سیستم های رفتار فروشنده و تجارت در بازار ارز خارجی

We study dealer behavior in the foreign exchange spot market using detailed observations on all the transactions of four interbank dealers. There is strong support for an information effect in incoming trades. The direction of trade is most important, but we also find that the information effect increases with trade size in direct bilateral trades. All four dealers control their inventory intensively. Inventory control is not, however, manifested through a dealer's own prices in contrast to findings by Lyons (J. Financial Econ. 39(1995) 321). Furthermore, we document differences in trading styles, especially how they actually control their inventories.

Short-term exchange rate fluctuations are notoriously difficult to explain (see, e.g., Frankel and Rose, 1995). After extensive research over many years, few stones have been left unturned when it comes to investigating the short-term explanatory power of macroeconomic variables. The microstructure approach to foreign exchange takes a different route and studies the agents that actually set the exchange rate: the dealers. Here, we study dealer behavior using a very detailed data set with the complete trading records of four interbank spot foreign exchange dealers during the week March 2–6, 1998. We first test models of price determination, and we then examine the dealers' trading styles. Our data set contains all relevant information about each trade such as transaction time, transaction prices and quantities, inventories, trading system used, and who initiated the trade. Despite the size and importance of foreign exchange (FX) markets, there are virtually no empirical studies using transaction prices and dealer inventories. A notable exception, however, is the study by Lyons (1995) that uses a data set from 1992 on transaction prices and dealer inventories for one dealer covering one week in August 1992. Other studies that should be mentioned are Yao, 1998a and Yao, 1998b and Mende and Menkhoff (2004). Much empirical work on market microstructure has focused on the specialists at the NYSE. However, due to its decentralized multiple dealership structure and its low transparency, the FX market is very different from the specialist structure on the NYSE. Non-bank customers trade bilaterally with dealers who provide quotes on request. The interdealer market has a hybrid market structure with two different trading channels available: direct (bilateral) trades and brokered trades (including both electronic brokers and the more traditional voice brokers). The FX market is also special in the sense that trading is largely unregulated. This means, for example, that low transparency has evolved endogenously. For direct interdealer trades and customer trades, details such as bid and ask quotes or the amount and direction of trade are only observed by the two transacting counterparties. Brokers are more transparent. Electronic brokers announce best bid and ask prices and the direction (not amount) of all trades (voice brokers announce a subset). This information is only available to the dealers, however. Electronic brokers have become very popular since their introduction in 1992 and are now the dominant tool for interdealer trading. As such, electronic brokers now provide some degree of centralization in an otherwise decentralized market. In contrast to the NYSE, at least two major stock markets, the NASDAQ and the London Stock Exchange, are organized as multiple dealership markets. Furthermore, electronic brokers were introduced relatively early in the FX market, and have recently been implemented by several stock markets. There are also many similarities between FX and bond markets, e.g., the U.K. gilt market studied by Vitale (1998) and the 5-year Treasury note interdealer broker market studied by Huang et al. (2002), and thus our results may apply more broadly than just to FX markets. Our first contribution with this paper is to test the two main branches of microstructure models, namely inventory control and adverse selection. Inventory control models (e.g., Amihud and Mendelson, 1980 and Ho and Stoll, 1981) focus on how risk-averse dealers adjust prices to control their inventory of an asset. The idea is that a dealer with a larger inventory of the currency than desired will set a lower price to attract buyers. This is called “quote shading”. Information-based models (e.g., Kyle, 1985, Glosten and Milgrom, 1985 and Admati and Pfleiderer, 1988) consider learning and adverse selection problems given that some market participants have private information. When a dealer receives a trade, he will revise his expectations (upward in the case of a buy order and downward in the case of a sell order) and set spreads to protect himself against informed traders. We use different methods to test the two main microstructure models. We start by testing whether dealer inventories are mean reverting. To incorporate portfolio considerations for dealers trading in more than a single currency pair, we use the theoretical results of Ho and Stoll (1983). We find strong evidence of mean reversion for all four dealers, which is consistent with inventory control. The median half-lives of the inventories range from less than 1–15 min. We then use two well-known models to test for inventory and information effects on price. The first, the Madhavan and Smidt (1991) model—similar to the model used by Lyons (1995)—receives no support. In addition, we use the indicator model suggested by Huang and Stoll (1997). The current paper is, to the best of our knowledge, the first to apply this model to FX markets. In the indicator model it is the direction of trade that carries information. Using this model we find much better support and, in particular, we find that adverse selection is responsible for a large proportion of the effective spread. Interestingly, we find no evidence of inventory control through dealers' own prices as predicted by the inventory models. The importance of private information in FX markets is further confirmed since cumulative order flows and price levels are cointegrated. Lyons (1995) finds evidence of adverse selection and, in contrast to our study, strong evidence of an inventory effect through price. Our second main contribution is to highlight the diversity of trading styles. In particular, we examine more closely how dealers use different trading options to control their inventories. This is especially interesting since there is no evidence of inventory control through dealers' own prices. To understand the lack of any price effect from inventory, it is important to remember the multiple dealer structure of the market. In a single dealer structure, such as the one in the Madhavan and Smidt (1991) model, the dealer must wait for the next order to arrive. His only possibility for inventory adjustment is to shade his quotes. On the other hand, in the hybrid structure of the FX market, dealers may submit limit or market orders to brokers (electronic or voice brokers), or trade at each others' quotes bilaterally. We find clear differences in trading styles among our dealers (despite the fact that all our dealers work in the same bank, i.e., they are not four independent draws from the population of dealers). The strong information effect and weak price effect from inventory is similar to evidence in Vitale (1998) for the UK gilt market and in several studies of stock markets, e.g., Madhavan and Smidt, 1991 and Madhavan and Smidt, 1993 and Hasbrouck and Sofianos (1993). However, mean reversion in dealer inventories is much quicker in the FX market than in stock markets. The extremely short half-lives of a few minutes documented here confirm that inventory control is the name of the game in FX. The evidence found in this study of strong mean reversion in dealer inventories but weak inventory effects through price is consistent with the findings in Manaster and Mann (1996) for futures dealers. Recent studies such as Evans and Lyons (2002) have integrated insights from microstructure to address the inability of macro models to explain exchange rate changes at frequencies higher than one year, and demonstrate that daily aggregate order flow may improve explanatory power significantly. This is a promising direction for FX research. Dealer analysis is likely to prove useful in the future for formulating realistic micro foundations for this microstructure-macro framework. It is comforting that the results presented here are consistent with the informational approach Evans and Lyons assume at the market-wide level. Dealer analysis also has a wider scope, however. For example, our results about inventory control have implications for an understanding of the large trading volumes in FX markets. The next section describes our data and some important market characteristics of relevance for our study. Section 3 provides an analysis of dealer inventories. Our investigation of price effects from information and inventories is presented in Section 4. Section 5 examines how the dealers actually control their inventories using other alternatives than price shading. The paper ends with conclusions and comments with respect to directions for future research.

This paper studies the behavior of four interbank spot foreign exchange dealers using a detailed data set for the week March 2–6, 1998, with transaction prices, trading quantities, dealer inventories, exact timing, and information regarding the trading system used for the transactions. The four dealers trade in different exchange rates and have different trading styles. Using our data, we study whether dealers set prices to protect against private information and how they control inventory to adjust their risk exposure. In a widely cited paper, Lyons (1995), using data from 1992, finds support for both information and inventory effects in the pricing of an FX dealer. Using a version of the Madhavan and Smidt (1991) model, Lyons finds, consistent with the model predictions, that the dealer increases his spread with trade size to protect against private information, and adjusts the midpoint in the spread (quote shading) to induce trade in a preferred direction to adjust inventory. Using the same model as Lyons for our dealers, we find no support for the types of information and inventory effects predicted by that model. Our results suggest that the Madhavan and Smidt model may not be as applicable to foreign exchange trading as first believed because of differences in trading styles among the dealers. A likely explanation is the change in the trading environment caused by the introduction of electronic brokers. The dealer tracked by Lyons may have been playing an interdealer risk-sharing role that the electronic brokers have come to dominate. Using an indicator model (the Huang and Stoll (1997) model), we show that private information is indeed important in the FX market. For DEM/USD, we find that private information is responsible for as much as 80% of the effective spread in the interdealer market. For NOK/DEM, roughly 50% of the effective spread is explained by private information. The finding that cumulative order flows and prices are cointegrated is also consistent with order flow as a carrier of information. Interestingly, we find strong evidence of mean reversion with half-lives of dealer inventories that range from less than one minute to fifteen minutes. However, little of this is manifested through dealers' own prices as predicted by the inventory models. We show how the dealers control their inventories and how different types of positions contribute to their overall profitability. In doing this we distinguish between three types of trades. We find that customer trades are highly profitable. This business is particularly important for the NOK/DEM Market Maker in our sample. The dealers also earn money from their direct incoming trades and from actively establishing positions through electronic brokers. Active position taking seems to represent an important share of the trading for three of the dealers. The dealers provide liquidity by submitting limit orders on electronic broker systems, and may thus earn money from the bid–ask spread. Or they submit market orders to establish speculative positions. Hence, the dealers do not use the interdealer market only to off-load unwanted positions from their direct incoming trades and customer business. For two of the dealers, we find that the share of outgoing trades is higher when they establish a position than when they unwind the same position. This finding may suggest that these dealers submit market orders when they have information. This paper is, to our knowledge, the first to provide strong and detailed evidence that (at least some) FX dealers engage in information-based speculation. This kind of behavior is also consistent with the standard signing convention that we use for the electronic broker trades. As mentioned, electronic brokering has become the dominant tool for interdealer trading since its introduction at the end of 1992. Interestingly, we do not find evidence that the price impact from direct trades is different from the price impact of electronic broker trades, in contrast to the results of Reiss and Werner (2002) for the London Stock Exchange. This may suggest that (at least during the week we study) electronic brokers can provide “sufficient” liquidity. This is also supported by the fact that the market share of electronic brokers has continued to rise since 1998 and now enjoys (according to practitioners) a market share of roughly 85% of all interdealer trading. What can we learn about FX trading from these four dealers? They are not dealers in one of the large US banks. However, the bank in question has a long history of FX trading and has been among the top 15 banks in DEM/USD over a long period. In NOK/DEM they are probably the largest bank. The introduction of electronic brokers has also made tight spreads available to more than just the key dealers in the largest banks, thus making the terms among dealers in the market more equal. Furthermore, their trading strategy seems quite successful. The fact that we document differences in trading strategy, roughly between the older market makers and the young electronic broker dealers, also means that we cover different aspects of FX trading even if the dealers are from the same bank. Still, there is a great demand for more knowledge about the microstructure of the FX market due to the lack of dealer-specific trading data and inventories. This is especially true for the new trading environment resulting from the introduction of electronic broker systems. In this respect, this study fills a gap in the literature.