بررسی جریان سفارش مشتری در بازار ارز

This paper examines the effect that heterogeneous customer orders flows have on exchange rates by using a new, and the largest, proprietary dataset of weekly net order flow segmented by customer type across nine of the most liquid currency pairs. We make several contributions. Firstly, we investigate the extent to which customer order flow can help to explain exchange rate movements over and above the influence of macro-economic variables. Secondly, we address the issue of whether order flows contain (private) information which explain exchange rates changes. Thirdly, we look at the usefulness of order flow in forecasting exchange rate movements at longer horizons than those generally considered in the micro-structure literature. Finally we address the question of whether the out-of-sample exchange rate forecasts generated by order flows can be employed profitably in the foreign exchange markets.

Currency markets are amongst the most liquid and economically important in the world but also, in terms of transaction information, amongst the most opaque. Over $3.2tn is traded on the foreign exchange (FX) market everyday according to the BIS,1 FX transactions facilitate international trade which, through the principle of comparative advantage, should be economically beneficial to all parties. The exchange rate is therefore very important for an international economy. It impacts on international competitiveness, growth and inflation through its effect on both import and export prices. Given their importance, currency markets have received a lot of attention in the academic literature. However, exchange rate determination and forecasting has remained something of an enigma ever since Meese and Rogoff’s seminal 1983 paper. In fact the so called “macro approach” (see Lyons, 2002) based on traditional exchange rate determination models has failed empirically. The failure of traditional empirical models has generated a body of research, led by Martin Evans and Richard Lyons, to identify micro-determinants of the exchange rates (i.e. order flows). This work aims to examine the micro-structure of the FX market to see if it has a better record in explaining and forecasting exchange rate movements. Evans and Lyons (2002) assert that order flow, that is, the detail on the size, direction and initiator of transactions, does have significant explanatory power on exchange rates, at least at a high-frequency, intra-day or daily level. The main conclusion of this research is that the FX market can act as an aggregator of information regarding the expectations and circumstances of participants, and order flow is the signal (i.e. it can be viewed as a variable mapping disperse information in the economy towards FX price discovery). Moreover, due to the nature of how this private signal is revealed, inferred from trades in the inter-dealer market, the effect on the spot price should not be transient and should improve the forecastability of exchange rates. Of course one would expect a lag2 between the time when the information contained in the order flow is formed and when it is fully revealed to the market.3 The objective of this paper is to explore and test some of these micro-structural relationships and examine their significance using weekly exchange rates and order flows. Specifically, it looks at customer order flow (a great majority of the present micro-structure literature has focused on inter-dealer or brokered markets). The reason for this focus is that customer order flow is the active side of the trade; the FX market is decentralized with market-makers who quote prices to a wide variety of customers. They then use the brokered market to adjust their inventory to the required level4 amongst themselves (thus adding “hot-potato effects” which greatly increase the total volume traded). Customer order flow can therefore be viewed as the ‘source’ of the transactions conducted in the inter-broker market. By definition all order flow must sum to zero, if we accept that dealers do not carry large inventory positions (see Bjønnes and Rime (2005) for evidence supporting this), therefore if there is a long term impact on FX rates, this must be due to a differential information content of individual orders, dependent on the (perceived) information of the person trading, the reason and size of the trade. The paper therefore examines the effect that heterogeneous customer orders (and the information contained in them) may have on exchange rates by using a unique dataset of weekly net order flow segmented by customer type across nine of the most liquid currency pairs over a six-year period. This is the largest order flow dataset ever used in the literature. It is important to make the distinction between ‘customer order flow’ and ‘inter-dealer order flow’. As described above, a large proportion of the empirical literature on micro-structure has focused on inter-dealer order flow (see Lyons (2001) and extensive references therein for examples). This data has been made available to researchers by some of the platforms used by market-makers to conduct their business. More recently (see Bjønnes et al., 2005b, Evans and Lyons, 2007 and Sager and Taylor, 2008 for examples), data on customer order flow, that is the initiated underlying trade that is given to one or several market-makers, has become available in various forms from some of the top FX trading investment banks in the world. As we shall discuss later in the paper, given market-makers’ risk-aversion to holding large inventory positions over-night (Bjønnes and Rime (2005)), customer order flow can be assumed to be the underlying information revealed through inter-dealer activity. For the rest of the paper the use of the term order flow is associated to this “source,” customer order flow unless we explicitly reference inter-dealer market. If order flow does indeed assist in the information transmission of heterogeneous agents’ expectations, there should be differential information signals from each customer segment. Presumably the motivation for trading of a large corporation will be very different from that of a leveraged hedge fund and therefore the information transmitted by the order should have a different impact on spot rates. Therefore we are interested in three separate issues. The first major issue follows from the previous literature and attempts to address the usefulness of order flow as a conduit through which private information becomes embedded within market prices. This involves an investigation of the extent to which order flow can help to explain exchange rate movements over and above the influence of macro-economic variables. The second issue is to assess the usefulness of order flow in forecasting exchange rates. Given our span of data we are able to shed some light on whether order flows are useful in forecasting exchange rate movements at longer horizons (one and two weeks ahead) than those generally considered in the micro-structure literature.5 Finally we address the question of whether order flow could be used to generate forecasts that can be employed profitably in the FX market (this approach is similar to that taken recently by Rime et al. (2010)). The paper is organised as follows. Section 2 provides a review of the main literature on the micro-structure approach to exchange rates. Section 3 describes our dataset of customer order flows and other macro-variables. Sections 4, 1 and 6 present the empirical results on the estimates and forecasting performance of the model with aggregate and disaggregate order flows. Section 7 examines the profitability of exchange rate forecasts from the order flow model via a simulating trading strategy. The final section summarises the main empirical findings.

This study uses a new proprietary dataset for nine of the most liquid currency pairs, the largest dataset ever used in the literature. Thus, this allows us to focus directly on the initiating customer trades, rather than inferring them from inventory-balancing trades undertaken in the inter-dealer markets. It addresses two important issues which have been investigated in the literature with contrasting empirical results. Firstly, it investigates whether (customer) order flows contain private information which helps to explain exchange rates returns. The in-sample analysis reported shows that this is the case. This result is in line with most of the previous empirical evidence. Additionally, the present study shows that the content of the private information is even more important if one has knowledge of the provenience of the transaction (i.e. with disaggregate data).26 It appears that it is not just that flow is informative but the reason for the flow is also critical. Our data is also disaggregated by customer type, which gives us the opportunity to examine the differential impact of different customer types. We find evidence that profit-motivated traders (leveraged or hedge fund investors and asset managers) have a greater impact on exchange rates and are more informed, and that corporate and private clients act more as liquidity providers, ‘leaning against the wind’ in response to price moves27 (confirming the results of Bjønnes et al., 2005a and Bjønnes et al., 2005b). This is an important result, it suggests that while the order flow is a determinant of exchange rates, it is the motive for the trade that is key. This supports the view that order flow is useful as a ‘backed-by-money” gauge of changes in investors’ expectations of macro-economic fundamentals put forward by Evans and Lyons (2007). The second issue the paper focuses on is using these (customer) order flows to forecast exchange rates. The paper addresses this issue by using both statistical and economic measures of forecasting power and the results are not encouraging, once publication and implementation lags are properly accounted for. This contrasts with the supportive evidence for order flows obtained from the econometric estimates. It may be that weekly data is not timely enough and the information contained within the order flow is already impounded into the exchange rate. How quickly the market discovers and absorbs this information remains an open question that requires a much richer dataset then those currently available. Other fruitful areas for further research include looking at non-linear models,28 the time-structure of order flows, causality between flow and price and the behaviour of flows in reaction to macro-economic variables and surprises.