The foreign exchange market is constantly changing

The foreign exchange market is constantly changing. Currently, only a maximum of 5% of transactions in this market are attributable to international trade. It is therefore natural that some macroeconomic models explain only a very small proportion of the variability of exchange rates. In addition, Meese and Rogoff (1983) have shown that over a period of twelve months, even the macroeconomic models that include variables of a financial nature, such as interest rates, or monetary, such as money, do not have an explanatory power higher than a random walk. It is therefore necessary to determine new models to explain the dynamics of exchange rates. The microstructure approach to exchange market allows to open new perspectives, so far ignored. This market has often been thought of differently than other financial markets, microstructure theory has more time to get involved. For example, it is only recently that the market presence of asymmetric information has been accepted. In fact, the microstructure theory restores the importance of the transaction process ignored by traditional macroeconomic models. Indeed, they provide no role in this process because the information publicly available, is incorporated immediately into prices without requiring the closing of transactions. The microstructure approach consists of two pillars interaction, namely information and institutions. Indeed, the market structure affects the way information is integrated into the price and the behavior of agents will also affect the organization of the market. By introducing new variables such as order flow, microstructure theory explores new horizons because the focus is on where prices are fixed and the behavior of agents who determines. In fact, this theory can be closer to the reality of the foreign exchange market on the way directly into the front office. This theory opens the "black box" where are the prices and trying to understand how it works. 


This new variable, order flow, is used in all models microstructures as proximate determinant of price. This flow is a measure of the pressure on the purchase or sale of assets. Also, keep in mind that this does not exclude the possibility that order flow are ultimately determined by macroeconomic fundamentals. Before beginning the study of the foreign exchange market microstructure point of view, it is interesting to note that this approach has some legitimacy. Indeed, in opening remarks of the symposium entitled Structure and dynamics of financial markets, Ron Parker states that "the Bank of Canada, as probably the most central banks, is interested in market microstructure." In addition, articles on the application of this theory to the foreign exchange market are published in newspapers the most famous. This new approach is known to economists and deserves the greatest interest. This thesis aims to use this new approach to find a model to explain the foreign exchange market. Therefore, a review of the literature, both empirical and theoretical, devoted to the analysis of market microstructure is carried out for a model approximating the reality of the foreign exchange market that improves the understanding of fundamentals governing the dynamics of exchange rates. The aim is to provide the theoretical tools and empirical theory on global foreign exchange market microstructure. Although this memory is clearly oriented towards the side informational, to achieve its objective, it is necessary to start with an analysis of institutions as the final specification proposed should fit best in the market that structural models. Above all, the reader must keep in mind that this brief focuses on the analysis of spot foreign exchange market (spot market) in the major currencies. This market is essentially characterized by a decentralized structure with multiple dealers. It is actually made up of a vast electronic network of investors at large and financial institutions. It is governed by a continuous market prices in the sense that a market maker (dealer) should be the consideration for each transaction. The players on the market are the dealers, brokerage firms (brokers) and customers. The dealers provide quotes to which they agree to buy or sell regardless of the amount proposed or requested by the counterparty. The dealer's quote is characterized by price range (spread), which measures the difference between the selling price (ask) and the purchase price (bid) offered by the dealer. The brokers are only intermediaries in the market because they to not handle on their own stock of assets. Customers conducting transactions for reasons of coverage or speculation, for example. In addition the market has some characteristics of its own. First, compared to other financial markets, a considerable volume is exchanged. Second, most transactions take place between dealers. Third, this market faces a relatively low transparency whose presence does not affect regulation. All of these institutional features can guide the evolution towards a model closer to the reality of the foreign exchange market. The methodology is to develop a model, give the pros and cons and refined by using a different specification. Therefore, from the rational expectations model in the range of macro and micro approach, a general model purely microstructure, namely the model of Kyle, is developed. Finally, the model for concurrent transactions, which considers many of the institutional features and information specific to the foreign exchange market, is advanced to fill the gap left by conventional macroeconomic models. At present, it is this specification that models the better the reality of the foreign exchange market the theoretical point of view. In addition, an empirical validation of the model to simultaneous transactions is essential and can measure the importance of the shift patterns microstructures. In addition, the empirical study can improve this model further by bringing it closer to the reality of the foreign exchange market. In this way, the effect of order flow on prices is broken down into the identity of market participants. This thesis has three parts. The first part provides an introduction to various concepts used throughout the memory relating to the exchange rate and the microstructure approach. This section is intended to familiarize readers with the essential characteristics of the foreign exchange market will be crucial in determining the ideal microstructure model explaining the dynamics of exchange rates and issues considered by the microstructure approach. The second part is the cornerstone of this thesis. Indeed, keeping in mind the importance of empirical studies, it is important to find out what the best theoretical model applies to the foreign exchange market. The three models mentioned above are developed in this section. Particular attention was paid to bring out all the intuitions supporting the model to simultaneous transactions because it is the theoretical model chosen for its ability to explain the dynamics of exchange rates. The third part is devoted to the empirical approach of the microstructure. Empirical validation of the model to simultaneous transactions is developed based on the Evans-Lyons model. Then, an empirical study on the impact on the prices of different types of actors is presented. Finally, another study combining the two approaches in a single model is proposed.

  1. The Foreign Exchange Market and Microstructure: Concepts and Definitions Before starting the analysis of the microstructure approach to exchange market via the explanation of models to explain the dynamics of exchange rates, it is necessary to understand the various concepts related to this market and the microstructure. The traditional macroeconomic models exclude the issue of pricing. Typically, these models assume a Walrasian auctioneer collects all orders and thus determines the equilibrium price under the law of supply and demand. The Walrasian auctioneer is a view of the mind to detach from the reality of price formation. Moreover, in these models, information is supposed to complete, free and public. Nevertheless, the foreign exchange market is a continuous market, fragmented and governed by the price. Moreover, it is a market with multiple dealers are competing. In fact, the Walrasian auctioneer is replaced by several dealers compete. The foreign exchange market is fragmented, the information is dispersed between the dealers and all of the available information is not observable by each individual dealer. It is therefore necessary to find new models to grasp what eludes the traditional macroeconomic theories, namely that the foreign exchange market is a market with multiple dealers and that there is asymmetric information. The microstructure approach captures these facts by entering the "black box" of the foreign exchange market, where prices are formed. The idea is to stand at the front office alongside traders (traders) to understand their behavior and their impact on prices. The microstructure approach not only interested in the institutional market. The other key dimension of this approach is information. Thanks to the microstructure approach, the presence of asymmetric information will be detected and used to understand the dynamics of exchange rates. To determine the effects of asymmetric information on exchange rates, it is necessary to understand how information travels in the system. The identification of these vectors of information will be achieved through a detailed study of the structure of the foreign exchange market. This section is organized into three sections. The first section aims to describe the institutional structure of the foreign exchange market. The microstructure approach to this market and the need to move this approach to explain the dynamics of exchange rates will be described in the second section. The third section will understand the structure of information on the foreign exchange market. The aim of this first part is to familiarize the reader with the essential characteristics of the foreign exchange market since the latter will be used in the next section to determine the theoretical specification models the determination of exchange rates. THE FOREIGN EXCHANGE MARKET The theoretical work of traditional ignore many of the details of price formation and market structure. Literature in the field of design focuses on the microstructure of markets that is used to determine how information is reflected in the price formation. Indeed, the structure of a market will have implications through the process of price formation. Therefore, it is essential to take a look at the foundations of the structure of the foreign exchange market before embarking on a further analysis of microstructure. Prior to characterize the foreign exchange market, it should clearly define the foreign exchange market. When the Bank for International Settlements (BIS) (2002) estimates the average daily trading volume in that market in April 2001 to USD 1.2 trillion, it is defined as the sum of the cash market (including small currencies) and that of Forex derivatives such as swaps (swap rate) and forwards. However, according to Lyons (2001), "the forex swaps did not affect the level of order flow in the foreign exchange market." Indeed, a swap that will bind two transactions in opposite directions. Net order flow from these two transactions will be zero (+ X - X = 0). However, as will be shown later in this section, order flow is the main variable used in the microstructure approach to foreign exchange market. The impact of forex swaps will be marked at the level of interest rates in the short term rather than directly on the foreign exchange market. For reasons of convenience and because the work on the microstructure of foreign exchange markets are heavily focused on the cash market, an approach identical to that of Lyons (2001) will be privileged. Unless otherwise noted, later in this paper, the term foreign exchange market will refer to the spot foreign exchange market on the major currencies. This section is devoted to the analysis of the various players in the market to determine their roles. Next, an overview of the architecture of the foreign exchange market will be developed. This will be followed by a study of its liquidity and the involvement of its structure on the latter. Finally, before concluding this section by analyzing the composition of trade, certain specific features in the foreign exchange market will be highlighted. This will, eventually, to better understand the mechanisms of information transfer through the variables used in the microstructure approach. Actors The major players in the foreign exchange market are, firstly, investors who make up the final demand for foreign currency and, on the other hand, operators who act as intermediaries between final demand and the market. Customers Lyons (2001) refers to investors, customers and operators. The same concept will be used later in this paper. Customers comprise, among others, institutional investors, individuals, speculators and arbitrageurs. Institutional investors and individuals seeking primarily to minimize their transaction costs. Figure 1 shows that a dealer is the counterparty to each transaction with a client. This results from the fact that the foreign exchange market is a market governed by the price. In other words, it is mandatory that a dealer is the counterparty to every transaction. In the remainder of this paper transactions between customer and dealer will refer to this relationship. For the sake of accuracy, it is important to note that, in reality, customers often do not communicate with dealers directly. They go through a commercial intermediary who is none other than the sales staff. Operators Several types of intermediaries are present in the foreign exchange market. They each have their specific role and often complementary. The two types of intermediaries having a particularly important role in this market, and thus the microstructure approach to exchange rates, are the dealers and brokers. As will be presented in the following section devoted to the architecture of the foreign exchange market, dealers continually provide a bid price (bid) and asked (ask) and are required to perform at this price transactions ordered by their customers. This puts them two major risks, the risk of excessive position and that of asymmetric information. They are therefore compensated for this risk taking. Their pay is the spread, ie the difference between the asking price and the price offered. Figure 1: Actors and channels of trade in the foreign exchange market Source: Mende and Menkhoff (2003), Different Counterparties In A Small Bank's FX Trading, University of Hannover, Germany, September, p. 27 The major difference between brokerage firms and dealers consists of the fact that brokers do not risk their own funds on the market. A broker buys and sells securities on behalf of customers (dealers) in exchange for a commission. The brokers do not determine prices, they just act as an interface between clients. Their role is to facilitate the exchange between dealers, which are actually clients of brokers. Lyons (2001) indicates that, in this way, the brokers provide a higher degree of centralization in the foreign exchange market that otherwise would be completely decentralized. Figure 1 describes the inter dealer transactions as those taking place between two dealers, either directly or indirectly, via through a broker. In the remainder of this paper, the same terminology is adopted. In order to understand the value of brokers, we must understand that to complete a transaction, the dealer has only two. The first is to call directly to another dealer, to ask the listing and, if agreed on price, to complete the transaction. This method is a direct exchange inter dealers. In the second method, the dealer may contact a broker and pass the order through him. According to Lyons (2001), in 1998 about half of the exchanges between dealers in major spot markets were direct whereas in 2000, only one tenth of this trade was. This confirms the important role of brokers. It is interesting to dwell a moment on the motivations that encourage dealers to move to a method of indirect exchange. Smaller dealers will have an interest to go through a broker, despite the commission that it will take. Indeed, small dealers do not have access to spreads as narrow as those enjoyed by major institutions. However, even the large institutions have an incentive to go through the brokers. With these, a large dealer will be able to indicate its willingness to buy or sell the whole market in a short time. Another advantage is extremely important in view of the memory and affects both large and small dealers is that the passage through a broker ensures anonymity before the conclusion of the exchange. In light of all these arguments, it is easier to understand the usefulness of indirect exchange even if it means paying a commission. Architecture After describing the actors on the foreign exchange market, it is essential to understand the structure of this market because the two points on which the microstructure approach focuses are information and structure. It is clear that when choosing a model explaining the best microstructure changes in exchange rates, the structural dimension of the market must be kept in the mind of the analyst. Madhavan (2002) defines the architecture of the market as "the set of rules governing the transaction process." This entire section devoted to the architecture of the foreign exchange market is based on Chapter Organizing Principles of Financial Markets from the book Market Microstructure - Institution, models and empirical tests of bias, Foucault and Hillion (1997) . In their book, they describe all the elements characterizing the structures of financial markets. A summary of their specific text on to the case of the foreign exchange market is developed in this section. The writings of Lescourret (2003), Lyons (2001) and Hamon (1995) have also contributed to the perspective of organizational principles of the exchange market. Continuous market A continuous market is, as its name implies, a market where transactions take place continuously. Agents have the opportunity to transmit and execute orders at any time during the opening of the meeting. The only restriction on the execution of a transaction is that we must find another party willing to make the exchange, that is why transactions are called bilateral. After each transaction, a new course is calculated. The price of the currency is not permanently fixed for the entire session, it evolves over the transactions. On the foreign exchange market, a number of purchase orders and sales are therefore associated with several awards. The major advantage of a continuous market is the flexibility it offers. Indeed, transactions can occur at any time, there is no deadline for placing orders. Obviously, this flexibility has a cost. Because of the bilateral trade, the date of actual completion of the transaction is endogenous. This is the result that it is necessary to wait a counterparty accepts the exchange. However, it is important to note that this market is facing some sources of discontinuity since no transaction can be executed during the closure. If an order is placed during the period of foreclosure, it will only be made to reopen it. This element is an important feature of the market when it comes to studying the behavior of traders. They try to close the day with a net position close to zero to prevent the most at risk of long (positive net position) or short (negative net position) during the closed season, the risk from the fact that in any transaction will be done during this period. Ruled by the Market Price Auction in the article versus Dealership Markets, Bennouri (2003) contrasts two main types of market, centralized auction markets (auction market), order-driven and market return (dealership markets). The currency market is characterized as a market counterparty, that is to say, governed by price. Agents must submit their orders to buy or sell at a dealer who will then return. It continuously displays a sale price (ask) and a purchase price (bid), the difference between the two is called the price range. Lescourret (2003) states that the exchange protocol of a typical market price is governed by two steps. First, the dealers display their prices. Then, customers submit their orders. The dealer is free to set and revise prices, but must serve the orders submitted to it at this price. A major difference compared to other markets governed by the price that the foreign exchange market, the dealer can not fix the amounts. Gravelle (2002) also stipulates that markets dominated by institutional investors and large orders are structured in markets governed by the price. These two elements characterize clearly the foreign exchange market. In addition, fewer counterparties are present on this market than in markets governed by the orders. The dealer's role is to absorb temporary imbalances and thus to ensure market liquidity. This means significantly reduced (see cancel) the risk of endogeneity of the date of execution made by the agents. Nevertheless, the risk has not disappeared, it is simply passed on to the dealer because it should play on its own stock to serve the orders. The risk of excessive position is the first type of risk faced by the dealer. Moreover, since the submission of orders takes place after the display of prices, it runs the risk that an agent with inside information may use it by buying a currency undervalued or her by selling overvalued and is the risk of asymmetric information. The dealer is compensated for these risks through the price range. Indeed, the purchase price is always lower than the sales price which enables it, all things being equal, to make a profit on each transaction of purchase and sale. However, Cheung and Chinn (2001) argue that the cost of risk of information asymmetry is not the main element in the formation of spread. Following their survey based on a questionnaire submitted directly by mailing list to American traders whose results were published in the article Currency Traders and Exchange Rate Dynamics: a Survey of the US Market, they found that reputation was the fundamental element in the formation of spread. A trader does not wish to make a listing with a wide spread because it reveals that he is currently in a negative net position, which would negatively affect its reputation, thereby depriving him of future transactions. Cheung and Wong (2000) confirms this argument and develop a reputation in their investigation A Survey of Market Practitioners' Views on Exchange Rate Dynamics. According to them, over 70% of dealers who responded to the survey believe that the convention are the major determinant of the spread. The dealers interviewed noted that frequent violations of these conventions lead to loss of reputation and ultimately a reduction in activity. In addition, it should be noted that much of the profits made by drug dealers come from movements in exchange rates. The spread also covers the processing costs (processing costs). Ultimately, these treatment costs are borne by the dealer counterparty. On the foreign exchange market, there are several dealers who compete, which is an element that can reduce the range compared to markets where the content is a monopoly. From the time when many dealers are in competition, the market return is called multi-dealer market that is opposed to the single market dealer.
  Fragmented market A centralized market where quotes from several dealers are available in a consolidated format. However, in a fragmented market, some quotes are not observable by all dealers. The foreign exchange market is by nature a fragmented market for order flow is distributed between different locations. Indeed, there is no one place where all agents wishing such currency are required to meet the foreign exchange market is, by his organization, an OTC market. On this type of market it is possible to observe several different prices at the same time to the same currency. A few years ago, dealers were broadcasting information primarily by telephone, but currently the computer systems Reuters Dealing 2000-1 (direct exchange inter-dealer), Dealing 2000-2 and EBS (exchanges between brokers) are the most used. This automation enables increased centralization of the currency market. For example, the system Reuters Dealing 2000-1 alone accounts for about 90% of direct inter dealer transactions. The BIS (2002), states that in 2001, "the increasing use of electronic brokers means that dealers have to make fewer transactions directly with each other." The brokers also play a role in centralizing serves as an interface between the dealers. The current trend is the centralization of the market originally fragmented. Moreover, the concentration has accelerated in recent years. However, at present, this market remains fragmented, which is important in the emergence of asymmetric information which will be discussed in the last section of this part. Liquidity Liquidity is a difficult concept to define. It is mainly influenced by transaction costs, the speed of execution and the impact of the transaction price. Sharpe, Alexander and Bailey (1999) define liquidity as "the ability to sell an asset quickly without requiring a substantial reduction in price." Lescourret (2003), in turn, defines liquidity as "the ability to quickly exchange of relatively large volumes with little impact on prices." Another dimension of liquidity is the speed of recovery in prices after the sale of a large volume (resiliency). In a market continuously, all order flow is distributed over time. In the case of a significant transaction volume, it may be difficult to find a counterpart without a concession on price. By cons in terms of speed, this market is ideal as a transaction can take place at any time. The dealer must serve the orders submitted to it by playing on its own stock. Thus, temporal imbalances between supply and demand are absorbed by the dealers. These characteristics are favorable foreign exchange market liquidity because they allow faster execution of orders while keeping an impact on prices relatively low. A centralized market is more liquid because, all things being equal, the volume will be larger than several fragmented markets. On this point, the foreign exchange market is not the most efficient. The increasing centralization of the market via the computer, however, could allow the currency market to increase liquidity. Specific features Lyons (2001) identifies three features specific to the foreign exchange market. First, trading volumes are considerable. In April 2001, the average daily trading volume was around USD 1.2 trillion in the foreign exchange market at large (that is to say, taking account of forex swaps), which is much greater than the volume of exchanges observed in other financial markets. By comparison the average daily volume (in value) traded on the New York Stock Exchange (NYSE) in 2003 was USD 38.5 billion. Madhavan (2002) also states that "the foreign exchange market is by far the largest market in terms of asset size." The enormous volume traded in this market due to the phenomenon of hot potato (hot potato) developed as part of a microstructure approach. A phenomenon of hot potato is defined as "a process that takes place when a trade is carried out unwanted positions dealer to dealer following an initial transaction from a client." The following example will help clarify the concept. Is a customer who buys a EUR 100 million to A dealer (the initial net position of zero), then A is short of EUR 100 million, so as not to bear too much risk, A will hedge its position by acquiring 100 million to another dealer B (initially along 30, by assumption), the net position of A returns to zero but B is found in turn short of EUR 70 million, it will then in turn get rid of the hot potato to another dealer, and so on. This small, highly simplified example shows that trading volume can easily grow very rapidly on the basis of an initial transaction from a client. This transition from dealer to dealer is the result of risk management and drug dealers from the fact that, ex ante, the dealer is not aware of the positions of other dealers. This contrasts with traditional macroeconomic models that assign a high volume of foreign exchange speculation. Second, another feature specific to the foreign exchange market is identified as transactions take place mainly between dealers. Indeed, the inter dealer transactions, direct and indirect (through a broker transactions) account for about two-thirds of total trading volume on the spot market. Third, the foreign exchange market is more opaque than other markets to multiple dealers. In the latter, transactions must be disclosed in the minutes. For example, Biais, Hillion and Foucault (1997) state that "the system broadcasts the ACC over the amount and timing of the last ten transactions." On the foreign exchange market, there is no disclosure requirement which implies that transactions generally are not observable. This particularly concerns the transactions between customers and dealers. Indeed, transactions between dealers are not so opaque. By the method of communication used, transactions between dealers via broker, are more transparent. Also, it is necessary to note that the foreign exchange market is not completely opaque, meaning that the transparency in this market is of no regulatory influence. Indeed, Williams (2000) confirms that "the foreign exchange market does not have a formal regulatory body." In fact, there are two forces. The first stems from the fact that increasing transparency accelerates the disclosure of information by the prices. However, the informed dealers do not wish to disclose information to the free market, this force tends to make the market more opaque. Madhavan (1995) also states that the dealers with an informational advantage will prefer a more fragmented market opaque, because the price competition is less strong and they can select the most profitable time to use their private information. The second force tends, in turn, enhance market transparency. In fact, the dealers do not want a fully opaque because, in this case, customers would have too little information to place orders. The dealer would then be unable to share risk with clients. It is generally accepted that transparency defined by O'Hara (1995) as "the ability of market participants to observe information about the transaction process" affects the market liquidity. In addition, it is generally accepted that prices have more information content when the market is transparent. However, full transparency is not necessarily pro-market operations. Traders with better information will have an interest to make trading in a market while ensuring anonymity uninformed traders prefer a more transparent market. Based on these arguments, a universally accepted level of transparency can not be determined. Transparency will be considered an advantage for some and a disadvantage for others who prefer opacity. Composition of trade In this institutional framework, it is interesting to measure the contribution to trade the various actors (customers, dealers and brokers). Indeed, as mentioned above, the intermediate transactions do not occur for the same reasons as those between customers and dealers. In addition, these transactions are less transparent than those made between intermediaries. It shows directly that the information contained in the various transactions will not necessarily be identical. Therefore, it is possible to find theoretical models to include these types of exchanges. Prior to linger a moment on the composition of trade itself, the decomposition of the volume of trade observed in the currency market will be addressed; table B.1. from the last triennial survey in 2001 concerning the activity of the foreign exchange market and derivatives made by the BIS (2002) is reproduced below (Table 1). Table 1: Total volume in the market for changes1 Daily average in April, in billions of USD
  Source: BIS (2002), op. cit. 1Adjusted for local and cross-border double counting. 2Revu since the last survey. 3The party transactions in currencies other than USD have been converted into USD amounts in the original currency amounts at the exchange rates of April for each annual survey. Then they were converted into amounts in USD at the exchange rate through April 2001. According to the survey in April 2001, about 30% of the daily volume of transactions (in value) on the foreign exchange market at large is traded on the spot market. However, at the beginning of this section it was stated that the forex swaps should be excluded because they did not affect the order flow. The volume of cash transactions is then USD 387 billion USD 544 billion of foreign exchange market, more than 70%. The hypothesis of the study be limited to the spot market in the context of this paper is reasonable. Regarding the evolution of the daily volume of transactions, it may be surprising that the foreign exchange market at large has seen its volume decline by 19% compared to 1998. Even looking at the volume at constant exchange rates (base 2001), a decrease in volume of 14% is found. However, although this contrasts sharply with all previous surveys, an explanation can be advanced that the expected decrease in volume is consistent with the spot market (which down nearly 32%). This market has undergone significant changes in recent years. In this context, the arrival of the euro has obviously precipitated the fall following the closure of foreign exchange markets between the countries of the Euro Zone. The very small decline in forex swaps may, in turn, be attributed to activity in the market for interest rate swaps. After this point is devoted to analyzing the composition of trade by different actors on the foreign exchange market. Lyons (2001) argues that transactions between intermediaries represent about two-thirds of the total volume traded. The rest are transactions between customer and dealer. The triennial survey of foreign exchange market activity in 1998 and 2001 conducted by the BIS will allow us to verify this. In 1998, USD 347.689 billion from USD 577.737 billion traded in the spot market, about 60% were due to inter-dealer transactions. In addition, Lyons (2001) explains that in reality the investigation conducted by the BIS tends to underestimate the part of the transactions as intermediaries, as defined by the BIS data, some intermediaries are also included in "other financial institutions ", it also contains currency brokers whose trading volume is not negligible. In 2001, over 56% (= 217619/386963) are related to transactions between intermediaries. According to the BIS (2002), some procedural changes between the two surveys can introduce bias in inter temporal comparisons. However, the decline between 1998 and 2001 could also be due to the fact that trade between intermediate passes increasingly by electronic brokers who may have been included in "other financial institutions." In 2001, nearly 29% of the volume of transactions on the spot market is recognized in the latter category, against only 21% in 1998. This argument is confirmed by the investigation conducted by Cheung and Chinn (2001) which found that over the past five years "in general, it appears that transactions via electronic broker increased primarily but not exclusively, at the expense of traditional brokers ". The data above are shown in the Annex (Table 5).
  MICROSTRUCTURE APPROACH Often, the microstructure approach is used to study problems on the securities markets. The originality of this approach to use the foreign exchange market is the fact that traditionally, the market is analyzed from a macroeconomic perspective. O'Hara (1995) defines the market microstructure approach as "the study of processes and outcomes of exchanging assets under explicit rules of trading". In practice this means that the approach microstructure of exchange rates is used to study the market structure but also the information content of transactions and their impact on price formation. In fact, it is important to keep in mind that on the currency market, prices are expected exchange rate. For example, the exchange rate of euro to dollar is 1.27 this means EUR 1.00 = USD 1.27. In other words, the price of the euro is USD 1.27 (and conversely the dollar price is EUR 0.79). In the remainder of this paper, the concepts of price and exchange rate will be used indiscriminately. In macroeconomic models used to explain the formation and evolution of exchange rates, it is often ignored the practical aspects of price formation. When the balance is determined, the price will be fixed automatically. The microstructure approach will help to understand how prices are formed. The goal is to return to the "black box" of the economy, especially in the foreign exchange market. In reality, this theoretical approach attempts to bring closer to where prices are formed, to return to the front office. Indeed, ultimately it is the traders that will determine the exchange rate. The microstructure approach to exchange market will therefore study the implication of market structure and the process of learning from information on training, at the micro level, exchange rates. This theory disrupts the established macroeconomic theories to explain the evolution of exchange rates. Micro Vision is brought into an area that traditionally has been considered the macroeconomic point of view. The theory of the microstructure of financial markets, the motivations of the agents to exchange are conducted primarily by the desire to share risk and enjoy better information than those available to other agents. The first motivation characterized models of stock while the motivation of asymmetric information is modeled using information models. It is essential at this stage to note that the theory applied to the microstructure exchange market does not only study the impact of institutions on the process of price formation, but also and above all, the analysis of the consequences of the presence of asymmetric information in exchange rates. In a market where transparency is low, as the foreign exchange market, information asymmetry has more opportunities to develop. For example, the sizes and prices of individual transactions are not observable by the entire market. Lyons (2001) shows that the model microstructures are not only used for high frequency data (high frequency data) but the results obtained by the type of informational models microstructures are also valid in the long term. Indeed, in economics, it is generally accepted that the price movements due to new information are persistent while the transient effects are due to pricing errors. Thus the identification of information channels and agents having access is essential. In fact, the theory microstructure research the causes of movement in the exchange rate at the front office. After describing the pillars of the microstructure approach, the end of this section will demonstrate the importance of the transition to a microstructure analysis of the dynamics of exchange rates. Variables in the Central Microstructure Approach In order to link the composition of trade in their information content microstructure mainly uses two variables. They are the two vectors of information most important in the microstructure approach. The first vector, the order flow is not used in the macroeconomic theories. This is explained by the fact that these theories make the assumption of complete information, free and public while the micro approach focuses on the problems of asymmetric information. The study of order flow allows us to understand the impact of this asymmetry of the price range which is the second vector information in the microstructure approach. Prices are determined by the dealers, the study of price formation is the study of the behavior of these dealers. Order flow Before determining how the order flow influences prices, it is necessary to clarify what exactly this new variable is the largest of the microstructure approach. Indeed, all models use the microstructure order flow as a proximate determinant of price. Transaction volume and order flow does not cover the same concept. Order flow is characterized by its sign (positive or negative). It depends on which party is responsible for the order. Figure 2 shows that if the buyer (seller) is the active part, the order flow is positive (negative). When the net order flow is positive (negative), the market as a whole is under pressure to buy (sell). According to this characteristic, the order flow can be interpreted as a variant of the concept of excess demand. Indeed, the two concepts are very similar although different. First, the equilibrium excess demand is zero, which is not necessarily the case in the net order flow. This is because the dealers are obliged to carry out the transactions that come before them. Second, order flow for the actual transactions as soon as new information is publicly available, demand adjusts without requiring the execution of transactions. Figure 2: Sign of order flow under Part initiating the order
  Source: Lyons (2001), op. cit., p 6 The role of order flow is to approximate the determinants of the dynamics of exchange rates. The intuition is that order flow carries information. This information can come from macroeconomic fundamentals (interest rates, inflation, employment, GDP, etc) but this is not limiting. From that point, the trader who observes a negative net order flow, can rationally believe that his return has received some bad news. Of course, that this reasoning is correct, it is necessary that the order flow does have an informational content. Evans and Lyons (1999) suggest that this is the case. So that order flow may convey information it is necessary to release one of two assumptions that the disconnect in prices. The first hypothesis, a priori difficult to challenge, indicates that the information relevant to the determination of exchange rates is publicly released. The second hypothesis states that the way this information affects the equilibrium price is also publicly known. However, this second hypothesis can be easily released due to lack of consensus on a model explaining the dynamics of exchange rates. It is therefore possible that the order flow carries information relevant for price formation. In addition, Cheung and Wong (2000) suggest that practitioners who participated in their survey were selected "information and a large customer base as the two main sources of competitive advantages of major players in the foreign exchange market." This confirms the importance attached to observing the order flow as a channel of information. Note that Evans and Lyons (1999) do not reject the role of informational rights. They only indicate that it is perhaps not the only relevant variables in determining exchange rates. Indeed, the information used by a customer who places an order from a dealer can come from his analysis of macroeconomic fundamentals and expectations he derives. In this sense, order flow could be seen as a conduit of information. They point out also that "it should be noted that the fact that order flow is a proximate determinant of prices in the model does not exclude microstructures macroeconomic fundamentals to be the underlying determinants." Evans and Lyons (2002a) conclude by demonstrating that their model, consisting of variables from the microstructure approach, such as order flow and macroeconomic variables such as interest rates, "explains about 60% of daily changes in the exchange rate DEM / USD "and over a period of four months. This is a new indicator of the importance of order flow as a channel of information. The major drawback of this variable is difficult to find empirical data. First, the foreign exchange market is a decentralized market. There is no body that centralizes all orders. On the other hand, if the order flow carries information, traders have no incentive to make public the order flow they observe as it would distribute information for free. Price Range The price range is the difference between the asking price and the price offered. Its usefulness consists in the dealer's expected earnings, traditionally, it will buy foreign currency at a price below that at which they would sell at the same time. However, when the time differential is taken into account, the range produced by the dealer will often be different from the range displayed. Indeed, the purchase and sale transactions generally have no place at the same time. During the time interval between the two transactions, the range may have changed. It has already been discussed above determinants of the price range. In summary, this is the compensation risk of asymmetric information, the position risk excessive transaction costs and, according to practitioners surveyed, the desire not to undermine their reputation. In order microstructure, the price range is one of the two main vectors of information. For example, a dealer that displays a wide range of prices the market will reveal that it is at a disadvantage, which will have a negative impact on reputation. A branch of the microstructure of financial markets focuses on the analysis of this price range. Some may equate the microstructure in this study. However, the microstructure approach to exchange market is not only interested in analyzing the bid-ask spread, which is not, in fact, only one dimension of the microstructure. Moreover, models microstructures have been developed in isolation from this price range. Unlike the order flow, the data on spreads are available in greater numbers, are more complete and series are longer. Microstructural models focusing on these variables can be tested more easily. However, as mentioned above, the microstructure approach to exchange market deals with the structure of the market but also and above all, problems of asymmetric information. This memory is centralized so the introduction of order flow as a predictor of the dynamics of exchange rates. Motivation Passage Models for Microstructures Hamon (1995) states that "the microstructure is concerned [...] to challenge the assumption of free information available without delay and without inequality", in other words, the microstructure is concerned about the effects the presence of asymmetric information on the foreign exchange market. However, before starting the analysis of this new approach it is essential to understand the basis for the introduction of asymmetric information on the foreign exchange market. The platform approach has been microstructure Article Empirical Exchange Rate Models of the Seventies: Do They Fit out of Sample? of Meese and Rogoff (1983). Following this article, the traditional macroeconomic approach has been undermined. Indeed, the authors determined that over a period of one to twelve months, a random walk has an explanatory power of the dynamics of exchange rate equivalent to that of macroeconomic models and this despite having used the fundamental values ​​actually observed rather than the anticipation. Note that, in their article Banking on Currency Forecasts: How Predictable Is Change in Money?, Chinn and Meese (1995) confirm that in the short term, the models based on macroeconomic fundamentals do not predict better exchange rate that a random walk. Based on these results, Evans and Lyons (1999) therefore indicate that "the proportion of the exchange rate explained by macroeconomic models is basically zero." Therefore it was necessary to find a new approach. The microstructure approach, introducing a variable approximating the determinants of exchange rates, was an innovative idea. Furthermore, according to Evans and Lyons (1999), "order flow explains most variations in nominal exchange rates over a period as long as four months." Meese and Rogoff (1983) tested several structural models based on the asset market approach. Under this approach, changes in exchange rates are from transactions in the goods market and asset markets. The idea is that an investor (client) European buying an asset (property) will have to buy U.S. dollars to pay its U.S. counterpart. Transactions on markets for goods and assets would result in a variation of the demand for foreign currency and, thus, a variation of the exchange rate to reach the new equilibrium. The general model tested by Meese and Rogoff (1983) takes into account the differential between domestic and foreign variables such as money supply, real income, interest rates, inflation and balance of payment:
  (Eq. 01)
When no constraint is present nullity of coefficients in the equation proposed by the authors (Eq. 01), it is representative of the most general model tested by Meese and Rogoff (1983), namely, that of Hooper- Morton. However, over a period of one to twelve months, this model fails to explain better the variation in exchange rates as random walk. To find the determinants of exchange rate dynamics, we must find new variables or change of thinking. The microstructure approach applies to change the way the transmission of information is collected on the foreign exchange market. Lyons (2001) states that the microstructure approach aims to release three assumptions commonly asked questions in macroeconomics. First, this approach recognizes that all the information is not public on the foreign exchange market. This implies the recognition of information asymmetry in this market. Then she admits that the agents affect different prices depending on their nature. Finally, market structure, the institutional side, also has an impact on the process of price formation.
  STRUCTURE OF THE MARKET INFORMATION EXCHANGE The microstructure approach to exchange market is characterized by the emphasis on the availability of market information. A more complete analysis of the structure of information on the foreign exchange market will find the links between theory and microstructure study of the dynamics of exchange rates. First, the differences between public and private information are identified and formalized. Second, based on current literature, it is shown that the foreign exchange market is facing a very asymmetric information. At this point, a link is established between the players passing orders on the foreign exchange market and in so doing, the structure of the information they reveal. Information Private versus Public Information Introduce the concept of asymmetric information on the foreign exchange market implies that all the information there is not public. This information will be called part of private information. Lyons (2001) defines private information as "information not known by everyone and produces better forecasts that public information alone." According to this definition, the order flow observed by a dealer is a source of private information. As a reminder, Lyons (2001) states that it is necessary that two assumptions are met for information to be considered public. Thus, information relevant to the evaluation of exchange rates must be transmitted publicly. In addition, the process of integrating this information into prices must be publicly known. It is precisely the non-compliance with this second hypothesis gave way to the introduction of private information. Based on a simple model with two trading periods, Lyons (2001) distinguishes two types of private information. An initial transaction (at t = 0) occurs at a price P0 and P1 at t = 1. Gain (payoff) to the value of V is achieved at t = 2. The first type of private information is related to the size of the payoff. For example, a dealer receives an order from a central bank has also received private information. In their article Sources of Private Information in FX Trading, Carpenter and Wang (2003) show, moreover, that "it is central banks that have the greatest impact on exchange rates", which confirms that the orders from central banks contain private information. The second type of private information on the prices at which transactions take place (P0 and P1). This includes all the variables for risk premiums. However, one dealer has a deeper knowledge of the risk premiums that the market as a whole. Indeed, data on order flow is not public, dealers, observing at least their own order flow, have a better ability to assess this risk. Variables related to the exchange capacity of traders also affect P0 and P1. Traders with a large exchange capacity can afford to wait any longer before changing their prices to return to a net position acceptable. In their article Asymmetric Information and Price Discovery in the FX Market: Does Tokyo Know More about the Yen?, Covrig and Melvin (2002) also identify the order flow from customers and knowledge of important government data output or 'political actions as two sources of private information. Information asymmetry Foreign Exchange Now that private information found on the foreign exchange market has been characterized, it is possible to demonstrate the presence of asymmetric information. The presence of asymmetric information will be recorded if there is private information. Examples of private information seen in the previous paragraph suggests that the foreign exchange market is faced with asymmetric information. The use of private information and the impact on prices that results are characterized by the fact that at constant volume, order flow from different types of players have different effects on prices. Indeed, the order flow carrying more information about future cash flows will have more influence on prices. Carpenter and Wang (2003) state that "the impact on prices reflects the information content of transactions from different groups and thus reveals the sources of private information in the foreign exchange market." The fact that several studies show that, depending on their characteristics, the actors on the foreign exchange market affect prices in very different ways it can be concluded formally in the presence of effects due to the phenomenon of asymmetric information. In Chapter 9 of his book, The Microstructure Approach to Exchange Rates, Lyons (2001), shows that the order flow of customers have different impacts on prices depending on the type of customers. He noted that non-financial corporations have no impact on the market USD / EUR. For cons, the orders from financial institutions have a positive effect. The players with the most market impact USD / EUR are, among others, pension funds, investment companies and insurance companies life. Lyons (2001) groups these players under the term "unleveraged Financial Institutions." A more complete analysis was conducted by Carpenter and Wang (2003). They not only compared the impacts of orders from different types of customers but they also analyzed the impact of orders between dealers. They split transactions inter dealer transactions in direct and indirect. Direct transactions are conducted by telephone or by the system Reuters Dealing 3000 Direct is actually the successor to the Reuters Dealing 2000-1 system mentioned above. Customers were divided into three categories: central banks, nonbank financial institutions and non-financial corporations. Carpenter and Wang (2003) distinguish the central banks of other types of clients because they have a very special role. Indeed, they have a monopoly on the supply of domestic currency. In addition, by their position, they have private information about macroeconomic fundamentals. Their analysis is based on transactions of a major Australian bank on the spot foreign exchange markets the Australian dollar against the U.S. dollar (AUD / USD) and euro against the U.S. dollar (EUR / USD) over a period of 45 days in 2002. Carpenter and Wang (2003) have determined that the transactions by the central banks had the largest impact on the exchange rate AUD / USD. This is explained by the fact that the Australian central bank (Reserve Bank of Australia) was very active in the market at that time. However, central banks did not affect the exchange rate EUR / USD. This does not, however, fits in contradiction with the previous result because the transactions were from non-OECD banks, so these transactions were not from the European Central Bank (ECB) or the U.S. Federal Reserve. In addition, the transaction volume on the market EUR / USD was not significant. The nonbank financial institutions have a significant impact on exchange rates in both markets. Given the small share of the total volume of foreign exchange transactions that account for non-financial corporations, it is logical to see the impact they have on the lowest price. In the inter bank market, dealers with more private information prefer a market characterized by low transparency after transaction. The indirect transactions that are revealed, at least partially, to the market via brokers have therefore less impact. The foreign exchange market is therefore facing the presence of asymmetric information. At least, these studies show that the actors believe that this market is faced with asymmetric information. The remainder of this paper will be devoted to the discovery of microstructure models capable of capturing this information asymmetry, while remaining consistent with the institutional characteristics of this market.

  2. Microstructure: Theoretical Study The second part is a central point of this brief. Indeed, keeping in mind the importance of empirical studies, it is important to find out what the best theoretical model applies to the foreign exchange market. However, this part does not aim to be exhaustive. Much work will not be considered because the goal of this second part is simply to provide a valid model for the microstructure analysis of the dynamics of exchange rates and to provide the main steps leading to the current model. The approach is based on a model closer to the traditional macroeconomic models and sharing certain characteristics of the microstructure model, to describe its operation and its results and then to target areas for improvement to better match the market structure exchange detailed in the previous section. Based on these factors, a model closer to reality is developed using the same principles to facilitate the comparison for the reader. Finally, a theoretical model appropriate to the foreign exchange market is highlighted. The principle used in this second part is therefore from a basic model, it graft extensions and correcting its defects lead to the final model. The rational expectations model, developed in the first section, was chosen as a starting point because it shares characteristics of traditional macroeconomic, such as the presence of an auctioneer implied, and microstructures, such as the importance given to the information. This model has the advantage that challenge the traditional macroeconomic models as the private information in the form of observing a signal about the future payoff of the risky asset plays a central role in determining the equilibrium price . However, some critics, such as the use of a Walrasian auctioneer, make this model must be completed to better conform to the reality of foreign exchange market. Therefore a second model correcting many of the shortcomings raised is analyzed. Thus the model of Kyle, known in the theory microstructure is analyzed in the second section. The same type of model analysis is adopted. The main quality of this model is that it includes an informational dimension and a structural dimension acting jointly in the process of pricing. In addition, the model explicitly addresses the issue of price formation by introducing market makers. In addition, this introduction allows us to appreciate a new informational dimension, market makers setting their prices based on their observation of order flow. However, some assumptions of this model should be lifted or modified. For example, it is essential that the final model characterizes a market governed by the price with more risk-averse dealers acting strategically. The third section is devoted to the study of such a model. The model is developed concurrent transactions. This is currently the best theory that models the exchange market. This model is essential in this paper as it will serve, including the basis for empirical models developed in the next section. Therefore, all the intuitions supporting this theory is developed in detail. Rational expectations models The beginning of the theoretical analysis is devoted to the study of a model of rational expectations. Although this is not exactly a model microstructure, it can be aware of the need to move to this approach. The main shortcoming of the model of rational expectations comes from the use of the Walrasian auctioneer that means to enforce the rule of price formation. Indeed, in this model, an imaginary agent, the Walrasian auctioneer, collecting orders before providing a price which clears the market and it will carry out the orders that were transmitted. Now this is a Walrasian auctioneer to mind that does not exist in the reality of foreign exchange market. In this regard, Williams (2000) states that "with this specification, economists are wondering how the prices work but not how they are established." However, even if the rational expectations model neglects the mechanism of price formation, it is interesting to begin with it because it takes into account the differential information may exist between the traders. Indeed, the notion of information, which, recall, is one of the two pillars of the microstructure, is central to this model. The version of the rational expectations model provided by Grossman and Stiglitz (1980) is representative of this type of model in the literature on the microstructure. Lyons (2001) simplifies the approach to target the most important features in microstructure. Therefore, the version of Grossman and Stiglitz (1980) simplified by Lyons (2001) will be used in this paper to reflect this type of model. Assumptions A risky asset is traded against a risk-free asset. The price of the risky asset is denoted P and the end of period payoff is denoted V. V is distributed according to a normal with zero mean and standard deviation  V (). Ask a mean of zero does not detract from economic reasoning and allows the hand to make it more intuitive. The transaction takes place over a single exchange. There are two agents, an informed trader and an uninformed trader, both of which are risk averse. In addition, they consider market prices as given, in other words, they behave non-strategically. In addition to being competitive, agents have rational expectations. The trader informed insider (I), has exclusive access to private information on the payoff that V will provide the risky asset. Advantage of this privileged information and cover are two motivations for this agent to conduct transactions. It receives this information on the private payoff V as a signal S. The informed trader and the uninformed trader know that S is distributed under a normal distribution with mean V and standard deviation  S (). The specification of this signal is:
  (Eq. 02)
Where  denotes the noise in the signal S.  has zero mean and variance equal to. The insider does not observe the ex ante payoff V but receives a signal allowing it to anticipate what payoff. His request for the risky asset will therefore depend on the price of the risky asset P and S signal he observes. The second agent is an uninformed trader (U) that trades at random or for hedging purposes. Although he does not observe the signal S, he knows, as an insider, the distribution of S. His request for the risky asset will depend on the price of the latter. However, since a rational expectations, they learn the relationship between prices and the distribution of S and uses it to derive the application. Each trader receives an initial endowment of risky asset is denoted as Xi. Xi is distributed according to a normal with zero mean and standard deviation  X (). XU XI and are distributed independently of one another and independently of the signal S and the payoff V. X represents the aggregate supply of risky asset:
  (Eq. 03)
All things being equal, if XI> XU, the insider will have interest in selling the risky asset on the grounds of coverage. Of course, the uninformed trader can not be derived simply by observing P S, a change in price may be due to a change in the information held by the insider or a change in aggregate supply. Generally, the microstructure models use exponential utility function to represent the utility of agents. In the model of rational expectations, each of the two agents has a utility function like this:
  (Eq. 04)
where Wi denotes the wealth of individual i at the end of the period. This utility function has an absolute risk aversion, constant and equal to unity. Each agent knows the rules of price formation, which implies that the uninformed trader can use this rule and the observation of price P to infer information from the informed trader. In the model of rational expectations, prices play a dual role to balance the market and to transmit information. The rule proposed pricing is as follows:
  (Eq. 05)
This rule of price formation must meet the conditions of rational expectations equilibrium. These conditions are two in number. First, the equilibrium price must balance the market, that is to say that the excess demand must be zero at equilibrium. Second, the function used by traders to determine their application must be the actual function of price formation that occurs in the market. As a result, expectations are correct and beliefs of traders are rational. Description of the Model The idea of ​​this model is that by carrying out transactions, the trader holding private information will be communicated, at least in part, to the uninformed trader. The latter will make its informational learning through observation of prices and knowledge of the rule of price formation. To determine equilibrium, we must first find the expressions of expectations of the payoff V for both traders. Then, on the basis of these expectations of V, the respective claims of the two traders for the risky asset will be determined. Then, using these applications and the rule of price formation, it remains to find the balancing market prices. Finally, check that the equilibrium prices satisfy the two equilibrium conditions in rational expectations above. Anticipation of Payoff V The informed trader learns only from its observation of the signal S. His beliefs about the payoff V, conditional on signal S, are normally distributed with:
  and (Eq. 06)
These results stem from the fact that the random variables are normally distributed and that the absolute risk aversion is constant. The Annex to Chapter 4 of Lyons (2001) provides more details on this modeling CARA-normal. The uninformed trader uses the price, containing information from an insider, and his knowledge of the rule of price formation to infer the signal S. Of course, the uninformed trader will not determine S perfectly, the aggregate supply is random and not observable. A price increase may come from a positive signal or a decrease in the supply of risky asset. Based on the rule of price formation (Eq. 05):
  Or, is distributed around an average S. To facilitate the notation, is set equal to Z. Since, and that S and X are independent, Z is distributed normally around S with a variance (). The uninformed trader learns from the rule of price formation. Knowledge, beliefs are normally distributed with:
  and (Eq. 07)
For the uninformed trader, knowledge of the coefficients   and is therefore essential to calculate and determine their expectations about the payoff V. Demand for the risky asset According to the CARA-normal model, the demand for the risky asset of each of the two traders are easily determined:
 
  (Eq. 08)
Substituting the values ​​of and (Eq. 06) in DI (Eq. 08) and and (Eq. 07) in (Eq. 08) gives:
 
  (Eq. 09)
Price Equilibrium Since excess demand must be zero at equilibrium:
  (Eq. 10)
Substituting the request of the insider trader and that of DI uninformed OF determined by their expression in the previous point (Eq. 09) and remembering that it is possible to obtain an expression for the price of the risky asset :
  (Eq. 11)
However, to achieve a balance of rational expectations and meet the second condition of rational expectations equilibrium, the price formation rule proposed (Eq. 05) must be the rule actually used (Eq. 11) to determine the price of the risky asset. However, to reiterate, the rule of price formation is proposed:
  (Eq. 05)
The values ​​of the parameters  and  matching rule effective pricing to that proposed are:
  and (Eq. 12)
In addition, these values ​​ensure that the excess demand is zero at equilibrium. Both conditions lead to a balance of rational expectations are met. The rule proposed pricing is therefore a rational expectations equilibrium. Because all information used by the uninformed trader consists of price observation and the rule of price formation, Bias, Hillion and Foucault (1997) emphasize that it is not possible to determine separately demand functions and equilibrium price. Obviously, the observation of prices is not part of all of the information used by the insider as it directly observing the signal S. A reading of the Annex to Chapter 4 in Lyons (2001) allows a better understanding of the mathematical details related to the simplified version developed above. For a complete and detailed mathematical description of the basic model the reader should consult the article by Grossman and Stiglitz (1980). Their most complete specification of the model includes a number of informed traders and uninformed traders more.  The proportion of informed traders is endogenous in the model. It will depend in particular to pay the cost c to acquire private information, the accuracy of the signal, the degree of risk aversion and volatility of the aggregate supply of risky assets. It is interesting to note that in reality, Grossman and Stiglitz (1980) show that for an equilibrium in which the proportion of informed traders is not zero, it is necessary that the equilibrium price is not very revealing. This is the case when there is private information that influences prices. In the version of the rational expectations model of Grossman and Stiglitz (1980), this is the signal S. Walrasian Auctioneer The presence of a Walrasian auctioneer is implicit in the model because no agent is responsible for training price. Although a rule of price formation is known, it is necessary that it be applied. In this model, its application must go through an auctioneer who collects implicit orders early in the period and sets the price according to the rule of price formation in order to balance the market. Critique of the Model The main positive point of the model of rational expectations is that it develops one of the pillars of the microstructure approach, namely information. This model allows the questioning of traditional macroeconomic models because besides the fact that prices are formed on the basis of rational expectations, the private information in the form of observing a signal about the future payoff is an central role in regulation of price formation. However, the rational expectations model suffers from certain imperfections which imply that it must be improved to better reflect the reality of the foreign exchange market. O'Hara (1995) states that the presence of a Walrasian auctioneer necessitated to enforce the rule of price formation is a problem with the microstructure literature. Indeed, the microstructure approach, the mechanism leading to the equilibrium price strongly influences the price dynamics and behavior of traders. The use of the Walrasian auctioneer will ignore this complication. In addition, the auctioneer's implicit intervention means that the rational expectations model is used to characterize a market while fixing the foreign exchange market is a continuous market. Lyons (2001) identifies four critical to this model. First, when the number of active and signal increases, he said that the existence of a rational expectations equilibrium is fragile. In such a situation, it is common for general models do not result in a balance or lead to an equilibrium does not satisfy the two conditions of rational expectations. In most cases, the use of a specific example of environment that satisfies the two conditions of rational expectations eliminates the problem of existence of equilibrium. Therefore Grossman and Stiglitz (1980) were first applied to a particular rule of price formation and then checked it was a good balance. The problem with this method is that it spreads easily. In addition, O'Hara (1995) states that "every change in the environment most likely change the balance, if it still exists." Second, the competitive behavior of agents is a highly restrictive assumption. Indeed, normally, the agents have an impact on prices, which is the case of agents knowledgeable, take into account the impact their decision. However, in this model, the agent informed takes market prices as given when it acts directly on it. It affects prices, but still acts competitively. In fact, the assumption of competitive informed agents is acceptable only when the number of agents is infinite. In the model discussed above, the hypothesis of strategic behavior is more realistic. Third, the knowledge of the rule of price formation is problematic. It is not clear how the uninformed trader becomes aware of this rule. However, it is used to determine their expectations and, ultimately, its demand for risky assets. Perhaps in the long run, by experience, the trader can learn to achieve this rule but it is not clear. Fourth, the order flow plays no role in the model of rational expectations. It is not clear what the active part in transactions. Thus, this model does not test the relevance of order flow in determining the price of the risky asset. In light of that disadvantage, the model of rational expectations is not strictly part of the microstructure approach, but nevertheless allows to introduce it. Finally, since implicit Walrasian auctioneer collects the orders early in the period and sets the price according to the rule of price formation in order to balance the market, the rational expectations model to a market characterized rather governed by the orders because the orders are not conditional price. Markets governed by prices, dealers quote their issue before receiving orders. MODEL KYLE Kyle's model is one of the first models showing the combined effects of asymmetric information and market structure on prices of financial assets through the explanation of the process of price formation. In this model, the pricing rule is not applied by an auctioneer implicitly but directly by some of the agents of the model. On this point, Kyle's model no longer resembles the reality that the model of rational expectations discussed earlier. The model at a time that Kyle described in Article Continuous Auctions and Insider Trading is developed below. Kyle (1985) also built a specification with several sequential periods and one in continuous time. Note however that the latter two models are essentially based on the model studied below. In addition, the intuition of this simple model in a period is identical to that of more complex models and the purpose of this memorandum is to provide the theoretical reasoning on the microstructure of the foreign exchange market and not to analyze in detail Kyle. It is therefore essential to understand the functioning of the model at a time. Therefore this model will be discussed further. Assumptions Over a single risky asset is traded against a risk-free asset. An end-period payoff equal to V is provided by the risky asset. V is distributed under a normal pV average and standard deviation  V (). All transactions take place at the same equilibrium price. Three types of agents are present in this model: a trader informed, uninformed traders and market makers. A trader informed insider, has exclusive access to private information on the ex post liquidation value of the risky asset (and therefore the payoff that will provide V). So if is large, the private information of the insider is more important because this information can be free from uncertainty about the payoff V. Order flow resulting from the transactions that pass is noted DI. Note also that the insider is risk neutral. Uninformed traders, the noisemakers, are the second type of agents. They conduct transactions at random or for reasons of coverage. Order flow resulting from their transactions noted OF. Since the reasons for the transaction noisemakers are not based on V, and V are independent. It is important to note that OF is not observed by the insider. It knows only the distribution of DU. DU is distributed under a normal distribution with zero mean and standard deviation  U (). The third type of agents is formed by the market makers, risk neutral, which set prices efficiently (meaning semi-strong efficiency) according to the information they have on the amounts exchanged by others. Description of the Model The purpose of this model is to understand the information contained in prices and to assess the transition in the prices of private information held by the insider. The innovative aspect of this model is that it explicitly addresses the issue of price formation. As a reminder, an equilibrium model where a single sequence is played auction is used as the basis for development. Kyle's model at a time is played in two stages. First, the insider trader and the uninformed choose the quantities they will exchange, determining in this way the order flow. Second, market makers set the price of their rule of price formation and exchange quantities balancing the market. Two Steps First Step During the first stage, the initiate and the uninformed traders simultaneously choose the quantities they will exchange. With this simultaneity, the insider may be camouflaged. Indeed, market makers are unable to distinguish the orders placed at random from those from the insider. The latter has the opportunity to make a profit at the expense of uninformed traders. By hiding, he will try to minimize the impact of orders on prices, because his command will cause a price change that will be unfavorable. The insider thus faces a dilemma between quantity effect and price effect. For example, if the insider has private information about the fact that the payoff will be higher than expected, then it will tend to place orders to purchase the assets, which, all things being equal, will effectively increase the price of the asset and therefore reduce the benefit of the insider. In fact, the insider has a monopoly because it is informational only have information on the implementation of the payoff V at end of period. It has a strategic behavior to maximize its profit conditional on the final value of the asset:
  (Eq. 13)
where the benefit of the insider is measured by its net income multiplied by orders placed:
  (Eq. 14)
Second Step Then the second stage, the market makers set a price and exchange the amounts that balance the market. Their decision is based solely on observations of order flow past and present. They not parse the fundamentals. This means that price changes always come from innovations in order flow. Although the price is set by the market makers, the model of Kyle treats instead of a market order-driven. Indeed, they are not conditional on quotes posted by market makers. Instead, orders are subject to market makers before they have issued quotes. Indirectly, so it's an insider who sets the equilibrium price using its private information. However, the initiate must take the rule of price formation of market makers as given. It can influence prices only through the quantity traded. In addition, market makers realize, on average, zero profit. This results from the fact that market makers compete and there is free entry. The equation of price formation is given by:
  (Eq. 15)
Prices depend only on the sum of the order flow of the insider and uninformed traders as market makers observe that flows together. They do not have the ability to distinguish the source of the stream. Note that under the rule of price formation, market makers are risk neutral because their net position is not involved in the rule. It is important to note that this model induces a link between prices and the protocol for the exchange since the rule of price formation and market makers are governed by their joint observation of order flow, the latter from one hand the insider and the other uninformed traders. Results Kyle (1985) states that there exists a unique equilibrium in which the function of the insider strategy (X) and the rule of price formation (P) are linear functions. Defining and   respectively, and the balance of P and X is given by:
 , (Eq. 16)
 where X characterizes the strategy of the insider by measuring the intensity with which the trader makes transactions on the basis of their private information. Indeed, the strategy initiated determines the order flow of the latter based on its expected profit. If  is small, the insider is less aggressive in carrying out their transactions to avoid the impact on prices of the latter. If high, the insider has better camouflage which will require it will be more aggressive. , meanwhile, characterized the rule of price formation by measuring the depth of the market that measures the market's ability to absorb large quantities without having a major impact on prices, that is to say the flow of orders necessary to vary the price of one dollar.  A small market is a deep, since by definition, if a market is liquid, it takes a lot of order flow to vary the price of one dollar. When is high, market makers adjust prices more aggressively as the private information of the insider is more likely to be substantial. The inverse relationship between   and explained by the fact that when  is high, for example, transactions have a strong impact on prices. This implies that the insider trading less aggressively to avoid the impact of its trading price. This attitude is characterized by a small . The opposite result is determined for a  high. It should also be noted that the uniqueness of equilibrium means that all market makers should give the same rating. This is achieved because the market makers compete, there is free entry and thus their profit is zero. In terms of mathematical description, it is useful to refer to Kyle (1985) and Lyons (2001). In addition, a version of the resolution mathematized model is developed in Annex Kyle order to clarify certain technical aspects not explained either in Kyle (1985) or in Lyons (2001). This initiative was taken to facilitate the reader's understanding of this model both intuitive and technical. Informational efficiency The informational efficiency of prices () is measured by the variance of V conditional on P:
  (Eq. 17)
This expression shows that only half of the private information held by the insider is incorporated into prices. This is in line with the proposal stating that for there to be informed agents, it is necessary that the equilibrium price does not fully reveal the information. In fact, it is because the insider was acting strategically as informational efficiency is reduced. In doing so competitive, the lack of constraints on the size of transactions, such as transaction costs, combined with the risk neutral traders result in immediate and complete incorporation of any new information. The market makers offset the bad trades by making the market less liquid. Indeed, the presence of an insider is costly for market makers because the insider is better informed about the market value of the payoff V. So for a given level of transactions from the uninformed traders, plus the insider has private information and its strategy is aggressive, less market makers make the market liquid. It is interesting that the insider anticipates this behavior and thus reduces the aggressiveness of its strategy to minimize the impact on prices of transactions it performs. Critique of the Model Kyle's model actually uses the microstructure approach because it includes an informational dimension and a structural dimension acting jointly in the process of pricing. In this way, even if the rule of price formation must be consistent with expectations as in the rational expectations model, the model of Kyle involves explicit auctioneers. This changes the nature of the rule of price formation as the act of setting prices is actually assigned to an agent of the model. In addition, the introduction of market makers can understand an informational dimension absent in the model of rational expectations that waited, prices are set by market makers on the basis of their observations. In reality, they are composed of the order flow therefore play a central role in the model of Kyle as in all models microstructures. Finally, the order flow are themselves determined following the protocol of the exchange. However, some assumptions of the model of Kyle are too restrictive or do not match the reality of the foreign exchange market. To further improve the model explaining the dynamics of exchange rates, it is therefore essential to target the defects of the model of Kyle. Indeed, after performing this step, it will be possible to find another model more specific to the foreign exchange market retaining the advantages of the Kyle and eliminating its shortcomings. First, the assumption that market makers derive all their information from the observation of order flow is too strong. This assumption stems from the fact that the information they learn by observing the order flow is not public. However, it is certain that these agents are attentive to information about fundamentals. In addition to observing the order flow, they also analyze the fundamentals. The information comes to them then a part of order flow (private information) and other fundamental analysis (information publicly released). Second, in the Kyle model, market makers are risk neutral, that is to say that they perform no transaction on the basis of their net position. However, market makers do not like situations where they face extreme positions as this increases their risk. It would be more realistic to make an assumption of risk aversion. Furthermore, neglecting risk aversion of market makers, Kyle (1985) only lists the first type of information asymmetry, ie that the payoff, but neglects the second type, ie the price. Indeed, the second type of private information on the prices at which transactions take place (P0 and P1). This includes all the variables for risk premiums. However, one dealer has a deeper knowledge of the risk premiums that the market as a whole. Indeed, the data on order flow is not public, the market makers observe at least their own order flow, can better appreciate the risk. Kyle (1985) states that the asymmetry of information comes from an insider, but on the foreign exchange market information asymmetry comes precisely from the observation of order flow. It is true that to observe more order flow can be inferred more information including the fundamental and the payoff V which is consistent with the model of Kyle because private information can be inferred indirectly through order flow. The fact remains that in this model, the asymmetry of information resulting from the fact that the mass of order flow can be observed between different market makers is not taken into account. Third, the model does not generate Kyle spread. This follows from the assumption that all transactions take place all at the same equilibrium price. However, prices in the foreign exchange market is characterized by a spread between the purchase price and the selling price. However, it is possible to determine an implied spread is defined as the marginal impact of a transaction unit price. In the model of Kyle, this impact is measured by . And since the impact on prices is   for purchase and for a sale, the implied spread is 2 . It should be noted that in the presence of an insider, the market makers make the market less liquid, which implies an increase of . So the market is less liquid and more important is the implied spread, which is consistent with the fact that the presence of an insider expensive will encourage market makers to increase their spread. Fourth, with a model like that of Kyle (1985), it is impossible to study the impact of an individual transaction prices. Indeed, it is necessary to perform an aggregation of pre-orders because the orders are aggregated () that affect price formation and not the size of an individual transaction. Derivations and Extensions Sequential multi-period model From the model explained above, Kyle (1985) removes the assumption of a single trading period. It yields a sequential equilibrium model of bidding. The major difference is that the initiate must optimize its strategy and taking account of future auctions and therefore the impact that its transactions on current prices but will also have on future prices. On the results, Kyle (1985) was also able to determine a unique linear equilibrium. Interpretations and   remain unchanged. As before, the market makers reduce liquidity in the market due to the presence of costly trading. The placing of a sequential auction model has an impact on the incorporation of information into prices. In fact, private information is incorporated into prices only gradually. The insider acting strategically in an attempt to minimize its impact on current prices in order to still benefit from his informational advantage during the following periods. Continuous Model To reduce the time between auctions, Kyle (1985) goes to a continuous model. It shows that the results are essentially identical. Balance is also linear and unique, the information is gradually incorporated into prices at a constant rate. However, Kyle (1985) found that in the continuous model, market depth  is constant over time which is not the case in the sequential model. It also shows that the sequential model converges to a continuous model when the interval between transactions becomes uniformly small. N and M Initiates Strategic Information and Strategies Not Speculators Kyle (1989) propose a model in which the insider is not a monopoly but where N strategic traders are informed and monopolistic competition. Another feature of this model is that the M uninformed speculators also act strategically. In addition to speculators, a third type of agents involved in transactions or random pattern coverage without taking into account their effect on prices, it is the noisemakers. The main findings are that prices reveal less information than in a competitive equilibrium. Moreover, given imperfect competition among informed traders, even at the limit where the uninformed speculators disappear, prices do not incorporate all the information. The fourth chapter of bias, Foucault and Hillion (1997) discusses in detail the question of the properties of equilibria obtained in the presence of informed agents acting competitively or strategically.
  MODEL simultaneous transactions To be closer to the reality of the foreign exchange market, it is fundamental that theoretical specification incorporates a number of features. Thus, it is necessary to include the presence of dealers setting their prices before observing the first orders, the foreign exchange market is governed by the price. Another element that both models mentioned above do not take into account is that about two-thirds of trading in the forex market are done between intermediaries. Therefore, a model with several dealers to be developed. In addition, the risk aversion of dealers marked by the desire not to run the risk of excessive position must also be introduced into the theoretical analysis. Indeed, dealers actively manage their stock to reduce their exposure. The removal of the assumption of risk neutrality allows to understand the interaction between risk management and disclosure of dealer information. Therefore, the final section of this second section on the theoretical study of the microstructure approach to deal with a model inserting in its specification, the three characteristics discussed above. This section is more detailed than previous ones because this new model uses a theoretical model for concurrent transactions that is currently closest to the reality of the foreign exchange market. Thus, this model as a basis for the empirical analysis presented in the next section. It is therefore important to understand all the intuition behind this theoretical modeling. This type of model involves the use of game theory and more specifically, to dynamic models of simultaneous games. The first effect of this remedy is that the dealer can not determine their transactions on the basis of those introduced by other dealers when their transactions are simultaneous. The second implication, also from the simultaneous inter dealer transactions, is the introduction of shocks to the stock of dealers. Indeed, transactions of dealers is not conditional on those of other dealers, it follows that the position of the dealer can change unpredictably. Thus, the phenomenon of hot potato as described in the first part will be captured by the model of simultaneous transactions. This is particularly interesting because other models did not allow microstructures to explain this particular feature of foreign exchange market, like the model of Kyle on the assumption that market makers are risk neutral and rational expectations model indicating that transactions are conditioned by the prices that balance the market. The two main references used in this paper to describe the modeling concurrent transactions with several dealers consist of section A Simultaneous Trade Model of the Foreign Exchange Hot Potato of Lyons (1997) and Section 4 of the fourth chapter of Lyons (2001). These two references show this model in two different specifications. To avoid confusion in the analysis, specification of Lyons (2001) will be studied in detail in this section. In fact, the article by Lyons (1997) to better understand the origin of the model and explain more fully the technical aspects. Assumptions The model for simultaneous transactions are two types of agents carrying out transactions. On the one hand, there are N dealers are risk averse. Their number is limited, dealers will act between them strategically. Each dealer has a customer base of the same size. On the other hand, the large number of customers is represented by a continuum of customers. They are characterized by competitive behavior. In other words, they engage in transactions primarily for reasons of coverage, cash, and so on. This model involves two periods. This is actually a dynamic simultaneous game in two periods that incorporates a risk-free asset and a risky asset, representing the currency. The final payoff of the risky asset, denoted V, is distributed under a normal distribution with mean zero and standard deviation  V (). The returns are made at the end of the second period and the return of the gross risk-free asset is normalized to unity, which means that interest rates are zero daily. Both types of agents have the same negative exponential utility function defined by the nominal wealth held at the end of the second period:
  (Eq. 18)
 is the coefficient of absolute risk aversion and represents the wealth of dealer i (where) at the end of period 2. Description of the Model Two Periods First Period The temporal dynamics model of simultaneous transactions is summarized in Figure 3. During the first period, each dealer receives a private signal, denoted Si, and a signal common to all dealers, noted S. At this stage, it is important to note that unlike the model of Kyle, the dealer may be fundamental analysts as they observe all the signal S that models publicly available information, ie public macroeconomic announcements. It is important to keep in mind that if is any private information beyond the observation of order flow. If possible, for example, represent the knowledge of the identity of customers. In addition, Carpenter and Wang (2003) show that dealers consider that a transaction initiated by a central bank has more information content. These two signals are distributed following a normal round the final payoff of the risky asset V. Based on these signals, each dealer i will issue a quotation for the first period, denoted by Pi1. At this point, all the information available to the dealer i, denoted by Pi1 , is:
  Note that the specification of  Pi1 is different from that presented in Lyons (2001). Indeed, Lyons (2001) defines  Pi1 as follows:
  However, after correspondence with Dr. Lyons, he confirmed that this was indeed a slight error in the implementation of the 1997 article in the modeling presented in his book. However, as will be seen later in the analysis, this change does not affect the trading strategies of balance. Figure 3: Temporal dynamics of the model to simultaneous transactions Source: Lyons (2001), op. cit., p. 97 Then each dealer receives orders from its own customer base that aggregates to Ci Ci is distributed under a normal distribution with mean zero and standard deviation  C (). Where Ci is positive, the dealer i observes a pressure to buy while when this is negative, the dealer i found that its customers will show a total sales. Of course, the variable is not observable by the N-1 other dealers. The variables S, Si and C are independently distributed. There are four rules for quotes Pi1 and Pi2. First, each quote is a single price at which the dealer agrees to sell or buy any quantity. Indeed, to reiterate, the foreign exchange market is different from other markets governed by the price including the fact that the dealer can not fix the amount. The first rule is very important because it implies that the model does not consider the price range. In fact, Lyons (2001) states that it would be possible to include this price in the quotation offered by the dealers, but that it would substantially increase the technical difficulty of the model, mainly in the inter dealer quotes. Then the ratings are, by definition, observable by all dealers. This assumption means that the search listings is free. However, although computerization has facilitated access to this information, the search listings remain expensive in reality. This assumption is too strong. Another rule governing the quotes is that they are simultaneous and independent. Here again the assumption is slightly stronger than the reality. However, electronic systems, replacing more and more voice systems in the trading rooms, can indeed make transactions and quotes simultaneously. The final rule is that a dealer must be the counterpart of each transaction to the proposed listing. This prevents a dealer goes out of play when it is in an informational disadvantage. This last rule is consistent with the practices of the currency market because if a dealer refuses to issue quotations, other dealers will consider it breaks the conventions of market and tarnish its reputation. Finally, the dealer has broken the convention market will be penalized because it will reduce its number of transactions. The dealers, obliged to serve the orders that have been submitted by customers, undergo unwanted inventory changes. To share their risk, dealers conducting transactions with each other. This step is represented by Ti1 reflecting order flow inter dealer net placed by the dealer i in the first period. reflects, in turn, the inter dealer order flow received by the dealer net i in the first period. For the sake of consistency, Ti1 will be positive when the dealer is a buyer i and negative when it is at the origin of the sale. A positive, which corresponds to a pressure to buy from other dealers, the dealer says i must be the seller. Two rules characterize inter dealer transactions. On the one hand, they are simultaneous and independent. This rule implies that a source of unwanted variability of its stock in the first period to be postponed until the next period. On the other hand, it is possible to conduct transactions with multiple partners. At this stage appears the phenomenon of hot potato from the risk aversion of dealers. These transactions to cover transactions with customer destabilized the net position of the dealer, will reduce the information revealed by prices. Indeed, on one hand, the phenomenon of hot potato reduces the information contained in the inter dealer transactions. On the other hand, it is these inter dealer transactions that are important in the formation of prices in the next period because the customer order flow is not observable by other dealers. As a result, the risk aversion of dealers, coupled with the possession of private information reduces the information revealed in prices. Calling Di1 net position desired dealer i in period 1, the inter dealer order flow net placed by the dealer i in the first period is given by:
  (Eq. 19)
The specification of Ti1 (Eq. 19) shows that purchases (sales) customers (C) must be redeemed (sold) to other dealers to restore the desired net position in period 1 (DI1). Orders placed by the dealer i in the first period (Ti1) must also consider the expectations of the dealer as to the orders he will receive N-1 other dealers. Indeed, international transactions are concurrent and independent dealers the dealer i can not observe the actual implementation of. Expectations regarding the dealer i are noted.  Ti1 represents all the information used by the dealer i in transactions in the first period. This set is detailed as follows:
  To ensure greater clarity, keep in mind that transactions inter dealer in period t are made at the listing of the same period. At the end of the first period, dealers observe the order flow global inter dealer (X):
  (Eq. 20)
X, considered as a signal of inter dealer order flow, as a net purchase of risky asset (where X is positive) and a net sale of risky asset (where X is negative). Lyons (2001) states that in reality, X consists of information relating to the sign of the net order flow submitted by the inter dealer brokers. However, all dealers receive this statistic, which is why the model to simultaneous transactions is assumed that this information is observed by all dealers. It should be noted, however, that customers do not have access to this information. Williams (2000) confirms that the dealers have several sources of information that customers do not have access. In fact, this model makes two implicit assumptions that apply to the transparency of dealers. First, the transparency of transactions between customer and dealer is zero because only the dealer conducting the transaction can actually remove the information. On the other hand, the specification of X does not include noise (Eq. 20). The difference in transparency between dealers and those international transactions between customer and dealer is maximum. However, in reality, the difference in transparency between the two types of transaction is lower because the effective transparency inter dealer is not complete. At the end of the first period, as the dealer and customer transactions are not observable, they are not reflected in the prices (quotes from dealer) until they have been reflected in the inter-dealer transactions as , only the latter can be observed by all dealers. Therefore what appears in the second period is critical. Indeed, in the model to simultaneous transactions, the source of private information, is clearly different from that specified in the two models discussed above because, in this model the dealers do not derive their information from a better understanding of the payoff at the end period. Personal information considered for each dealer is rather the result of the observed order flow from customers and net present position, ie its inventory of risky assets. This is also consistent with the reality of the foreign exchange market in which to hold private information on the implementation of end-period payoff is almost impossible because in reality, this would hold privileged information about future interest rates . Second Period Early in the second period, each emits a dealer quote for the risky asset. Pi2 and denotes the rating for the dealer the risky asset i in period 2. A very important point to note is that all  Pi2 information used by the dealer for determining its listing i Pi2 contains the aggregation of order flow to its customers (C) and observing the order flow global inter dealer (X) which acts as a proxy of total customer order flow:
  Thereafter, the dealer transacts i N-1 with other dealers. Inter dealer transactions in the second period Pi2 are the price determined in previous step. Ti2 denotes the net order flow inter dealer placed by the dealer i in the second half and can be presented as follows:
  (Eq. 21)
The inter-dealer order flow received by the dealer net in the second half i was noted. Di2 means the net position desired by the dealer i in period 2. The introduction of terms Di1 (Eq. 19) and Di2 (Eq. 21) modeling the dealer i's requests on the grounds of speculation, can understand the conflicting roles of two drug dealers in a market with multiple dealers. On the one hand, they act as intermediaries between the information contained in the flow of client orders and the subsequent market price. On the other hand, they are speculators who adopt a rational strategic behavior, which distorts the information that they spend in their dealings inter dealer and thus reduces the informational efficiency of prices.  Ti2 represents all information available to the dealer when i determines Ti2:
  The position achieved in the first period must be reversed. The latter consists of three elements, namely Di1 and (Eq. 21). The term represents the unanticipated shock on the stock of risky assets from dealer i. At the end of the second period, the model ends with the realization of the payoff of the risky asset (V). Intuition about the objectives of Dealers As in most microeconomic models, the agent determines its equilibrium strategy by maximizing its utility function. It will be the same in the model to simultaneous transactions. The dealer i maximizes its utility function to determine the equilibrium strategies of international transactions and dealer quotes on the risky asset. This has a negative exponential form and is defined by Wi2, the nominal wealth at the end of the second period (Eq. 18):
  (Eq. 22)
Such that
 
 i mean all the information available at each decision node. is a listing received by dealer i in period t. Wealth at the end of second period (Wi2) is the sum of six different words. The first is the wealth of dealer i at the beginning of the game (wi0) is a constant. The second term represents the profit (or loss) realized by the dealer i order flow on its own customers. For example, a Ci negative means that, overall, customers sell to the dealer the risky asset i. It buys the risky asset at the price he offered, ie Pi1, and sells the same amount to other dealers at the price they offered, ie. If the dealer sells more than it buys, its wealth increases. In other words, if Ci <0 and Pi1 < then
  The third term represents the capital gain from the dealer i in the first period on speculation (DI1) and demand for coverage (). Indeed, the dealer i includes its expectations in transactions that place among the other dealers to reduce the importance of the impact on its stock prompted by the realization of. Since the dealer i is the applicant, the amount of risky assets will be bought (sold) in the first period, the price offered by other dealers (). Note that these requests can be positive (purchase of risky assets) or negative (sell). The positions taken in the first period will be reversed in the next period the price (). Thus, when the dealer is a buyer i and the price received in the second period is higher than that of the first period, the dealer reap a profit on its claims on the grounds of speculation and hedging because it has purchased in the first period, of the risky asset at a price below which it will be sold in the second period. The same intuition can be transposed to set the fourth term representing therefore the capital gain in the second period. The fifth and sixth terms can reflect unwanted transactions from other dealers (and). These two terms measure and the disruption brought by the characteristic of simultaneous transactions. As a result, the dealer can not meet perfectly, not observable in the first period. and will be purchased (sold) prices issued by the dealer i (Pi1 and Pi2, respectively) and reversed the prices quoted by other dealers (and V). Indeed, the dealer who i request to reverse his position. At the end of the second period, the game ends and the price is then given by V. It should be noted that empirical studies of Cheung and Wong (2000) and Cheung and Chinn (2001) explain that the dealer i want to reverse his positron in the next period. In fact, at the end of the day, most dealers are trying to return to a positron net close to zero to reduce their risk during the closure. The intuition behind this behavior is given by Cheung and Chinn (2001) who state that "the dealers attach a higher level of risk in the open position for long periods." Results The results of the model to simultaneous transactions are obtained by solving a Bayesian Nash equilibrium. Lyons (1997) defines a perfect Bayesian Nash equilibrium as a profile belief - a strategy that, for any period, produces a result and satisfied, for all i, the following two conditions: (1) Bayes law is used to update beliefs (2) strategies for listing and trading are sequentially rational given the beliefs. A pair (P, T) strategies of quotation and transaction is sequentially rational if for any pair () alternative, the following condition holds:
   i (Eq. 23)
Ui (.) Denotes the utility function of player i. Profiles and trading strategies for trading are respectively defined as follows:
 
  While alternative strategies () are given by:
 
  The intuition of the above condition (Eq. 23) is quite simple. A pair (P, T) strategies of quotation and transaction is said to be sequentially rational if the expected utility of player i, conditional on all information available at each decision node is greater than the hope of usefulness of the same player if he plays another strategy. In addition, this condition must be verified by all the players present. Strategies Training Price Balance In this model, the ratings of balance must be identical between dealers to prevent the occurrence of arbitrage opportunity. Indeed, all dealers have access to other quotes and they are valid regardless of the amount. This feature of the model to simultaneous transactions, coupled with the fact that a quotation is a single price at which the dealer agrees to sell or buy any quantity, implies that it is impossible to directly communicate private information through quotations issued. In addition, since each dealer offers a listing of balance equal to that of N-1 other dealers, it must be established on the basis of publicly available information, this means that: P11 = ... ... = = = Pi1 PN1 = P1 All the information available for each dealer in the first decision node, one for P1, consists of the private signal if and public signal S. But at this point, the only information common to all the dealers are, by hypothesis, S. As a result, for the first time, the training strategy price equilibrium is a function
linear signal S:
  (Eq. 24)
The constant is a coefficient of extraction of the signal that produces an unbiased estimator of the value of V conditional on S. The equation determining P 1 (Eq. 24) verifies the uniqueness of the listing for all dealers. During the formation of P1, the fact that all the available information does not contain does not affect the result because even if the dealers were able to condition their trading on the orders they receive individual customers, issue a listing at a price that is not conditional on private information is a Nash equilibrium. As a result, state that  is given by Pi1 instead of not really matter but the end result, however, allows to better match the model to the reality of the foreign exchange market. For the same reasons as before (no-arbitrage condition, no spread, and so on), the strategy of scoring in the second period must also be common to all dealers: Pi2 = P12 = ... = ... = PN2 = P2 The difference arises because in this case, the rating depends on two information common to all dealers. Indeed, the common signal S and the order flow global inter dealer X is data that all dealers observe. In addition, X is actually part of all of the information used in the determination of P2. The relationship expressing P2 according to S and X is given by:
  (Eq. 25)
Since X is the sum of inter dealer order flow (Eq. 20), it contains information that was not yet public. X is used as a signal flow of client orders and also incorporates information at the private signal Si, however, X also depends on the risk aversion of dealers decisive in the development of their application on the grounds grounds for speculation and hedging. Moreover, as explained above, this risk aversion is the source of the phenomenon of hot potato reducing the information in prices. Thus, X does not capture all the information contained in all the Ci and Si Another argument supporting this assertion comes from Madhavan (1995) which states that the dealers may have an interest not to integrate their private information in their scoring. Any private information not included in the price then serves as a basis for determining the demand for speculative reason in the second half (Di2). Strategies for inter Dealers Transactions Balance Given the trading strategies described in the preceding paragraph, the optimal strategies for inter dealer transactions corresponding to a symmetric linear equilibrium is defined by:
 
  (Eq. 26)

 , Ti1 and Ti2 establish a perfect Bayesian Nash equilibrium. It is important to keep in mind that the listing rules P1 and P2 are linear, respectively, and. The inter dealer trading strategies thus have a structure derived from the corresponding linear form of the utility function and the assumption of normality that generate a linear demand function. Expressions of Ti1 and Ti2 are mutually consistent and sequentially rational. The terms of the coefficients,,,  11,  21,  31,  41,  12,  22,  32,  42,   52 and 62 are detailed in Table 1 of the article by Lyons (1997 , p. 286). The details of mathematical proofs can be found in Appendix A of this article. Mathematical derivations will be complete and clearly explained, their reproduction in this memory is not critical because, as mentioned above, the emphasis was deliberately placed on the intuition behind the theoretical modeling. At equilibrium, the model produces a phenomenon of hot potato that provides a rational explanation for the observed volume of the foreign exchange market. Because of the simultaneity of trade, the dealers have, ex ante, any knowledge relating to the net position of other dealers. Transactions are therefore in no time, conditional transactions other dealers. Therefore, a perfectly efficient risk sharing among dealers is impossible. The model thus allows concurrent transactions to capture a feature of the foreign exchange market that the model of rational expectations does not encompass. In fact, in equilibrium, prices do not adjust imbalances conditional on individual stocks but risky assets are determined based on the imbalance of the aggregate stock of risky assets (Eq. 25). The model for simultaneous transactions proposed by Lyons (1997) and also developed by Lyons (2001) is the first theoretical model of the specificity of the foreign exchange market. Critique of the Model The model is simultaneous transactions, currently, the model closest to the reality of the foreign exchange market. In fact, it incorporates many distinctive features to this market. N dealers are risk averse and act strategically. In addition, the theoretical specification characterizes a market governed by the price. All the information is not public. The dealers are in fact a private signal about the final payoff of the risky asset and have the ability to observe the flow of client orders made with their own customers. Moreover, in equilibrium, a rational phenomenon occurs hot potato. Therefore, this model will be chosen to serve as a support to empirical models that attempt to explain the dynamics of exchange rates. However, by introducing a structure to multiple dealers, theoretical analysis becomes more complex. To keep an explanation as clear as possible so it was necessary to exclude certain specific characteristics of the foreign exchange market. Thus, the model still suffers simultaneous transactions in several imperfections. Lyons (2001) identifies three critical related to this model. The first relates to the fact that it does not portray the spread expected that each quotation is a single price at which the dealer agrees to sell or buy any quantity. Well as adding a fee for the initial transaction with customers is fairly straightforward, it is much more technically complex to include a spread in inter-dealer transactions as arbitrage opportunities eliminate, the greater the price dispersion. The second criticism points out that the model does not transmit signals via prices. The dealers can not collect private information held by other than through the observation of inter-dealer order flow. As explained above, private information will be reflected in prices only when it is first translated into the inter-dealer order flow. This review is also deepened by Dominguez (2003a) indicates that Lyons (2001) has tended to focus almost exclusively on the informational role of order flow. All informational signal on price is impossible because of the arbitrage opportunities that would result in the absence of spread. But in reality, price monitoring can, too, to infer private information held by other dealers. The lack of modeling of transactions executed via inter dealer brokers is the third criticism of the model. The choice of an exchange mechanism especially in a market with several mechanisms may have an impact on the information incorporated into prices. Bjones and Rime (2000) show that transactions processed on a bilateral (inter dealer) have an informative effect when the transactions electronically via brokers have no impact on the level of the information contained in prices. However, by observing the cumulative flows, exchange some sequences may be informative for prices. This is consistent with the intuition that X, the order flow global inter dealer is in the formation of prices in the second period because it consists of information relating to the sign of the inter net order flow submitted by dealers brokers. Bjones and Rime (2000) also indicate that the introduction of electronic brokers has changed the behavior of dealers who use these new systems to control their net position. One final note should be advanced. Indeed, although the model for concurrent transactions meet a lot of characteristics of the foreign exchange market, keep in mind the above limitation specifying that, in this paper, the concept of foreign exchange spot market is reduced to about major currencies. But "The market structures are being established to meet (often private) operators. This suggests that there is no unique structural form that is optimal for each market and each operator in a given market. " This implies that if the model correctly describes simultaneous transactions in the currency markets the main, it will not necessarily true in markets for exotic currencies. Finally, even if the model is simultaneous transactions currently closest to the foreign exchange market, some features of this market are not modeled in the theoretical specification. This is one aspect of the theory microstructure analyzing the dynamics of exchange rates that should be developed in future research.
  3. Microstructure: An Empirical Study The purpose of this section on the microstructure approach empirically is twofold. First, a validation of the theory microstructure devoted to foreign exchange will be offered. Then, empirical studies can also complement the theoretical analysis on certain aspects which are currently not modeled consistently. Therefore, it is interesting to dwell on these studies. The main results are the empirical validation of the model to simultaneous transactions and highlighting the importance of the identity of the participants about the impact that the order flow they have on prices. It is also question of the endogeneity of order flow and the causal link between them in price. The first section is devoted to the empirical validation of the model to simultaneous transactions discussed above. In this context, the Evans-Lyons model is used to demonstrate the explanatory power of order flow in the formation of prices in the foreign exchange market. The second section presents the work of Carpenter and Wang (2003) using an empirical model breaking down the order flow by source. The advantage of this model is to prove the identity of the counterparties is a significant informational. MODEL EVANS-LYONS The first section focuses on empirical validation of the model to simultaneous transactions using the specification of Evans and Lyons (2002a) presented in their paper Order Flow and Exchange Rate Dynamics. The intention not to overlook relevant variables to explain the dynamics of exchange rates, Evans and Lyons (2002a) adds another dimension to the macroeconomic model. Figure 4 illustrates graphically the process of integration of information into prices for the different approaches developed in the theory of the foreign exchange market. In the macroeconomic approach, all information is public. The ads are directly reflected in prices. The traditional approach microstructure is characterized by the introduction of private information essential in determining prices. This information is aggregated in the order flow that becomes, in this way, a signal picked up by the market maker stating that prices must be adjusted. This perspective is developed in the model of Kyle. The hybrid view reconciles the two above approaches, stating that the dealers get their information from order flow and public announcements. This representation depicts the foreign exchange market more realistically. Figure 4: Process integration of information into prices for the different approaches developed Source: Lyons (2001), op. cit., p. 175 The specification of Evans and Lyons (2002a), using the two types of information in order to understand the dynamics of exchange rates, thus characterized a hybrid model directly inspired by the concurrent transactions where the introduction of economic fundamentals aims to make the model more realistic compared to the foreign exchange market. The development of the Evans-Lyons model presented in this paper comes from Evans and Lyons (2002a) and is supported by the writings of Lyons (1997), Evans and Lyons (1999) and by the work of Lyons (2001). Theoretical Foundations of the Model The model presented in Evans and Lyons (2002a) is a variation model portfolio (portfolio shifts model) referring largely to model concurrent transactions proposed by Lyons (1997). At the beginning of each trading period, the public demands are made uncertain currency, producing orders that are not publicly observable. The information they contain to be calculated and incorporated into prices through the transaction process. It makes sense that these applications have an impact on prices because, according to the law of supply and demand, the market demands a price movement to absorb them. The aim is to understand the mechanism by which the public demands will affect the price and validate it empirically. The Evans-Lyons model has T trading periods, a risk-free asset with a gross return per unit and a risky asset modeling the currency. The public demands that currency are not correlated with future interest rate differentials. These include, for example, applications for reasons of liquidity or hedging. The sum of T 1 Payoffs on the risky asset is denoted R, where R is itself composed of a series of increases noted:
  (Eq. 27)
Increments are identically and independently distributed according to a normal with zero mean and variance () and are publicly observed at the beginning of each trading period. The market which is traded on the risky asset (foreign exchange) market is governed by the dealers price consists of N indexed by i. In addition, the large number of clients with respect to N dealers, is represented by a continuum of customers indexed by. All agents are risk averse and have the same utility function
  negative exponential defined by the wealth of a T:
  (Eq. 28)
 is the coefficient of absolute risk aversion and represents the wealth of the agent in T 1. Each trading period is a three laps as the temporal dynamics shown in Figure 5. In the first round, simultaneously and independently, each dealer makes a quotation. This rating, determined on the basis of observation and other available information, is a single price at which the dealer agrees to buy or sell any amount to its customers. denotes the rating issued by the dealer i the first round of the exchange period t. It is important to note that the characteristic governed by the market price is met, each dealer is looking at the order flow from customers () after a proposed listing. Again, a positive (negative) indicates that, overall, customers sell (buy) from the dealer the risky asset i (the price). The N customer order flow is distributed independently of each other and independently, according to a normal with zero mean and standard deviation  C (). is observable only by the dealer i. Figure 5: Temporal dynamics of the model of Evans-Lyons Source: Evans and Lyons (2002a), op. cit., p. 173 In the second round, the dealer set out, simultaneously and independently, scoring only one available for the N-1 other dealers. Then, simultaneously and independently, dealers conducting transactions at prices offered by other dealers. Net transactions initiated by the inter dealer dealer in the second round are noted Tit. As defined by the order flow shown in the first part, a positive Tit shows a net purchase of dealer i. At the end of the second round, the agents observe the flow
 order of the inter dealer trading period, noted:
  (Eq. 29)

  is the sum of order flow inter individual dealers (Eq. 29). If X as the sum of the inter dealer order flow across all T periods in the Evans-Lyons model, only the flux per period are taken into account. Indeed, although the order flow of customers ultimately affect prices, they have no impact until they have been previously incorporated into the inter dealer order flow. This is also explained in the model to simultaneous transactions (Eq. 25). In the third round, the dealer may share their overnight (overnight risk) with customers. The risk is defined as the overnight risk of holding an open position between the closing of the session and reopened. As part of the modeling of Evans-Lyons, this reflects the risk of retaining an open position between the end of the third round of the trading period t-1 and the beginning of the first round of the exchange period t. Empirical studies of Cheung and Wong (2000) and Cheung and Chinn (2001) confirm that late in the day the dealers are trying to return to a zero net positron to reduce their risk. First, each dealer has a listing available to customers. Assuming that the total public demand, ie the customer is not completely price inelastic, it is possible to write as a linear function of expected returns:
  (Eq. 30)
 is a positive coefficient that captures the overall ability of customers to bear the risk. The information provided publicly is represented by 3 , including and. The equilibrium relationship between the inter dealer order flow and price alignment comes from the results obtained in the analysis of the model to simultaneous transactions. The key to remember is that prices are determined on the basis of information common to all the dealers because they set their trading simultaneously and independently. In the modeling of Evans and Lyons, are made of this information during training and in conjunction with the determination. To recap, the inter dealer order flow () contains information to be included in the price () because in reality it is a signal of the overall variation of the portfolio of customers in the first round ( ):
  (Eq. 31)
If dealers wish that, in the third round, customers reabsorb this variation portfolio, prices must adjust. In particular, dealers align their prices as the net change in overall customer portfolio is zero at the end of each trading period t:
  (Eq. 32)
The price change ( Pt) between the end of the trading period t-1 and the end of the trading period t is thus given by:
  (Eq. 33)
 and  are positive constants.   and depends on and ensures that the net change in overall customer portfolio is zero (Eq. 32). Indeed, at the end of each period, the price includes all past increases and the amount of overall changes in customer portfolio prior periods whose size is provided by the inter dealer order flow past. The term represents the term modeling the change in portfolio. Depending on the assumptions, the definition of the temporal dynamics of the model and determining the rule of price formation that results, modeling Evans-Lyons takes a kind of simultaneous transactions. The remainder of this section is devoted to the empirical validation of the model proposed above. To empirically test the function of price change (Eq. 33), it is necessary to make two changes. First, the increase is replaced by
variation in the nominal interest differential:
  (Eq. 34)
it is the nominal interest rate in the dollar and the nominal interest rate in foreign currency (here in YEN or DEM). Lyons (2001) states that the change in differential nominal interest rate is clearly incomplete as a measure of movements in fundamentals. However, the interest rate is, to date, the only macroeconomic variable calculated at daily frequencies. In addition, interest rates are a crucial variable in the dynamics of exchange rates. Indeed, in the traditional macroeconomic models, an increase in interest rates of the currency implies, all things being equal, an inflow of capital as deposits held in that currency are more profitable. So the demand for domestic currency increases and it is appreciated. In other words, its price increases. A final argument for the use of the variation in the differential nominal interest rate is that price shocks are caused by unanticipated shocks in the differential. Second, the dependent variable is replaced by the change in the logarithm of the spot exchange rate ( pt). In fact, as stated D'Souza (2002),  pt represents the return of the exchange rate. This change is intended to make the empirical specification of Evans and Lyons (2002a) comparable to conventional macroeconomic models. The empirical specification of the Evans-Lyons model is thus: