«Microstructure Effects, Bid-ask Spreads and Volatility in the Spot Foreign Exchange Market Pre and Post-EMU Frank McGroarty University of Southampton ...»
&KUEWUUKQP 2CRGTU KP
#EEQWPVKPI CPF (KPCPEG
Microstructure Effects, Bid-ask Spreads and Volatility in
the Spot Foreign Exchange Market Pre and Post-EMU
University of Southampton
Owain ap Gwilym
University of Wales
University of Southampton
1EVQDGT ISSN 1356-3548
Microstructure effects, bid-ask spreads and volatility
in the spot foreign exchange market pre and post-EMU
Frank McGroarty1, Owain ap Gwilym2 and Stephen Thomas3 Abstract This article examines how microstructure effects, evident in high frequency data, influence bid-ask spreads and volatility in transaction price series. It uses the event of European Monetary Union (EMU), and the upheaval that this entailed, as an opportunity to empirically investigate these relationships in the electronic inter-dealer spot FX market. These microstructure effects relate to both price and time. There are two price effects: 1) price discreteness and 2) price clustering, and two time effects: 1) the time elapsed between sample periods and 2) the time-gap between successive trades or quoted price submissions. Strong evidence emerges that all four factors are important in the determination of bid-ask spreads.
JEL Classification: F31; G12; G15; D4 Keywords: Price clustering, Price discreteness, euro, Market microstructure, Bid-ask spreads.
Corresponding author. Lecturer in Finance, School of Management, University of Southampton, SO17 1BJ, UK. F.J.McGroarty@soton.ac.uk Tel: +44 23 8059 2540 Fax: +44 23 8059 3844 Professor of Finance, School of Management & Business, University of Wales, Aberystwyth, SY23 3DD. UK.
Professor of Financial Markets, School of Management, University of Southampton, SO17 1BJ. UK.
We are very grateful to EBS for supplying the data used in this study.
This version: 14th October 2004
1. Introduction This study uses a unique and rich foreign exchange (FX) dataset of global inter-dealer electronic transactions to examine microstructural effects in the spot foreign exchange market.
This dataset enables us to shed new light on the debate surrounding the observations that trading volumes have fallen and bid-ask spreads have widened in inter-dealer spot FX markets following European Monetary Union (EMU). Our work provides a more detailed account of the changes that actually occurred at this time, because our data is more comprehensive than has previously been available. We are able to demonstrate that much of the change in bid-ask spreads and some of the change in price volatility can be explained by four technical features of high-frequency data, namely price discreteness, price clustering, the time elapsed between sample periods and the time-gap between successive traded or quoted prices. This explanation is in contrast to the hypothesis of market maker response to exogenous changes in volume as proposed by Hau, Killeen and Moore (2000 and 2002).
Price discreteness means that prices or exchange rates are not an infinite number of digits long, but rather they are truncated to a small number of digits. In the case of the FX market, exchange rates are specified to five digit accuracy. Price clustering refers to the fact that traders may not use all available exchange rates uniformly. In practice, rates ending in 0 or 5 tend to be used more than other rates. The time elapsed between the sample periods is important for a very obvious reason - price levels can differ radically if data is sampled from periods that are far apart in time. On the other hand, the time-gap between successive individual prices is also important because it allows these prices to drift apart. When the successive prices are transaction prices, this effect increases volatility. When they are successive bid and ask prices, the bid-ask spread is increased.
EMU brought widespread change to financial markets. Much of this change is directly due to the re-denomination of certain instruments from Deutschemarks (DEM) to euros (EUR).
Since these currency units are of different values, the nature of the price discreteness affecting instruments which are now denominated in EUR will be different from what it was under DEM. This point is exemplified by the fact that the smallest sized bid-ask spread and smallest price increment for the EUR are both 74% greater than that for the DEM, after controlling for changes in currency values.
It is an acknowledged fact that volumes have decreased in the inter-bank spot FX market since EMU (see BIS, 2001). It has also become increasingly accepted that the EUR/USD bidask spread widened at the same time. Goodhart, Love, Payne and Rime (2002) attracted much interest when they argued that the change in denomination from DEM to EUR is sufficient to account for observed increases in bid-ask spreads, post-EMU. Their “price granularity” hypothesis combines the difference in price discreteness between the DEM and the EUR, with trader inertia in setting bid-ask spreads. These authors argue that the fall in volume is a coincidence primarily due to the increased use of electronic inter-dealer broking and to banking industry consolidation. This model was put forward in response to the controversial work of Hau et al. (2000 and 2002), who had suggested that lower FX trading volumes and wider bid-ask spreads since EMU, are both due to a decrease in “market transparency”. The latter hypothesis centres on the idea that the availability of fewer currency pairs after EMU makes risk management harder to implement. Hau et al. suggest that this causes market makers to quote wider bid-ask spreads, which result in lower volumes. Detken and Hartmann (2002) used a wider but lower frequency dataset and reached a broadly similar conclusion to Goodhart et al (2002).
The remainder of this paper is organised as follows. Section 2 discusses the importance of price discreteness and price clustering. Section 3 explores the role of time and duration.
Section 4 reviews market structure and the unique dataset, while Section 5 discusses the empirical methodology. Section 6 presents the results and Section 7 concludes.
2. The Importance of Price Discreteness and Price Clustering Prices move in discrete units. The exact size of these discrete units may be imposed by a regulator or an exchange, or it may arise as a market convention. Price discreteness is clearly important for bid-ask spreads because the minimum tick size places a lower bound on a (nonzero) bid-ask spread. It also determines the increments by which this can increase.
Campbell, Lo and MacKinlay (1997, p.113) used a sequence of stock return plots with progressively finer scales to illustrate graphically how discreteness imposes structure on the sequence of returns. Gottlieb and Kalay (1985) found that “…the variance and…the higher order moments of the rate of return of stocks are upward biased due to the discreteness of observed stock prices”. Harris (1990) elaborated on the Gottlieb and Kalay (1985) result by showing that discreteness induces negative serial correlation in high frequency data. This occurs because discreteness causes the high frequency data to overstate the volatility in economic value. More recently, Osler (2003) found that discreteness in FX rates could contribute to excess kurtosis.
The preceding discussion addresses how price discreteness alone affects the bid-ask spread and volatility. In its own right, price discreteness is obviously an important factor in shaping bid-ask spreads and price volatility. However, as Harris (1991) pointed out, if certain prices are used more than others, then the price discreteness effect will be compounded. This is how price clustering exerts an influence over bid-ask spreads and over volatility. This draws us to the question of why price clustering behaviour arises.
Until recently, price clustering studies on FX markets were rare and primarily used Reuters indicative quotes data. Goodhart and Curcio (1991) was the only widely cited paper. The main reason for the paucity of research is the lack of appropriate FX data. This situation has now eased. Besides Goodhart et al (2002), Sopranzetti and Datar (2002) produced a price clustering study of the FX market, using data from the Federal Reserve Bank of New York.
Using Royal Bank of Scotland data, Osler (2003) found that the clustering of “ stop loss” and “ take profit” orders at certain price points explained why certain technical analysis forecasts have predictive power. However, none of these studies have the richness of data that the present paper offers. This is because we have been able to explore global inter-dealer electronic transactions, which are the largest source of trade volume in the spot foreign exchange market.
The vast bulk of previous research on price clustering focuses on the equity markets. The most notable contributions include Harris (1991), Christie and Schultz (1994), Christie, Harris and Schultz (1994), Kleidon and Willig (1995), Aitken, Brown, Buckland, Izan and Walter (1996), Grossman, Miller, Fischel, Cone and Ross (1997), Woodward (1998), Weston (2000), Kandel, Sarig and Wohl (2001) and Brown, Chua and Mitchell (2002). Additionally, some prior work focuses on futures markets, including papers by Ball, Torous and Tschoegl (1985), Brown, Laux and Schachter (1991), ap Gwilym, Clare and Thomas (1998a and 1998b) and ap Gwilym and Alibo (2003).
Over the past decade, most papers which address the subject of price clustering make early reference to Christie and Schultz (1994). In the early 1990s, these researchers caused a stir with a price clustering study, when they suggested that widespread NASDAQ market makers avoidance of odd-eights quotes could amount to tacit collusion to maintain wider bid-ask spreads. As they put it themselves, “ [they] are unable to envision any scenario in which 40 to 60 dealers who are competing for order flow would simultaneously and consistently avoid using odd-eighth quotes without an implicit agreement to post quotes only on the even price fractions”, and they consider this evidence of an “ … apparent lack of competitiveness of the NASDAQ market”. This collusion hypothesis has gathered huge support, largely because the use of odd-eight quotes increased following the publication of their results and increased again after subsequent rule changes (see Christie et al, 1994).
Several competent and convincing rebuttals of the Christie and Schultz (1994) result have emerged since their paper was published. These include Kleidon and Willig (1995), Grossman et al (1997) and Woodward (1998). On the other hand, other sources like the US Securities and Exchange Commission (1996) and Weston (2000) seem to support the Christie and Schultz (1994) case. The debate is still open.
There are other plausible reasons, besides collusion, that could give rise to cluster patterns in price data. Yule (1927) observed that numerical clustering arose systematically from errors that people made when asked to read numbers from a scale. In the 1960s, a series of papers (Osborne, 1962; Niederhoffer, 1965; Niederhoffer, 1966; Niederhoffer and Osborne, 1966) began to explore the issue of clustering in the context of financial prices. Niederhoffer (1966) argued that price clustering could even be at odds with market efficiency. Niederhoffer and Osborne (1966) went on to specify profitable trading rules based on cluster frequencies.
Utilising rules laid out by Harris (1991) for NYSE stocks when price increments of $1/8 were in place, Goodhart and Curcio (1991) compare “ attraction” with “ price resolution” as possible explanations of observed price clustering in the FX market, where the price increment is 1 and the full range of ten final digits is available. The attraction hypothesis focuses on the kind of rounding behaviour that Yule (1927) identified. The price resolution hypothesis, put forward by Ball et al (1985), asserts that price clustering is a natural occurrence in a market which has reached “ the optimal degree of price resolution”. This arises from the trade-off between increased price accuracy on the one hand, and the inconvenience of longer prices on the other.
Aitken et al (1996) found evidence of attraction-type clustering in the Australian stock market. Kandel et al (2001) found the same effect in the Israeli IPO market. ap Gwilym et al (1998b) also find it for international bond futures. Ball et al (1985) not only proposed the price resolution hypothesis, but they also found evidence for it in the gold market. Goodhart and Curcio (1991) find evidence in favour of price resolution in their FX data and of attraction in their bid-ask spreads data. Harris’ (1991) results for NYSE stocks are consistent with the price resolution hypothesis, and this is also supported by Grossman et al (1997) for a variety of global markets.
Harris (1991) and Brown et al (1991) both suggest another explanation for price clustering, namely the negotiation hypothesis. This suggests that there could be a two-tier price system, which would give the appearance of resolution-type price clustering when combined. The idea is that large trades result in harder bargaining and progressively more finely tuned trade prices, while small trades are transacted with cruder pricing from a reduced grid of prices.
Grossman et al (1997) demonstrate that price clustering is a both a common and a variable feature across global financial markets. These authors argued that, instead of NASDAQ being out of step with NYSE as Christie and Schultz (1994) had alleged, it is actually NYSE that is anomalous for its lack of price clustering compared with other markets. They go on to show that a wide variety of factors contribute to price clustering. Market structure can play a part.
Also, whether the quotes are binding or not matters. Once again, these conclusions do not contradict the final-digit price clustering pattern associated with price resolution, but are attempts to explain why this occurs.