First Draft: March, 2004
Current Draft: January, 2005
Preliminary Draft: Please do not quote
The Information Content of Trader Identification
Alex Frino
David Johnstone
and
Hui Zheng*
Abstract. This paper develops an information-based sequential trade model which predicts that sequential transactions executed by the same broker provide a rational basis for market reassessment of the value of a security, and that this effect is stronger in the presence of greater information asymmetry. Both theoretical propositions are tested using a sample of transactions drawn from the Australian Stock Exchange – a market in which broker identity is transparent. Evidence is provided that consecutive buyer-initiated (seller-initiated) transactions by the same buying (selling) broker have a greater permanent impact on security prices than matching transactions made through different buying (selling) brokers. The price effect of repeat trades by the same broker is more pronounced for stocks with higher expected adverse selection costs and during intraday time intervals when adverse selection costs are highest. These results are consistent with the proposition that trader identity conveys information to market participants.
* Finance Discipline, Faculty of Economics and Business, University of Sydney, NSW, 2006, Australia.
Mail: Finance Discipline, Faculty of Economics and Business, University of Sydney, NSW 2006 Australia. Phone: (61 2) 9351 6451. Fax: (61 2) 9351 6461.
Email:
This research is funded by the Capital Markets Cooperative Research Centre. The authors would like to thank the Australian Stock Exchange for the provision of data and the Securities Industry Research Centre for technical assistance.
1. Introduction.
A number of previous studies demonstrate that trader anonymity influences market behaviour and has economic consequences. For example, Forster and George (1992) develop a theoretical model in which they assume that the identity of a trader provides information about the direction and magnitude of liquidity trade. They demonstrate that anonymity influences the cost of executing trades and market depth. Garfinkel and Nimalendran (2003) demonstrate that in less anonymous markets, market makers set tighter spreads in the presence of informed traders suggesting that they are able to identify informed traders. Grammig, Schiereck and Theissen (2001) demonstrate that informed traders prefer trading in anonymous markets. While these papers recognise that anonymity is important and has economic consequences, they do not specifically identify the mechanism by which the motivations of traders are revealed through knowledge of the identity of traders.[1]
Benveniste et.al. (1992) identifies one way in which the identity of informed traders is revealed. They argue that relationships can develop amongst brokers in a market setting with lower anonymity. This, in turn, can lead brokers to reveal the motivations of their clients. This paper extends this research by identifying another way in which informed traders are identified in market settings with lower anonymity. Specifically, we develop and test a model in which two successive buyer or seller initiated trades executed by the same broker imply that the underlying trader is informed and results in a reassessment of the value of a security.
This paper is also related to the literature examining the way in which information is revealed through trading. It is well established that permanent stock price movements are due largely to private information revealed through trading rather than public information releases [eg. French and Roll, 1986; Barclay, Litzenberger and Warner, 1990]. Furthermore, a number of studies recognise that various characteristics of transactions can be used to infer the motivations for trade and consequently result in permanent price movement. These include trade size [eg. Easley and O’Hara, 1987; Halthausen, Leftwich and Mayers, 1987]] and trade frequency [eg. Easley and O’Hara, 1992; Chan and Fong, 2000]. This paper extends this literature by identifying another specific characteristic of transactions through which information is revealed – trader identity.
Given that anonymity has economic consequences, it is an important element in the design of markets. In contrast to predominantly floor traded markets (eg. NYSE and CME), screen traded systems are capable of delivering greater anonymity. In fact, there is considerable diversity in this aspect of market design around the world. While some electronically traded markets disclose the identity of the parties involved in a transaction (eg. Australian Stock Exchange and Hong Kong Stock Exchange) others are completely anonymous (eg. SETS on the London Stock Exchange and Superdot on NYSE). Furthermore, a number of markets have recently moved to anonymous trading. For example, on 1 July, 2003 the Toronto Stock Exchange stopped releasing real time transaction data containing broker identifiers and now provides brokers with the option of choosing whether they want their identity to be disclosed.[2] Similarly, in September 2003, the SEC approved a proposal by Nasdaq to implement fully anonymous trading for certain transactions executed on SuperMontage.[3] In another twist to this diversity, some markets allow traders to choose whether to remain anonymous (eg. Swiss Exchange). Foster and George (1992) recognise that research examining the impact of anonymity is important to assessing the appropriateness of less anonymous markets (eg. floor traded markets or less anonymous electronic trading systems). Accordingly, the evidence provided in this paper is relevant in assessing, at least partly, the appropriateness of anonymous trading.[4]
The empirical analysis in this paper is based on the Australian Stock Exchange [herein ASX]. The ASX provides an opportunity for examining the market reaction to trader identity. Unlike electronic trading systems used in US markets, the ASX fully discloses the identity of brokers executing transactions, which in turn permits an examination of the way in which the market reacts to such disclosure. This is not possible in anonymous markets. Furthermore, the fully electronic nature of the ASX produces near perfect data. This allows a great deal of precision in pinpointing transactions and classifying trades into buyer and seller initiated, as well examining price behaviour using a very fine observation interval – all of which prove critical to detecting the price reaction to trader identity.
In this paper we demonstrate one of the ways in which broker identity can convey information to the market. We demonstrate that 2 sequential transactions by the same broker, in the same direction, has a permanent impact on security prices.[5] Furthermore, we demonstrate that the impact of such transactions is greater for stocks with higher estimated adverse selection costs and in the first half-hour of the trading day These results are consistent with the notion that trader identity conveys information, and that information content is greater in the presence of higher adverse selection.
The remainder of the paper is organised as follows. In Section I, some theory which explains the way in which trader identity conveys information is developed. Section II describes the data and method used in testing the impact of trader identity on securities prices. Section III presents the results, while the final section concludes and provides suggestions for future research.
2. Theory
We develop an information-based sequential trade model that predicts the direction and magnitude of stock price shifts over two trades in succession in the same direction by a single identified trader. The characteristic feature of information-based models is that the market maker and at least one set of traders are seen as rational (Bayesian) decision makers who condition their probability distributions on all information available to them or inferred from the trades of others. A fundamentally important insight, formalized by Copeland and Galai (1983), is that the market maker sets bid and ask prices that anticipate the information implicit in their being accepted. The prices quoted are thus “regret free” in the sense that on completion of the trade the expected loss, conditioned using Bayes theorem on whatever information is implied by the trade having occurred, is non-negative, or zero given a risk-neutral market maker.
To model the effects of strategic considerations and other factors that decide whether and how frequently an informed or uninformed trader trades twice in succession, we introduce two additional parameters to those characterizing the stochastic trader arrival process in existing sequential trade models:
l = p(same trader at second trade|informed trader at first trade)
and
g = p(same trader at second trade|uninformed trader at first trade).
In words, l (g) is the probability of the same trader re-appearing at the second trade given that he is informed (uninformed). It is assumed that this trader first appeared (after some absence) at the previous trade – otherwise there would have been three or more trades in succession by the same trader. In general, l is expected to clearly exceed g since a randomly selected informed trader has, by definition, a profit motive to repeat any trade already made. There are, however, other relevant factors (including, for example, the overall proportion of informed traders) that determine whether any particular trader, informed or not, succeeds in trading a second time ahead of all other willing traders. The role of l and g in the model is to capture the effects of strategic and other factors that combine to determine the probability of a repeat trade by a given randomly selected informed or uninformed trader.
Our model is essentially as per Glosten and Milgrom (1983) and Easley and O’Hara (1987, 1992). The market maker and all informed traders are assumed rational risk-neutral utility maximizers. Uninformed traders are “liquidity traders” (Milgrom and Stokey, 1982) who trade for reasons independent of stock value. The unknown stock value V is either high VH or low VL, and is constant during any repeat-trade sequence. The market maker quotes both bid and ask prices. All trades are of one unit of stock at either the bid or ask, and are made in series. There are no commissions, inventory holding costs or costs of carrying short positions. Any newly arrived trader (i.e. not the same trader as in the preceding transaction) is informed with probability m and uninformed with probability (1-m). When selected, an uninformed trader chooses to buy with probability h, and sell with probability (1-h), independent of the stock value V. Contrary to Easley and O’Hara (1992), uninformed traders when selected do not have the option to not trade. Otherwise the stochastic arrival process would have to be modelled over a third, fourth or potentially infinite trade sequence so as to generate two completed trades. The Easley and O’Hara (1992) model was developed to demonstrate the information content of a “non-trade” (a time interval containing no trade) whereas our study considers the information content of two completed trades in succession with no condition on the time elapsed between trades (except that it is short, as in any microstructure study).
A further assumption, not relevant in the previous models, is that a trader who trades twice in succession makes the second trade in the same direction as the first. This is true with probability one for informed traders (the true asset value is constant over repeat trades) and we assume with probability one for uninformed traders. On this assumption, our model’s implications (see below) are conservative. Intuitively, a probability less than one for an uninformed trader would mean that fewer repeat trades in the same direction are made by uninformed traders, and hence that such occurrence is stronger indication of VH (VL) in the case of two buys (sells).
We consider the effect on the probability of VH of two buys (buyer initiated trades) in succession. Results are the same (mutatis mutandis) for two sells in succession, as is easily verified. There are three possible conditioning events, defined as follows. Event xx represents two buys in a row by the same trader, xy represents two buys in a row by two different traders, xxÈxy represents two buys in a row with no information concerning the traders’ identities. The posterior probability of VH, p(VH| . ) can be found conditional on any of these three events. Our primary focus is on p(VH|xx) and p(VH|xy), so as to understand the relative evidential effects of two buys in succession by the same trader and by two different traders. Note that p(VH|xxÈxy) is a weighted average of p(VH|xx) and p(VH|xy) since by the law of complete probability p(VH|xxÈxy)=p(xx |xxÈxy) p(VH |xx) + p(xy |xxÈxy) p(VH |xy).
Given the assumptions listed above, the probability tree representing the sequential trade process that generates stochastically two buys in succession (immediately following a sell) is as shown in Figure 1. Using this tree, we arrive at the following empirically testable propositions. Derivations are provided in the appendix.
Figure 1 about here
Proposition 1. Two buys in succession by the same trader constitute stronger evidence for VH than two buys in succession by different traders. That is, p(VH|xx) > p(VH|xy).
By Bayes theorem, this requires p(xx|VH)/p(xx|VL) > p(xy|VH)/p(xy|VL), which occurs for l greater than
lc =.
The relative magnitudes of l and g are critical to the relationship between p(VH|xx) and p(VH|xy). In general, a randomly selected informed trader is systematically more disposed to trade a second time in succession than an uninformed or liquidity trader, implying lg. The fundamental reasons for this were put by O’Hara (1995, pp.57-9, 74-75). In particular, an informed trader has a profit motive to follow one buy with another (the first trade would not have been made otherwise) whereas the motivation of an uninformed trader to trade a second time is unclear:
…informed and uninformed traders are both assumed willing to continue to transact. For an informed trader, such repeat trading is certainly optimal, but it is less believable that a randomly selected uninformed trader’s behavior remains the same after trading as it was before. (O’Hara 1995, p.74)
The determinants of l and g are partly strategic. An informed trader may disguise information by delaying rather than trading again immediately, however with short-lived information, waiting will often have negative marginal expected value. Since other informed traders have the same motivations, l can be seen as the frequency with which a given (randomly selected) informed trader opts to trade again immediately and succeeds in making the very next trade against all other willing buyers (informed and uninformed). Proposition 1 relies on the insight that this conjunction is systemically more likely for an informed than uninformed trader.
The price path followed by the traded asset over two buys in a row by the same trader is shown in Figure 2. The evidential effect of the first buy is to increase the price to its revised expected value E(V|x) from its starting point of E(V). The effect of the second buy is to shift the asset price to E(V|xx), the value of which depends on the parameters l and g. If g=0, an uninformed trader never makes two buys in succession and hence the trader observed must have been informed. The price E(V|xx,g=0) is then VH. At the other extreme, if l=0 the trader must have been uninformed, in which case there is no evidence about V implied in the two trades taken as a pair, and the asset price E(V|xx,l=0) reverts to its starting point E(V). Finally, if l=lc, the asset price conditional on two trades in a row by the same trader, E(V|xx,l=lc), equals E(V|xy), which is shown (see appendix) to exceed E(V|x)>E(V). Provided that llc, E(V|xx)>E(V|xy), as per Proposition 1.