Liquidity and traders' behaviors in financial markets

Liquidity and traders' behaviors in financial markets

HARVARD U N I V E R S I T Y Graduate School of Arts and Sciences

DISSERTATION ACCEPTANCE CERTIFICATE

The undersigned, appointed by the Committee for the PhD in Business Economics

have examined a dissertation entitled

Liquidity and Traders' Behavior in Financial Markets

presented by Laura Elena Serban

candidate for the degree of Doctor of Philosophy and hereby

certify that it is worthy of acceptance

Liquidity and Traders' Behavior in Financial Markets

A dissertation presented by

L a u r a Elena Serban

to The Committee for the PhD in Business Economics

in partial fulfillment of the requirements

for the degree of Doctor of Philosophy

in the subject of Business Economics Harvard University Cambridge, Massachusetts September 2010

UMI Number: 3435462

All rights reserved INFORMATION TO ALL USERS The quality of this reproduction is dependent upon the quality of the copy submitted

In the unlikely event that the author did not send a complete manuscript

and there are missing pages, these will be noted Also, if material had to be removed,

a note will indicate the deletion

UMI Dissertation Publishing

UMI 3435462 Copyright 2010 by ProQuest LLC

All rights reserved This edition of the work is protected against

unauthorized copying under Title 17, United States Code

©20io — Laura Elena Serban

All rights reserved

Dissertation Advisors: John Y Campbell Laura Elena Serban Erik Stafford, Shawn A Cole

Liquidity a n d T r a d e r s ' Behavior in Financial M a r k e t s

ABSTRACT

This thesis consists of three essays on the liquidity characteristics and traders' behavior in the main market for agricultural commodity futures in India, the Na-tional Commodity and Derivatives Exchange This electronic trading platform was launched at the end of 2003 and subsequently became the third largest agricultural futures market globally

The first essay estimates the impact of speculators' capital constraints on their willingness to provide liquidity as measured by trade participation, and on overall market liquidity as measured by bid-ask spread To overcome the standard identifi-cation problem, the study exploits exogenous variation in trading performance in the form of losses in one asset unrelated to the fundamentals of another asset The study finds that a small number of traders accounts for an overwhelming share of trading activity and participate in the market for a large number of commodities Consis-tent with theoretical predictions, a negative shock to these active traders' aggregate capital causes an increase in future bid-ask spread, but the economic magnitude of the estimated effect is small Changes in competition to provide liquidity explain a considerable fraction of the variation in subsequent market liquidity The effect is non-linear: the bid-ask spread is smallest around a natural level of competition, but increases as competition intensity deviates away from this point

iii

Using the same dataset, the second essay investigates sources of traders' superior returns in local commodities Investors bias their portfolios towards local commodi-ties, crops that are differentially grown within lookm of their location, and earn returns in these commodities that are 3.2% higher than in their non-local commodi-ties, even amongst traders who turnover positions frequently This differential is greatest in crops that are weather sensitive and for which India has a high percent-age of world production The results are consistent with traders possessing superior domestic supply information on local commodities because their proximity to crop production causes information acquisition costs to be lower

The third essay analyzes the trading decisions and performance of all three trader categories - individuals, brokers, and commercial institutions - participating in agri-cultural commodity markets in India In contrast to U.S commodity markets, individ-uals represent about 80% of participants by number, and contribute between 40-50%

of trading activity and open interest in the market Client commercial institutions account for less than 5% of overall trading activity, but for up to 35% of open interest; although fewest by number, broker proprietary trading desks account for a large por-tion of trading activity Brokers are the most active group in spread strategies, while both brokers and individuals engage frequently in day-trading activities Broker pro-prietary accounts are highly diversified across commodities trading 14 commodities

on average, compared to about 4 traded by the other types In aggregate, brokers make the largest amount of profits, and they do so consistently over time The mean broker account's profits from both intra-day and overnight profits is almost 40 to 60 times larger than the corresponding profits obtained by the mean client institution or individual In contrast, individuals lose significant amounts of money Trading ac-tivity, open interest and profitability are concentrated within each market participant

group

This study also analyzes the impact of market-wide characteristics, and beyond that, the impact of peer actions and outcomes on individuals' decisions to enter into com-modities futures market Aggregate entry rates of both individuals and companies in the commodity futures market are positively serially correlated, and increasing with trading volume and commodity market returns The actions and market outcomes of local peers affect entry decisions The number of new individual traders in a zip-code

is highly positively serially correlated, and zip-codes with more active participants perience higher entry rates in the future Moreover, the recent returns of individual traders in a zip-code are positively correlated with the future number of individual entries in that zip-code; the influence of peer returns is restricted to situations when neighbors experience negative returns Our findings suggest that information about negative peer performance is more likely to spread among individuals than informa-tion about positive peer performance, or that the individuals in our sample react only

ex-to learning about negative peer returns

v

Table of C o n t e n t s

Abstract iii Acknowledgments viii

i Active Traders, Capital and Liquidity 1

1.1 Introduction 1

1.2 Institutional Context and Data 10

1.3 Identification and Characteristics of Liquidity Providers in an Order

Driven Market 13

1.4 Measuring Liquidity and Capital Availability 24

1.4.1 Liquidity Measures 24

1.4.2 Exogenous Capital Shocks as Trading Revenue in

Fundamen-tally Unrelated Commodities 28 1.4.3 Intra-day Round Trip versus Overnight Inventory Related Rev-

enues 30 1.5 Active Traders' Aggregate Revenues and Individual Security Liquidity 33

1.6 Active Traders' Competition and Individual Security Liquidity 50

2.4.1 Measurement of Returns and Capital 93

2.4.2 Performance for Local Traders 94

2.4.3 Performance and Information on Domestic Supply Shocks 99

2.5 Discussion 108

2.6 Conclusion 110

3 The Trading Decisions and Performance of Various Investor Types: an Anatomy of a

Large Commodity Futures Market 112

3.1 Introduction 112

3.2 Literature Review 119

3.3 Data and Market Rules 128

3.3.1 NCDEX Market Rules 130

3.3.2 India's Commodity Markets: World Placement and Brief History 132

3.4 Market Participants' Types and Characteristics 136

3.5 Trading Activity over Time 145

3.6 Share of Trading Activity of Aggregate Trader Type 150

3.7 Performance of Aggregate Trader Types 157

3.8 The Trading Strategies of Aggregate Trader Types 163

3.8.1 Attrition Levels 164

3.8.2 Common Futures Trading Strategies 168

3.9 Trading Activity Concentration by Trader Type 173

3.10 Heterogeneity of Trading Activity and Performance by Trader Type 181

3.11 Market-Wide and Neighborhood Determinants of Entry Decisions 190

3.12 Conclusion 207

Appendix 211

A Appendix to Chapter 1 2x2

A.i NCDEX Trading Platform 212

A.2 Trading Revenue Decomposition: Intra-Day Round-Trip Trades vs Overnight

Inventory 215

B Appendix to Chapter 3 219

A C K N O W L E D G M E N T S

This thesis owes its existence to the help, support, and advice of many people My deepest gratitude goes to the chair of my thesis committee, Prof John Y Campbell, for his patient guidance, encouragement and excellent advice throughout this study His thoroughness, rigor, efficiency, wealth and generosity of thought will always be an inspiration for me I am forever indebted to my long-standing advisor, Prof Shawn Cole, for his constant support and for his unflinching confidence in my ability to succeed! I have learned a great deal from our frequent research discussions where he pushed me to be refine my ideas, improve my analysis, clarify my interpretations, and always work harder The generosity of his time, as well as the constructive comments

in the final stages of this thesis will never be forgotten; his down-to-earth attitude and positive energy are a lesson for life I was also lucky to benefit from Prof Erik Stafford's advice His imagination, high research standards, insightful comments, and writing advice have helped me to become a more creative and sophisticated researcher I have also benefited greatly from discussion and advice from Prof Jeremy Stein, Prof Josh Coval, and Prof Robin Greenwood I am grateful for their insightful suggestions on the various topics covered in this thesis, as well as for pushing me to think of relevant, hard and out-of-the-box research questions

My academic life was also shaped by wonderful teachers and mentors I was fortunate to have the opportunity to talk extensively with my general examiners Prof Alvin Roth and Prof David Parkes, as well as with Susan Athey They all taught me about the relevance of models and mathematics in the real world, and have kindled a passion for market design, a field to which I hope to contribute some day I would also

like to acknowledge Gary Chamberlain and Jim Stock from whom I learned statistics and econometrics I would have never started a Ph.D in Economics without the inspirational advice of Prof David Laibson, who guided my first steps in economics research

This thesis would have taken a very different shape without my colleague and authors, Stefan Hunt, who generously introduced me to the dataset on commodity futures on which the results in this thesis rely, and to the management team at the National Commodity and Derivatives Exchange in India, which whom he had devel-oped a prior relationship I am thankful for our numerous and long discussions that sharpened my thinking and iorced me to become more organized, as well as for his help with my presentation skills! Despite some rough times in our working relation-ship, our co-authorship has been an invaluable experience for me! My colleagues Erik Budish, Daniel Carvalho, Paul Niehaus, Thomas Mertens, Soojin Yim, Tarek Hassan, Justin Ho, Itay Fainmesser, Elias Albagli, Fuhito Kojima, and Mihai Manea have also provided inspiration and a pleasant work environment

co-I am also indebted to the National Commodity and Derivatives Exchange for their support and assistance in obtaining the data set used in this thesis, and for sharing their extensive experience and insights with me Special thanks are due to the Ramalinga Ramasehsan, Jagdish Choudhry, Anand Iyer, Somesh Vaidya, Ankur Garg, Raj Benahalkar, Nirmalendu Jajodia, Uma Mohan, Ravinder Sachdev, and Sid-dharth Surana I would also like to acknowledge financial support from the South Asia Initiative Fellowship and the Warburg Research Funds, as well as from the Har-vard Business School Doctoral Programs Office Special thanks are due to Janice McCormick, John Korn, Debra Hoss, LuAnn Langan and Jennifer Mucciarone for their care, promptness, and efficiency in running the program

IX

I could not imagine my life in Cambridge and Boston without my Romanian friends A few words cannot summarize the impact they have had on my life I would like to thank Tatiana Truhanov for her optimism and resourcefulness; Cristina Bucur for her kind care; Florin Morosan for his tireless advice, insightful discussions, and wonderful cooking; Alex Salcianu and Emanuel Stoica for always sharing the depth and wealth of their knowledge, and for perfectly organized MIT events, especially the Romanian parties; Andreea Balan-Cohen and Charles Cohen for their original entertainment recommendations; Florin Albeanu for many wonderful coffee breaks; and Mihaela Enachescu for her long and always supportive friendship I am most grateful to Emma Voinescu and Crisii Jitianu for their unquestioning and consistent support when I needed it most, and for all the great times we spent together! I am also grateful to my boyfriend, Dan Iancu, for being next to me throughout my Ph.D journey His perseverance, passion for detail, and kindness have certainly made me

a better person! I will always carry with me his inspired and perfectly timed gifts, including my favorite Starbucks bears, and cherish the memories we built together over the last five years

I would have never been able to complete this Ph.D without my family Their tireless love, unconditional support, and constant encouragement have kept me afloat through the toughest times of my life I owe my direction in life to my father, who has taught me to ask questions, to insist on finding a solution to any problem, to persevere

in spite of any difficulties, and to always believe in myself! I have always tried to emulate his unbounded optimism, as well as his passion for work and research I

am grateful to my mother for her immense love and care, and for never tiring in trying to make me a better person Her kind advice has prevented me from making many mistakes! My sister has always been there for me for the good as well as the

bad, and I could not imagine a better sibling! Her thoughtfulness, reserved manner, and refined taste have always been great resources for me I am also grateful to my brother-in-law for his optimism and humor, and to my cutest baby nephew, who has truly enlightened our lives ever since he came into this world! This thesis is dedicated

to my beautiful family!

XI

i Active Traders, C a p i t a l a n d Liquidity

1.1 Introduction

Many asset classes exhibit significant cross-sectional and time-series variation in liquidity Understanding the causes of variation in liquidity is important for a num-ber of reasons First, according to recent extensions of the capital asset pricing model, liquidity risk is a systematic determinant of asset prices Second, liquidity is impor-tant for our understanding of how traders affect asset prices Third, efficient trading requires liquid markets; thus, understanding what determines liquidity is crucial to the design of financial markets

While there is significant literature on the cross-sectional determinants of ity (Stoll, 2000, 2003), the time-series variation in liquidity received less attention Yet, there is considerable daily variation in the liquidity of a security when measured

liquid-by bid-ask spread, market depth, price impact, or price reversals Moreover, recent studies have suggested that there are common liquidity components within an asset class (e.g., stocks) as well as across asset classes (e.g., stocks and bonds) What are the sources of high-frequency variation in liquidity? To what extent are these due

to changes in adverse selection costs faced by liquidity providers, shocks to their gregate capital, or related variation in the intensity of their competition? Are there liquidity spillovers due to market-making arbitrageurs trading across multiple secu-rities?

ag-In this paper, I attempt to provide answers to these questions using a unique dataset and an innovative empirical strategy The data consist of the entire history of trader-identified order and transaction activity records along with institutional types

of all participants on the National Commodity and Derivatives Exchange (NCDEX)

of India, globally the third largest exchange for agricultural commodity futures from inception on the 15th of December 2003 to the end of the first quarter in 2008 This data-set is particularly suited for my research questions for several reasons First, the NCDEX trading platform is a fully electronic limit order book allowing for the computation of comprehensive liquidity measures Second, over the sample period, close to 85% of trading activity in commodity futures in India occurred on NCDEX providing a centralized, rather than a fragmented picture of liquidity Third, there are

no designated market makers on NCDEX, a feature common across the majority of financial exchanges nowadays, allowing me to investigate the questions of interest in

a general setting Fourth, India is an environment where trading capital resources are likely scarce on a regular basis: NCDEX is a young trading platform in a developing country where banks, and institutional and foreign traders are restricted from trad-ing Most importantly, the combination of the large variety of commodities traded

on NCDEX with the existence of multi-commodity high-frequency traders allows for

an innovative empirical design that isolates relatively exogenous shocks to trading capital

Traditional models useful for the characterization of liquidity point to two main sources The first relates to asymmetric information and suggests that liquidity providers should rationally shade quoted prices to account for the probability of trading with a better informed trader (Kyle, 1985; Glosten and Milgrom, 1985) The second relates to either inventory holding costs and risks in dealer markets (Ho and

2

Stoll, 1983) or to waiting costs and risks in the execution of limit orders (Rosu, 2005)

in order-driven markets These models assume that market makers' trading budgets are infinite Recent work by Brunnermeier and Pedersen (2007) focuses on specula-tors' capital availability and shows that when capital constraints are tight - because of higher margins or trading losses - speculators reduce positions and market liquidity declines Orthogonally, a number of models emphasize that the level of competi-tion among liquidity providers is causally related to market liquidity Grossman and Miller (1988) show that because market makers face fixed costs of monitoring and maintaining a presence in the market and aggregate profits from liquidity provision are bounded, there is an optimal number of such traders Their model implies that deviations away this point may cause a decrease in liquidity Chacko, Jurek, and Stafford (2008) find that a simple imperfect-competition (and full-information) model

of market-making is able to fit a wide range of features of real-word transaction costs Transaction costs are modeled as the rents that a quasi-monopolistic market maker extracts from impatient investors who trade via limit orders Importantly, the magni-tude of these rents depends on the competition from opposing order flow

In this paper, I use a novel empirical strategy to investigate the extent to which speculators' capital constraints and the intensity of their competition cause time-series variation in liquidity First, although, in limit order markets, no trader has an affirma-tive commitment to provide an option to trade at all times, the natural market-maker candidates are participants who trade frequently and persistently - I refer to these traders as active If market-making arbitrageurs arise naturally, they almost surely fall within this group

I find that a small fraction - close to 4% (about 7,400 traders) - of all traders on NCDEX account for slightly more than 60% of the average daily traded volume I

characterize the portfolios, institutional features, trading behavior and performance

of active traders Broadly, I find that the median active trader participates in the ket for about 15 commodities and at least two different commodity types The mean active trader is classified as such in at least 2 commodities with some being active in

mar-as many mar-as 33 distinct commodities About 54% of broker proprietary accounts and 11% of institutional clients are active traders, while only 4% of individual traders are

so Active traders perform particularly well in commodities in which they are fied as such, while losing especially on open positions in other commodities held in their portfolio

classi-I proxy for changes in speculators' wealth by trading revenues on the NCDEX trading platform; specifically, I am able to precisely reconstruct the history of profit and losses of each trader in all commodities in which he participates However, using changes in capital from trading in a certain commodity to identify the link between capital and the liquidity of that commodity is problematic For example, suppose I observe that a wheat trader who suffered a major loss in wheat futures becomes sub-stantially less willing to provide liquidity in wheat Is this because he suffered a neg-ative capital shock, because he updated his view on whether others possessed better information than he did, or because his long-term view on the expected price evolu-tion of wheat has changed? Even with precise trading data, it is nearly impossible

to discern between these motives The ideal way to address such issues is to identify exogenous shocks to speculators' capital and competition intensity and to examine market liquidity around such events My novel empirical strategy draws on this in-sight and uses three features of the data I use the fact that active traders participate

in the market for multiple, diverse commodities For a commodity, such as wheat, I can separate the remaining contracts traded into fundamentally related futures, e.g.,

4

other cereals, and fundamentally unrelated futures such as metals or spices I then proxy for an exogenous shock to a liquidity provider's capital with a gain or loss in one market unrelated to the fundamentals of another market Commodities are the appropriate asset class for my methodology In contrast to equities and bonds, which co-move with the market portfolio in commodity space, I can track pairs of securities that are fundamentally unrelated by measuring their price correlation and selecting low correlation pairs Finally, I decompose a trader's daily revenue into a component due to intra-day round-trip trades within a contract and a component due to holding

a position overnight for at least one day As trading revenue from intra-day, trip trades may be mechanically correlated with future liquidity measures 1 , I argue

round-that an active trader's second type of revenue, derived from fundamentally lated commodities, serves as a source of plausibly exogenous variation in his capital, which allows me to directly identify the relationship between capital and willingness

unre-to trade and provide liquidity The identification restriction is that changes in prices

of unrelated commodities have no effect on a trader's desire to trade in a commodity, except through affecting that trader's capital

I quantify the effect of plausibly exogenous variation in active traders' aggregate capital on overall individual security liquidity as measured by average daily bid-ask spread While my estimates are consistent with theory - an exogenous negative shock to active traders' aggregate capital causes an increase in next-day liquidity - the economic magnitudes are small To my knowledge, this paper is the first to provide credible causal estimates of the capital channel for liquidity in a competitive market

be larger on days when the bid-ask spread is large If liquidity measures are persistent over time, then intra-day revenues may be mechanically positively associated with short-run liquidity The concern extends to intra-day, round-trip revenue from uncorrelated commodities, if for any reason, liquidity across price uncorrelated commodities has common components due to causes other than constraints

or competition among multi-commodity trading arbitrageurs

environment In order to further investigate the impact of capital shocks on liquidity provision, I turn to active traders' participation decisions Although I do find that

a decrease in capital due to trading in an unrelated security, lowers participation as measured by a participation indicator, number of trades or traded value and that this effect is non-linear, the magnitude of the estimated coefficients is again economically small

Interestingly, I also find that the effects of intra-day round-trip trading losses on liquidity as well as on participation levels in a particular commodity are considerably larger than those due to revenues from overnight positions even when the loss comes from trading in unrelated commodities However, as pointed out earlier, the former effect could potentially be due to a mechanical correlation

I next ask whether changes in competition for liquidity provision cause variation

in subsequent market liquidity Intuitively, profitable active traders are natural petitors for liquidity provision As such, for each commodity, I measure competition intensity as the ratio of profitable active traders to the total number of active traders Following a similar empirical design as above, I find that plausibly exogenous shocks

com-in competition for liquidity provision due changes to com-in the number of wcom-inncom-ing tive traders in uncorrelated commodities explain a large fraction of the variation -between 20-30% - in subsequent bid-ask spreads The effect is non-linear: the bid-ask spread is smallest around a naturally optimal level of competition, but increases

ac-as competition intensity deviates away from this threshold An increac-ase in the ratio

of winning active traders in uncorrelated commodities to the total number of active traders from o to 0.4 corresponds a decrease in next day bid spread of 20 basis points (50% of a standard deviation) However, an identically sized increase in this ratio beyond the 0.5 cutoff corresponds to an increase in next day bid-ask spread of 18 to

6

20 basis points

In general, the magnitudes and statistical significance of the estimated effects of interest are larger when predicting liquidity in the non-nearby contracts than in the nearby contract 2 This is encouraging because the degree of non-synchronicity in natural customer order flow is higher in non-nearby contracts and hence shocks to liquidity supply from market making intermediaries are expected to have a more pronounced impact on the aggregate liquidity of these contracts

My results suggest that although on the NCDEX market, intermediary-speculators' capital constraints do generate spill-overs in participation and liquidity from one se-curity to another, such spill-overs are economically small In contrast, competition spill-overs defined as competition among active traders who made profits in other, unrelated commodities appears to be an economically important channel

My paper is most closely related to a study by Camerton-Forde, Hendershott, Jones, Moulton, and Seasholes (2008) who show that aggregate market and specialist-firm level effective bid-ask spreads widen following periods when the NYSE special-ists have large positions or lose money, suggesting that market makers' financing constraints are associated with lower future liquidity Several earlier studies have also suggested that this channel is important for the time-series variation of liquidity Hatch and Shane (2002) show that specialist firm acquisitions are followed by a lower bid-ask spread in stocks assigned to the acquired firm potentially due to an easing

of financing constraints Coughenour and Saad (1998) compare the liquidity teristics of assigned stocks across specialist firms with different organizational forms showing that the stocks assigned to partnerships whose access to capital is tighter relative to firms capitalized by larger parent companies exhibit worse liquidity fea-

month On a given day, the contract records the highest traded volume

tures There are at least three drawbacks to these studies First, they all focus on NYSE specialist firms and hence are relevant only to this specific market structure Second, the sample period examined ends before 2007 when the NYSE adopted its electronic hybrid market structure, significantly decreasing the importance of spe-cialists for liquidity provision Most importantly, none of these estimates can be interpreted causally My main contribution to this literature is to provide plausible causal estimates of the capital-channel for liquidity variation In addition, my setting

is that of an electronic, limit order market with no designated market makers, the market structure most common nowadays

My study also relates to an extensive theoretical (Amihud and Mendelson, 1980;

Ho and Stoll, 1981, 1983) and empirical literature (Hasbrouck, 1998a; Hasbrouck and Sofianos, 1993; Hasbrouck, 1998b; Naik and Yadav, 2003; Wahal, 1998) that character-izes the trading behavior and market quality impact of designated dealers focusing

on specialists on the NYSE and dealers on NASDAQ For example, the empirical study of Hasbrouck and Sofianos (1993) investigates the trading activity of specialist

on the NYSE, finding that they make profits largely from short-term market making activity rather than from long-term value positions Wahal (1998) finds that the entry and exit of dealers on NASDAQ has a significant effect on spreads in addition to other known determinants However, he does not have access to dealers' positions and performance and hence cannot directly estimate the link between capital and liquidity

My work indirectly contributes to the recent literature that exhibits common tors in the liquidity of stocks (Chordia and Subrahmanyam, 2005a; Huberman and Halka, 2001; Hasbrouck and Seppi, 2001) and across different asset classes such as stocks and bonds (Chordia and Subrahmanyam, 2005b) While studies such as that

fac-8

of Chordia and Subrahmanyam (2005b) and Hameed, Kang, and Viswanathan (2008) point to macro-economic factors as driving such commonality, others suggest that it could be due informational spill-overs or capital constraints of speculators trading across different asset classes For example, Coughenour and Saad (1998) find that stocks traded by the same specialist firm share a common liquidity component and suggest that this might be due the specialist firm's capital constraints I contribute

to this debate through my empirical design, which strives to distinguish between the informational and capital constraints channel

Through the use of data from commodity futures markets, my study also relates

to an older article that examines the activities of floor traders in CBOT's futures pits (Manaster and Mann, 1996) The authors focus on cross-sectional relationships be-tween market makers' inventory positions and document that they control inventory throughout the trading day However, they also show that floor traders' behavior con-tradicts typical inventory control models as their inventories and reservation prices are positively correlated consistent with active position taking My work examines naturally arising rather than designated market makers and focuses on the effect of capital constraints and competition intensity on liquidity

The paper proceeds as follows Section 1.2 briefly describes the institutional text of commodity futures markets and the National Commodity and Derivatives Exchange in India, the data and my sample selection criteria Section 1.3 describes

con-my procedure for the selection of active traders as well as these traders' portfolios, trading and performance characteristics Section 1.4 describes the main features of my methodology, which isolates exogenous shocks to active traders' capital Sections 1.5 and 1.6 focus on the impact of active trades' capital and competition intensity on the liquidity of the securities in which they trade Section 1.8 evaluates the effect of

capital availability on active traders' participation decisions In section 1.7 I perform robustness checks Section 3.12 concludes and outlines future research directions

1.2 Institutional Context and Data

NCDEX is one of the 3 national exchanges recognized by the Indian commodity regulator, the Forward Markets Commission, and the Government of India, after the policy decision taken early 2003 to fully lift the ban on commodity futures trading

3 NCDEX functions as an electronic limit order book providing commodity futures

in 98 commodities during my sample period broadly in 4 categories - agricultural produce, precious metals, non-ferrous metals and energy Additional details on the the rules governing the NCDEX trading platform are available in Appendix A.i At the end of 2008, NCDEX accounted for 25% of the total Indian commodity futures market, but more than 85% of the agricultural commodities market MCX, its main competitor, dominates trading in metals and energy

Table 1.1 presents the evolution of the exchange by year and commodity type over the entire sample period between 2004Q1-2008Q1 During this period, the av-erage daily one sided open interest value was $711 million and average daily traded value $583 million; 98 commodities were traded on the exchange, of which 75 were agricultural commodities In total, 160,045 traders and 835 members participated in the market Members of the exchange act as brokers and may trade on proprietary accounts Approximately 85% of the traders are classified as individuals, with the

launched in Ahmedabad in November 2002, followed by two exchanges in Mumbai, the Multi ity Exchange (MCX) launched in October, 2003 and the National Commodity and Derivatives Exchange (NCDEX) launched on December 15, 2003 During my sample period, commodity futures traded on the 3 national exchanges and 22 local exchanges However, trading volume concentrated on national exchanges with 99% of traded value occurring on NCDEX and MCX by 2006

Commod-10

remaining accounted for by domestically registered commercial companies, financial securities firms and brokers In contrast to markets in developed economies, partici-pation in commodity futures markets in India was severely restricted over my sample period 4 with banks, mutual funds, hedge funds and foreign investors forbidden from trading

My study uses proprietary data including the entire trading, holding, order, and best bid-ask records of the exchange since its inception on December 15, 2003 to March 31, 2008 The data come from the Surveillance Files of the trading platform, and thus they are complete to the best extent possible as well as highly reliable In the following, I describe my data and sample selection

Each trading record consists of a timestamp (seconds granularity), contract bol, buy/sell indicator, quantity traded, and transaction price For each trade, I can identify both parties to the trade as well as the unique order sequence number that generated the trade These data allow me to precisely compute the open position value and trading revenue for each market participant at high frequency

sym-Bid-ask data by contract are available for the period July 14, 2004 and March 31,

2008 I employ all the bid-ask observations to compute average daily percentage and dollar bid-ask spreads by commodity and contract Liquidity measures are computed separately for the nearby contract and other active contracts traded concomitantly For commodities that began trading after the first date of the bid-ask series, I remove the first 20 days of trading in order to exclude potential outliers due to thin markets at contract launch In addition, I remove observations on trading days when the bid-ask spread is available, but no trades took place in the nearby contract

Information on the institutional types of approximately 90% of the accounts is

barriers

Table 1.1: Exchange Summary Statistics

The table displays summary statistics on the trading activity, traded commodities and participants on NCDEX over the sample period, and commodity type

(agricultural, metals and energy)

Total Agri Metals Energy

4 5 6

10,193 1,113

6 3 7

3 4 , 5 2 4

3 , 5 1 2 9,652

3 , 6 4 9

7 6 7

5 9 , 8 7 0

4 , 1 5 8 1,964

available as follows: individual/retail trader, company and its organizational form, broker-member proprietary account I use these data to characterize the active traders

in my sample

Finally, daily spot prices5 and close futures prices have been collected from the NCDEX website I use both prices series to determine zero price-correlations com-modities and close futures prices to mark-to-market end of day traders' positions For my analysis, I restrict the sample to the subset of agricultural commodities with successful futures contracts for which liquidity measures based on the NCDEX data are more likely to be representative Specifically, I select commodities with at least 100 days of trading, and 5 trades and 10 participants daily This filter yields

36 commodities Of these, I drop 5 symbols that correspond to contracts for minor commodity varieties that trade concomitantly to the main contract Thus, my final sample consists of 31 agricultural commodity symbols Table 1.2 provides summary statistics on the trading activity in these commodities Importantly, while I analyze active traders' liquidity provision and overall liquidity impact on selected commodi-ties, capital availability proxies are based on their entire activity on the exchange

1.3 Identification and Characteristics of Liquidity Providers in an Order Driven Market

Identifying traders who might voluntarily engage in market making strategies in

an order driven market is not trivial Neither the theoretical market microstructure literature, nor exiting empirical evidence on such markets provide definite guidance

in this respect In keeping with the traditional literature characterizing the nants of liquidity provision of designated dealers and specialists, the theoretical work

determi-5 Polled by the exchange at the main delivery location of a commodity

Table 1.2: Trading Activity Summary Statistics: Representative Commodities

The table includes activity summary statistics for the most highly traded

commodities on NCDEX It shows the number of participants (# Accts.), traded value

(Trd Val.), number of trades (Trd Val.) and open interest value (OI Val)

pepper pigeon p e a / t u r soybean

chilli rape-mustard guar gum sugar wheat turmeric cotton sugar/gur wheat corn mentha oil raw cotton/kapas castor seed

jute barley potato cotton lentil soybean meal gold

silver steel copper brent crude oil light sweet oil

# Accts

('000) 27.22 25.50

2 1 0 1 20.46 9.18 23.90 13.80 12.85 14.90

9.48 7.63

16.76 14.32 10.74 9.08

5-93

10.61 6.29 4-73

4 2 1 1-73 2.64 0.93

Trd Val

($m)

156.54 111.20

7 2 6 3

35-47 34-49 33-54

2 1 2 8 19.37 16.80 16.52 13-57 12.63 11.60

8.45

5.70

4.69 3.98 3.76

3.28 3.22 1.63 1.07 0.79 0.68 0.50 0.21 0.15

4 5 8 2

6 3 7 1

4 0 9 6 31-85

68.64

19.67

4 1 7 8 25-75

48.48 38.96

14.45 12.15 12.13

2 1 8 5 11.99

6.95 6.88

6.52 4-34 2.39 1.56 1.67 0.86 0.92

35-33

3 9 0 7 12.28 2.60 3.92 1.70

OI Val ($m)

181.96 108.96 74-59

1 8 8 1

7-95

9.04 6.14

6.66 6.99

3-25 2.35

2 4 7

2 8 2 0.52 0.30 0.31 17.24

4 7 1 1

4-85 3-93 3-47

1.81

14

on order driven markets refers to the party that submits the limit order (as opposed

to the market order) within a transaction as the liquidity provider for that tion However, as emphasized by Hasbrouck and Saar (2005) in real markets, whether

transac-a limit order's ustransac-age conforms to its theoretictransac-al perspective transac-as transac-a ptransac-atient supplier of liquidity is idiosyncratic and depends on the technology and trading culture of the market For instance, the authors find that, on INET, a substantial portion of limit orders are canceled within an extremely brief time thus more resembling market or-ders that search for immediacy from hidden liquidity rather than supply liquidity themselves Thus, it is not evident that an empirical criterion for selecting voluntary market makers should be solely based on limit order usage Moreover, even if fea-tures of limit order usage are taken into account, one should consider the contribution

of a trader's limit order volume relative to the overall depth of the market as well as the share of limit orders relative to total orders submitted by that trader For example,

it is not at all clear that an infrequent trader who always uses limit orders in a futures market should be classified as a market making intermediary; rather, it may simply

be a patient trader who engages in the market for natural hedging reasons Further, any trader may optimally choose to trade with limit or market orders depending on the state of the order book as characterized by the size of the bid-ask spread and the depth on each side of the market (Parlour, 1998) While it is unlikely that a success-ful market making strategy can be sustained by trading with market orders, even a market maker may choose to employ market orders when constraints imposed by capital availability or risk aversion dictate quick inventory reduction through trading

in a specific direction In addition, much of the compensation for agents who vide liquidity in price pressure situations may come through price reversal Clearly,

pro-if an agent uses limit orders to accommodate price pressure, he gains the full bid-ask

spread on round-trip trades and hence the realized profit will be higher However, it

is still conceivable that, in price pressure situations, an agent may use market orders

to accommodate increasing aggressive limit orders on the other side of the market Another potential promising line of inquiry attempts to draw on the behavior

of designated market makers and specialists (Hasbrouck and Sosebee, 1993) Such traders have an obligation to post quotes for a specified volume on both sides of the market Thus, one might think that voluntary market makers should exhibit similar behavior However, both the academic views and empirical evidence are split on this subject On the one hand, Hamao and Hasbrouck (1995) refer to member proprietary account market making on the Tokyo Stock Exchange as placing limit orders on both sides of the market simultaneously or within a short period of time On the other hand, Aitken and Mclnish (2007) provide evidence that hedge funds following mar-ket making strategies on the Australian Stock Exchange do not simultaneously post limit orders on both sides of the market Rather they post multi-price limit orders

on the same side of the order book at different points in time Thus, the existing evidence suggests that posting buy and sell orders simultaneously may be too strict

of a requirement for voluntary market makers

Finally, one may think that certain institutional types are more natural candidates for market making than others However, again, the existing empirical evidence is not consistent in this regard On the one hand, two recent articles (Kaniel and Liu, 2006; Kaniel, Liu, Saar, and Titman, 2008) suggest that in the U.S equity markets, indi-viduals behave as risk-averse, uninformed liquidity providers to institutions and that their performance is explained by compensation for this service On the other hand, Hamao and Hasbrouck (1995) suggest that proprietary broker a n d / o r hedge funds are superiorly positioned to provide liquidity in the light of their large scale, lower

16

fixed costs of monitoring the market as well as smaller transaction costs Further, Menkhoff and Schmeling (2009) provide evidence that in currency markets informed traders dominate the dynamics of liquidity provision, while uninformed traders are insensitive to spreads, volatility and market depth However, in an experimental elec-tronic market, Bloomfield and Saar (2004) find that informed traders sometimes take liquidity with market orders (earlier in the trading day) so as to profit from their private information and sometimes use their knowledge of the true price to act as dealers by switching to submitting limit orders and earning the bid-ask spread (later

in the trading day) As such, the institutional characteristics of a trader may not proxy well for market making activity

Thus, the current literature at best suggests a number of order and trading egy features that are indicative of liquidity provision First, liquidity providers are likely to persistently facilitate a large fraction of trades Second, on regular days they are likely to trade a lot relative to end of day inventory positions as well potentially to trade against the order flow in instances of price pressure. 6 Third, they are likely to employ relatively more limit orders than market orders; more importantly, the limit orders of market makers are likely to be aggressively priced and placed at or beyond current quotes as well as canceled infrequently Ideally, one would construct a trader-commodity pair specific, daily index of liquidity provision combining the measures above, potentially based on the first principal component of the measures While it would be interesting to study the empirical properties of this index - for example, whether it persistent over time and across commodities, its calculation at our desired level of granularity is computationally challenging

and after the order submission or modification measures this feature Negative covariance indicates trading as a liquidity provider, while positive covariance indicates price chasing The stronger the absolute value of the correlation, the stronger the particular strategy followed

Instead, in this study, I opt for a sequential procedure for the identification of market makers by first using the most essential feature of their trading activity That

is, I consider as potential liquidity provider any trader who persistently participates

in a large fraction of daily traded volume in the recent past Specifically, I identify

po-tential liquidity providers in each commodity on each trading day based on the level

of trading activity in past 60 trading days (approx 3 months of trading) by requiring

a trader to be in the top decile of trading volume (based on contracts of all delivery months) for at least 10 trading days 7 I reclassify traders into the liquidity providing category daily In effect, I focus the analysis on active traders My choice is justified

by the fact that all views of market making imply frequent and consistent trade ticipation Moreover, it is important that I focus the analysis on active traders who are likely to be influential for the quality of the market

par-The empirical behavior of voluntary market makers is as yet unstudied in the literature The comprehensive nature of my data set allows for an understanding

of the market and its intermediaries in a way that has not been feasible in previous studies As such, in the following, I examine the features of persistent, active traders,

a group likely to be a tight approximation to the set of voluntary market makers

on the NCDEX platform, in terms of market share, institutional features, portfolio characteristics, trading behavior, and overall performance

Table 1.3, Panel A justifies my selection focus on the top decile of traders ranked

by volume of trading as they account for a large portion of trading volume On the average commodity trading day, the top decile of traders accounts for 62% of trading volume; and the cutoff for being classified in the top decile of trading is, on average, 13% of trading volume While there is some variation across commodities, the top

7 In robustness checks, I also consider a 30 day alternative to the 10 trading days cutoff However, the basic findings of the study do not change

18

decile of traders accounts for at least 50% of traded volume on the average day and the cutoff for the top decile is at least 8% However, only a small fraction of traders stay in the top decile for a large number of days as the 50th decile of maximum number of days spent in the top decile of trading of a commodity by a trader during the sample is 56 trading days (approx 3 months)

Table 1.3, Panel B provides evidence on the institutional types of the large, tent, active liquidity providers in my market Several observations are in order First, 54% (417) of the member proprietary accounts have been classified in the liquidity providing category By and large, members are institutions About 11% (1,113) of the non-member institutions are also classified in the liquidity providing category Fi-nally, there are 4,777 individuals, but they represent only 3.7% of the total individual accounts that trade on the exchange 8 This finding is consistent with the empirical hypothesis in the market microstructure literature that a "group of traders would naturally gravitate towards functioning as de facto dealers" (Hamao and Hasbrouck, 1995) and that such traders are most likely to be members of the exchange due to their structural advantages in terms of lower trading costs (as they do not pay the additional brokerage fees), lower fixed costs of monitoring the market and access to customer order flow

persis-Table 1.3, Panel C display statistics on the total number of active traders and their share of the marker overall and cross-sectionally over sample commodities For the overall values, on each day I average the relative percentages accounted for by the liquidity providers in each commodity and I present the daily averages of these quantities Within a particular commodity I average the relevant values across trading days My selection procedure yields 7,387 traders representing only 4.7% of all the

traders in the sample There is considerable variation across commodities in terms

of the number of liquidity providers, with a minimum of 12, a maximum of 2013 and a median of 269 traders per commodity, with the more liquid commodities by traded value and number of participants having a larger number In percentage terms, between 1% and 4% of all traders in a commodity are classified as liquidity providers On the average trading day, these traders participate in 64% of the trading volume and 61% of the trades However, only 30% of the trading activity comes through executed limit orders Perhaps surprisingly, active traders account for as much as 55% of total open interest on a daily basis and do not necessarily hold end of day positions on the same side of the market as their net open interest accounts for only 11% of total open interest on the average trading day Interestingly, active trades

do trade considerably with each other: 80% of their trading activity (by volume or number of trades) occurs within the group of active traders While there is some variation across commodities, the overall averages are not driven by any particular commodity

Table 1.3, Panel D shows that the median (average) active trader trades in 20 (30) different commodities during my sample, across all commodity sub-types and in at least 2 commodity types Moreover, the mean active trader is classified as such in at least 2 distinct commodities and some are classified as such in at least 5 commodities (one standard deviation away from the mean) Thus, the majority of active traders derive trading revenue from a large number of diverse commodities The mean active trader does not necessarily specialize in a single commodity The average Herfindahl index of active traders' portfolios also indicate a solid level of diversification I also confirm that some traders are active across multiple commodities at the same time

2 0

Table 1.3: Active Traders' Characteristics Panel A provides information on the guiding cutoffs used to select active traders

Panel B presents the institutional break-down of selected active traders The main

categories are broker-proprietary trader versus client trader, which is split further

into individuals and commercials traders Panel C displays summary statistics on the

trading activity of active traders as a share of the overall market The first column

shows equally weighted means across all sample commodities and trading days

The remaining columns display cross-sectional variation across commodities by first

computing the commodity-specific daily averages for each measure and then

reporting statistics over these averages Panel D presents evidence on the degree of

diversification of active traders' trading portfolios: number of commodities traded

by active traders, number of commodities in which they are classified as active and

non-active, and the Herfindahl index for portfolio concentration

Panel A: Cutoffs used in Active Traders' Selection

Trd Vol % Cutoff (%) Dist # Days Top Decile

Top Decile Top Decile 25% median 75% 95%

61.79 13-87 12 56.5 370 1010

Panel B: Active Trader of Institutional Type

# Prop # Client # Client Inst # Client Ind # Client Unknown (% All Prop.)(% All Clients)(% All Client Inst)(% All Client Ind) (% All Unknown)

417 6970 1113 4777 1080 (54-09) (4-39) (11.81) (3-72) (4-93)

Table 1.3 (Continued)

Panel C: Active Traders' Share of Market Trading Activity

All Comm Variation Across Sample Commodities

min mean median std max

% Num Trd from Limit Order 29.32

% Net Open Interest 10.94

% Abs Open Interest 54-75

% Vol Trd other Active Traders 83.72

% Num Trd other Active Traders 80.47

12 508.70 269 552.71 2013 0.96 2.21 2.22 0.59 3.62 32.92 62.33 62.47 11.27 82.07 12.65 29.74 30.27 6.25 40.69 33.40 59.77 60.97 10.51 77.24

4.19 10.17 9.65 4.26 21.61 29.74 54-24 55-73 10-07 71-75 62.47 87.15 86.23 8.47 100.00 60.00 78.76 81.61 16.87 100.00

Panel D: Active Traders' Portfolios

min mean median std max

# Comm Traded

# Comm Sub-Types Traded

# Comm Types Traded

# Comm as Active Trader

# Comm as non-Active Trader

HHI based on Traded Value

Table 1.4 presents summary statistics on the trading revenue and end of day sitions of active traders For each trader, the results are split across commodities in

po-an active traders' portfolio in which he is classified as active (C-LQ) or non-active

(C-NonLQ), respectively Further, gross trading revenue is disaggregated according

to revenue from intra-day round-trip trades and revenue from end of day positions held for at least one night The exact definitions of these variables can be found in the appendix End of day notional position values are disaggregated as follows: the absolute value of daily positions, the absolute value of net open position value and the net open position value of a trader I find that active traders perform signifi-cantly better and hold larger overnight positions in ihe commodities in which they are classified as active than in other commodities in their portfolio In fact, the entire distribution of trading revenues and end of day positions by trader for active com-modities is shifted to the right relative to that for non-active commodities Notably, for non-active commodities traders lose large amounts on overnight positions

1.4 Measuring Liquidity and Capital Availability

1.4.1 Liquidity Measures

I use the average daily percentage bid-ask spread as my main measure of liquidity For a given commodity, this is computed daily as the average of the ask less the bid over the midpoint value for all the observations available during the trading day and is expressed in percentage terms The average daily bid-ask spread is computed separately for the nearby contract and the remaining contracts trading concurrently Robustness checks are executed using average daily dollar bid-ask spread

Figure 1.1 displays the time-series variation in daily proportional and dollar

bid-ask spread for eight selected commodities Table 1.5, Panel A presents summary

24

statistics for the percentage bid-ask spread and dollar bid-ask spread for the sample

of agricultural commodities split by nearby and other contracts The mean dian) daily proportional bid-ask spread in the nearby contract is 0.451% (0.207%) and 0.813% (0.424%) for the remaining contracts with a standard deviation of 0.651% and

(me-9,874%, respectively The results in table 1.5, Panel B show that commodity fixed

effects explain less than 0.5 of the variation in the daily average bid-ask spreads thermore, crop-season fixed effects explain as little as 0.003 °f m e variation in bid-ask spread Nevertheless, the coefficient on the seasonal dummy, albeit small, is positive and statistically significant, indicating that on average bid-ask spreads are slightly higher during growing season consistent with the theoretical prediction that liquidity decreases during periods with higher information asymmetry (in this case, supply related)

0 02 04 06 08 1 Avg Daily S Spr

.1 2 3 4 Avg Daily S Spr