DOC

Information-Based

By Raymond Perez,2014-07-31 08:19
11 views 0
Information-Based ...

Intertrade Duration and Information-Based Trading

    on an Electronic Order-Driven Market

    *Jiekun HUANG

    #

    Department of Finance, Xiamen University

    Downloadable from the SSRN website

    http://papers.ssrn.com/sol3/papers.cfm?abstract_id=389741

    This research is done under the guidance of Prof. Langnan Chen.

    Many read the manuscript and provided valuable comments

    including Ruey S. Tsay, Jevons Lee, Frank M. Song, Liu Feng and

    Paul Brockman. Special thanks are given to Ouyang Yongwei and

    Huang Houchuan who helped collect the data and provided

    computational tips. Special thanks also go to Daniel Pan who

    provided two references and some helpful advice. Discussions with

    Wu Jiangming helped improve this paper. I’m indebted to my family

    and my dearest Zhang Shuo whose love and encouragement is

    invaluable to me and this research. All remaining errors are my own

    responsibility.

     * Corresponding author: P. O. Box 817, Xiamen University, Xiamen, Fujian 361005, P. R. China; Home Phone:

    +86-592-5911218; Mobile: +86-13015914677; Email: jkhuang77@hotmail.com # 通讯地址:福建省厦门大学817号信箱(361005),电话:0592-5911218,移动电话:(0)13015914677

    电子邮件:easist@163.com

     1

    Intertrade Duration and Information-Based Trading on an

    Electronic Order-Driven Market

    Abstract: Microstructure literature and empirical models offered conflicting prediction regarding the relationship between trading intensity and information-based trading. In

    this paper, I undertake an empirical investigation that is motivated by this conflicting

    theory. Firstly, I develop an asymmetric specification that attempts to capture the

    asymmetric effect of good news and bad news on intertrade durations. One interesting

    point that emerges from the analysis is that good-news-based trading will generally lead

    to increased trading intensity, while bad-news-based trading will generally contribute to

    longer durations. Then I ask whether long durations are associated with bad news. It

    turns out that long durations will lead to declining prices and low volatility; moreover,

    the commonly assumed leverage effect is rejected at the transaction data level.

    Keywords: ACD model, UHF-GARCH model, microstructure, trading intensity, information-based trading, volatility, asymmetric effect, Shanghai Stock Exchange (SSE)

    JEL classification: G10, G15, C10, C41

Contents

1. Introduction .......................................................................................................... 1

    2. Literature Review ................................................................................................ 2

    2.1 Theoretical Models and Some Testable Hypotheses ............................... 2

    2.2 Empirical Studies ....................................................................................... 3

    3. Econometric Framework .................................................................................... 5

    3.1 The ACD Model ......................................................................................... 5

    3.2 The UHF-GARCH Model ......................................................................... 6

    4. Institutional Background and Data Description ............................................... 8

    4.1 Institutional Features of the Shanghai Stock Exchange......................... 8

    4.2 The Data ................................................................................................... 10

    4.3 Preliminary Data Manipulation ..............................................................11

    5. Are Impatient Trades Information-Based? ..................................................... 14

    5.1 Some Peripheral Evidence ...................................................................... 14

    5.2 Auxiliary Regressions .............................................................................. 15

    5.3 A Trading Strategy ................................................................................... 15

    6. Model Specifications .......................................................................................... 17

    6.1 Duration Impact of Information-Based Trading .................................. 17

    6.2 Information Content of Intertrade Duration ........................................ 18 7. Estimation and Results ...................................................................................... 19

    7.1 Estimates of the Asymmetric-WACD model ......................................... 19

    7.2 Estimates of the GJR-version UHF-GARCH model ............................ 23 8. Conclusion .......................................................................................................... 25

    References .................................................................................................................. 26

     3

1. Introduction

    Microstructure literature and empirical models have so far offered conflicting prediction regarding the relationship between trading intensity (the reciprocal of intertrade duration) and information-based trading. Diamond and Verrecchia (1987) consider the impact of short sale constraints on informed trader and conclude that non-trading is indicative of bad news. In Easley and OHara (1992), increased numbers of transactions are due to

    information events and the naturally increased numbers of informed traders. Thus information events will lead to short durations. However, Admati and Pfleiderer (1988) demonstrate that uninformed traders will refrain from trading when there is evidence for informed trading, which implies that informed trading lead to long durations.

    The present paper has twofold purposes. Firstly, I examine the duration impact of information-based trading. So far little empirical studies have explored the causality running from information-based trading to intertrade duration. The lack is partially attributable to the difficulty of distinguishing between information-based trading and liquidity-based trading. An innovation of the paper is employing impatient buys/sells as

    an indictor of information-based trading. Using recently proposed ACD framework, I develop an asymmetric specification that attempts to capture the asymmetric effect of good news and bad news on intertrade durations. One interesting point that emerges from the analysis is that good-news-based trading will generally lead to increased trading intensity, while bad-news-based trading will generally contribute to longer durations.

    On the other hand, does it necessarily mean long durations are associated with bad news and short durations are associated with good news? This has been examined by Engle (2000) but with conflicting results. Thus the second purpose of this paper is to investigate the information content of durations, or more specifically long durations. A GJR-version UHF-GARCH is estimated. It turns out that long durations will lead to declining prices and low volatility; moreover, the commonly assumed leverage effect is rejected at the transaction data level.

    The analysis is based on transaction data for individual stocks comprising SSE 180 Index. As one of the largest emerging markets with an electronic order book, the Shanghai Stock Exchange (SSE) certainly possesses some institutional features to warrant interest.

    The remainder of this paper is structured as follows. In Section 2, I review the microstructure literature and empirical literature, while some testable hypotheses are formulated. Section 3 provides the econometric framework proposed by Engle and Russell (1998) and Engle (2000). Institutional features of Chinas stock market and data

    set used in this research is described in section 4. Section 5 presents evidence that impatient trades are information-based. Model specifications and estimation results are presented in section 6 and section 7 respectively. Section 8 concludes.

     1

2. Literature Review

This section provides a review of theoretical and empirical works on the relationship

    between trading intensity and information-based trading. Some testable hypotheses are

    derived herein.

2.1 Theoretical Models and Some Testable Hypotheses

    The informational role of time between trades is first developed by Diamond and

    Verrecchia (1987) and Easley and OHara (1992). These papers provide partially

    contradictory results regarding the relationship between trading intensity and

    information-based trading.

The intuition behind Diamond-Verrecchias model is that if traders are unable to transact

    in certain states of the world, then non-trading can be informative of the underlying state.

    At the beginning of the trading day, one of the two possible events happens, either good

    news or bad news. If it is good news, informed traders will always buy; while if it is bad

    news and short sale constraints are imposed, informed traders who do not own the stock

    will not trade. Uninformed traders face a similar decision problem but differ from

    informed traders in that their trades are liquidity based, rather than information based.

    This trading behavior means that an absence of trade can occur for three reasons. First,

    the trader selected to trade simply does not want to transact. This decision is independent

    of information on the assets value and so there is no information content to the absence

    of trading arising for this reason. Second, an absence of trade can occur if an uninformed

    trader facing positive liquidity demands is unable to short sell because of constraints.

    Again, this decision is not information-related and so also provides no information to the

    market. Finally, a trader informed of bad news may be unable to trade if short sales are

    prohibited. In this case, observing a non-trading outcome may signal that there is bad

    news about the value of the asset. Two hypotheses that emerged from

    Diamond-Verrecchias model can be summarized as,

H1. Bad news will contribute to long durations.

H2. No trade is indicative of bad news.

    Easley and OHara (1992) introduce event uncertainty into the analysis. In the Easley-OHara model informed traders trade only during days with information events

    that influence the asset price. If the news is good (bad), informed traders buy (sell). In

    contrast, uninformed traders buy and sell at a constant rate regardless of circumstances.

    So if there is no information event, only the uninformed traders trade. Thus traders

     2

watching the market can interpret long durations as evidence that there is no news. Again,

    there are two hypotheses, which are

    H3. Information events, either good news or bad news, will lead to increased transaction intensity. Or equivalently,

H4. No trade means no news.

Admati and Pfleiderer (1988) consider a more sophisticated behavior pattern on the part

    of uninformed traders. Their models include two types of liquidity traders.

    Nondiscretionary liquidity traders must trade a particular number of shares at a particular

    time. In addition, discretionary liquidity traders also have liquidity demands but can be

    strategic in choosing when to trade within a given period of time. A major finding is that

    the discretionary liquidity traders concentrate trading activity in periods where there is no

    indication for informed trading, while they avoid to trade when there is evidence for

    informed trading. In this case, the trading is not related to information event and hence

    volatility would be low just when the market is active. Two hypotheses can be derived.

    These are

    H5. Information-based trading will deter uninformed traders from trading, thus leading to long durations.

    H6. Slow trading is indicative of informed trading and high volatility.

These hypotheses are interrelated and some are contradictory, e.g. H2 vs. H4, H4 vs. H6,

    H3 vs. H5. They can be placed in two groups, with H1, H3 and H5 placed in a group

    labeled duration impact of information-based trading and H2, H4 and H6 placed in a

    group labeled information content of durations.

2.2 Empirical Studies

    With recently developed econometric techniques, these hypotheses are testable. The

    Autoregressive Conditional Duration (ACD) model proposed by Engle and Russell (1998)

    focuses on the time elapsed between the occurrences of trading events and is perfectly

    suited for the analysis of irregularly-spaced high frequency data. Engle (2000) proposes

    adapting the GARCH model for application to irregularly spaced transaction by

    transaction data, thus provides ultra-high-frequency measures of volatility. Following

    1these seminal work, a large body of ACD/UHF-GARCH literature is emerging.

Like the theoretical models, empirical studies so far have not provided unambiguous

    results regarding the relationship between trading intensity and information-based

     1 Engle and Russell (2002) provide a comprehensive review of recent econometric techniques for high frequency

    financial data.

     3

trading.

    Ghysels and Jasiak (1998) develop an ACD-GARCH model, where the intertrade durations determine the parameter dynamics. The results obtained using IBM transaction data show the existence of Granger causality between past volatility and duration and the values of cross-correlation function are positive. Conceiving volatility as an indicator for informed trading, this is consistent with H5.

    Engle and Lunde (1999) make a generalization of the ACD model and formulate a bivariate point process to jointly analyze transaction and quote arrivals. Their empirical application shows that when quotes have not been revised for a long time, i.e. volatility was low, transaction intensities increase, which is consistent with H5.

Dufour and Engle (2000) extend Hasbroucks (1991) VAR model for the dynamics of

    trades and quote revisions to allow the coefficients to vary with time. They use the ACD model for the transaction arrival time. The results obtained using transaction data of 18 NYSE traded stocks show that intertrade duration and price impact of trades are negatively related, which can be interpreted as consistent with H3.

    Zhang, Russell and Tsay (2001) provide a new perspective for the issue. They develop a Threshold ACD (TACD) model to allow the expected duration to depend nonlinearly on past information variables. The model is applied to IBM transaction data. Strong evidence is provided suggesting that fast trading regime is coupled with wide spread, large volume and high volatility, all of which proxy for information-based trading. Thus, the results are consistent with H3.

    In Engle (2000), several specifications of the UHF-GARCH model are proposed and applied to IBM transaction data. A major result of Engles empirical application is that

    both H2, i.e. long durations will lead to declining prices and H4, i.e. long durations contribute to low volatility are supported. Engle does not provide any explanations for this obvious discrepancy.

Grammig and Wellner (2002) propose to extend the recursive UHF-GARCH

    specifications to an interdependent specification in which the transaction intensity impacts on the volatility process, and vice versa. They term it the interdependent duration-volatility (IDV) model. The model is applied to secondary market trading after the Deutsche Telekom IPO. They find that lagged volatility (which is conceived as an indicator for informed trading) has a significantly negative impact on transaction intensity, which is in support of H5.

    Russell and Engle (2002) propose the Autoregressive Conditional Multinomial (ACM) model for price changes which is discrete-valued. Several results emerge from their empirical application to the Airgas (ARG) transaction data. Firstly, long durations are

     4

associated with declining prices, as predicted by H2. Secondly, long durations contribute

    to low volatility per unit time, which is in line with H4.

So far, no empirical work, if any, has explored the asymmetric impact of good news and

    bad news on intertrade duration.

3. Econometric Framework

The availability of high frequency financial data makes it possible for empirical

    investigators to take a close look at the functioning of the market. The analysis of such

    data, however, is complicated by the fact that they are irregularly spaced in time. This

    issue is first addressed by Engle and Russell (1998). They treat the event (transaction or

    quote) arrival times as random variables which follow a marked point process.

In this section, I start by a brief review of the ACD model introduced by Engle and

    Russell (1998). This model shares some features of the GARCH model and is

    particularly well suited to the analysis of irregularly spaced data, such as stock market

    data, where the time elapsed between two trades conveys information.

    In the second part of this section, I discuss Engles (2000) UHF-GARCH model, which is a variant of the GARCH model adapted to irregularly spaced transaction by transaction

    data.

3.1 The ACD Model

    x?t?tLet be the interval of time between event arrivals which will be called the iii?1

    ?duration. Let be the expectation of the duration given the past arrival times which is i

    given by

    E(x|x,x,...,x)??(x,x,...,x;?)?? (3.1) ii?1i?21ii?1i?21i

    Furthermore, let

    x??? (3.2) iii

    ?~i.i.d.where with density p(?;?) with non-negative support, and and are ??i

    variation free. A simple ACD (1, 1) parameterization for the expectation is given by

    ?????x??? (3.3) ii?1i?1

     5

    with the following constraints on the coefficients: , , and . ??0????1??0??0The last constraint ensures the existence of the unconditional mean of the durations; the

    others ensure the positivity of the conditional durations.

Popular choices for the density include the exponential and the Weibull p(?;?)

    distributions. The Weibull distribution has greater flexibility than the exponential one. If

    , the model exhibits a decreasing hazard function: long durations will be less likely. ??1

    If ??1, long durations will be more likely.

The parameters are estimated by MLE. The log likelihood function for the Weibull ACD

    (WACD) is

    ?NT()????????(1?1?/)x?(1?1?/)xii???????ln?ln? (3.4) ?????????xi?1iii??????

    ?for some parameterization of . i

As suggested by Engle and Russell (1998), the performance of the ACD model in

    capturing the autocorrelation structure of the data can be evaluated by examining the

    standardized residuals

    xie ?(3.5) iˆ?i

    ˆ?where is given by ML estimates. The ACD model successfully captures the i

    autocorrelation of the durations if the residuals look like white noise. This can be tested

    with Ljung-Box Q-statistics.

The most basic application of the ACD model to financial transactions data is to model

    thxithe arrival times of trades. In this case it denotes the arrival of the transaction and i

    thth(i?1)idenotes the time between the and transactions.

3.2 The UHF-GARCH Model

    thth(i?1)riLet the return from the to the transaction be denoted by . Define the iconditional variance per transaction as

     6

    V(r|x)?h (3.6) i?1iiiwhere this variance is defined conditional on the contemporaneous duration as well as the

    past returns and durations. The variance of interest, however, is the variance per unit time.

    This is related to the variance per transaction as

    r2i V(|x)??1(3.7) ?iiixi

    2h?x?so that the relationship between the two variances is . iii

The volatility per unit time is then modeled as a GARCH process. Engle proposes an

    rieARMA(1,1) model for the series . Let denote the innovation to this series. If the ixi

    durations are not informative about the variance per unit time, then the GARCH(1,1)

    model for irregularly spaced data is simply

    222?????e??? (3.8) ?1?1iiiEngle terms this model the UHF-GARCH model or Ultra-High-Frequency GARCH

    model.

As suggested by Engle, additional variables, such as contemporaneous duration, lagged

    expected duration, spread and volume, might be introduced into the conditional variance

    equation to test various microstructure hypotheses. In particular, Engle considers the

    following specification

    rrii?1 ???e??e??xii?1ixxii?1(3.9)

    1222?????????? e11??iiixi

    xThe inclusion of in the mean equation has two economic meaning. The longer the i

    interval over which the return is measured, the higher the expected return since both the

    risky and riskless rates are measured per unit of time. However, if no trades is bad news

    as in Diamond and Verrecchia (H2), then long durations would imply declining prices.

    The reciprocal of contemporaneous duration is introduced into the conditional variance

    equation to test the volatility impact of duration. As suggested by Engle, is expected ?

    to have a positive sign under Easley and OHara hypothesis that long durations indicate no news and lower volatility (H4).

     7

Report this document

For any questions or suggestions please email
cust-service@docsford.com