I. FX Market Microstructure Overview

Transcription

1 I. FX Market Microstructure Overview Focus on agents behavior and market characteristics *heterogeneity: information, horizons *institutional constraints: market makers liquidity traders risk sharing price discovery Goods & asset market approaches empirical failures *goods trade small fraction of FX market *asset markets move quickly in response to expected changes "news" quickly incorporated in price expectations unobservable M. Melvin,

2 "Macro" vs. "Micro" Approaches to FX macro models focus on macro fundamentals and representative agents *public info. moves market *all share same info. micro models focus on trading process and participants *risk sharing ("hot potato" passing) generates pyramiding of volume *trades are informative private info. conveyed incomplete risk sharing no Walrasian auctioneer *dealers don't see all orders *price quotes precede orders M. Melvin,

3 macro models estimated at low frequency and include "fundamentals" *public info. doesn't explain much *no "price discovery" of private info. micro models include order flow & inventory *examine exchange rate between transactions *even with macro fundamentals included, order flow is important (Cai, Cheung, Lee & Melvin; Evans & Lyons) fundamentals expectations in order flow? micro models have role for spreads that is missing in macro models *may reflect information, inventory, uncertainty M. Melvin,

4 MICROSTRUCTURE THEORY auction markets *best price defined by submitted orders *auctioneer collects orders & sets price *orders batched & simultaneously executed at single market-clearing price dealership markets *best price defined by dealer quotes *price quotes precede orders *sequential trade of individual orders desirable FX theory *decentralized dealers *risk-averse actively manage positions *interbank vs. customer trades M. Melvin,

5 RATIONAL-EXPECTATIONS AUCTION MODEL (Grossman & Stiglitz, AER, 1980) insights players *price clears markets and conveys info. *1 informed trader (risk averse, perfect competitor) *1 uninformed trader (" " ") information protocol *informed receives signal of final payoff of risky asset *uninformed only sees current market-clearing price *single trading period *batch clearing (all trades at single price) *known pricing rule allows uninformed to infer informed's signal from market price M. Melvin,

6 THE KYLE MODEL: AN AUCTIONEER (Kyle, Econometrica, 1985) Now introduce an explicit auctioneer or marketmaker to rational expectations model sets prices sees order flow and takes positions *insights *players marketmakers count order flow not fundamentals marketmakers cannot differentiate informed vs. uninformed orders informed traders exploit latter liquidity (market depth) affects strategic behavior 1 risk-neutral marketmaker 1 risk-neutral informed trader many uninformed, non-strategic traders *information informed sees payoff value of risky asset informed does not observe uninformed orders marketmaker sees total orders, not components M. Melvin,

7 *protocol single trading period batch clearing, all trades at single price marketmaker prices to earn zero expected profit M. Melvin,

8 SINGLE DEALER SEQUENTIAL TRADE MODEL (Glosten & Milgrom, JFE, 1985) Now specify a single dealer whose prices are conditioned on order flow and who is a counterparty to all trades with randomly selected traders *insights *players spreads quoted to equate loss expected to informed with gains from uninformed dealer learns from sequential arrival of individual orders (price discovery) 1 risk-neutral dealer many informed risk-neutral non-strategic traders many uninformed non-strategic traders *information informed know if payoff on risky asset is high or low dealer knows uncond. prob. of payoff dealer knows prob. that next trader is informed dealer sees sequence of incoming orders M. Melvin,

9 *protocol sequential trading, 1 trade per period dealer one side of all trades potential trader randomly selected from pool each period dealer quotes bid & offer to potential trader bid & offer set so expected profit equals zero M. Melvin,

10 MULTIPLE DEALERSHIP SIMULTANEOUS TRADE MODEL (Evans & Lyons) Now have interdealer trades involving simultaneous-move games. Now have inventory shocks from incoming orders and get "hot potato" phenomenon. Earlier models had no undesired inventory as dealers are either risk neutral (single dealer and Kyle models) or trades are conditioned on the market-clearing price (R.E. auction model). *insights *players dealer inventories & customer order flow are sources of private info. dealer speculation affected by former private info. & strategic dealer behavior reduce info. revealed by price a continuum of risk-averse, non-strategic customers n risk-averse and strategic dealers M. Melvin,

11 *information each dealer receives a private signal of payoff value of risky asset all dealers receive a common signal of payoff value of risky asset each dealer receives customer orders after trading, dealers observe signal of interdealer order flow *protocol dealer quoting is simultaneous, independent, & required quotes are available to all dealers quotes are single price at which any amount may be bought or sold trading is simultaneous & independent can trade with multiple partners Reference: Lyons, The MicrostructureApproach to Exchange Rates, MIT Press. M. Melvin,

12 *Trading venues MARKET CHARACTERISTICS 5 years ago: interbank and traditional brokers each ½ today: interbank 1/3; traditional brokers 1/6; electronic brokers ½ *Counterparties Customer trades 1/3 Interbank trades 2/3 both today and 5 years ago *Conventional spreads ("pips") USD/GBP 5 DEM/USD 3 JPY/USD 3 CHF/USD 5 adopted by most "to maintain an equitable & reciprocal trading relationship" important for reputation "Dealers make the majority of their profit on rate movement, not spread." M. Melvin,

13 *Deviations from conventional spread 1. Thin & hectic market 2. Before & after major news release 3. Increased market volatility *What gives dominant players advantage? 1. Large customer base 2. Better information *How much time is required for full adjustment to macro announcements? 1/3 say < 10 seconds 2/5 say < 1 minute price adjusts quickly to new info. *What news matters most? Today: unemployment, interest rates, inflation 5 yrs ago: trade deficit, interest rates, unemployment so expect parameter instability in "fundamentals" models of exchange rates References: Cheung & Chinn, "Currency Traders & Exchange Rate Dynamics," JIMF, Aug M. Melvin,

14 Public Information Arrival, Exchange Rate Volatility, and Quote Frequency Michael Melvin Arizona State University and Xixi Yin American Express Co. M. Melvin,

15 Introduction Is the FX market in need of regulation? Is trading 'self-generating' with 'excess volatility'? Examine how quote frequency and volatility are related to public info arrival I. The Data and Seasonality JPY/USD & DEM/USD to Public info: number of news headlines Intradaily patterns in quotes and news: scale by hourly means for each day of week M. Melvin,

16 Figure 1: Average Hourly Quotes: Mark/Dollar Quotes per Hour Monday Tuesday Wednesday Hour (GMT) Thursday Friday Figure 4: Correlogram: Mark/Dollar Quotes Autocorrelations Unadjusted Seasonally Ad -0.8 Hourly Lag M. Melvin,

17 II. Public Information Arrival and Exchange Rate Changes N traders Market arrives at sequence of equilibria each hour *new info arrives new equilibrium *length of time between within-period equilbrium determined by info arrival IIa. A Mixture of Distributions Model 1) P in = γ ( s in * s i ) desired position of nth trader at i determined by perceived 'true value' of currency relative to current spot rate N 2) P 0 n 1 in = = market clearing condition 3 ) 1 N s i = s N n 1 in *. = market cleared by average of reservation spot rates 4 ) 1 N ds 0 2 i = N ds in 1 *, ds i ~ N(, σ i ). n = change in spot rate from info arrival I 5) ds = ds, ~ ( 0, 2 i 1 i ds N σ i I). = M. Melvin,

18 I info events in hour is random hourly change is mixture of indep. normals with mixing variable I quote frequency records I 6) I t = a+ bi u t 1 + t, with autocorrelated news, get autocorrelated variance of ds 7) h 2 2 t = σ a+ bh u t 1 + σ t. IIIb. Public Information Arrival and Quote Frequency implication of model: 10) q t = α + α I 0 1 t + ε t h t = b + b h b t 1 t + ε. with S.A. data: when more than normal amount of news arrives is there more than normal amount of quotes? *answer: yes M. Melvin,

19 IIc. Public Information Arrival and Exchange Rate Volatility Model implies volatility should also depend on I *estimate with and without I 8) ds t = α + ε t h t = b + b h b t 1 t + ε. 9) ds t = α + ε t h t = b + b h b t 1 bi t + +. ε 3 t 1 When there is more than normal amount of news, is there more than normal volatility? *answer: no If no S.A., then find strong positive effect of I on h IId. Joint Estimation estimate bivariate model to increase efficiency given potential correlations across quotes and returns Yt H t = βxt + εt = C ' G A ' ' 0 C 0 + εt 1ε t 1A + mean equations as before N ' I t N + B H t 1 now find I significant in both mean for quotes and variance for returns M. Melvin,

20 III. Implications of Results and Conclusions More than the normal amount of news more than the normal amount of quotes and volatility *no support for "self-generating" trading *no support for regulation FX trading is probably adjusting prices and quantities to efficiently allocate resources Reference: Melvin & Yin, "Public Information Arrival, Exchange Rate Volatility, and Quote Frequency," The Economic Journal, July 2000 or M. Melvin,