis

the nyse can be inferred from similarly comparing the bid-ask spreads on the tsx and

the nyse, respectively. It is natural to hypothesize

H0 : spreadtsx = spreadnyse versus H1 : spreadtsx < spreadnyse,

and it can be checked by defining for each firm, i, and for each month, t,

d(i, t) spreadnyse(i, t) spreadtsx(i, t),

which yields an exceedingly significant Wilcoxon test-statistic V0 = 680698 with a

p-value < 2.2 1016, that overwhelmingly agrees with the alternative hypothesis as

seen in the second column of Table 2.

The statistical significance testing of the relative dominance of the tsx over the nyse

in terms of information shareequivalently, the tsx leads the nyse in relative contri-

bution to price discovery of the cross-listed pairscan be conducted as follows.

H0 : istsx = isnyse versus H1 : istsx > isnyse,

16


18/50

then, again, for each firm, i, I define an appropriate difference measure

d(i) istsx(i) isnyse(i),

and the information share of the tsx far exceeds that of the nyse such that V0 = 1413

with a p-value = 3.99 109 as seen in the third column of Table 2.

In summary, I find that, on average, the tsx (the lead market) usually shows a higher

pin than the nyse (the lag market). In other words, the exchange with heavier intensity

of informed trades contributes more to price discovery of cross-listed pairs. This result

is in line with that of Eun and Sabherwal (2003) and can be understood in that informedtraders prefer to trade in the lead market where more original information can be found

and entry and/or exit strategies benefit from the abundant home market liquidity.

It provides an empirical evidence that informed traders contribute to cross-border

price discovery and that the pin is an appropriate proxy for asymmetric information

on cross-listed pairs. The trades executed on the lead exchange, the tsx, are more

likely to be information-based than the trades executed on the lag exchange, the nyse.

1. The investors on the TSX posses more private information on Canadian cross-

listers than their counterparts on the NYSE.

2. The TSX contributes more to price discovery than the NYSE does.

3. The market makers on the TSX are more competitive in quote spreads than their

competitors on the NYSE.

17


19/50

4.2 Convergence speed and informed traders

The Canadian cross-lists on the nyse are found to carry, on average, positive cross-

listing premia against their home-listed stocks on the tsx through the sample period.31

Provided that Canadian shares and their cross-lists in the U.S. are intrinsically iden-

tical, and the markets and the economies they trade within are integrated (so are the

costs of capital), an average positive premium on the U.S.-listed replica against its syn-

chronously traded home-listed stock is puzzling within the conventional international

asset pricing literature.

Easley, Hvidkjaer, and OHaras (2002) cross-sectional analysis shows that, in equi-

librium, a higher pin stock commands a higher required return. As I previously sta-

tistically confirmed that the tsx, on average, is heavier in information-based trades

(higher in the pin) than the nyse, the original Canadian stock can be relatively dis-

counted against its cross-list on the nysewhich translates into a positive cross-listing

premium.32

It is natural to ask how quickly and by whom that positive cross-listing premium

reduces to the parity. Following Gagnon and Karolyi (2004), I estimate the convergence

speed parameter in the model33 in a daily frequency and the averaged it across the

estimates for each firm. The pin effect on the convergence speed can be inferred from

regressing the convergence speed parameter (speedconv ) onto the average pin on

both exchangessince convergence speed is a mutual conceptand average spread

31Based on the daily cross-listing premia of 55 cross-listed pairs traded January 1998 throughDecember 2000, the arithmetic mean, the median, and the standard deviation are 0.00864, 0.00023,and 0.21614, respectively.

32In Appendix A5, based on an extended version of the noisy rational expectations model withinformation asymmetry introduced by Grossman and Stiglitz (1980), I provide a sufficient conditionof home market liquidity dominance under which a higher-pin tsx-listed stock must be priced lowerthan its nyse-listed replica in no-arbitrage equilibrium.

33DRi(t) = i + i DRi(t 1) +1

j=1 USj R

U SM (t + j) +

1j=1

Cj R

CM(t + j) +1

j=1 F Xj RF X (t + j) + i(t).

18


20/50

(on both exchanges), controlled for firm size (size),34 industry dummy (industry),35

and volume36 as follows

speedconv = 1 pinavg + 2 spreadavg + 3 size + 4 industry + 5 volume + .

According to the regression model, the dynamics of synchronous cross-listing pre-

mia is explained by the asymmetric information component (pin) and market friction

(pin) while holding idiosyncracies (size and industry) and liquidity constraint (volume)

constant. Surprisingly, in Table 3, a higher pin on either exchange very significantly

decelerates the convergence speed of a tsx-nyse pair to the implied parity in all

specifications. Counter to intuition, the uninformed traders appear to deplete cross-

listing premia faster than their informed cohort! The pin appears robust controlled

for liquidity on cross-listed pairs in models 2 and 3. The higher the spread on either

exchangethe higher the average spread as a resultthe slower the convergence speed

in models 1 and 2, since the convergence speed parameter is reciprocal to actual speed.

Practitioners executing statistical arbitrages (pairs trades) and profiting from cross-

listing premia need not be informed of the issuers fundamental value. Timely execution

and unwinding their positions will suffice. Thus, statistical arbitrageurs are believed

to be discretionary liquidity traders who are responsible for quickly converging and

low-pin cross-listed pairs. Relating the dynamics of synchronous cross-listing premia

to asymmetric information is novel in the cross-listed shares literature.37, xx

34Normalized average market capitalization on the tsx and the nyse.35Equals one if the cross-lister is a manufacturing firm, and zero otherwise.36Normalized average per-trade volume on the tsx and the nyse.37Asymmetry may exist in a traders preference on either tsxs or nyses side of cross-border

arbitrage against the other side for liquidity reason. For example, it may be easier to short-sell on thetsx than on the nyse. The cross-market quoted spreads are defined as follows

spreadnt offernyse bidtsx

us$can$

ask

spreadtn offertsx

us$can$

bid

bidnyse.

19


21/50

I further explore the cross-sectional relationship between the average spread across the

exchanges against the average pin on both exchanges, the convergence speed parameter,

controlling for firm size and industry dummy. In Table 4, the average pin is very

significantly positively associated with the average spread which is consistent with the

finding of Easley, Kiefer, OHara, and Paperman (1996).

spreadavg = 1 pinavg + 2 speedconv + 3 size + 4 industry + .

4.3 Cross-listing effects on PIN

Table 5 shows fifteen Canadian firms that cross-listed on the nyse through the sample

period (January 1998December 2000). Thirteen firms had been listed on the tsx

before they cross-listed on the nyse. The firms without the pin either have cross-

listing dates are too near the end of the sample period (December 31, 2000) or are

insufficiently liquid. For the pin estimates before and after cross-listing events, there

are eight pairs with a six-month window, six pairs with a twelve-month window, and(monthly average) nine pairs with an exhaustive window through the sample period.

The arithmetic means of the columns of the pin show that there are increments

on cross-listing events. The pre- versus post-cross-listing scatter plots are provided for

each window pairs in Figures 3, 4, and 5. The pin on the tsx, on average, rises upon

cross-listing on the nyse within all event windows. The significance of pin increase

(rise in the intensity of adverse selection) around cross-listings can be verified by the

An arbitrageur may prefer to short-sell on the nyse and to long on the tsx, while another to short-

sell on the tsx and to long on the nyse. The first strategy narrows down spreadnt, while the second

option squeezes spreadtn. Either strategy may turn out more lucrative than the other due to existenceof bid-ask spread in the U.S.-Canada foreign exchange rate. If the two cross-arbitrage returns appearstatistically indifferent, then my case of using foreign exchange rate mid-quotes is reasonable.

20


22/50

Wilcoxon signed-rank test with the difference in pin before and after cross-listing for

each firm, i,

d(i) pinafter(i) pinbefore(i).

In Table 6, all the null hypotheses against the alternative hypotheses

H0 : pin+3m = pin3m versus H1 : pin+3m > pin3m,

H0 : pin+6m = pin6m versus H1 : pin+6m > pin6m,

H0 : pinafter = pinbefore versus H1 : pinafter > pinbefore

are rejected at a 10% right-tail significance level. The tsx-listed firms, on average,

post negative cumulative abnormal returns (cars) within all event windows around

cross-listings in Table 5. It is reasonable that Canadian firms which cross-list in the

U.S. do not benefit from lower costs of capital unlike those in the emerging market

economies, Canadian financial managers can easily diversify their financing risk across

the border. There appears to be no discernable cross-listing premium due to diminished

market incompleteness (Merton (1987)) for Canadian cross-listers in the U.S.

While the conventional theory of market integration may look lackluster in the

trans-Niagara setting, the implications of information-based microstructure theory

stand conspicuously. The result in Table 6 that the pin risesor that the intensity of

private information gets heavier on the home exchangeupon cross-listing unifies and

extends the existing claims in the cross-border finance literature. Cross-listings lower

transaction costs and narrow the spreads on the tsx and, resultantly, reduce noise

trader risk (Eun and Sabherwal (2003), Fleming, Ostdiek, Whaley (1996), and Jones

and Seguin (1997))or subdue excessive volatility borne by liquidity trades.

The more perverse degree of adverse selection on the home market shown in Table

6 is the first documentation ofrelative cross-listing effects on the home exchange infor-

21


23/50

mation environment.38 Previous articles only mention the decrease in noise trades in

absolute magnitude. The higher post-cross-listing pin also explains the negative event

study returns on the home-listed stocks in Table 5. As the original stocks become more

infested with private information, they must reflect relative discounts in equilibrium,

in line with Easley, Hvidkjaer, and OHara (2002).

Table 7 shows that the bid-ask spreads narrow after cross-listing events in tandem

with Foerster and Karolyi (1998) and the effect is evidently over the exhaustive thresh-

old window.xxi Whether Canadian firms cross-listings on the nyse facilitate enhanced

volume39 on the home exchange is shown in Table 8. Statistically speaking, the incre-

mental effect of cross-listing on home market liquidity is not vivid. This opacity maybe

due to the limited sample size or may reflect Karolyis (2006) summarizing comment

that ... I may, in fact, learn that price discovery does not necessarily originate in the

markets with the highest relative turnover, but rather where the informed traders are

going with limited market impact.

The above findings suggest that, as least within integrated economies, cross-listings

boost intensity of private information-based trades in home-listed stocks. A higher pro-

portion of informed trades is a double-edged sword since while it fosters price discovery,

adverse selection exacerbates on the inferior side of trades. This regime-switching in-

formation ambience lends support to the claim of Bailey, Karolyi, and Salva (2006)

that cross-listings may not alleviate information asymmetry.xxii The result herein may

contradict the bonding hypothesis (Coffee (1999), and King and Segal (2004)) which

states that insiders are less incentivised to trade after cross-listings.40

38In a comparable case, Chan, Menkveld, and Yang (2008) report that the pin on b-shares inChinathat had only been traded by foreign investorsrises on opening access to locals.

39Per-trade number of tsx-listed shares of nyse-cross-listed Canadian firms.40A testable hypothesis on this conflict may be that the emerging market cross-lists may warrant

higher event study returns on the home exchange than the developed market ones, since the formergroups bonding effect dominates while the latter groups adverse selection exacerbates.

22


24/50

5 Conclusion

The theory of international asset pricing suggests that stocks cross-listed within inte-

grated economies do not entail significant cross-listing premia. Yet, I find that there

are, on average, sizable premia stemming from asymmetric information amongst the

stocks that are listed and traded on multiple exchanges separated by the Niagara Falls.

I provide a sufficient condition under which cross-border dominance in private infor-

mation leads to positive premia. The theoretic prediction is empirically supported as

I analyze pairs of the Canadian shares co-listed on the Toronto Stock Exchange (tsx)

and on the New York Stock Exchange (nyse) in the three-year period, January 1998

through December 2000.

My first empirical finding reveals that, on average, the tsx leads the nyse in price

discoverymeasured by information shareand shows a higher pin. In other words,

the exchange with heavier intensity of informed trades contributes more to the price

discovery of cross-listed pairs. It is an explicit cross-border evidence that informed

traders stoke price discovery.

As the first finding justifies positive cross-listing premia, I report that lower-pin

pairs are quicker in parity-convergence of the premia. Specifically, discretionary liquid-

ity traders are represented by less pin and they live on cross-listing premia by making

timely executions and unwinding in the absence of fair-value information on the cross-

listers. This is the first notable documentation of relating the dynamics of synchronous

cross-listing premia to asymmetric information.

Finally, on average, the pinthe likelihood of adverse selectionon a tsx-listedstock rises upon cross-listing on the nyse. This result of relative cross-listing effects

on the home market information environment not only explains negative event study

returns, but also unifies and extends existing claims in the literature that focus on

23


25/50

noise trader risk in absolute magnitude.

The three key results effectively address how information asymmetry is priced

across the border, across the timeline, and across the event horizon. Overall, the pin

proves to be a pithy and robust measure of asymmetric information on fragmented

stocks across the border.

24


26/50

References

[1] Admati, A. and P. Pfleiderer. 1988. A Theory of Intraday Patterns: Volume and Price Variability. Review ofFinancial Studies 1: 3-40.

[2] Akaike, Hirotugu. 1974. A New Look at the Statistical Model Identification. IEEE Transactions on AutomaticControl 19(6): 716-723.

[3] Bagehot, W. 1971. The Only Game in Town. Financial Analysts Journal 22: 12-14.

[4] Bailey, Warren B., G. A. Karolyi, and C. Salva. 2006. The Economic Consequences of Increased Disclosure:Evidence from International Cross-Listings. Journal of Financial Economics 81(1):175-213.

[5] Bailey, Warren B., Connie X. Mao, and Kulpatra Sirodon. 2006. Locals, Foreigners, and Multi-Market Tradingof Equities: Some Intraday Evidence. Working Paper, Cornell University, Temple University and ThammasatUniversity.

[6] Bailey, Warren B., Connie X. Mao, and Kulpatra Sirodom. 2007. Investment Restrictions and the Cross-BorderFlow of Information: Some Empirical Evidence. Journal of International Money and Finance 26: 1-25.

[7] Bank of New York. 2006. The Depositary Receipt Markets. New York: Bank of New York.

[8] Blume, M. E. and M. Goldstein. 1991. Execution Costs on the NYSE, the Regionals and the NASD. WorkingPaper, University of Pennsylvania.

[9] Brennan, M. J., and H. Cao. 1997. International Portfolio Flows. The Journal of Finance 52: 1851-1880.

[10] Chan, Kalok, Albert J. Menkveld, and Zhishu Yang. 2008. Information Asymmetry and Asset Prices: Evidencefrom the China Foreign Share Discount. The Journal of Finance 58(1): 159-196.

[11] Chen, Haiqiang, Paul Moon Sub Choi, and Hyunseob Kim. 2008. American Depositary Receipts: Asia-PacificEvidence on Convergence and Dynamics. Journal of Multinational Financial Management 18(4): 346-368.

[12] Choe, Hyuk, Kho, Bong-Chan, and Rene M. Stulz. 2005. Do Domestic Investors Have an Edge? The TradingExperience of Foreign Investors in Korea. The Review of Financial Studies 18: 795-830.

[13] Chouinard, Eric and Chris DSouza. 2003. The Rationale for Cross-Border Listings. Bank of Canada Review,Winter 2003-2004. Available at URL http://www.bank-banque-canada.ca/en/review/winter03-04/chouinarde.html

[14] Chowdhry, Bhagwan, and Vikram Nanda. 1991. Multimarket Trading and Market Liquidity. The Review ofFinancial Studies 3: 483-511.

[15] De Jong, A., L. Rosenthal, and M. van Dijk. 2003. The Limits of Arbitrage: Evidence from Dual-Listed Compa-nies. Working Paper, Erasmus University.

[16] De Long, J.B., A. Shleifer, L.H. Summers, and R.J. Waldmann. 1990. Noise trader risk in financial markets.Journal of Political Economy 98: 703-738.

[17] Dickey, David A., and Wayne A. Fuller. 1981. Likelihood Ratio Statistics for Autoregressive Time Series with aUnit Root. Econometrica 49: 1057-1072.

[18] Doidge, Craig, G. Andrew Karolyi, and Rene M. Stulz. Has New York Become Less Competitive in GlobalMarkets? Evaluating Foreign Listing Choices over Time. Working Paper, Ohio State University.

[19] Easley, David, Robert F. Engle, Maureen OHara, and Liuren Wu. 2008. Time-Varying Arrival Rates of Informedand Uninformed Trades. Journal of Financial Ecnometrics 6(2):171-207.

[20] Easley, David, and Maureen OHara. 1987. Price, Trade Size, and Information in Securities Markets. Journal ofFinancial Economics 19: 69-90.

[21] Easley, David, and Maureen OHara. 1992. Time and the Process of Security Price Adjustment. The Journal ofFinance 47: 577-604.

[22] Easley, David, Soeren Hvidkjaer, and Maureen OHara. 2002. Is Information Risk a Determinant of Asset Returns.The Journal of Finance 57: 2185-2221.

[23] Easley, David, Nicholas M. Kiefer, and Maureen OHara. 1996. Cream-Skimming or Profit-Sharing? The CuriousRole of Purchased Order Flow. The Journal of Finance 51(3): 811-833.


27/50

[24] Easley, David, Nicholas M. Kiefer, and Maureen OHara. 1997. The Information Content of the Trading Process.Journal of Empirical Finance 4: 159-186.

[25] Easley, David, Nicholas M. Kiefer, and Maureen OHara. 1997. One Day in the Life of a Very Common Stock.The Review of Financial Studies 10(3): 805-835.

[26] Easley, David, Nicholas M. Kiefer, Maureen OHara and Joseph B. Paperman 1996. Liquidity, Information, andInfrequently Traded Stocks. The Journal of Finance 51(4): 1405-1436.

[27] Engle, R. F. and C. Granger. 1987. Co-integration and Error Correction: Representation, Estimation, and Test-ing. Econometrica 55(2): 251-276.

[28] Engle, R. F. and B. S. Yoo. 1987. Forecasting and Testing in Co-integrated Systems. Journal of Econometrics35: 143-159.

[29] Eun, C. and S. Sabherwal. 2003. Cross-Border Listings and Price Discovery: Evidence from U.S. Listed CanadianStocks. Journal of Finance 58: 549-574.

[30] Fleming, Jeff, Barbara Osdiek, and Robert E. Whaley. 1996. Trading Costs and the Relative Rates of PriceDiscovery in Stock, Futures, and Option Markets. Journal of Futures Markets 16: 353-387.

[31] Foerster, Stephen R., and Andrew G. Karolyi. 1998. Multimarket Trading Liquidity: A Transaction Data Analysisof Canada-U.S. Interlistings. Journal of Internatinal Financial Markets, Institions and Money 8: 393-412.

[32] Froot, K. A. and E. Dabora. 1999. How are Stock Prices Affected by the Location of Trade? Journal of Financial

Economics 53: 189-216.

[33] Gardner, A, and A. Subrahmanyam. 1994. Multi-Market Trading and the Informativeness of Stock Trades: AnEmpirical Intraday Analysis. Working Paper, Columbia University.

[34] Gagnon, L. and G. A. Karolyi. 2004. Multi-Market Trading and Arbitrage. Working Paper, Ohio State University.

[35] Glosten, L., and P. Milgrom. 1985. Bid, Ask and Transaction Prices in a Specialist Market with HeterogeneouslyInformed Traders. Journal of Financial Economics 14: 71-100.

[36] Grossman, S., and J. Stiglitz. 1980. On the Impossibility of Informationally Efficient Markets. American Eco-nomic Review 70: 393-408.

[37] Harris, Frederick H. deB.,Thomas H. McInish, and Robert A. Wood. 2002. Security Price Adjustment acrossExchanges: An Investigatino of Common Factor Components for Dow Stocks. Journal of Financial Markets 5:277-308.

[38] Harris, Frederick H. deB.,Thomas H. McInish, Gary L. Shoesmith, and Robert A. Wood. 1995. Cointegration,Error Correction and Price Discovery on Informationally Linked Security Markets. Journal of Financial andQuantitative Analysis 30: 563-579.

[39] Hasbrouck, Joel. 1991. Measuring the Information Content of Stock Trades. The Journal of Finance 46: 179-208.

[40] Hasbrouck, Joel. 1995. One Security, Many Markets: Determining the Contribution to Price Discovery. TheJournal of Finance 50(4): 1175-1199.

[41] Hasbrouck, Joel. 2007. Empirical Market Microstructure: The Institutions, Economics, and Econometrics ofSecurities Trading. New York: Oxford University Press.

[42] Hong, Yongmiao. 2006. Nonlinear Time Series Analysis: Econometric Theory and Analysis. Lecture Notes, CornellUniversity.

[43] Johansen, Sren. 1991. Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian Vector Autore-gressive Models. Econometrica 59(6): 1551-1580.

[44] Jones, Charles M., and Paul J. Seguin. 1997. Transaction Costs and Price Volatility: Evidence from CommissionDeragulation. American Economic Review 87: 728-737.

[45] Karolyi, G. Andrew. 2006. The World of Cross-Listings and Cross-Listings of the World: Challenging ConventionalWisdom. Review of Finance 10(1): 73-115.

[46] Karolyi, G. Andrew, and Stulz, Rene M. 1996. Why Do Markets Move Together? An Investigation of U.S.-JapanStock Return Comovements. The Journal of Finance 51(3): 951-86


28/50

[47] Kato, K., S. Linn, and J. Schallheim. 1991. Are There Arbitrage Opportunities in the Market for AmericanDepository Receipts? Journal of International Financial Markets, Institutions & Money 1: 73-89.

[48] Kim, M., A. C. Szakmary, and I. Mathur. 2000. Price Transmission Dynamics between ADRs and Their UnderlyingForeign Securities. Journal of Banking and Finance 24: 1359-1382.

[49] King, Michael R., and Dan Segal. 2004. International Cross-Listing and the Bonding Hypothesis. Working Paper,The Bank of Canada.

[50] Kyle, A. 1985. Continuous Auctions and Insider Trading. Econometrica 53: 1315-1335.

[51] Lee, Charles M. C. 1993. Market Integration and Price Execution for NYSE-Listed Securities. The Journal ofFinance 48(3): 1009-1038.

[52] Lee, Charles M. C. and Mark J. Ready. 1991. Inferring Trade Direction from Intraday Data. The Journal ofFinance 46(2): 733-746.

[53] Lee, Tae-Hwy, Halbert White and Clive W. J. Granger. 1993. Testing for Neglected Nonlinearity in Time SeriesModels: A Comparison of Neural Network Methods and Alternative Tests. Journal of Econometrics 56: 269-290.

[54] Maldonado, W. and A. Saunders. 1983. Foreign Exchange Futures and the Law of One Price. Financial Man-agement 12: 19-23.

[55] Merton, R. C. 1987. Presidential Address: A Simple Model of Capital Market Equilibrium with IncompleteInformation. The Journal of Finance 42: 483-510.

[56] Miller, P. D., and R. M. Morey. 1996. The Intraday Pricing Behavior of International Dually Listed Securities.Journal of International Financial Markets, Institutions and Money 6: 79-89.

[57] Park, J., and A. Tavokkol. 1994. Are ADRs a Dollar Translation of Their Underlying Securities? The Case ofJapanese ADRs. Journal of International Financial Markets, Institutions, and Money 4: 77-87.

[58] Sarkissian, S. and M. Schill. 2004. The Overseas Listing Decision: New Evidence of Proximity Preference. TheReview of Financial Studies 17(3): 769-809.

[59] Sengupta, Sunando. 2005. Implications of European Trading for the New York Stock Exchange Open. WorkingPaper, Arizona State University. Available at URL http://www.fma.org/SLC/Papers/NYSEOpen.pdf

[60] Seppi, Duane J. 1990. Equilibrium Block Trading and Asymmetric Information. The Journal of Finance 45:73-94.

[61] Sims, Christopher A. 1980. Macroeconomics and Reality. Econometrica 48: 1-48.

[62] Wahab, M., M. Lashgari, and R. Cohn. 1992. Arbitrage in the American Depository Receipts Market Revisited.Journal of International Markets, Institutions and Money 2.

[63] Wang, J. 1993. A Model of Intertemporal Asset Prices under Asymmetric Information. Review of EconomicStudies 6: 249-282.

[64] Werner, Ingrid M., and Allan W. Kleidon. 1996. U.K. and U.S. Trading of British Cross-listed Stock: An IntradyAnalysis of Market Integration. Review of Financial Studies 9: 619-664.

[65] Wilcoxon, F. 1945. Individual Comparisons by Ranking Methods. Biometrics 1: 80-83.

[66] Zhao, Xin and Kee H. Chung. 2006. Decimal Pricing and Information Based trading: Tick Size and InformationEfficiency of Asset Pricing. Working Paper, State University of New York at Buffalo.


29/50

Appendix A1: Tables

Table 1: tsx-listed cross-listers on the nyse

TSXlistedcrosslistersonNYSE NYSE US$*

CAN$

value* PINNYSE Spread NYSE IS NYSE TSX CAN$* PINTSX Spread TSX IS TSXAbitibiConsolidated,Inc. ABY 2.79 3.22 0.151 0.018 41.7% A 3.21 0.184 0.005 58.27%

AdvantageEnergyIncomeFund AAV 10.47 12.08 0.372 0.117 50.0% AVN.UN 12.24 0.482 0.131 49.95%AgnicoEagleMinesLimited AEM 36.22 41.78 0.188 0.026 50.0% AEM 42.11 0.421 0.080 50.01%

AgriumInc. AGU 37.48 43.23 0.190 0.020 43.0% AGU 43.24 0.202 0.007 56.96%

AlcanInc. AL 52.73 60.82 0.147 0.006 40.7% AL 60.91 0.169 0.003 59.29%

BankofNovaScotia(The) BNS 46.67 53.83 0.234 0.063 49.9% BNS 53.53 0.188 0.003 50.09%

BarrickGoldCorporation ABX 28.90 33.33 0.190 0.008 38.6% ABX 33.52 0.215 0.003 61.42%

BCEInc. BCE 27.71 31.96 0.112 0.006 49.1% BCE 30.13 0.174 0.002 50.85%

BiovailCorporation BVF 21.63 24.95 0.181 0.008 49.5% BVF 25.08 0.220 0.006 50.53%

BMOFinancialGroup BMO 60.80 70.13 0.160 0.007 41.4% BMO 70.90 0.204 0.002 58.65%

BrookfieldPropertiesCorporation BPO 39.78 45.88 0.267 0.020 45.3% BPO 45.68 0.226 0.016 54.67%

CamecoCorporation CCJ 40.47 46.68 0.223 0.020 38.0% CCO 46.15 0.197 0.009 62.02%

CanadianImperialBankofCommerce CM 87.05 100.40 0.308 0.017 49.9% CM 100.35 0.160 0.002 50.13%

CanadianNationalRailwayCompany CNI 44.23 51.01 0.139 0.007 48.4% CNR 51.81 0.215 0.003 51.61%

CanadianPacificRailwayLimited CP 55.88 64.45 0.206 0.007 43.8% CP 64.49 0.173 0.003 56.24%

CelesticaInc. CLS 5.99 6.91 0.186 0.010 44.4% CLS 6.98 0.225 0.005 55.63%

CGIGroupInc. GIB 8.66 9.99 0.195 0.028 49.9% GIB.A 0.280 0.018 50.13%

ComptonPetroleumCorporation CMZ 10.18 11.74 0.110 0.010 50.0% CMT 11.76 0.253 0.023 50.01%

CorusEntertainment,Inc. CJR 38.57 44.49 0.311 0.016 46.0% CJR.B 0.210 0.012 54.02%

CottCorporation COT 13.48 15.55 0.147 0.012 50.0% BCB 15.30 0.223 0.014 50.00%

DomtarCorporation UFS 9.18 10.59 0.199 0.010 50.0% DTC 0.206 0.007 50.00%

Energy

Metals

Corporation

(Listed

NYSE

Arca) EMU

12.00

13.84

0.203

0.059

50.0% EMC

13.71

0.274

0.047

49.98%EnerplusResourcesFund ERF 42.63 49.17 0.261 0.020 46.2% ERF.UN 49.43 0.286 0.019 53.82%

FairfaxFinancialHoldingsLimited FFH 225.79 260.43 0.308 0.012 50.0% FFH 260.40 0.254 0.007 50.04%

FourSeasonsHotelsInc. FS 80.98 93.40 0.202 0.009 46.5% FSH 0.214 0.009 53.50%

GildanActivewearInc. GIL 59.25 68.34 0.239 0.019 83.6% GIL 69.62 0.800 0.018 16.36%

GoldcorpInc. GG 24.25 27.97 0.354 0.072 41.4% G 28.54 0.178 0.011 58.58%

IntertapePolymerGroupInc. ITP 4.12 4.75 0.246 0.020 40.8% ITP 4.96 0.277 0.014 59.22%

IPSCOInc. IPS 129.82 149.73 0.301 0.027 48.7% IPS 136.44 0.215 0.010 51.29%

KinrossGoldCorporation KGC 13.63 15.72 0.247 0.059 44.8% K 15.94 0.231 0.012 55.15%

MagnaInternationalInc. MGA 75.42 86.99 0.153 0.006 42.6% MG.A 0.179 0.004 57.44%

ManulifeFinancialCorp. MFC 34.47 39.76 0.223 0.011 41.4% MFC 39.66 0.222 0.031 58.61%

MDSInc. MDZ 18.81 21.70 0.218 0.024 33.8% MDS 21.80 0.323 0.038 66.15%

MeridianGoldInc. MDG 24.98 28.81 0.205 0.042 42.7% MNG 28.95 0.267 0.019 57.33%

Nexen,Inc. NXY 62.07 71.59 0.168 0.014 44.9% NXY 70.92 0.160 0.004 55.15%

NortelNetworksCorporation NT 24.23 27.95 0.205 0.006 47.9% NT 28.31 0.188 0.002 52.10%

NOVAChemicalsCorporation NCX 31.26 36.06 0.245 0.015 38.8% NCX 36.18 0.275 0.006 61.18%

PengrowthEnergyTrust PGH 17.09 19.71 0.247 0.025 49.2% PGF.UN 0.183 0.007 50.82%

PetroCanada PCZ 39.21 45.22 0.238 0.017 42.5% PCA 45.18 0.196 0.004 57.54%

PotashCorporationofSaskatchewanInc. POT 156.97 181.05 0.128 0.007 44.2% POT 179.74 0.170 0.005 55.84%

PrecisionDrillingTrust PDS 23.01 26.54 0.167 0.011 36.5% PD.UN 0.190 0.005 63.49%

Quebecor

World,

Inc. IQW

12.61

14.54

0.230

0.013

45.4% IQW

14.40

0.183

0.004

54.64%RBCFinancialGroup RY 50.22 57.92 0.163 0.007 46.3% RY 58.26 0.173 0.002 53.69%

RogersCommunicationsInc. RG 32.24 37.19 0.179 0.017 37.4% RCI.B 36.96 0.238 0.006 62.59%

ShawCommunicationsInc. SJR 37.38 43.11 0.195 0.012 49.2% SJR.B 0.187 0.007 50.76%

StantecInc. SXC 27.31 31.50 0.158 0.010 50.0% STN 35.51 0.394 0.020 50.00%

SunLifeFinancial,Inc. SLF 45.46 52.43 0.233 0.042 50.0% SLF 52.40 50.00%

SuncorEnergyInc. SU 77.22 89.07 0.185 0.010 47.4% SU 89.17 0.184 0.004 52.63%

TalismanEnergyInc. TLM 17.80 20.53 0.190 0.013 39.4% TLM 20.21 0.164 0.005 60.64%

TELUSCorporation TU 49.94 57.60 0.199 0.014 43.0% T.A 25.04 0.228 0.005 57.04%

TheThomsonCorporation TOC 41.96 48.40 0.290 0.034 49.6% TOC 48.61 0.175 0.005 50.40%

TimHortonsInc. THI 30.50 35.18 0.202 0.017 50.0% THI 34.90 0.536 0.124 50.04%

TorontoDominionBank TD 60.32 69.57 0.152 0.010 26.0% TD 69.68 0.203 0.002 74.05%

TransAltaCorporation TAC 21.58 24.89 0.308 0.081 49.9% TA 25.04 0.180 0.005 50.05%rans ana a orporation 33.37 38.49 0.157 0.012 48.6 38.69 0.211 0.004 51.37

*ClosingpriceasofMarch28,2007

NoteI.PINsandspreads(=(askbid)/[(ask+bid)/2])arearithmeticmeansofmonthlyestimates,January1998throughDecember2000.

NoteII.InformationsharesarenumuericallycomputedbyVectorErrorCorecctionModel,January1998throughDecember2000.

The pin is the probability of informed trading, following Easley, Kiefer, OHara, and Paperman (1996). The

bid-ask spreads are defined: 1. spreadnyse asknyse bidnyse(asknyse +bidnyse)/2; and 2. spreadtsx asktsx bidtsx(asktsx +bidtsx)/2

.

The information share is exchange-specific relative contribution to price discovery of a security traded onmultiple exchanges, following Hasbrouck (1995, 2007).


30/50

Table 2: The Wilcoxon signed-rank test results

pin Spread Information Share

H0 pintsx = pinnyse spreadtsx = spreadnyse istsx = isnyseH1 pintsx > pinnyse spreadnyse > spreadtsx istsx > isnysed pintsx(i, t) pinnyse(i, t) spreadnyse(i, t) spreadtsx(i, t) istsx(i) isnyse(i)

V0 424250 680698 1413p-value 0.001458 < 2.2 1016 3.99 109

The pin is the probability of informed trading, following Easley, Kiefer, OHara, and Paperman (1996). The

bid-ask spreads are defined: 1. spreadnyse asknyse bidnyse

(asknyse +bidnyse)/2; and 2. spreadtsx

asktsx bidtsx(asktsx +bidtsx)/2

.

The information share is exchange-specific relative contribution to price discovery of a security traded onmultiple exchanges, following Hasbrouck (1995, 2007). The Wilcoxon signed-rank test is a non-parametricpair-wise comparison test, following Wilcoxon (1945). The coordinates (i, t) denote each firm and eachmonth, respectively. d is a differential measure defined for the estimates of each quantity of interest. Theyare defined as: 1. d(i, t) pintsx(i, t) pinnyse(i, t); 2. d(i, t) spreadnyse(i, t) spreadtsx(i, t); and3. d(i) istsx(i) isnyse(i). The Wilcoxon test-statistic is defined as: V0

{(i,t)} 1{d(i,t)>0} it,

where it is the rank of{|d(i, t)|}.


31/50

Table 3: A cross-sectional association of speedconv against key variables

speedconv = 1 pinavg + 2 spreadavg + 3 size + 4 industry + 5 volume +

Model 1 Model 2 Model 3

pin 1.281 0.919 1.138

s.e. 0.341 0.369 0.330t 3.761 2.488 3.445p-value 0.000 0.017 0.001

spread 4.606 3.034s.e. 2.323 2.355t 1.982 1.288p-value 0.054 0.205

size 0.021 0.006 0.074s.e. 0.249 0.240 0.236t 0.085 0.024 0.313p-value 0.933 0.981 0.756

industry 0.165 0.205 0.227

s.e. 0.076 0.075 0.074t 2.168 2.717 3.058p-value 0.036 0.009 0.004

volume 0.406 0.483s.e. 0.191 0.183t 2.123 2.643p-value 0.040 0.011Adj. R2 0.606 0.635 0.629

Following Gagnon and Karolyi (2004), I estimate the reciprocal speed parameter () of the parity-convergence of cross-listing premium. According to their model, each firm is relative cross-listing pre-mium can be explained by its first-order lag, and its respective time-distributed risk exposure to thereturns on the host and home markets and the foreign exchange rate: DRi(t) = i + iDRi(t 1) +

1j=1

USj R

USM (t+j) +

1j=1

Cj R

CM(t+j) +

1j=1

FXj RFX(t+j) + i(t). The forward-lag is due

to information leakages and market impact, and is deemed standard in event models. For each cross-listedpair, i, 1. speedconv ( i) measures the reciprocal speed of convergence to the parity, following Gagnonand Karolyi (2004); 2. pinavg is the arithmetic average of the pins of the pair on the tsx and the nyse;3. spreadavg is the arithmetic average of the bid-ask spreads of the pair on the tsx and the nyse; 4.size is the proxy of normalized firm size and defined as the average market capitalization on the tsx andthe nyse; 5. industry equals one if the cross-lister is a manufacturing firm, and zero otherwise; and 6.volume is the normalized per-trade average volume of the pair on the tsx and the nyse.


32/50

Table 4: A cross-sectional association of average spread against key variables

spreadavg = 1 pinavg + 2 speedconv + 3 size + 4 industry +

Estimate s.e. t p-value

pinavg 0.074 0.022 3.415 1.38103

speedconv 0.0178 0.009 1.982 0.054

size 0.017 0.012 1.421 0.163industry 0.002 0.005 0.351 0.727Adj. R2 0.657

For each cross-listed pair, i, 1. speedconv (= (i)) measures the reciprocal speed of the parity-convergenceof cross-listing premium, following Gagnon and Karolyi (2004); 2. pinavg is the arithmetic average of thepins of the pair on the tsx and the nyse; 3. spreadavg is the arithmetic average of the bid-ask spreadsof the pair on the tsx and the nyse; 4. size is the proxy of firm size and defined as the average marketcapitalization on the tsx and the nyse; and 5. industry equals one if the cross-lister is a manufacturingfirm, and zero otherwise.


33/50

Table 5: tsx-listed cross-listers on the nyse, 1998 through 2000

Company Industry TSX Code TSX Listing NYSECode NYSEListing ListingSequence

CelesticaInc. ElectricalandElectronicProducts CLS 707,1998 CLS 630,1998 NYSETSX

ShawCommunicationsInc. Communications&Media SJR.B 325,1983 SJR 701,1998 TSXNYSE

NOVAChemicalsCorporation Chemicals NCX 703,1998 NCX 706,1998 TSXNYSE

CGIGroupInc. Consulting GIB.A 421,1992 GIB 1007,1998 TSXNYSEBrookfieldPropertiesCorporation PropertyManagementandInvestment BPO 627,1985 BPO 602,1999 TSXNYSE

IntertapePolymerGroupInc. PackagingandContainers ITP 106,1993 ITP 816,1999 TSXNYSE

GildanActivewearInc. HouseholdGoods GIL 624,1998 GIL 901,1999 TSXNYSE

ManulifeFinancialCorp. Insurance MFC 930,1999 MFC 924,1999 NYSETSX

SunLifeFinancial,Inc. Insurance SLF 329,2000 SLF 323,2000 NYSETSX

MDSInc. MedicalServices MDS 625,1973 MDZ 407,2000 TSXNYSE

CorusEntertainment,Inc. EntertainmentServices CJR.B 903,1999 CJR 510,2000 TSXNYSE

CanadianNaturalResources,Ltd. OilandGasProducers CNQ 514,1976 CNQ 731,2000 TSXNYSE

TELUSCorporation TelephoneUtilities T.A 201,1999 TU 1017,2000 TSXNYSE

Nexen,Inc. OilandGasProducers NXY 714,1971 NXY 1114,2000 TSXNYSE

EnerplusResourcesFund*** OilandGasProducers ERF.UN 311,1987 ERF 1117,2000 TSXNYSE

Company 3M +3M 6 M + 6M B efo re A ft er 3M +3M 6M +6M Before After

CelesticaInc. 0.186

ShawCommunicationsInc. 0.237 0.164

NOVAChemicalsCorporation 0.329 0.326 0.329 0.268 0.006 0.007 0.006 0.007 0.006 0.007

CGIGroupInc. 0.183 0. 28 3 0 .17 6 0 .27 7 0. 25 6 0. 22 6 0.268 0.151 0.181 0.123 0.150 0.055

BrookfieldPropertiesCorporation 0.223 0.206 0.068 0.194 0.020 0.017 0.069 0.016IntertapePolymerGroupInc. 0.218 0. 24 7 0 .20 9 0 .20 9 0. 26 6 0. 26 2 0.034 0.025 0.026 0.025 0.027 0.031

GildanActivewearInc.

ManulifeFinancialCorp. 0.035 0.150 0.003 0.004 0.221 0.003

SunLifeFinancial,Inc.

MDSInc. 0.156 0. 19 2 0 .15 6 0 .15 4 0. 10 2 0. 238 0.008 0.009 0.008 0.008 0.097 0.008

CorusEntertainment,Inc. 0.098 0. 21 2 0 .13 4 0 .18 0 0. 06 7 0. 201 0.029 0.051 0.028 0.042 0.025 0.036

CanadianNaturalResources,Ltd. 0.142 0. 12 7 0 .15 9 0 .34 8 0. 15 2 0. 12 8 0.004 0.003 0.004 0.004 0.007 0.004

TELUSCorporation 0.120 0.338 0.559 0.047 0.336 0.004 0.005 0.006 0.005 0.006 0.005

Nexen,Inc. 0.100 0. 16 3 0 .10 0 0 .16 3 0. 01 8 0. 13 4 0.009 0.005 0.009 0.005 0.009 0.005

EnerplusResourcesFund***

Average 0.168 0.235 0.180 0.259 0.112 0.207 0.045 0.028 0.033 0.024 0.062 0.017

Spread Spread** PIN PIN PIN* Spread

Company [2,+2] [ 5,+5] [10,+10] [ 10,+250]

CelesticaInc.

ShawCommunicationsInc. 0.002 0.156 0.242 0.024

NOVAChemicalsCorporation

CGIGroupInc. 0.204 0.269 0.204 0.757

BrookfieldPropertiesCorporation 0.045 0.041 0.075 0.358IntertapePolymerGroupInc. 0.031 0.040 0.083 0.740

GildanActivewearInc. 0.046 0.029 0.124 0.477

ManulifeFinancialCorp.

SunLifeFinancial,Inc.

MDSInc. 0.018 0.006 0.037 0.341

CorusEntertainment,Inc. 0.033 0.087 0.047 0.684

CanadianNaturalResources,Ltd. 0.042 0.011 0.019 0.287

TELUSCorporation 0. 027 0 .0 33 0.011 0.615

Nexen,Inc. 0.025 0 .01 2 0 .00 1 0.364

EnerplusResourcesFund*** 0. 014 0 .0 07 0.027 0.287

Average 0.017 0.020 0.020 0.444

**Arithmeticmeanofmonthlyspreadestimates

***PriortoJuneof2001,EnerplusResourcesFundtradedunderERF.G.UponthemergerwithEnerMark,thesymbolbecameERF.UN.

CumulativeAbnormalReturn

*ArithmeticmeanofmonthlyPINestimates.ForderivationandestimationalgorithmofPIN,seeAppendixA3.

The pin is the probability of informed trading, following Easley, Kiefer, OHara, and Paperman (1996).

The bid-ask spreads are defined as: 1. spreadnyse asknyse bidnyse

(asknyse +bidnyse)/2 ; and 2. spreadtsx asktsx bidtsx

(asktsx +bidtsx)/2. When estimating the cumulative abnormal returns (cars) around cross-listings on the

nyse, 1. the market model uses the s&p tsx Composite Index as the explanatory market return throughthe pre-run-up period ([250,11]) prior to listings; then 2. the products of gross residuals within theevent windows are subtracted by one to yield cars.


34/50

Table 6: The Wilcoxon signed-rank test results of cross-listing effect on the pin on the tsx

Window 3 months 6 months Threshold

H0 pin+3m = pin3m pin+6m = pin6m pinafter = pinbeforeH1 pin+3m > pin3m pin+6m > pin6m pinafter > pinbefored pin+3m pin3m pin+6m pin6m pinafter pinbefore

V0 33 14 30p-value 0.01953 0.05282 0.05469

The pin is the probability of informed trading, following Easley, Kiefer, OHara, and Paperman (1996). dis a differential measure defined for the estimates of each quantity of interest. The Wilcoxon signed-ranktest is a non-parametric pair-wise comparison test, following Wilcoxon (1945). The Wilcoxon test-statisticis defined as: V0

{(i,t)} 1{d(i,t)>0} it, where it is the rank of{|d(i, t)|}, and the coordinates (i, t)

denote each firm and each period, respectively.

Table 7: The Wilcoxon signed-rank test results of cross-listing effect on bid-ask spread


H0 spread+3m = spread3m spread+6m = spread6m spreadafter = spreadbeforeH1 spread+3m < spread3m spread+6m < spread6m spreadafter < spreadbefored spread3m spread+3m spread6m spread+6m spreadbefore spreadafter

V0 45 48 72p-value 0.34820 0.25740 0.05260

The bid-ask spreads are defined as: 1. spreadnyse asknyse bidnyse

(asknyse +bidnyse)/2; and 2. spreadtsx

asktsx bidtsx(asktsx +bidtsx)/2

. The Wilcoxon signed-rank test is a non-parametric pair-wise comparison test, following

Wilcoxon (1945). d is a differential measure defined for the estimates of each quantity of interest. TheWilcoxon test-statistic is defined: V0

{(i,t)} 1{d(i,t)>0} it, where it is the rank of{|d(i, t)|}, and

the coordinates (i, t) denote each firm and each period, respectively.

Table 8: The Wilcoxon signed-rank test results of cross-listing effect on volume


H0 volume+3m = volume3m volume+6m = volume6m volumeafter = volumebeforeH1 volume+3m < volume3m volume+6m < volume6m volumeafter < volumebefored volume3m volume+3m volume6m volume+6m volumebefore volumeafter

V0 42 39 58p-value 0.7293 0.6285 0.9433

volume is per-trade number of the tsx-listed stock of a nyse-cross-listed Canadian firm. The Wilcoxonsigned-rank test is a non-parametric pair-wise comparison test, following Wilcoxon (1945). d is a differentialmeasure defined for the estimates of each quantity of interest. The Wilcoxon test-statistic is defined:V0

{(i,t)} 1{d(i,t)>0} it, where it is the rank of {|d(i, t)|}, and the coordinates (i, t) denote each

firm and each period, respectively.


35/50

Appendix A2: Figures

Figure 1: Monthly estimates of pin on tsx and nyse

0.15

0.20

0.25

0.30

0.35

0.40

9801

9803

9805

9807

9809

9811

9901

9903

9905

9907

9909

9911

0001

0003

0005

0007

0009

0011

PIN

TSX

annuala

vg.

NYSE

annuala

vg.

The pin is the probability of informed trading, following Easley, Kiefer, OHara, and Paperman (1996).Figure 1 shows the average monthly pin of the sample firms co-listed on the tsx and the nyse. The annualestimates for the pin on the tsx are {0.242, 0.213, 0.206} in 1998, 1999, and 2000, respectively, while thecorresponding estimates for the nyse are {0.204, 0.212, 0.196}, over the same period.


36/50

Figure 2: Impulse response function plotscross-border responses of quote changes


37/50

Each of the above four consecutive impulse response function plots of Aibiti Consolidate (co-listed on thetsx and on the nyse) specifies the source of innovation by two standard deviations. The quotes on thenyse rarely affect the quotes on the tsx. To the contrary, positive increases in ask and bid prices on thetsx are followed by changes in ask and bid prices on the nyse, respectively.


38/50

Figure 3: Cross-listing effects on the pin on the tsx, six-month ([-3,+3]) window

Above scatter plot describes various coordinates of the pin on the tsx before (horizontal axis) and after(vertical axis) nyse-listing. A coordinate in the upper 45-line region denotes a rise in the pin, whereasone in the lower region a decline.


39/50

Figure 4: Cross-listing effects on the pin on the tsx, twelve-month ([-6,+6]) window



40/50

Figure 5: Cross-listing effects on the pin on the tsx, threshold monthly



41/50

Appendix A3: PIN estimation algorithm

My pin estimation algorithm is based on symmetric Poisson intensity for arrivals of both unin-formed buyers and sellers. Information events occur at the market open with probability and, on arealization of such event, informed traders who arrive with intensity perceive a binary signal withprobability either P {share price falls} or 1 = P {share price rises}.

The probability of informed trading (pin) is the relative degree of private informationor a mea-sure of asymmetric informationweighed on a randomly chosen transaction executed by an informedtrader

pin

E [B(uy) + S(ell)]=

+ b + s=

+ 2 ,

given the assumed symmetry of intensity in uninformed trader arrivals, either buyers or sellers. (SeeFigure 6.) Empirically, a trade is considered buyer-initiated if it is higher than the five-second earliermid-quote, or seller-initiated if lower (Lee and Ready (1991)).

I adopt a log-likelihood factorization from Easley, Engle, OHara, and Wu (2008) as follows

L ln P{Bt, St}

Tt=1 |,,,

=T

t=1

[2 + M ln (x) + (Bt + St) ln ( + )]

+T

t=1

ln

(1 ) exp () xStMt + exp() xBtMt + (1 ) xBt+StMt

,

where 1. Mt min(Bt,St)+max(Bt,St)

2 ; and 2. x

+ . Thus, the parameters are estimated bymaximum likelihood method such that

, , ,

= arg max

L | (, ) > 0, (, ) [0, 1]2

,

hence the resulting pin estimator is

pin = + 2

.

Figure 6: Derivation of pin


42/50

Appendix A4: Information shares of stock exchanges

Consider a Canadian cross-listed pair (pt, pn) traded (fragmented) on the tsx (t) and the nyse (n).The time series of the pair shares a common efficient price41 (mt) such that

pt,t

pn,t = 1

1 mt + ct qt,t

cn qn,t ,where ct and cn, and qt and qn are market-specific cost coefficients and their associated trade volumes,respectively. Trade directions in the two markets may be contemporaneously associated as

Var

qt,tqn,t

=

1 q

q 1

.

An attractive trait of the common efficient price is that the securities with same underlying assetstraded on distinct exchanges are linked by no-arbitrage condition in equilibrium. An implied-vectormoving average (vma) formulation for the differences of prices is

pt,t

pn,t = t,t

n,t + qt,t

qn,t = tt tn

nt nn t,t1

n,t1 ,then

Et

pt,t+1pn,t+1

=

pt,tpn,t

+

tt tnnt nn

t,t1n,t1

,

thus

Et

pt,t+1pn,t+1

Et1

pt,tpn,t

=

pt,tpn,t

+

tt tnnt nn

t,t1n,t1

.

given that the two prices share the same efficient underlying price (1 + tt, tn) = (nt, 1 + nn) .

Following Eun and Sabherwal (2003), the augmented Dickey-Fuller (1981) unit root test is conductedto each daily-price time series of the 55 tsx-nyse cross-listed pairs with appropriate lag length, perAkaike (1974), to verify first-order integration (I(1)). Applying Johansens (1991) either trace test oreigen-value test yielded one cointegrating42 equation for each tsx-nyse pair.

As a result, an econometric impasse is that since the cross-listed pairs are cointegrated, a vectormoving average (vma) representation cannot be recovered by Simss (1980) vector autoregressive (var)structural formulation. Subsequently, in the absence of accounting for sources of shocks to fragmentedshares, decomposing exchange-specific relative contribution to price discovery of the tsx-nyse pairsposes an unwieldy task.

A breakthrough was introduced by Engle and Granger (1987) and Engle and Yoo (1987), andHasbrouck (1995) adopts their error correction model (ecm) to arrive at information shareanexchanges percentage share in price discovery of shares whose orders are executed from many markets.The vector error correction model (vecm) for the cointegrated trade-level quote prices is

pt = (L)pt + ( zt1) + t,41A security price time-series

{mt}

t=0

is efficient if, by definition, the conditional expectation of the first-order

difference is zero. In other words, an efficient price is unpredictable given the presently available information. Equiva-lently, the increment of the price follows a martingale difference sequence: mt = mt1 +ut E

mt| {ms1}

ts=1

=

Eut| {ms1}

ts=1

= 0. See Lee, White, and Granger (1993).

42Security prices are cointegrated if there exists a linear combination of the non-stationary prices that can be tonedstationary. A time series is strongly stationary if its probability distribution is time-invariant, and weakly stationary ifup to its second momentsmean, variance, and covariance.


43/50

where 1. (L)pt are vector autoregressive terms; 2. is a vector of cointegrating coefficients; 3. > 0 is a vector of long-run cross-border bid-ask dollar spreads; and 4. zt is a vector of cross-borderdollar spreads in bid

pbt,t

A Trans-Niagara Tale Of Informed Traders

Documents

Transcript of A Trans-Niagara Tale Of Informed Traders