«Volume Title: The Microstructure of Foreign Exchange Markets Volume Author/Editor: Jeffrey A. Frankel, Giampaolo Galli, Alberto Giovannini, editors ...»
This PDF is a selection from an out-of-print volume from the National
Bureau of Economic Research
Volume Title: The Microstructure of Foreign Exchange Markets
Volume Author/Editor: Jeffrey A. Frankel, Giampaolo Galli, Alberto
Volume Publisher: University of Chicago Press
Volume ISBN: 0-226-26000-3
Volume URL: http://www.nber.org/books/fran96-1
Conference Date: July 1-2, 1994
Publication Date: January 1996
Chapter Title: Foreign Exchange Volume: Sound and Fury Signifying Nothing?
Chapter Author: Richard K. Lyons Chapter URL: http://www.nber.org/chapters/c11365 Chapter pages in book: (p. 183 - 208)
Foreign Exchange Volume:
Sound and Fury Signifying Nothing?
Richard K. Lyons Volume in the spot foreign exchange market dwarfs that in any other financial market. But is all this trading informative? This paper provides some empirical evidence. At the broadest level, my results help clarify why trading volume in this market is extraordinarily high. At a narrower level, I provide some sharp results regarding the relation between the intensity of trading and the informa- tiveness of trades.
Specifically, I provide results that discriminate between polar views of trad- ing intensity, to which I refer as (1) the event-uncertainty view and (2) the hot potato view. The event-uncertainty view holds that trades are more informative when trading intensity is high; the hot potato view holds that trades are more informative when trading intensity is low. In general, theory admits both possi- bilities, depending primarily on the posited information structure.
To understand the event-uncertainty view—that trades are more informative when trading intensity is high—consider the work of Easley and O'Hara (1992). In contrast to earlier models where new information is known to exist, in Easley and O'Hara (1992) new information may not exist. That is, there is some probability, say p, of new information and probability (1 — p) of no new information. In the event of new information, there is some probability, say q, that an informed trader has received good news and probability (1 — q) of having received bad news. They demonstrate that, if there is no trade at time t, then a rational dealer raises the probability that she attaches to the no- Richard K. Lyons is associate professor in the Haas School of Business at the University of California, Berkeley, and a faculty research fellow of the National Bureau of Economic Research.
The author thanks the following for helpful comments: George Constantinides, Mark Flood, Jeffrey Frankel, Antonio Mello, Carol Osier, and seminar participants at Berkeley, LSE, North- western, NYU, UBC, MIT, the IMF, and the NBER. He also thanks Jeff Bohn for valuable research assistance and Merrill Lynch and Lasser Marshall for access to dealers and brokers while trading.
Financial assistance from the National Science Foundation and the Berkeley Program in Finance is gratefully acknowledged. Any errors are his own.
183 184 Richard K. Lyons information event and lowers the probability of news having occurred. Put differently, if trading intensity is low, an incoming trade of a given size induces a smaller update in beliefs since it is less likely to be signaling news. On the flip side, trades occurring when intensity is high should induce a larger update in beliefs.
To understand the term the hot potato view—that trades are more informative when trading intensity is low—consider the ideas of Admati and Pfleiderer (1988). Key to their model is the presence of discretionary liquidity traders: in order to minimize their losses to informed traders, rational liquidity traders clump together in their trading. (The reason that informed traders cannot fully offset this advantage to clumping is that information is short-lived.) Owing to this clumping of liquidity traders, trades occurring when intensity is high tend to be less informative.
The metaphor of the hot potato offers a link between this discretionary liquidity trading and foreign exchange trading. Foreign exchange dealers use the metaphor in referring to the repeated passage of idiosyncratic inventory imbalances from dealer to dealer following an innovation in customer order flow. These interdealer liquidity trades are clearly discretionary as to timing— hence the connection between discretionary liquidity trading and the hot potato view of order-flow information. To clarify the hot potato process, consider the following crude but not unrealistic example. (Keep in mind that roughly 85 percent of foreign exchange trading is interdealer, a much higher share than in other multiple-dealer markets.) Suppose that there are ten dealers, all of whom are risk averse, and each currently with a zero net position. A customer sale of $10 million worth of deutsche marks is accommodated by one of the dealers.
Not wanting to carry the open position, the dealer calculates his share of this inventory imbalance—or one-tenth of $10 million—calls another dealer, and unloads $9 million worth of deutsche marks. The dealer receiving this trade then calculates his share of this inventory imbalance—or one tenth of $9 million—calls another dealer, and unloads $8.1 million worth of deutsche marks.
The hot potato process continues. In the limit, the total interdealer volume generated from the $10 million customer trade is $9 million /(I — 0.9) = $90 million. Thus, the example produces an interdealer share of 90 percent, roughly matching the empirical share.
Here are two possible reactions to the example given above, neither of which vitiates its message, (a) Shouldn't the multiplier be infinite since risk-averse dealers would not choose to retain any of the imbalance? The answer is that, in equilibrium, price will adjust to induce dealers to hold some of the perceived excess supply. The 10 percent rule of the example is a crude approximation of a much richer short-run clearing mechanism.1 (b) Interdealer trades can reduce idiosyncratic inventory imbalances—which reduces idiosyncratic risk rather
1. For an optimizing model in which hot potato trading arises between dealers, see Lyons (1995a). Flood (1992) examines simulation experiments that allow for hot potato trading.
185 Foreign Exchange Volume than simply bouncing it—and this will mute the multiplier. This is true, particularly if the trades are brokered. It is therefore more reasonable to think about the example in terms of net customer orders rather than gross.
The role of time in the empirical microstructure literature has only recently emerged. Two important contributions are Hasbrouck (1991) and Hausman, Lo, and MacKinlay (1992). Hasbrouck decomposes the variance of stock price changes into trade-correlated and trade-uncorrelated components and finds that trades are more informative at the beginning of the trading day. Also working with stocks, Hausman et al. test for exogeneity of the length of time between transactions, which they reject at conventional significance levels. However, they argue that their estimates do not change when endogeneity is addressed using instrumental variables. On the basis of this, they forge ahead with the assumption of exogenous intertransaction times.
Empirical microstructure work in foreign exchange has been constrained until recently by a lack of transaction-level data. The paper most closely related to the analysis here is Lyons (1995b), which uses a transactions data set that is a subset of the data used here (namely, it uses dealt quotes only). That paper presents evidence supporting both of the two branches of microstructure theory: the asymmetric-information branch and the inventory-control branch. Although many papers have provided evidence supporting the asymmetricinformation branch, little or no direct evidence had previously been found in support of the inventory-control branch (see, e.g., Madhavan and Smidt 1991;
Manaster and Mann 1993; and the overview in O'Hara 1995). The fact that they are both present provides further impetus for the application of microstructural models to the foreign exchange market. The application here extends previous work by addressing the informational subtleties of order flow.
The chapter is organized as follows: section 5.1 presents a model of transaction prices that includes a relation between trading intensity and the information content of trades; section 5.2 describes the data; section 5.3 presents the results; and section 5.4 concludes.
5.1 A Model in Which Time Matters The following model extends the model of Madhavan and Smidt (1991) by incorporating a role for intertransaction time. As they do, I will exploit the model's ability to disentangle the information effects of trades from the inventory-control effects. The result is a richer characterization of the effect of trades on price.
There are two assets in a pure exchange economy: one riskless (the numeraire) and one with a stochastic liquidation value (representing foreign exchange). The foreign exchange market is organized as a decentralized dealership market with n dealers. Here, we focus on the pricing behavior of a representative dealer, denoted dealer i. A period is defined by a transaction effected against dealer i's quote, with periods running from t — 1, 2,..., T.
186 Richard K. Lyons Signal S t Receive Observe Signal C. Quote P Trade Q Fig. 5.1 Sequencing in each period Note: S, is a public signal of the full information value Vr; Cjt is dealer j's private signal of Vr, where j denotes the dealer requesting the quote from dealer i; Pit is dealer i's bilateral quote to dealer j, a schedule matching each transaction quantity with a price; QJt is the signed quantity traded, positive for dealer j's purchases, negative for sales; and r, is the period t increment to Vr Let dealer j denote the dealer requesting dealer i's quote in any period. Figure
5.1 summarizes the timing in each period.
5.1.1 The Information Environment The full information price of foreign exchange at time T is denoted by V, which is composed of a series of increments—for example, interest differentials—so that V = Xf=0 ?i, where r0 is a known constant. The increments are i.i.d. mean zero. Each increment rt is realized immediately after trading in period t. Realizations of the increments can be thought to represent the flow of public information over time. The value of foreign exchange at t is thus defined as Vt — Xj=0 rr At the time of quoting and trading in period t, that is, before f, is realized, Vt is a random variable. In a market without private information or transaction costs, the quoted price of foreign exchange at time t, denoted Pt, would be equal to V,_,, which is the expected value of the asset price conditional on public information available at t.
The following two signals define each period's information environment
prior to dealer i's quote to dealer j :
(1) S, = V, + fj,, (2) Cj, = Vt + bjt, where the noise terms r\t and oo^ are normally distributed about zero, are independent of one another and across periods, and have variances o^ and u2w, respectively. At the outset of each period t, all dealers receive a public signal S, of the full-information value V,. Also at the outset of each period t, dealer j — the dealer requesting a quote—receives a private signal Cjt of Vt. In the foreign exchange market, one potential source of private signals at the dealer level is order flow from nondealer customers; because each dealer has sole knowledge of his own-customer order flow, to the extent that this flow conveys information it is private information, which can be exploited in interdealer trading (see, e.g., Goodhart 1988, 456; and Lyons 1995a).
187 Foreign Exchange Volume Dealer i conditions on St and then quotes his schedule as a function of possible Qjt. The schedule's sensitivity to Qjt ensures that any realization of QJt will be regret free for the quoting dealer, in the sense of Glosten and Milgrom (1985). That is, the quote takes account of the adverse selection arising from dealer j's additional information Cjt. Of course, the realization of Qjt still provides dealer i a signal of Cjt. As is standard, the signed quantity that dealer j chooses to trade is linearly related to the deviation between dealer j's expectation and the transaction price, plus a quantity representing liquidity demand Xjt
that is uncorrelated with Vt:
where ix^ is the expectation of Vr conditional on information available to dealer j at t, and the value of Xjt is known only to dealer j. (The demand function that supports eq.  requires either exponential utility defined over a single period or mean-variance optimization over multiple periods.) I introduce a role for time in the model via equation (3) and the liquidity demand Xjt. The hot potato hypothesis of order-flow information associates liquidity demand XJt with inventory-adjustment trading. In foreign exchange— according to the hypothesis—innovations in nondealer order flow spark repeated interdealer trading of idiosyncratic inventory imbalances. This rapid passing of the hot potato generates a relatively large role for liquidity trades in periods of short intertransaction times. The event-uncertainty hypothesis, in contrast, associates short intertransaction times with a relatively large role for informative trading: in the presence of event uncertainty, intense trading is a
signal that an information event has occurred. To summarize, for given precisions of the signals Cjt and St, we can characterize these views as follows:
Hot potato hypothesis:
low when intertransaction times are short;
high when intertransaction times are long.
This change in the relative intensity of liquidity trading will alter the signal extraction problem faced by the quoting dealer, to which we now turn.
5.1.2 The Formation of Expectations Dealer i's quotes depend on his conditional expectation of V, at the time of quoting, which I denote |xlV. This expectation, in turn, is a function of the variables described above: Sr and Qjr; the third variable described above, Cjr, is communicated (noisily) to dealer i via Qjt.
188 Richard K. Lyons I now address the determination of this expectation |x.,. Dealer i's prior belief regarding V, is summarized by the public signal St. Dealer i then considers the "what if" of various possible Qjt's. In particular, from any Q.t dealer i can form