big players’ aggregated trading and market returns in istanbul stock exchange
TRANSCRIPT
This article was downloaded by: [Universitat Politècnica de València]On: 21 October 2014, At: 04:25Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,37-41 Mortimer Street, London W1T 3JH, UK
Applied Financial EconomicsPublication details, including instructions for authors and subscription information:http://www.tandfonline.com/loi/rafe20
Big players’ aggregated trading and market returns inIstanbul Stock ExchangeNuman Ülkü aa Central European University, Business School , Frankel Leo ut. 30-34, Budapest 1023,HungaryPublished online: 22 Nov 2011.
To cite this article: Numan Ülkü (2012) Big players’ aggregated trading and market returns in Istanbul Stock Exchange,Applied Financial Economics, 22:6, 491-508, DOI: 10.1080/09603107.2011.619492
To link to this article: http://dx.doi.org/10.1080/09603107.2011.619492
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) containedin the publications on our platform. However, Taylor & Francis, our agents, and our licensors make norepresentations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of theContent. Any opinions and views expressed in this publication are the opinions and views of the authors, andare not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon andshould be independently verified with primary sources of information. Taylor and Francis shall not be liable forany losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoeveror howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use ofthe Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematicreproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in anyform to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions
Applied Financial Economics, 2012, 22, 491–508
Big players’ aggregated trading and
market returns in Istanbul Stock
Exchange
Numan Ulku
Central European University, Business School, Frankel Leo ut. 30-34,
Budapest 1023, Hungary
E-mail: [email protected]
This study uses a special data set, derived from member brokers’
transactions, as a proxy for big players’ trading. Big players as represented
by this variable include institutional, big individual and foreign traders,
and these groups are not mutually exclusive. The interaction between big
players’ trading and market returns is analysed using a structural Vector
Autoregressive (VAR) model. Big trader flows are strongly positively
associated with contemporaneous returns, exhibit persistence (possibly
indicative of herding), positive feedback trading and little forecast ability.
The tendency to herd is stronger than to positive feedback trade. Big
players’ trading is correlated with information, and the apparent positive
feedback trading seems to result from delayed response to information
rather than naively following past returns. Asymmetric price impact of
buys versus sells is driven by the underlying market conditions.
Keywords: market microstructure; big players’ trading in stock markets;
feedback trading; price impact
JEL Classification: G14; G15
I. Introduction
Literature related to big players’ trading in stock
markets consists of several different paths: trades
sorted by size, block trades, institutional trading and
foreigners’ trading. These paths are not mutually
exclusive: for example, Griffin et al. (2003) document
an overlap between being institutional and having
large trade size; several articles such as Campbell
et al. (2009) use ‘large-size’ as a proxy for ‘institu-
tional’ (see Lee and Radhakrishna, 2000); foreign
investors are mostly institutional; and big individual
traders behave like institutions (Ng and Wu, 2007).
The current article pertains to an intersection of these
paths. The common conclusion of these strands of
literature is that big players, whether they are wealthy
individuals, institutions or foreign investors, do have
a strong impact on stock prices. They are more
likely to herd together and pursue positive feedback
trading, which do not appear to be irrational. Some
studies find forecast ability in big players’ trading,
however this does not necessarily translate into entry-
to-exit profitability.The significant price impact along with tendencies
of positive feedback trading and herding makes big
players price setters, hence raises important issues to
be investigated: Is their price impact temporary or
permanent (i.e. reflects price pressure or informa-
tion)? Do they destabilize stock markets by herding
together and pursuing positive feedback trading
Applied Financial Economics ISSN 0960–3107 print/ISSN 1466–4305 online � 2012 Taylor & Francis 491http://www.tandfonline.com
http://dx.doi.org/10.1080/09603107.2011.619492
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014
strategies? Does their trading bear forecast ability?
Does their price impact differ in buys and sells?However, the literature that investigates the interac-tion between big players’ trading and stock returns is
handicapped by limited availability of trading datawith trader identity. Research is confined to specificsamples privately obtained from stock exchanges (e.g.
short sample periods, reported block trades), quar-terly or annual institutional holdings or quite imper-fect proxies.
This study uses a special type of data on Istanbul
Stock Exchange (ISE) that permits to identify biginvestors’ aggregated net trading endogenously.Unlike small samples and delayed availability in the
extant literature, these data are publicly availablecontinuously, on an intraday real-time basis for asmall fee. These data are used by market participants,
including big individual speculators and professionalfund managers, to infer big players’ trading in ISE.Specifically, the data set contains buying and selling
value of member brokers. In this study, market-wide-aggregated figures at the daily frequency are used.1
While the link from brokers’ cumulative trading
volume to identification of big players may seemindirect at first sight, the derivation explained inSection III ensures that these data pick big players’
net trading. The category of big players as picked bythese data includes institutions, big individual tradersand foreign investors. In fact, as extant literature
documents, these categories are not mutually exclu-sive, especially as far as their behaviour is concerned.
Given the repeated failure of economic models toexplain, let aside to forecast, exchange rate changes,
Lyons (2001) proposes order flow analysis as aneffective tool to explain and, to some degree, forecastexchange rate changes. That tool, however, is only
available to major dealers. The data employed in thisstudy can be considered as the stock-market counter-part of the order flow approach to Foreign Exchange
(FX) markets, with the additional advantage of beingavailable to the public on a real-time basis.2
The unique features of this data set help contributeto the literature in several ways: first, in earlier studies
of trades sorted by trade size, potential links betweenorders (i.e. serial orders) are omitted, and fixed
arbitrary trade size categories ignore variation intrade size resulting from big traders’ dynamic stealth
trading tactics. Menkhoff and Schmeling (2010)provide evidence that medium-size trades of largetraders convey the most information, and the relationbetween trade size and permanent price impact isnonlinearly intermediated by trader size. Moreover,
the relation between trade size and trader size hasbroken in recent years as a result of increased ordersplitting thanks to computerized trading (Hvidkjaer,2008; Campbell et al., 2009). The data used in thisstudy identify big net traders without referring to thesize of individual trades, and critically capture the
interaction between trade size and trader size.Second, most empirical studies in the literature areconfined to specialist dealer systems. With theadvance of computerized trading, all major stockmarkets are now migrating towards electronic con-
tinuous auction systems with no specialist dealers,where not only order execution strategies but also,and perhaps more importantly, the mechanism bywhich the information content of trades is incorpo-rated into price might differ. However, there are only
few studies of big players’ trading under the contin-uous limit-order book, blind-matching system. Asunder the specialist system an order’s price impactdepends on specialist’s assessment of its informationcontent and inventory effect, it needs to be seenwhether the results would differ under blind electronic
systems without a specialist where the ‘crowd’ issupposed to fulfill the same function (see Bloomfieldet al., 2005). The current article fills this gap. Third, thesample period corresponds to a symmetric and signif-icant V-shaped price action around the recent global
crisis, enabling clean empirical tests of long-debatedhypotheses about the price impact asymmetry such asChiyacanthana et al. (2004) and Saar (2001). Fourth,this article fills a gap as there is a scarcity of studies onbig players’ trading in European, especially emergingEuropean, stock markets.3 Finally, the way our key
variable is derived from broker-level data, thoughsimple, is new to the literature, and can inspire similarstudies in other markets.
This study employs a structural VectorAutoregressive (VAR) model to portray the dynamic
1 See Chordia et al. (2002) for a market-wide-aggregated study of order imbalance at the daily frequency on New York StockExchange (NYSE). However, Chordia’s article differs from the current one in that the order imbalance variable is actually aproxy for market orders (active orders executed against the limit order book), while the variable derived here is a proxy for bigplayers’ trading. In other words, they focus on trading aggressiveness, while this study focuses on size. Interestingly, bothvariables exhibit similarities such as positive relation to current returns and persistence, which may be an indication of bigplayers being more likely to use market orders. See Visaltanachoti and Luo (2009) for a study of order imbalance on Thailand.2 Interestingly, in October 2010 while this article was under review, ISE administration stopped making these data available topublic. Currently, the appropriateness of this decision is heavily debated in the media. Hence, findings of this study maypotentially guide policy makers.3One exception is Voronkova and Bohl (2005), who use data on semi-annual and annual holdings of pension funds in Poland.
492 N. Ulku
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014
interaction between big players’ trading and marketreturns, augmented with world market returns whichenter the system exogenously. As global marketreturns are a relevant information variable with highexplanatory power on ISE returns, particularly duringthe sample period which corresponds to the recentglobal crisis, our specification enables to condition thereturn-flow interaction with an information variable.This provides a unique contribution to the literatureby distinguishing between positive feedback tradingand delayed reaction to information.
Results suggest a strong positive contemporaneousprice impact, which confirms, under a differenttrading system, previous conjectures that link infor-mation content to trade- or trader size. By the end ofthe trading day, the information contained in bigplayers’ net buying is almost fully reflected in marketprices although there is no specialist dealer whoderives information from order size, sequence andidentity of traders. It takes longer, however, for bigplayers’ selling to be fully priced-in, probably becauseof the practical absence of short selling in ISE. Thisindirectly suggests that some market participants inISE infer information from observing trades (possiblyusing this data set) to accelerate the incorporation ofinformation contained in trades, thus fulfilling theinformational role of a specialist.
Big player flows exhibit persistence, which may beindicative of herding. The relation of current bigplayer flows to past big trader flows is stronger thanto past returns. Moreover, big player flows are muchmore strongly affected by past world returns than bypast local returns. As world returns are considered ahighly-relevant information variable during oursample period, these results can be interpreted asbig players’ responding to information rather thannaively to past returns.
Thanks to the V-shaped price action during oursample period, this study provides a clean confirma-tion of Chiyachantana et al.’s (2004) suggestion thatthe asymmetry in price impact documented in earlierstudies may have simply been driven by the under-lying market conditions.
Section II relates this article to the extant literature.Section III describes the unique data set employedin this study and the methodology. Results arepresented in Section IV, and main conclusions aresummarized in Section V.
II. Related Literature
In the microstructure literature, large-size trades havebeen associated with significant price impact and
informed trading both theoretically and empirically(Easley and O’Hara, 1987; Easley et al., 1997).Barclay and Warner (1993) modify this associationwith the stealth trading hypothesis (i.e. under aspecialist dealer system, privately informed traderswill mostly concentrate in medium-size trades forstrategic reasons). Chakravarty (2001) confirms thestealth trading hypothesis on a 63-day TORQ sample,and further documents that most of the price impactis due to medium-size trades of institutions. Chan andFong (2000) find that the order imbalance in largetrade-size categories affects returns more than that insmaller trade-size categories. Using a 6-day sample ofall trades with trader identity from the MICEXcurrency exchange, Menkhoff and Schmeling (2010)show that most of the price impact is due to medium-size trades of big traders. The literature on regulartrades sorted by size employs special short samples oftrading data privately obtained from stockexchanges.
Major international replications of the analysis oftrades sorted by size are on China and Taiwan, whereindividual investors dominate. Using a VAR model,Lee et al. (1999) find that big individual trades inTaiwan Stock Exchange (TSE) are strongly positivelycorrelated with contemporaneous returns, leadreturns over the next 15-minutes interval, and arethemselves independent of past returns or any othercategory of trades. Small individual trades exhibitcontrarian behaviour. Using detailed audit trail datafrom Shanghai Stock Exchange for April 2001–August 2002, Ng and Wu (2007) find that institutionsand large-size individuals exhibit momentum trading,while small-size individuals exhibit contrarian trad-ing. Only trading by institutions and largest-sizeindividuals affect future volatility, while none of theinvestor groups’ trading has forecast ability. Usingthe same data set, Wongchoti et al. (2009) confirmthat when past market returns are high, investorswith larger (smaller) trade-size tend to buy (sell).These results are relevant for the setup of this study asthey document that large individual traders behavelike institutions.
The literature on block trades documents a strongprice impact proportional to trade size, and a priceimpact asymmetry: the price impact of block pur-chases is larger in magnitude and more permanentthan block sales, suggesting that the former might beassociated with information whereas the latter mightreflect price pressure (Kraus and Stoll, 1972;Holthausen et al., 1987, 1990). Seppi (1992) docu-ments private information content of block tradesprior to earnings announcements. Bozcuk and Lasfer(2005) find that trade size, trader’s resulting level ofownership and the type of investor behind the trade
Big players’ aggregated trading and market returns in Istanbul Stock Exchange 493
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014
are major determinants of the permanent price
impact. Keim and Madhavan (1996) document
significant price movements up to 4 weeks prior toupstairs block trades, positively related to trade size.
The only analysis of block trades under a nondealer
auction system, Ball and Finn (1989) on Sydney, findsno post-block reversals, which supports the informa-
tion hypothesis. Moreover, there is no significantrelationship between block size and post-block rever-
sal which rules out price pressure hypothesis. The
continuation following sales is stronger, similar toour finding in this article.
Clearly, institutional traders and trades are of large
size, and constitute a major component of big players.
The literature on the interaction between institutionaltrading and stock returns is mainly confined to US
(using quarterly holdings data) and Asian markets.
The earliest study of institutional trading, Klemkosky(1977), documents the price impact of large institu-
tional trading imbalances. Chan and Lakonishok(1995) analyse trades of 37 large investment manage-
ment firms, using trade packages (sequence of trades)
as the unit of analysis. They document a price impactasymmetry: average price impact is about 1% for
buys and �0.35% for sells, and the subsequent
reversal for buys is much smaller. The price impactis proportional to trade-size. Analysing orders of 21
institutions, Keim and Madhavan (1995) find hetero-
geneity in feedback trading strategies varying withinvestment style, with the overall effect likely to be
offsetting. Feedback trading is often not symmetric
for buys versus sells, and buys take longer to execute.Nofsinger and Sias (1999) report strong positive
contemporaneous relationship between stock returns
and annual changes in institutional ownership, whichis not reversed during the following 2 years, and some
evidence of institutional positive feedback trading,
mostly on smaller firms. Using a short sample of dailydata, they conclude that the contemporaneous rela-
tion reflects the impact of institutional trading onreturns rather than intraperiod positive feedback
trading. Lakonishok et al. (1992) find weak evidence
of momentum trading by pension funds, whileGrinblatt et al. (1995) find stronger evidence of
momentum trading by mutual funds.4 Cai and Zheng
(2004) find that returns Granger-cause institutionaltrading rather than vice versa, and are negatively
related to lagged institutional buying, suggesting that
institutions are uninformed positive feedback traders.
Jinjarak and Zheng (2010) report that emerging
market mutual funds’ positive feedback trading is a
tranquil-period phenomenon. Using data at daily andintradaily frequencies, sorted by brokers specializing
with institutions and individuals, Griffin et al. (2003)
find that institutions positive-feedback-trade. Most of
the daily positive contemporaneous association is dueto intraday positive feedback trading. Campbell et al.
(2009) find that institutional trades are highly persis-
tent, respond positively to recent daily returns but
negatively to longer-lag past daily returns.Institutional trades, particularly sells, consume liq-
uidity. Their trading anticipates both earnings sur-
prises and post-earning announcement drift.An important aspect of institutional trading is
herding given the potentially destabilizing effects.
Wermers (1999) reports significantly higher levels of
herding by mutual funds in small stocks and among
growth-oriented funds, especially on the sell side, andthat mutual fund herding accelerates the price-
adjustment process. While the overall degree of
institutional herding reported by Lakonishok et al.(1992), Grinblatt et al. (1995) and Wermers (1999) is
not very strong, Sias (2004) provides compelling
evidence of institutional herding by demonstrating
that net institutional demand is more strongly relatedto lag institutional demand than to lag returns. Lee
et al. (2004) find that institutions in Taiwan have
more persistent order imbalances, inducing continu-
ation in price pressure. Their detailed data permit todistinguish between herding and order splitting, and
they conclude that both herding and order splitting
appear to cause persistence. Sias and Starks (1997)document that return autocorrelation of individual
stocks is increasing in institutional ownership, after
controlling for size. As high-institutional-ownership
stocks tend to lead, institutional trading seems toreflect information. Employing a similar methodol-
ogy, Dennis and Strickland (2002) provide evidence
that institutions (except banks) are more likely to
herd and add to volatility by joining momentum.Dasgupta et al. (2010) find that multi-quarter herding
by institutions negatively predicts long-term stock
returns. Li and Wang (2010), however, show that
institutional informed trading is negatively related tovolatility in the retail-investor-dominated Chinese
stock market.5
One of the unresolved issues in the literature is the
price impact asymmetry, that is, the stronger
4 Badrinath and Wahal (2002) confirm that investment advisors and mutual funds, particularly growth funds, exhibit strongertendency of positive feedback trading, while overall positive feedback trading is modest. Institutions exhibit momentum(contrarian) trading in entering (exiting) a position. Average abnormal entry-to-exit returns are close to zero.5 They argue that when the percentage of institutional investors in a market reaches a certain level, the effect ofnoninformational institutional trading may prevail and increase volatility.
494 N. Ulku
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014
permanent price impact of institutional/block buysversus sells. Saar (2001) proposes an explanationbased on information search and trading constraintsto explain the asymmetry. Chiyachantana et al.(2004), using data from 37 countries and two differentsample periods with bull versus bear market charac-teristics, document that underlying market conditions(whether the market is in a bull or bear trend) is amajor determinant of the price impact of institutionaltrading, and can explain the asymmetry in priceimpact of institutional buy and sell orders. Theirfindings imply that previous results on this asymme-try might have been driven by sample specific marketconditions.
Another member of big players is foreign investors,especially in emerging markets. Foreign investors’trading displays a positive contemporaneous associ-ation with returns and positive feedback trading(Dahlquist and Robertsson, 2004; Griffin et al., 2004;Richards, 2005; Reis et al., 2010).
On the other hand, several studies document acontrast between large and small traders and betweeninstitutional and individual traders: Henker andHenker (2010) provide convincing evidence thatretail investors have no impact on stock prices.Small players’ trades are usually negatively relatedto contemporaneous and past returns (Lee et al.,1999; Hvidkjaer, 2008). Unlike institutions, individ-ual traders are more prone to disposition effect andcontrarian trading (Odean, 1998; Kaniel et al., 2008).Small trades negatively predict medium- and long-term stock returns (Hvidkjaer, 2008; Barber et al.,2009), while positively predict short-term returns(Barber et al., 2009). Kaniel et al. (2008) find thatindividual trader imbalances earn positive excessreturns in the next month, and attribute this tocompensation for liquidity provision to meet theinstitutional demand for immediacy. Indeed, Leeet al. (2004) using a detailed data set on TSE showthat liquidity appears to be provided predominantlyby small individuals who ‘often lean against thewind’. Several studies note an overlap between ‘smalltrade-size’ and ‘individual investor’ and use theformer as a proxy for the latter (Lee andRadhakrishna, 2000; Barber et al., 2009). Hvidkjaer(2008) and Campbell et al. (2009), however, warn thatthis relationship has disappeared in recent years as aresult of increased splitting of institutional ordersthanks to computerized trading.
The current study is similar to Griffin et al. (2003)as their data are also identified at the broker level.Hence, inferring trader category from broker-leveldata is not new to the literature. Their broker-level
classification shows a strong correspondence betweentrade-size and being executed by a broker specializingwith institutions (trades they classify as institutionalbased on the broker that executed the trade make up86% of block trades of 10 000 shares or more). Theyaddress the same questions employing daily data andVAR methodology, as in this article. However, thecurrent article differs in that the investor categoryunder investigation is big traders instead of institu-tions. Griffin et al. (2010) document that big andsophisticated investor groups herd on each other’strading. Hence, combining the institutions, foreignersand wealthy individuals under one category of ‘bigplayers’ is warranted given the established results inrecent literature.
III. Data and Methodology
The data set used in this study consists of the netbuying (i.e. purchases minus sales) value by thelargest net buying and net selling member brokersover a unit period of time. A positive (negative)reading in a particular broker’s figure implies netbuying (net selling) by that broker. These data arederived from cumulative trades of each and everymember broker in ISE. Specifically, all broker mem-bers6 are ranked in terms of their cumulative netbuying during a unit period of time, then n largest netbuyers at the top and n largest net sellers at thebottom of the ranking are identified. Let us call thesum of the net buying values of the top n largest netbuyers LNBn
t and the sum of the net selling values ofthe largest n net sellers as LNSn
t . Then, the key figureof interest, which will be notated as Nt throughoutthis article, is computed as Nn
t ¼LNBnt �LNSn
t . Mostmarket participants and commentaries in ISE inac-curately refer to N5 or N10 as the ‘net money inflow’,with a negative number implying ‘money outflow’.In reality, the sum of net buys of all brokers in ISE isalways zero; in other words, there can be no in- oroutflows, as ISE operates under an order-drivenelectronic system with no specialist dealers. Whatthey refer to as inflow (outflow) is, in fact, that largestnet buyers (sellers) have bought from (sold to) therest of the market participants (‘crowd’). Thus,the ‘net buys’ figures are a proxy for big players’trading.
We do not group individual trades by trade-size, orby the identity of the parties (e.g. institutional versusindividual). Rather, we employ an ingenious deriva-tion technique utilized by market practitioners.
6As of April 2010, there are 103 member brokers in ISE.
Big players’ aggregated trading and market returns in Istanbul Stock Exchange 495
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014
While N is functionally a good proxy for big players’trading, one can argue, for example, that a brokerwith a large number of small investors, all of thembeing simultaneously net buyers (sellers) in a partic-ular period, may appear as a big net buyer (seller),although in reality it reflects small players’ trading.However, that all small traders of a particular broker,and not of the others, trade in the same direction in aperiod is highly unlikely. Most typically, a few bigtraders’ transactions far outweigh in value the sum ofmany small traders’ transactions.7 Small traders of aparticular broker making it appear in the list of topnet buyers or sellers might have been a likely case,had some large brokers specialized on small (retail)clients. However, statistics from ISE suggest just theopposite: the highest percentage of domestic individ-ual investors is with the smaller brokers, and thepercentage of big player groups (foreign, proprietaryinstitutional) are higher with larger, bank- and/orforeign-owned brokers (the Association of CapitalMarkets Intermediary Institutions of Turkey, 2009statistics). Thus, trader size appears to be in parallelwith broker size.8 Therefore, these figures typicallyprovide an accurate vision of big versus small players’direction of trading. Any counter-arguments fail topass tests of logic: if all small traders respond topublic news, then none of the brokers of small traderswould stand out as a large net trader. That all smallclients of a particular broker, but not those of otherbrokers, trade in the same direction to make theirbroker appear as a large net trader would be possibleonly if all of the small clients of this broker faithfullyand strictly follow a broker-specific signal such astheir broker’s private investment advises. Then,trading by this broker would be functionally nodifferent than an institutional portfolio manager.
As this data set is argued to reflect the trading ofinstitutions, big individual traders and foreigners,external checks would help verify this proposition.However, in ISE no data on institutional orindividual trading are made available to public. Oneopportunity, however, is presented by the market-wide foreign ownership ratios published by theClearing and Custody Bank on a daily basis.
The correlation between N10 and the changes in thepercentage market value held by foreigners over oursample period is þ0.54, which is significant atp < 0.001.9 As foreigners represent 83.8% of totalinstitutional holdings as of December 2009, thisimplies that our data set captures a large portion ofinstitutional trading, as well.
A risk of misinterpreting this data set may,however, result from big players’ strategic behaviour.A big client in the stock market may trade vianumerous brokers, even simultaneously buying viaone and selling via another. The electronic contin-uous-auction trading system of ISE with irreversiblelimit orders, no market-making specialists, progres-sively lower brokerage commission rates and a highlevel of transparency encourage such fictive trades inISE. Many experienced traders believe that bigplayers sometimes try to conceal their selling inten-tions by appearing as large net buyer through abroker known to have foreign clientele and beingsmall net sellers through a number of other brokers,misleading those who try to infer information fromthis data set. A remedy for this problem could be toinclude a higher number for n, since the number ofdifferent brokers a trader may use in the same periodhas practical limits. We therefore experiment withthree different values for n: 5, 10 and 15. The resultswith the three versions are similar, which suggeststhat our results were not significantly affected by suchstrategic behaviour. Throughout the article, resultsfor n¼ 10 are reported.
The data were obtained from Euroline�, a domesticdata vendor who redistributes data from ISE.Cumulative net buys data are summarized overperiods of 1 day, so the study is at the dailyfrequency.10 Our sample period spans from 1August 2007 to 16 April 2010 (679 trading days).11
This period corresponds to the sharp downtrend dueto the recent global crisis and the symmetric recoveryfollowing the bottom in March 2009, hence presentsan excellent opportunity to incorporate the globalcrisis as a significant information event into theanalysis, in particular to exploit world market returnsas a significant information factor. Furthermore,
7 The high concentration is well-documented in other markets as well. For example, Menkoff and Schmeling (2010) reportthat 100 out of 723 traders in their sample account for 50% of total trading volume.8 In a similar manner, Barber et al. (2009) document that small-trade-size order imbalance correlates well with orderimbalance from retail brokers.9 It is desirable that this correlation is significantly smaller than þ1, as big individual investors and domestic institutions aretwo other important components of big players. Otherwise, this article would collapse into a study of foreigners’ trading.10 In general, this data set can be obtained over any frequency, hence intraday analysis is possible. However, data at intradayfrequencies are not stored, nor data for individual stocks. Ulku (2008) manually collects a short sample of market-wideintraday data and daily data for 15 individual stocks, and finds that results are overall similar to those with daily market-widedata.11 These data are broadcast on a real-time basis, and lost unless stored. Euroline stores these data only for moving windows ofseveral months. The sample used in this study is obtained by combining windows collected at different time points.
496 N. Ulku
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014
dividing the sample into two parts by the March-2009
bottom, one obtains an almost perfect V-shape in ISE
index which enables a clean test of Chiyachantana
et al.’s (2004) argument by comparing the price
impact of big traders’ purchases and sales in bull
versus bear markets. The price/time chart of the ISE-
100 index is plotted in Fig. 1.Unit root tests remain indecisive on whether Nt is
stationary or not. Specifically, Augmented Dickey-
Fuller (ADF) test rejects the null of a unit root only
at 10% level, while Phillips–Perron (PP) test safely
rejects at 1%. It is natural to think that N may trend
together with market capitalization. Hence, a prudent
approach would be to scale N by market capitaliza-
tion (MC). This would also enable us to standardize
big investors’ net trading relative to the value of
shares in circulation. Thus, we define the key variable
used in this study as follows12:
Nt ¼ ðLNBnt � LNSn
t Þ=MCt ð1Þ
We employ VAR methodology to portray the
dynamic interaction between big trader flows and
returns. In particular, this framework has the ability
to simultaneously test feedback trading, information
(predictive) content and persistence in big player
flows; to distinguish between temporary and perma-
nent price impact, hence to test price pressure and
information hypotheses.13 Our VAR model includes
N and R as endogenous variables in the system, where
R is returns of the ISE-100 index, defined as
Rt ¼ lnðISE100Þt � lnðISE100Þt�1 ð2Þ
Since global market returns are a major factor that
strongly affects both ISE returns and big players’
trading, we augment the bivariate-VAR model with
world returns that are affected only by their own lags.
The advantage of this specification instead of a
conventional VAR is that big player flows and ISE
returns are not permitted to affect the world returns,
but are affected by the instantaneous and lag values
of world returns. Thus, world returns are treated as
an exogenous variable within the system. This ensures
a more accurate characterization of the joint dynamic
relationship between big trader flows and returns.
Hence, we run the following VAR model:
Nt
Rt
Wt
264
375¼
an
ar
aw
264
375þ
b11ðLÞ b12ðLÞ b13ðLÞ
b21ðLÞ b22ðLÞ b22ðLÞ
0 0 b33ðLÞ
264
375
Pp1
Nt�p
Pp1
Rt�p
Pp1
Wt�p
266666664
377777775
þ
"nt"rt"wt
264
375 ð3Þ
Dat
e
ISE-100
20 000
25 000
30 000
35 000
40 000
45 000
50 000
55 000
60 000
13.0
9.20
07
30.1
0.20
07
12.1
2.20
07
29.0
1.20
08
17.0
3.20
08
30.0
4.20
08
13.0
6.20
08
28.0
7.20
08
09.0
9.20
08
27.1
0.20
08
16.1
2.20
08
29.0
1.20
09
13.0
3.20
09
28.0
4.20
09
12.0
6.20
09
27.0
7.20
09
08.0
9.20
09
23.1
0.20
09
10.1
2.20
09
25.0
1.20
10
09.0
3.20
10
Fig. 1. The daily closing levels of the ISE-100 index during the sample period
12An alternative way of normalization can be obtained by dividing by the total trading value. Results with this version aresimilar. However, variation in trading volume appears to add some noise, hence dividing by market cap is preferred. This wayof normalization is also compatible with previous articles such as Griffin et al. (2003).13 See Hasbrouck (1991) who was the first to suggest the interaction of trades and returns be modelled as a VAR system.
Big players’ aggregated trading and market returns in Istanbul Stock Exchange 497
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014
where the a’s represent intercept terms, b(L) denotespolynomial in the lag operator L, and "nt , "
rt and "
wt
are zero-mean residuals that are assumed to beintertemporally uncorrelated, and W is the worldmarket returns. The block exogeneity is representedby the zero entries in the coefficients matrix.
We use Morgan Stanley Capital International(MSCI)-Europe index as a proxy for W. Use ofMSCI-Europe index instead of MSCI-World index orUS indices avoids time-zone differences which mightconfound the analysis at the daily frequency. The lagorder is nine as suggested by Akaike InformationCriterion (AIC). In the impulse response analysis,standard Choleski factorization and asymptotic con-fidence bands are employed. Naturally, world returnsare placed first in the Cholesky ordering. In line withthe common treatment in the literature, we place netflows before ISE returns, which implies that flowshave contemporaneous effect on returns but not viceversa; ISE returns can only affect net flows with a lag.
As previous literature finds big players’ trading tobe correlated with information, and as global marketreturns constitute a major source of information forISE during our sample period, a comparison ofresults under our augmented model in Equation 3and a bivariate model will provide useful insight.Specifically, it will show whether big player flows’positive response to past returns reflects naıve pos-itive feedback trading or reaction to information withdifferential response time. For example, Griffin et al.(2003) document a positive response of institutionalflows to lagged intraday returns, however cannotdistinguish whether it is a response to lagged returnsthemselves, or the information therein. By controllingfor world returns as an information factor stronglycorrelated with returns, we are able to provide ananswer to this question left unanswered in the extantliterature.
IV. Results
We present results by studying Impulse-ResponseFunctions (IRFs), which show the dynamic behav-iour of a variable due to a shock in another variablein the system. In all IRF graphs to follow, the solidline in the middle represents a point estimation ofimpulse responses. A 2-SE confidence interval isshown by the upper and lower dashed lines.
Statistical significance is implied when neither of theconfidence bands crosses the x-axis. The results fromthe bivariate specification, which is not augmented bythe world returns, are presented in the Appendix. Acomparison confirms that omission of world returnswould lead to misspecification and biased results.
The behaviour of big trader flows
The determinants of big players’ trading are charac-terized by studying IRFs of big trader net flows (N)to a shock in global returns (W), local returns (R) anditself (N), respectively, in Panels A, B and C of Fig. 2.
Panel A suggests that big trader flows are signif-icantly correlated with global returns both contem-poraneously and at the first lag. While this isconsistent with the results on foreigners’ trading(Griffin et al., 2004; Richards, 2005), in the context ofinstitutional or big trader flows this relation might beconsistent with both ‘positive feedback trading’(including the possibility of intraday positive feed-back trading) and ‘reacting to information’. Previousliterature is unable to distinguish between these twoalternatives. Recalling that the sample period corre-sponds to the recent global crisis which originatedfrom developed markets, global returns can beconsidered as a significant information variable.Hence, specific characteristics of our sample periodand our specification enable us to distinguish betweenthese two alternatives, as seen below.
Panel B suggests a much weaker tendency ofpositive feedback trading with respect to local returns(see period 2). As big trader flows are significantlyrelated to global returns but only weakly related toreturns of the underlying market, our results favour‘reacting to information’ over ‘positive feedbacktrading’. Hence, it appears that big traders do herdon information with slightly differential responsetimes rather than naively conditioning on pastreturns. This is an important result answering amain question which has not been satisfactorilyaddressed in the previous literature because it hasbeen difficult to condition the return-flow interactionon an information variable of high relevance forwhich a continuous time series is available.14
Note that failure to augment the VAR specificationwith exogenous global returns (see the Appendix)would lead to a misleading inference of significantpositive feedback trading as ISE returns are highlycorrelated with global returns. In a small emerging
14A caveat applies here: foreign investors have been shown to respond to global market returns (Griffin et al., 2004) due toportfolio rebalancing rather than new information. However, as foreign investors constitute only a portion of the big playerflows proxy used in this study, our finding goes beyond the relationship already-documented for foreign investors.
498 N. Ulku
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014
market context, the need to include global returns as
an information factor is obvious. In previous
research, however, any possible conditioning infor-
mation variable in the return-flow interaction has
been simply omitted. Thus, pervasive findings of
positive feedback trading by institutions may in
reality be a reflection of institutions’ herding on
information signals with differential response and
order execution times.The negative responses on 7th and 8th days are
worth mentioning, as they more-than-offset the initial
positive response to ISE returns. It seems that any
weak positive feedback trading by big players is
reversed shortly, or some big players tend to take
advantage of return reaction after several days. It alsoimplies that local information factors have littleenduring effect on big players’ trading in ISEduring our sample period.
Panel C suggests significant persistence of bigtrader flows extending up to 4 days. Several articlessuch as Sias (2004) measure institutional herding bythe positive correlation between current and laggednet demand by institutions. In our context, theStructural VAR (SVAR) model employed in thisarticle is particularly informative, as the impulseresponse of N to its own shocks reveals the correla-tion between current and lagged net big trader flowsafter controlling for momentum (feedback) andinformation (world return) effects. Our results sug-gest significant herding by big players. As theresponse of N to lagged N is much more significantthan to lagged R, the tendency to herd appears to bemore significant than the tendency to positive-feed-back-trade, consistent with Sias’ (2004) result in across-sectional context, and Griffin et al.’s (2010)conclusion that big players herd on each other’sinformation.15
The variance decomposition in Table 1 suggeststhat about 11% of forecast error variance of bigtrader flows can be explained by global returns. While3% can be explained by local returns, it should benoted that most of it comes in the form of contrarianaction after the 6th day.
Further, we distinguish big players’ response topositive and negative past returns to see if thefeedback trading tendencies differ. Results in Fig. 3suggest a visible asymmetry: big players respondmore strongly to negative past returns,16 consistent
–100
0
100
200
300
1 2 3 4 5 6 7 8 9 10
Panel A: Response of N to W
Response to Cholesky 1-SD innovations 2-SE
–200–100
0100200300400500600700
1 2 3 4 5 6 7 8 9 10
Panel C: Response of N to N
–100
–50
0
50
100
150
200
1 2 3 4 5 6 7 8 9 10
Panel B: Response of N to R
Fig. 2. The impulse responses of big trader net flows
Notes: The impulse response of net big trader flows (N) toa 1-SD shock in world returns (W), ISE returns (R) anditself (N) is portrayed in Panels A, B and C, respectively.The x-axis shows the number of days (day 1 refers tothe contemporaneous period). The solid line in the middlerepresents the point estimates of the impulse responsecoefficients, while the dashed lines represent 2-SE confi-dence bands.
Table 1. Variance decomposition of big trader flows
Period SE W N R
1 616.86 7.90 92.10 0.002 638.10 11.14 88.57 0.293 645.35 11.01 88.64 0.344 652.46 10.78 88.85 0.375 654.84 10.99 88.33 0.686 656.35 11.31 87.96 0.737 660.12 11.19 86.99 1.818 664.37 11.12 85.92 2.969 665.77 11.09 85.58 3.3310 670.09 11.01 85.70 3.29
Notes: SE is the forecast error of net big player flows (N).The figures in columns W and R show the percentage offorecast error variance of N explained by world market andISE returns, respectively, up to 10 days. The remainingforecast error variance is shown under column N.
15Order splitting is not a likely explanation in a market-wide analysis.16 In unreported analysis, I have confirmed that this finding applies both to bull and bear market sub-periods.
Big players’ aggregated trading and market returns in Istanbul Stock Exchange 499
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014
with O’Connell and Teo’s (2009) result thatinstitutional investors are less prone to thedisposition effect and tend to aggressively reducerisk following losses but mildly increase risk follow-ing gains.
The price impact of big players’ trading
Big players’ price impact is characterized by
studying IRFs of ISE market returns (R) to a
shock in big trader flows (N) in Panel A of Fig. 4.
To complement this analysis, responses of ISE
market returns to a shock in global returns (W) and
itself (R) are also presented in Panels B and C,
respectively.The contemporaneous relation of ISE returns to a
shock in big trader flows is significantly positive,
as expected. The responses at the first three lags are
positive and borderline-significant, and are driven by
the persistence in big player flows.17 A 1-SD shock in
big player flows (approximately US$ 18 million) is
associated with a 0.75% change in ISE index on
the contemporaneous day, and followed by 0.42%
further cumulative change over the next 3 days.
However, the latter is subsequently reversed. Thus,
while the bulk of the price impact of big trader flows
is incorporated contemporaneously, a nontrivial
follow-through occurs due to future big trader flows
signalled but not arbitraged away. This implies some
predictability, however as the follow-through part is
subsequently reversed it does not imply information.
The positive contemporaneous response is not
reversed subsequently, which is consistent with
information rather than pure price pressure.
–150
–100
–50
0
50
100
150
200
250
1 2 3 4 5 6 7 8 9
Response of N to positive world returns
–80
–40
0
40
80
120
160
200
1 2 3 4 5 6 7 8 9
Response of N to negative world returns
–160
–120
–80
–40
0
40
80
120
1 2 3 4 5 6 7 8 9
Response of N to positive ISE returns
–100
–50
0
50
100
1 2 3 4 5 6 7 8 9
Response of N to negative ISE returns
Fig. 3. The impulse response of big trader net flows to positive and negative past returns
Notes: World and ISE returns data are partitioned by the sign of the returns. Other explanations are the same as in Fig. 2.
17A regression analysis that controls for current Nt values, as in Chordia and Subrahmanyam (2004), confirms this conclusion(available from the author).
500 N. Ulku
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014
The true permanent information content of big
players’ trading is priced-in within the contempora-
neous period.18
Panel B shows that ISE returns incorporate global
market information to a large extent instantaneously,
with very little reaction left to the second day. Panel
C shows no significant autocorrelation in ISE market
returns. Thus, ISE seems to incorporate all types of
shocks within 1 day but not price-in the persistence in
big trader flows.The variance decomposition in Table 2 suggests
13.5% of ISE returns to be attributed to big trader
flows, while global market returns is the mostimportant factor accounting for 46% of ISE returns’forecast error variance.
Differential price impact of big players’ net pur-chases and sales
One of the unresolved issues in this line of research isthe price impact asymmetry (stronger price impact ofinstitutional/block purchases compared to sales). Bycomparing the price impact of institutional trades intwo different sample periods characterized by bullversus bear market conditions, Chiyachantana et al.(2004) raised doubt on earlier findings of price impactasymmetry, which were all obtained in bull marketsample periods (Kraus and Stoll, 1972; Holthausenet al., 1987, 1990; Chan and Lakonishok, 1993, 1995;Keim and Madhavan, 1996). Chiyachantana et al.find that the price impact of institutional buys isstronger in a bullish market period and the priceimpact of institutional sells is stronger in a bearishmarket period. Saar (2001) develops a model toexplain permanent price impact in relation to pastprice performance of the stock. Specifically, he arguesthat the longer the run-up in a stock’s price the less isthe permanent price impact asymmetry between buysand sells. Implications of his model hold when theasymmetry is defined in terms of daily net order flowsinstead of individual blocks.
The large V-shaped price action during our sampleperiod provides an excellent opportunity to replicateChiyachantana et al.’s (2004) finding. Chiyachantanaet al. compare results on the January 1997–March1998 and the January 2001–September 2001 sub-periods, which are of unequal length and 3 years
–0.002
0.000
0.002
0.004
0.006
0.008
0.010
1 2 3 4 5 6 7 8 9
Panel A: Response of R to N
–0.004
0.000
0.004
0.008
0.012
0.016
1 2 3 4 5 6 7 8 9
Panel B: Response of R to W
–0.004
0.000
0.004
0.008
0.012
0.016
1 2 3 4 5 6 7 8 9
Response of R to R
Response to Cholesky 1-SD innovations 2-SE
Fig. 4. The impulse responses of ISE returns
Notes: The impulse response of ISE returns (R) to a 1-SDshock in big trader net flows (N), world returns (W), anditself (R) is portrayed in Panels A, B and C, respectively.Other explanations are the same as in Fig. 2.
Table 2. Variance decomposition of ISE returns
Period SE W N R
1 0.01683 47.79 12.04 40.172 0.01686 47.76 12.32 39.923 0.01701 47.56 12.68 39.764 0.01709 47.58 12.86 39.565 0.01733 47.54 12.94 39.526 0.01739 47.56 13.07 39.377 0.01758 46.81 13.09 40.108 0.01759 46.28 13.01 40.729 0.01761 46.27 13.01 40.7210 0.01783 45.94 13.53 40.54
Note: See notes in Table 1.
18 Lee et al. (2004) reach a similar conclusion on Taiwan Stock Exchange operating under a batch-processing call auctionsystem without designated market makers: ‘price pressures created by order imbalances are effectively absorbed and do notpersist beyond one day’.
Big players’ aggregated trading and market returns in Istanbul Stock Exchange 501
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014
apart from each other. As the V-shape in our sampleperiod is symmetric and as both sub-periods areadjacent, our sample offers a healthy test ofChiyachantana et al.’s (2004) argument, avoidingpossible structural changes when two sub-samples arefar apart from each other. We also provide the firstempirical test of Saar’s (2001) model by comparingthe price impacts of big player net buys and net sellsduring the early and late parts of the run-up sub-period (the right wing of the V-shape) in our sample.
As our data are market-wide-aggregated, however,a caveat applies here: Chan and Lakonishok (1993)and Keim and Madhavan (1995) explain the asym-metric information content of block/institutionalpurchases versus sales by the possibility that pur-chases in a specific stock are more informationally-motivated than sales as the purchase of a specificstock involves a choice among many potential assetswhereas sales are mostly due to liquidity motives andlimited to assets already held. Thus, the asymmetryhinges on stock selection. In our context, a purchasedecision does not involve a choice among manypotential assets, thus we may not observe an uncon-ditional (full-sample) asymmetry if information-motivated stock selection in purchases is the majorsource of the asymmetry. However, another factorcontributing to the price impact asymmetry is theshort-sale restrictions, as argued by Saar (2001). InISE, short sales are practically nonexistent, hencesales are less likely to be informationally-motivatedthan purchases. In sum, the full-sample asymmetrymay be less visible or nonexistent in our case.However, this does not prevent us from testing
alternative hypotheses: the testable implications ofChiyachantana et al.’s (2004) and Saar’s (2001)arguments lie in the time-variation in the degree ofasymmetry.
To check for price impact asymmetry of bigplayers’ net buys versus net sells, we partition Nt byusing a dummy variable. Because IRFs portray theresponse to a 1-SD shock and SD of positive andnegative Nt’s are unequal, we report results bycomparing the coefficients in the return equation ofthe SVAR system in Panel A of Table 3. Onlycontemporaneous and first-lag coefficients arereported as the other lags are insignificant. Thecontemporaneous price impact of buys and sells arenot significantly different. Thus, no price impactasymmetry is observed. This suggests that the asym-metry may not be present when the purchase decisiondoes not involve a choice among many potentialassets to buy, favouring Chan and Lakonishok’s(1993) and Keim and Madhavan’s (1995) hypotheses.At lag 1, the impact of net sells exhibits asmall continuation whereas the impact of net buysexhibits a small reversal. This differential behaviourat lag 1 is consistent with short-sales restrictionsslowing down the incorporation of information inbig-trader sales (discussed in more detail in the nextsubsection).
To test Chiyachantana et al.’s (2004) argument thatthe price impact of purchases and sales simplydepends on the underlying market conditions duringthe sample period, we partition the sample: the first(second) half of our sample period represents bear(bull) market. Panel B of Table 3 compares the price
Table 3. The differential price impact of big trader net purchases and sales
Nb Ns Sig(D) Nb(�1) Ns(�1) Sig(D)
Panel A: Full sample (testing price impact asymmetry)0.0000116 0.0000137 0000 �0.0000025 0.0000025 0.05
Panel B: Testing Chiyachantana et al.’s (2004) argumentThe First half (bull market)0.0000091 0.0000165 0.05 �0.0000018 0.0000045 0.10
The second half (bull market)0.0000148 0.0000071 0.05 �0.0000017 �0.0000028 0.78
Panel C: Testing Saar’s 3 (2001) argumentThe first 1/3 of the run-up period0.0000134 0.0000065 0.29 �0.0000019 �0.0000026 0.92
The third 1/3 of the run-up period0.0000147 �0.0000009 0.23 �0.0000006 0.0000277 0.03
Notes: Nb (Ns) is the contemporaneous coefficient of net buys (net sells) in the return equation.Nb(�1) and Ns(�1) are the coefficients of previous day’s net buys (net sells), respectively. Sig(D) is at-test for the equality of the Nb and Ns coefficients. A significantly larger Nb implies positive priceimpact asymmetry. In Panel B and C, Chiyachantana et al.’s (2004) and Saar’s (2001) arguments,respectively, are tested by comparing sub-period results.
502 N. Ulku
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014
impact of big players’ net purchases and net sales inbull and bear market conditions. Chiyachantana
et al.’s (2004) argument that the price impact asym-
metry is driven by underlying market conditions isclearly confirmed, as the price impact of big players’
buys is significantly larger than that of sells duringthe bull market (second half) whereas it is signifi-
cantly smaller during the bear market (first half).These results imply that, notwithstanding informa-
tion-motivated stock selection in purchase decisions,
the previously-reported price impact asymmetry mayhave been driven by bull market conditions prevailing
in respective sample periods, as argued byChiyachantana et al. (2004).
We test the implications of Saar’s (2001) model by
dividing the second half (bull run) into three sub-
periods and comparing the first and third sub-period
results. Saar’s model implies that the price impact of
big players’ buys should be much stronger during the
early part of a bull market than during the late part,
when the price impact asymmetry may even turn
negative. Thus, Panel C of Table 3 compares price
impact coefficients of big players’ net buys and sells in
the first and last third of the second half.
Contemporaneous coefficients do not seem to sup-
port Saar’s hypothesis, as price impacts are similar,
even more positive, during the last third of the bull
market period compared to the first third. However,
lagged coefficients are significantly different in the
late part of the run-up period. The cumulative price
impact asymmetry turns negative as predicted by
Saar. It may be that the information content of sells is
recognized with a lag during the later stages of a bull
market.
Predictive information content of big players’trading
The full-sample impulse response of ISE returns to ashock in big trader net buys in Panel A of Fig. 4suggests borderline significant coefficients at first,second and third lags (periods 2, 3 and 4), whichimply statistical forecast ability. However, the break-down in Fig. 5 suggest that this result is mainly drivenby net sells. There is a clear asymmetry in the forecastability of big players’ net buys versus net sells. In ISEshort sales are practically nonexistent. As the data setused in this study is vigilantly utilized by marketparticipants in ISE on a real-time (intraday) basis, itis legitimate to assume that many short-term tradersrevise their trading decisions accordingly. The infor-mation in big players’ purchases are faster incorpo-rated into prices on the same day, as exploitingrequires only the availability of cash; however, theincorporation of information in big players’ salestakes time as selling, in the absence of short-sales inISE, is an option for only those who own the stock.Thus, this asymmetry is to be explained by structuralcharacteristics of the ISE, that is, practical absence ofshort sales.
The implication of these results for practitioners isthat net selling by big players offers trading signalsand possibly (to the extent that the index futures basisis not correlated to these signals) profitable arbitrageopportunities.
Event study methodology
While the SVAR methodology employed in thisarticle proves to be more informative, it may still be
–0.004
–0.002
0.000
0.002
0.004
0.006
0.008
1 2 3 4 5 6 7
Response of R to Cholesky1-SD N *IN innovation
–0.004
–0.002
0.000
0.002
0.004
0.006
0.008
1 2 3 4 5 6 7
Response of R to Cholesky1-SD N *OUT innovation
Fig. 5. The impulse response of returns to big player net buys and net sells
Notes: See the explanations in notes of Fig. 3. The left (right) panel shows the response of ISE returns to a 1-SD shock inpositive (negative) big trader net flows.
Big players’ aggregated trading and market returns in Istanbul Stock Exchange 503
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014
useful to repeat the event study methodology
employed in earlier studies on block trading for the
sake of comparison. For this purpose we define net
big trader inflows and outflows larger than a specified
size19 as event days, and portray the market returns
around those days. Results are depicted in Fig. 6.Large net flows of big traders are preceded by
returns of the same sign, and followed by a small
magnitude of return continuation. There is no
evidence of reversal, hence price pressure hypothesis
is rejected over information. The significant returns
on day (�1) can be consistent with positive feedback
trading, delayed reaction to (possibly private) infor-
mation, or leakage of information on large big trader
flows and ‘front running’. However, because the
magnitude of negative returns preceding block sales is
no smaller than that of positive returns preceding
block purchases, the ‘leakage and front running’
explanation is not supported: as short sales are
practically absent in ISE, one would expect less
front running activity ahead of block sales compared
to block purchases. The asymmetry between days
(�1) and (þ1) suggests a larger tendency of past
returns to shape big trader flows rather than big
trader flows shaping future returns. Positive feedback
trading and delayed response to information are two
possible explanations, and event study methodology
mostly used in the previous literature is not able to
distinguish between them. Our SVAR results in the
previous section, however, have shown that delayed
response to information, rather than positive feed-
back trading, is more likely to be the main explana-
tion behind the positive relationship between big
trader flows and previous days’ returns.To have another view, we define extreme return
days as event, and monitor average big trader flows
surrounding event days. We define days with a return
greater than 3% in absolute value as an event.
Figure 7 portrays average normalized net big trader
flows surrounding those event days.Big trader flows both before and after extreme-
return days have the same sign as the event-day
Panel A: Returns around large positive net big trader flows
–0.002
0
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
Panel B: Returns around negative large net big trader flows
–0.022
–0.018
–0.014
–0.01
–0.006
–0.002
0.002
–6 –5 –4 –3 –2 –1 3 4
–6 –5 –4 –3 –2 –1 0
0 1 2
1 2 3 4 5
Fig. 6. Returns around large net big trader flows
Note: x-axis shows the days (0 is the event day), and y-axis shows average returns.
19 The size is set at 0.01% of market cap for net buys and 0.007% for net sells. The asymmetry is due to the positive daily meanbig trader flow over the sample period. This differential size criterion leaves us with 128 large net buying and 87 large netselling days.
504 N. Ulku
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014
return, suggesting heterogeneous reaction time toinformation. Net big trader flows following extreme-return days are much larger compared to thosepreceding extreme-return days. This confirms that,rather than big players predicting future returns,previous day’s returns (or information containedtherein) predict big trader flows, and part of the bigtraders’ reaction comes with a 1-day delay. In otherwords, positive feedback trading or delayed responseto information seems to be more prominent thantrading on private information. There is also evidenceof some contrarian trading a few days later after largenegative returns.
V. Conclusions
The fact that all typical results on big players’ tradingwere replicated in our empirical analysis shouldconvince any reader skeptical about the nature ofthe data set used in this study. It appears that the way
market participants in ISE utilize this data set picksan important driver of the market. The results of thisstudy raises the possibility that important informa-tion can be derived from similar broker-level data andbe utilized by market participants in other stockmarkets around the world. Furthermore, as trade sizeis shown to matter in interaction with trader size, thekey variable N used in this study successfully pickshighly relevant information, which is not easy to pickvia order-size breaks used in earlier research.
Our results confirm that big players’ trading isstrongly positively correlated with market returns.Conditioning by an information variable in ouranalysis makes clear that this relationship existsmainly because their trading is correlated withinformation, rather than due to naıve intraperiodpositive feedback trading or pure price pressure. Bigplayers’ trading both leads and follows returns,however ‘following’ is more significant than ‘leading’which would give an impression of positive feedbacktrading in the absence of a relevant conditioninginformation variable. Our specification enables to
Panel A: Average big trader flows around extreme–positive–return days
–200
–100
0
100
200
300
400
500
600
700
800
900
Panel B: Average big trader flows around extreme–negative–return days
–400
–300
–200
–100
0
100
200
–6 –5 –4 –3 –2 –1 0 1 2 3 4 5
–6 –5 –4 –3 –2 –1 0 1 2 3 4 5
Fig. 7. Average big trader flows around extreme-return daysNote: x-axis shows the days (0 is the event day), and y-axis shows average normalized net flows of big traders.
Big players’ aggregated trading and market returns in Istanbul Stock Exchange 505
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014
distinguish delayed response to information conveyedby world market returns from positive feedbacktrading.
Big players’ trading does not significantly predictfuture returns. Their trading co-moves with stockprices, and exhibits persistence. The apparent forecastability is accounted for by the persistence in bigplayer flows. Hence big players’ interim portfolioperformance may appear to be more superior to theirrealized entry-to-exit returns.20 As our variable Npartitions market participants as ‘big players’ versus‘the rest of the market’, an important implication ofour results is the contrast between big players’- andothers’ trading: while big players’ trading is corre-lated with information, smaller players seem toprovide liquidity for them, as suggested by Leeet al. (1999, 2004). The information contained inbig players’ buying is incorporated into prices on thesame day. The delay in incorporating the informationcontained in big players’ selling, which is apparentlydue to practical absence of short selling in ISE,suggests that traders monitoring big player flows in ablind-matching, continuous auction system fulfill theinformation aggregation role of a specialist.
Finally, our cleaner test confirms Chiyachantanaet al.’s (2004) argument that the price impact asym-metry between block/institutional buys and sells mayhave been driven by underlying market conditions(bull market sample periods). In the first empiricaltest of Saar (2001), we find that the lagged reaction tobig players’ net sells in the late stage of a bull run isstronger than that in the early stage.
Acknowledgements
I thank Euroline� for providing data. Financialsupport by Central European University (Grantnumber: REG/412/200910/R) is gratefullyacknowledged.
References
Aragon, G., Bildik, R. and Yavuz, D. (2007) Do institu-tional investors have an information advantage?,Working Paper, Arizona State University.
Badrinath, S. and Wahal, S. (2002) Momentum trading byinstitutions, Journal of Finance, 57, 2449–78.
Ball, R. and Finn, F. (1989) The effect of block transactionson share prices: Australian evidence, Journal ofBanking and Finance, 13, 397–419.
Barber, B., Odean, T. and Zhu, N. (2009) Do retail tradesmove markets?, Review of Financial Studies, 22,151–86.
Barclay, M. J. and Warner, J. B. (1993) Stealth trading andvolatility: which trades move prices?, Journal ofFinancial Economics, 34, 281–305.
Bloomfield, R., O’Hara, M. and Saar, G. (2005) The ‘makeor take’ decision in an electronic market: evidence onthe evolution of liquidity, Journal of FinancialEconomics, 75, 165–99.
Bozcuk, A. and Lasfer, A. (2005) The information contentof institutional trades on the London Stock Exchange,Journal of Financial and Quantitative Analysis, 40,621–44.
Cai, F. and Zheng, L. (2004) Institutional trading and stockreturns, Finance Research Letters, 1, 178–89.
Campbell, J., Ramodorai, T. and Schwartz, A. (2009)Caught on tape: institutional trading, stock returns,and earnings announcements, Journal of FinancialEconomics, 92, 66–91.
Chakravarty, S. (2001) Stealth trading: which traders’trades move stock prices?, Journal of FinancialEconomics, 61, 289–307.
Chan, K. and Fong, W. (2000) Trade size, order imbalance,and the volatility-volume relation, Journal of FinancialEconomics, 57, 247–73.
Chan, L. and Lakonishok, J. (1993) Institutional trades andintraday stock price behavior, Journal of FinancialEconomics, 33, 173–99.
Chan, L. and Lakonishok, J. (1995) The behavior of stockprices around institutional trades, Journal of Finance,50, 1147–74.
Chiyachantana, C., Jain, P., Jiang, C. and Wood, R. (2004)International evidence on institutional trading behav-ior and price impact, Journal of Finance, 59, 869–98.
Chordia, T., Roll, R. and Subrahmanyam, A. (2002) Orderimbalance, liquidity, and market returns, Journal ofFinancial Economics, 72, 111–30.
Chordia, T. and Subrahmanyam, A. (2004) Order imbal-ance and individual stock returns: theory and evidence,Journal of Financial Economics, 65, 485–518.
Dahlquist, M. and Robertsson, G. (2004) A note onforeigners trading and price effects across firms,Journal of Banking and Finance, 28, 615–32.
Dasgupta, A., Prat, A. and Verardo, M. (2011)Institutional trade persistence and long-term equityreturns, Journal of Finance, 66, 635–53.
Dennis, P. J. and Strickland, D. (2002) Who blinks involatile markets, individuals or institutions?, Journal ofFinance, 57, 1923–49.
Easley, D., Kiefer, N. M. and O’Hara, M. (1997) Theinformation content of the trading process, Journal ofEmpirical Finance, 4, 159–86.
Easley, D. and O’Hara, M. (1987) Price, trade size andinformation in securities markets, Journal of FinancialEconomics, 19, 69–90.
Griffin, J., Harris, J., Shu, T. and Topaloglu, S. (2010) Whodrove and burst the tech bubble?, Working Paper,University of Texas at Austin.
Griffin, J., Harris, J. and Topaloglu, S. (2003) Thedynamics of institutional and individual trading,Journal of Finance, 58, 2285–320.
20 This implication is in line with Aragon et al. (2007), who find, using privately obtained data on complete trading history ofall stocks in ISE over the period 1999–2003 that realized risk-adjusted returns of institutions are no better than individuals.
506 N. Ulku
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014
Griffin, J. M., Nardari, F. and Stulz, R. M. (2004) Aredaily cross-border equity flows pushed or pulled?,Review of Economics and Statistics, 86, 641–57.
Grinblatt, M., Titman, S. and Wermers, R. (1995)Momentum investment strategies, portfolio perfor-mance and herding: a study of mutual fund behavior,American Economic Review, 85, 1088–105.
Hasbrouck, J. (1991) Measuring information content ofstock trades, Journal of Finance, 46, 179–207.
Henker, J. and Henker, T. (2010) Are retail investors theculprits? Evidence from Australian individual stockprice bubbles, European Journal of Finance, 16,281–304.
Holthausen, R., Leftwich, R. and Mayers, D. (1987) Theeffect of large block transactions on security prices:a cross-sectional analysis, Journal of FinancialEconomics, 19, 237–67.
Holthausen, R., Leftwich, R. and Mayers, D. (1990) Large-block transactions, the speed of response and tempo-rary and permanent price effects, Journal of FinancialEconomics, 26, 71–95.
Hvidkjaer, S. (2008) Small trades and the cross-sectionof stock returns, Review of Financial Studies, 21,1123–51.
Jinjarak, Y. and Zheng, H. (2010) Financial panic andemerging market funds, Applied Financial Economics,20, 1793–805.
Kaniel, R., Saar, G. and Titman, S. (2008) Individualinvestor trading and stock returns, Journal of Finance,63, 273–310.
Keim, D. B. and Madhavan, A. (1995) Anatomy of thetrading process: empirical evidence on the behavior ofinstitutional traders, Journal of Financial Economics,37, 371–98.
Keim, D. B. and Madhavan, A. (1996) The upstairs marketfor large-block transactions: analysis and measurementof price effects, Review of Financial Studies, 9, 1–36.
Klemkosky, R. (1977) The impact and efficiency ofinstitutional net trading imbalances, Journal ofFinance, 32, 79–86.
Kraus, A. and Stoll, H. (1972) Price impacts of blocktrading on the New York stock exchange, Journal ofFinance, 27, 569–88.
Lakonishok, J., Shleifer, A. and Vishny, R. W. (1992) Theimpact of institutional trading on stock prices, Journalof Financial Economics, 32, 23–43.
Lee, Y., Lin, J. and Liu, Y. (1999) Trading patterns of bigversus small players in an emerging market: anempirical analysis, Journal of Banking and Finance,23, 701–25.
Lee, Y., Liu, Y., Roll, R. and Subrahmanyam, A. (2004)Order imbalances and market efficiency: evidence fromthe Taiwan Stock Exchange, Journal of Financial andQuantitative Analysis, 39, 327–42.
Lee, C. and Radhakrishna, B. (2000) Inferring investorbehavior: evidence from TORQ data, Journal ofFinancial Markets, 3, 83–111.
Li, W. and Wang, S. (2010) Daily institutional trades andstock price volatility in a retail investor dominatedemerging market, Journal of Financial Markets, 13,448–74.
Lyons, R. (2001) New perspective on FX markets: order-flow analysis, International Finance, 4, 303–20.
Menkhoff, L. and Schmeling, M. (2010) Whose tradesconvey information? Evidence from a cross-section oftraders, Journal of Financial Markets, 13, 101–28.
Ng, L. and Wu, F. (2007) The trading behavior ofinstitutions and individuals in Chinese equity markets,Journal of Banking and Finance, 31, 2695–710.
Nofsinger, J. R. and Sias, R. (1999) Herding and feedbacktrading by institutional and individual investors,Journal of Finance, 54, 2263–95.
O’Connel, P. and Teo, M. (2009) Institutional inves-tors, past performance and dynamic loss aversion,Journal of Financial and Quantitative Analysis, 44,155–88.
Odean, T. (1998) Are investors reluctant to realize theirlosses?, Journal of Finance, 53, 1775–98.
Reis, L., Meurer, R. and Da Silva, S. (2010) Stock returnsand foreign investment in Brazil, Applied FinancialEconomics, 20, 1351–61.
Richards, A. (2005) Big fish in small ponds: The tradingbehavior and price impact of foreign investors in Asianemerging equity markets, Journal of Financial andQuantitative Analysis, 40, 1–27.
Saar, G. (2001) Price impact asymmetry of block trades: aninstitutional trading explanation, Review of FinancialStudies, 14, 1153–81.
Seppi, D. (1992) Block trading and information revelationaround quarterly earnings announcements, Review ofFinancial Studies, 5, 281–305.
Sias, R. (2004) Institutional herding, Review of FinancialStudies, 17, 165–256.
Sias, R. and Starks, L. (1997) Return autocorrelation andinstitutional investors, Journal of Financial Economics,46, 103–31.
Ulku, N. (2008) Do big investors’ trades have predictivepower? A note on Istanbul stock market, BRSAJournal of Banking and Financial Markets, 2, 85–108.
Visaltanachoti, N. and Luo, R. (2009) Order imbalance,market returns and volatility: evidence from Thailandduring the Asian crisis, Applied Financial Economics,19, 1391–9.
Voronkova, S. and Bohl, M. (2005) Institutional traders’behavior in an emerging stock market: empiricalevidence on Polish pension fund investors, Journal ofBusiness Finance and Accounting, 32, 1537–60.
Wermers, R. (1999) Mutual fund herding and the impact onstock prices, Journal of Finance, 54, 581–601.
Wongchoti, U., Wu, F. and Young, M. (2009) Buy and selldynamics following high market returns: evidencefrom China, International Review of FinancialAnalysis, 18, 12–20.
Appendix: Results of the Bivariate VARModel
IRFs from the bivariate model indicate significantdifferences from the SVAR model employed
throughout the article. For example, Fig. A1 belowshows that the response of big player flows to past ISEreturns is highly significant. However the significancedisappears once global returns are controlled for (asshown in Panel B of Fig. 2). Hence, failure to control
Big players’ aggregated trading and market returns in Istanbul Stock Exchange 507
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014
for global returns in an emerging market context, ormore generally failure to control for information,might have caused misspecification and biased resultsin the previous literature. In particular, it may give a
false impression of positive feedback trading by bigplayers at high frequency, when in fact they are simplyresponding to information with differential responseand order execution times.
–200
–100
0
100
200
1 2 3 4 5 6
Fig. A1. Results of the bivariate model: impulse response of big trader net flows to ISE returnsNote: See notes for Fig. 2.
508 N. Ulku
Dow
nloa
ded
by [
Uni
vers
itat P
olitè
cnic
a de
Val
ènci
a] a
t 04:
25 2
1 O
ctob
er 2
014