Identifying speculators in the FX market: A microstructure approach

Accepted Manuscript Title: Identifying speculators in the FX market: A microstructure approach Author: Ben Z. Schreiber PII: DOI: Reference:

S0148-6195(14)00013-7 http://dx.doi.org/doi:10.1016/j.jeconbus.2014.02.001 JEB 5673

To appear in:

Journal of Economics and Business

Received date: Revised date: Accepted date:

19-5-2012 27-1-2014 9-2-2014

Please cite this article as: Ben Z. Schreiber Identifying speculators in the FX market: A microstructure approach (2014), http://dx.doi.org/10.1016/j.jeconbus.2014.02.001 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Identifying speculators in the FX market:

January 2014

Ac

ce pt

ed

M

an

(Second revision)

us

Ben Z. Schreiber*

cr

ip t

A microstructure approach

* Bank of Israel and Bar-Ilan University Office: (972)2-6552595, Mobile: (972) 503-810031, [email protected]. Keywords: Foreign Exchange Markets, Microstructure, Speculation JEL code: F31, F41, G15. I thank Gad Nathan (who passed away last year, God bless his soul), Zvi Wiener, Tzahi Frankovits, and the participants of the seminars: ISI Dublin, 2011, Research Dept. Bank of Israel, 2011, and the Hebrew University, Jerusalem 2011. All remaining errors are mine.

Page 1 of 39

ip t

*Manuscript, excluding Author Details

an

January 2014

us

A microstructure approach

cr

Identifying speculators in the FX market:

M

Abstract

This paper suggests a methodology for identifying speculators in FX (foreign exchange) markets. A player is identified as a speculator only if his speculative

ed

characteristics are extreme compared with those of other players and his influence on exchange rates on outlying days is significant. Implementing the proposed

ce pt

methodology on Israel's FX market, which includes 366 large players, identified 58 potential speculators—almost all of them nonresident entities, local banks, and financial companies. Examining their activity based on a unique dataset for 2008-09 revealed speculators that purchased foreign currency before and/or on outlying depreciation days and sold foreign currency before and/or on outlying appreciation

Ac

days. Thus, some speculators joined or initiated the trend before the outlying appreciation or depreciation days. Based on these speculators found during 2008–09, it was possible to identify similar behavior before and on outlying days during 2010, which was defined as an out-of-sample period. The proposed methodology may help market makers and regulators track speculators before and on outlying days.

Page 2 of 39

1. Introduction Although speculators have a major influence on FX markets, there is no consensus in the literature or among practitioners as to the definition of a speculator or of a method to identify

ip t

these important market players. There is a wide range of definitions for the term speculation, ranging from broad definitions that essentially include all day-to-day investment activity to

cr

those with only a very narrow scope. Moreover, the term speculation (or speculator) in a FX market is even more problematic than in other markets due to the sophistication of the players,

us

the cross-border nature of FX activity (e.g., ―carry trade‖1), and the relative scarcity of

an

information regarding trading and players. Hence, although the FX market is a very large market, Menkhoff et al. (2013) call it a 'dark market'. It is argued that speculators have a greater

M

influence on the FX markets in emerging markets (see Chang et al. 2009), but even in those economies, it is quite difficult to identify and track them in the absence of an agreed-upon

ed

definition of a speculator. Melvin and Taylor (2009) describe several crises caused by "carry trade" players that substantially affected the FX markets. Moreover, Burnside et al. (2007) find

ce pt

that the returns to currency speculation in emerging markets are positive and much higher than in developed countries. Thus, it is important to be able to identify and track speculators especially in emerging markets. This is particularly the case on outlying days, which are

Ac

defined in this paper as days with sharp movements (to be defined later) in the Israeli Shekel/Dollar exchange rate (hereafter ER). This paper is related to Osler (2008), Menkhoff and Schmeling (2010), Moore and Payne (2011), and Menkhoff et al. (2013). All of these studies found different motives, sophistication, trading pattern and style, and price impact, by various player groups in the FX market. There are, however, several differences between this paper and those studies, as follows: (1) This paper examines the activity of key players

1

Taking a loan at a low interest rate in one market and simultaneously depositing the funds at a higher interest rate in another market.

2 Page 3 of 39

regardless of their sectoral affiliation, e.g., whether they are financial companies or households; (2) the paper focuses on outlying days (to be defined later); and, (3) the unique data set consists of a detailed data on all trades in the Israeli FX OTC market—thus, the relationships between the order flows and the ER ought to be more significant compared to a sample of disaggregated

ip t

order flow as in those studies or a sample based on aggregated data, as in many other studies. The paper suggests an empirical methodology to identify speculators using the following two

cr

non-exclusive conditions:

us

(1) Whether a player's Net Buying Pressure (hereafter NBP, defined as buying foreign currency minus selling foreign currency) has a significant impact on the ER. This relationship

an

is determined using the three-stage-least-square (3SLS) regressions.

(2) Whether a player‘s trading activity has substantial speculative characteristics (to be defined

M

herein). A player is identified as a potential speculator in this paper by fulfilling both conditions, i.e., having extreme speculative trading characteristics relative to those of other

ed

players and having a significant influence on the ER on outlying days. The main finding is that the behavior of the financial sector (domestic banks, financial

ce pt

companies and foreign institutions) differed significantly from that of the commercial sector (exporters, importers, and institutional investors), particularly on outlying days. This evidence is consistent with the current evidence that the majority of players can be distinguished by their

Ac

core business, i.e., financial versus commercial players. However, not all players in the financial sectors were found to be speculators, and not all players in the commercial sectors were non-speculators.2 The implementation of the proposed methodology on the Israeli FX market identified 58 large players (out of 366) whose NBP significantly influenced the ER and

2

Although not all speculation is due to speculators (some speculative activity is carried out by, for example, exporters' or importers' CFOs), the paper focuses on speculators who actively trade in the FX market for financial profit.

3 Page 4 of 39

had trading characteristics that were extreme compared to those of the other large players.3 Thus, these players fulfilled conditions (1) and (2) above during the sample period (2008-09). The list of speculators is not random as the probability of selecting the very same players according to both independent conditions was found to be very small.

ip t

The suggested methodology also makes it possible to distinguish between players who trade in the same direction as the ER (appreciation/depreciation) and those who trade in the opposite

cr

direction. As I shall see later, such a categorization may be of importance to various

us

participants in the FX market, like regulators and market makers. Of the 58 players that were found to satisfy the potential 'speculator' definition, four potential speculator types were

an

identified:

(a) 12 of the players usually traded with the direction of the ER, i.e., they sold (bought)

M

foreign currency on outlying appreciation (depreciation) days (hereinafter, Pro).4 These players are supposed to be the most successful investors. Menkhoff et al. (2013) argue

ed

that such a behavior is typical of trend chasers or momentum traders. (b) 16 of the players usually traded in the opposite direction of the ER, i.e., they sold

ce pt

(bought) foreign currency on outlying depreciation (appreciation) days (hereinafter, Con). These players behaved like contrarians during those days. Menkhoff et al. (2013) argue that such a behavior is typical of households.

Ac

(c) 12 usually sold foreign currency on average regardless of appreciation or depreciation outlying days (hereinafter Sell). One can expect exporters and institutional investors whose assets are exposed to exchange rate appreciation to follow such an investment pattern. Notice that Sell players are different from Pro and Con players since they do not 3

The methodology can be implemented on any player, particularly a small one. However, for practical purposes the paper focuses on large potential speculators who have the ability to affect the exchange rate. Although small players can statistically cause the exchange rate e.g., as revealed by Granger's causality tests they do not actually affect it. Therefore, large sophisticated players are better candidates to be speculators than small naïve ones. 4 An appreciation (depreciation) day is a day when the Israeli shekel is substantially strengthened (weakened) relative to the USD.

4 Page 5 of 39

change their behavior before or on appreciation/depreciation days, i.e., they usually sell foreign currency on both types of outlying days. In spite this difference they behave as potential speculators; namely, they fulfill the above conditions of potential speculators. In addition, exporters and institutional investors usually hedge their position by selling

ip t

foreign currency against their FX position in advance, particularly on or after outlying days. Thus, most exporters and institutional investors who behave as hedgers are

cr

supposed to react to outlying appreciations/depreciations but not initiate them as

us

speculators do.

(d) 18 usually purchased foreign currency, on average, regardless of outlying appreciation or

an

depreciation days (hereinafter, Buy). As a mirror image of exporters, importers are basically exposed to exchange rate depreciations.

M

Tracking the above four types of potential speculators during an out-of-sample period (2010) revealed, in most cases, similar behavioral patterns to those in the in-sample period.

ed

There are some implications for regulators and large dealers (market makers). The former are responsible for the well-functioning of financial markets and particularly, the FX market. Thus,

ce pt

tracking the behavior of potential speculators can help regulators to limit the impact of speculators on outlying days, either by taking administrative means or differential taxation, or any kind of intervention. On the other hand, market makers that track speculators can benefit

Ac

from the private information (see Menkhoff et al., 2013; Osler, 2010) regarding their NBP and join them before outlying days - days that are responsible for a large part of the annual profit and loss (hereafter P&L) of any market player. The rest of the paper is organized as follows. Section 2 surveys the literature. Section 3 presents basic statistics on the activity of the main groups of players in the Israeli FX market on outlying days and on non-outlying (normal) days. Section 4 presents the regression results and examines the behavior and the P&L of speculators during both in-sample and out-of-sample periods.

5 Page 6 of 39

Section 5 discusses policy implications of the proposed method and presents robustness checks for the determination of outlying days, while Section 6 concludes.

Literature review

ip t

Few papers deal with the identification of speculators in the FX market, as there is no unique definition for speculation and the data which can be used to track their activity is uncommon.

cr

According to Graham and Dodd (1951), speculation is any risky investment. Osler (2006)

us

characterized a foreign currency speculator as a player who focuses on changes in exchange rates, in contrast to a commercial player, such as an exporter or importer, whose activity is

an

based on transaction needs, or an institutional investor who is interested in the level of exchange rates while investing abroad. Accordingly, a speculator will prefer shorter periods of

M

investment, will be characterized by greater leverage and will change his position from long to short and vice versa with a high frequency. His goal is to profit from short-run changes; thus,

ed

generally speaking, financial traders (particularly speculators) initiate short-term movements in exchange rates while commercial traders react to such movements. A substantial difference

ce pt

between Graham and Dodd's definition of speculators and Osler's definition is that the former definition focuses on the activity while the latter focuses on the player. This distinction raises the question of whether regulators should track or even restrict speculative activity (focusing on the activity) or speculators (focusing on the player). Speculators provide liquidity to the

Ac

2.

markets thus lowering bid-ask spreads, and may help the market to discover the efficient price by getting more information into prices and trading against pricing errors. In contrast, manipulative strategies and predatory trading could reduce price efficiency especially via High Frequency Trading (HTF). Moreover, speculators sometimes intensify trends, and thus destabilize the markets, e.g., by following momentum trading strategies. In an exhaustive survey of the foreign currency markets, Osler (2008) noted that speculators in a typical foreign

6 Page 7 of 39

currency market account for up to 80–90% of the total activity, and commercial customers rarely speculate, though the inclusive definition she uses for a speculator is only one of many alternatives. Recently, Menkhoff et al. (2013) disaggregated the NBP (positive order flows minus negative order flows) of a mega bank's customers into four different groups, and

ip t

assessed each group‘s impact on the future exchange rate. They found that the NBP of asset managers—particularly institutional investors and hedge funds—positively influences the

cr

future exchange rate, and that the NBP of corporations has no impact on the future exchange

us

rate, whereas, private customers‘ NBP is negatively related to the future exchange rate. Additionally, they found that institutional investors are trend-chasers, and that the impact of the

an

NBP of hedge funds on the future exchange rate is short-lived (transitory), while private investors behave as contrarians. Consequently, Menkhoff et al. argue that disaggregated NBPs

M

carry substantial economic values for the dealer observing those flows (see also Osler, 2008). Chang et al. (2009) found that foreign investors have a significant influence on the prices of

ed

shares traded on the Taiwan Stock Exchange, even though they do not account for a large proportion of trading. Furthermore, in contrast to the activity of other investors, it was possible

ce pt

to use the trades of foreign residents to produce fairly accurate one-week-ahead predictions of stock prices. These findings are consistent with the ―clientele effect‖, which claims that for every asset in the financial markets, there are various types of investors who are differentiated

Ac

from one another by: preferences for risk-return trade-off, trading style, sophistication, and the terms of investments. Such clientele effect was found in the Israeli FX OTC option market by Galai and Schreiber (2013). In this regard, this study categorizes the players included in the sample into seven sectors: exporters, importers, institutional investors, households, financial companies, local banks, and nonresidents. The last three sectors are considered financials, while the first four are non-financials. However, as the definition of a speculator is ambiguous and since the dataset contains disaggregated data we can analyze the short-term behavior of the

7 Page 8 of 39

players, on an individualized basis, regardless of sectoral affiliation. Regarding the horizon of the analysis, the commonly held belief (Osler 2008) is that speculators influence foreign currency market variables in the short run, while in the long run, prices are determined by longterm fundamentals. This is confirmed by Menkhoff et al. (2013) who found that the private

ip t

information contained in daily flows is impounded very quickly into exchange rates (up to three

us

2.1 Identifying speculative activity by trading characteristics

cr

days).

To determine which variables should be used to define speculative characteristics, a small-scale

an

verbal survey was carried out among professionals from ten domestic and foreign banks that are active in the Israeli FX market. According to the survey results and the literature (mainly

M

Osler, 2008; Menkhoff and Schmeling, 2010; Moore and Payne, 2011; Menkhoff et al., 2013), speculative activity was characterized according to the following parameters:

ed

(1) A relatively short-term investment horizon;

(2) A massive use of leverage (NBP over the FX position);

ce pt

(3) Rapid shifts from a long to a short position regardless of fundamentals; (4) The number and size of trades tend to increase in a volatile market; and (5) The players either influence or are influenced by market variables. The latter characteristic

Ac

is taken into consideration separately from the other four characteristics at a later point.

3. Descriptive statistics of the speculative characteristics by sector The dataset used in this study is composed of two independent sources of data: the exchange trades system for players‘ trading characteristics and the exchange rate system for ER. The proprietary database of the trades system is unique, as every single FX trade (including Spot, FRA, and Options mostly denominated in US dollars) conducted through domestic banks must

8 Page 9 of 39

be reported to the Bank of Israel. This type of data coverage is usually an obstacle for most FX datasets and market empirical studies where the information regarding a particular player is limited, on one hand while the relationships between aggregated order flows and the ER are biased for small samples, on the other hand. The analysis herein, therefore, focuses on players

ip t

who act as speculators and who affect the ER.5 To determine which sectors influenced the exchange rate, or were influenced by it, (as NBP

cr

and the exchange rate are simultaneously determined), this section focuses on sectors and

us

financial instruments that are relevant for the Shekel/Dollar exchange rate (ER) changes. Therefore, all FX swap transactions (approximately 135,000 trades), which do not directly

an

affect the exchange rate6, were omitted, as well as exotic options, which account for a low proportion of trade (less than 1%). After omitting those observations, the remaining

M

approximately 481,000 records were grouped by both player and trading day.7 In contrast with international definitions (e.g., the SNA sectoral classifications), I

ed

differentiated between exporters and importers8 as they influence the exchange rate (and are influenced by it) in opposite directions. In addition, interbank trades, which have different

ce pt

characteristics (see Osler, 2008), were omitted, as well as trades of ―other‖ institutions9 due to their low proportion of trading.10 The database includes the following sectors: exporters,

5

Ac

There is also an offshore FX market; however, the volume of that market during the sample period was quite small compared to the local FX OTC market. Experts estimate that market to be less than 10% of the domestic FX market. 6 FX Swap transactions contain two legs: spot and forward (FRA) in the opposite direction, i.e., either long spot and short FRA or short spot and long FRA. In both cases there is no direct impact of FX swaps on the ER as the two legs cancel each other out. 7 Of approximately 2,500 players in the market, the 366 large players selected accounted for approximately 90% of the total volume during the sample period. 8 An exporter/importer is defined as an entity whose exports/imports account for more than 10% of their sales (exports + imports). A small number of manufacturers (with respect to the value of their trades) were defined as ―others‖ and were therefore omitted from the data. 9 These trades are, in general, derived from the surplus demand and supply of each bank with respect to its customers (the rest of the sectors included in the database). Additionally, one bank whose reports to the Bank of Israel were not complete during the sample period was omitted from the data. 10 These players are usually small ones and do not substantially affect the ER. Yet, the correlation coefficient among their NBPs was found to be approximately zero. The analysis of the interbank market whose influence on the ER is presumably not negligible, is different from the market of non-bank players (Osler, 2008) thus, it is beyond the scope of this paper.

9 Page 10 of 39

importers, financial companies, institutional investors, nonresidents, local banks, and households. Financial companies, and in almost all cases nonresidents, contain hedge funds, mutual fund companies, and portfolio managers, while according to the classification of the Bank of Israel, institutional investors include provident and compensation funds, pension funds,

ip t

advanced study funds and insurance companies. The database also includes the following financial instruments: spots, forwards, call options and put options (plain vanilla).

cr

According to the literature, trading is more intensive on volatile days. Therefore, outlying

us

days were defined according to the daily changes in the Israeli Shekel/Dollar (ER). Using Hampel's method for outlier identification (see Appendix), 24 outlying days were found during

an

the in-sample period (2008-09) and 15 more during the out-of-sample period (2010). The total number of outlying days represents 5.4% of the total (720) trading days during the period; 3.1%

M

depreciation outlying days and 2.3% appreciation outlying days. Based on both the literature

identify potential speculators: 1.

ed

and the small survey mentioned above, the following trading characteristics were selected to

Number of trades on outlying days in excess of normal days (NUM): A large number of

ce pt

trades may be an indicator of speculative activity (day traders for example). Furthermore, a large number/volume of trades on days with higher volatility (such as outlying days) is additional indirect evidence of speculative activity. Thus, NUM was defined as:

Ac

Mean(#trades on outlying days) - Mean(#trades on normal days), and it may reflect the unexpected number of trades. 2.

Number of days until expiration (DTE - weighted average of days for spot, forward and

option trades): It is conjectured (see, for example, Osler, 2008) that speculators prefer shortterm financial instruments to reduce Bid-Ask Spreads (BAS) costs. 3.

Heterogeneity (HET): calculated as Mean[(buy - sell)/(buy + sell)] = NBP/VOL, where

VOL reflects the volume of a player's trades. As 1≥HET≥-1, values close to 0 indicate a

10 Page 11 of 39

potential speculator. For example, an importer (exporter) usually buys (sells) foreign currency to hedge his exposure to a depreciation (an appreciation) of the ER, thus HET will be close to +1 (-1). In contrast, speculators sometimes buy foreign currency and some time sell thus, their HET will be closer to 0. Number of changes in the direction of the position during the sample period (CHG):

ip t

4.

Frequent changes in the direction of the net position (hereinafter, POS), i.e., from a long

5.

us

that is not for the purposes of hedging or long-term investment.

cr

position in foreign currency to a short position or vice versa, may be an indicator of activity

Leverage (LEV): relative to the overall position (in absolute terms) reflects the degree of

an

leverage in financial activity (Mean[Abs(NBP/POS)]. Speculators are usually characterized by relatively high leverages.

M

The percentiles of the above five trading characteristics were calculated for each player, and a simple mean of these percentiles was calculated. The players were then sorted from high to low

ed

according to these scores (except for DTE and HET which were sorted from low to high). Thus, a higher score for player j {j = 1..366} indicates more speculative characteristics. Table 1

ce pt

presents the above parameters for all players and days, broken down by sector. [Enter Table 1 here]

The differences between the sectors are quite noticeable. For example, 45% of the 366 large

Ac

active players were non-residents, while another 40% were financial companies, exporters, and importers. The large number of foreign investors is consistent with Chang et al. (2009) and Bjonnes et al. (2009), who found that foreign investors have a significant influence on local markets. The differences in trading characteristics between sectors were even more pronounced. For example, local banks and financial companies were active on more days relative to their proportion of total players (13% versus 7% and 17% versus 14%, respectively) than were export and import firms (10% versus 14% and 8% versus 12%, respectively).

11 Page 12 of 39

Additionally, the average size of a transaction and the position (panels a and b) of the financial sectors, i.e., local banks, financial companies, and foreign residents (mainly international mega banks), differ from those of the commercial sectors, i.e., exporters, importers, and institutional investors. In this regard, the negative position of exporters and institutional investors (hedging

ip t

future FX income) and the positive position of importers (hedging future FX expenditure) are much larger than the positions of the financial sectors (in absolute terms). The differences

cr

between the financial and commercial sectors can also be seen in DTE (panel c), HET (panel

us

d), CHG (panel e), and LEV (panel f). The financial sectors can be characterized as follows: They tend to trade in short-term instruments (DTE < 10 days), their heterogeneity is close to

an

zero, they frequently change positions (CHG > 0.06), and their relative net flow is greater than that of other sectors (LEV > 0.74). The evidence in Table 1 shows how trading characteristics

M

differ among the sectors, but it does not reveal to what extent the players' behavior changes on outlying days, if at all. Based on both Osler (2008), Menkhoff and Schmeling (2010), and the

ed

small-scale oral survey, speculators tend to increase their activity around outlying days and

ce pt

sometimes initiate a trend or cause an outlying day.

3.1 Speculative characteristics on outlying days and on non-outlying (normal) days This subsection describes how the players' behavior changed during outlying days, which are

Ac

characterized by sharp movements in the Israeli Shekel/Dollar exchange rate (ER). Figure 1 presents the outlying days (red dots) by Hampel's method, which is robust to jumps (see Appendix), and the ER during both in-sample and out-of-sample periods. As a robustness check, the figure also presents outlying days (black octagons) by a quantile method such that the total number of outlying days by both methods is the same (39 days). [Enter Figure 1 here]

12 Page 13 of 39

There were 39 outlying days and 681 normal days during the entire sample period of 2008–10. As can be seen, most outlying days are concentrated in clusters, particularly in 2008. In addition, there were more depreciation outlying days than appreciation outlying days (22 and 17 depreciations and appreciations, respectively). Although both methods indicated the same

ip t

number of outliers—39, only 28 outlying days were common to both methods while another 11 days were non-overlapping. In what follows outlying days are determined by the Hampel's

cr

method (except for robustness checks). Table 2 shows how players' behavior changed on

[Enter Table 2 here]

us

outlying days.

an

The table presents the means of trading characteristics on normal versus outlying days by sector. One can see a picture similar to that shown in Table 1, where the differences between

M

the financial sectors (except nonresidents) and the commercial sectors are substantial both on normal days and on outlying days. For example, differences in daily order flows (NBP) and

ed

position (POS) of the financial sectors in absolute terms on outlying days were smaller than those of the commercial sectors, as expected. Similarly, the differences between the financial

ce pt

sector and the commercial sector on outlying days regarding days to expiration (DTE), heterogeneity (HET), changes from long position to short position and vice versa (CHG), and leverage (LEV) were of the same magnitude as the differences on normal days. However, the

Ac

change rates in absolute terms between outlying days and normal days (Gap) are usually greater for the financial sector than for the commercial sector (except for CHG, and LEV). This is probably the consequence of different relationships between these two sectors and market variables as financial sectors initiate trends and speculate on currency movements while commercial sectors react (Osler, 2008). The next section examines the relationship of player j's NBP with the ER regardless of the player‘s sectoral affiliation such that, together with the

13 Page 14 of 39

above speculative characteristics, the relationships enable us to identify potential speculators in the FX market.

4. The proposed method's regression results

ip t

Speculators have a number of unique trading characteristics; however, their activity can also be correlated with ER, especially around outlying days. In other words, speculators may initiate a

cr

move toward a sharp appreciation or depreciation, but they can also be influenced by ER. To

us

examine the contemporaneous relationships between NBP and ER, a two-equation system was estimated for each of the 366 players using 3SLS regressions, as follows:

an

(1a) ERt  1  1 NBPt  1 Appt  NBPt  1Dept  NBPt  1t

(1b) NBPt   2  2 ERt  2 Appt   2 Dept  2 gap3mt 1  2 IVt 1   2t

M

where gap3m is a three month interest rate differential between the local and the US markets, IV is the implied volatility derived from one-month FX options, t is the current day; and  is the

ed

noise term. App and Dep are dummy variables for outlying appreciation and depreciation days, respectively determined by the Hampel's method. As is common, the variables in lag were

ce pt

selected as instrumental variables.11 Also, an additional AR(1) term, which is not presented in the table, was sometimes added to the regressions to take care of autocorrelation.12 In equation (1a), the ER is explained by NBP of the jth player {j = 1.366} and by extra NBPs on outlying

Ac

appreciation and depreciation days. The variables in lag: gap3mt-1 and IVt-1 control for carry trade and volatility heterogeneity effects (see Menkhoff et al., 2013) and the correlation between them and the ER was very low.

11

Several instrumental variables such as the Shekel/Euro exchange rate, the effective exchange rate (weighted average of the Israeli major trade partners), a player's position rather than NBP, etc. were examined. In all cases the results did not yield better estimates. 12 The addition of the AR(1) terms was needed in case of serial correlation. Thus, out of 71 acceptable regression results (58 potential speculators and 13 potential non-speculators), AR(1) was added only to 31 players' equations.

14 Page 15 of 39

The advantage of using a two-equation system is to avoid simultaneity. This approach overcomes the 'simultaneity bias' when an endogenous variable is treated as an exogenous variable. In particular, if a player influences the ER on outlying days, both ERt and his NBPt are simultaneously determined. However, this approach can be implemented only when there

ip t

are disaggregated trades data, such as this paper's proprietary database, regarding the particular player.

cr

Obviously, there may be four possibilities concerning the relationships between players‘ NBP

us

(positive/negative) and ER (appreciation/depreciation) on outlying days. Therefore, in this study, I define four types of potential speculators:

an

(1) Pro (with the trend) - players who usually sell foreign currency on outlying appreciation days and buy foreign currency on outlying depreciation days;

M

(2) Con (contrarian) - players who usually buy foreign currency on outlying appreciation days and sell foreign currency on outlying depreciation days;

ed

(3) Sell – players who usually sell on both outlying appreciation and depreciation days; and (4) Buy – players who usually buy on both types of outlying days.

ce pt

These new definitions, which discard the sectoral affiliation as in other studies, are necessary to identify speculators (especially Pro players) who affect the ER on or before outlying days.

Ac

Table 3 presents the regression results broken down by speculator type. [Enter Table 3 here]

The table presents the coefficient means and t-statistics of the 3SLS regressions for the four speculator types. By construction, the main exogenous variables in equation (1a) (given the constraint of a level of significance greater than 0.95), i.e., AppNBP and DepNBP, are significant. For example, on outlying appreciation days, the mean coefficient of the variable AppNBP related to Pro players was 0.136 (Prob. < 0.01), that of Con players was -0.192 (Prob. < 0.01), and similar coefficients were found for outlying depreciation days. Compared to the

15 Page 16 of 39

small insignificant coefficients of the NBPs (1 = -0.002 and 1 = -0.013 for Pro players and Con players, respectively) on non-outlying days, these coefficients confirm the substantial impact of the NBP of some speculators on ER, on outlying days. Although all the selected speculators fulfilled both conditions, i.e., they had speculative

ip t

characteristics and they influenced ER, Pro players are the ones who initiate trends or join them and are the main beneficiaries of that. Yet, the relationships between ER and NBP of all

cr

speculators on non-outlying days were insignificant; a phenomenon emphasizing the

us

importance of outlying days to market participants such as regulators and market makers. In contrast, the coefficients of all 58 speculators and non-speculators whose regression results

an

were acceptable (Adj. R2 >0, D.W. between 1.65 and 2.35, and number of trading days > 50)13 were insignificant. Osler (2006) argues that high volatility attracts speculators as they take

M

advantage of the rapid changes in the exchange rates. Therefore, one should expect more speculative activity around outlying days when ER changes are extreme (see Table 2). In

ed

general, the endogenous coefficients in equation (1b) were not significant, implying that ER did not directly influence speculators' NBP. The results indicate that various speculators behaved

ce pt

differently on outlying days and on normal days. This phenomenon is explained by the 'clientele effect' of the players participating in the FX market (see Galai and Schreiber, 2013). The question that consequently arises is, are these results random? Table 4 presents some basic

Ac

statistics of the four speculator types and tests the randomness of the various players. [Enter Table 4 here]

Panel A presents the number of players in each category. Of the 366 regressions (one for each player), only 58 were defined as potential speculators, i.e., yielded acceptable regression results and reflected speculative characteristics. Another 116 players had speculator characteristics but their regression results were not acceptable, while 186 players had neither speculator

13

Changing the conditions for acceptance did not affect the results substantially.

16 Page 17 of 39

characteristics nor acceptable regression results. Of the 58 potential speculators, 12 were Pro players, 16 were Con players (contrarians), 12 were Sell players, and 18 were Buy players. The latter two types of potential speculators, i.e., Buy and Sell players, were consistent in their (average) behavior across the sample period regardless of whether a day was an outlying

ip t

appreciation or depreciation day. These numbers are not a random result, as the probability of selecting the same players that fulfill both conditions is close to zero. This is reflected in both

cr

Fisher's and Barnard's Exact tests for all potential speculator types (see a description of these

us

tests in Mehta and Senchaudhuri, 2003).

Panels B and C show the sectoral affiliation of potential speculators and non-speculators,

an

respectively. Among Pro and Con players there were foreign residents (17.2% and 20.7%, respectively), especially international mega banks, and 3.4% local banks. Recall that these two

M

types of potential speculators changed their behavior on outlying appreciation and depreciation days, accordingly. This prominent result is consistent with the literature described above

ed

concerning the benefits of market makers from the private information they hold regarding order flows of their customers and the major rule foreign residents play in local FX markets. In

ce pt

addition, except for 2 import firms (3.4%), all other 56 potential speculators (96.6%) belong to financial sectors. This is consistent with Osler (2008) who argued that most commercial firms are hedgers rather than speculators. Yet not all firms affiliated with the financial sector are

Ac

speculators, as shown in Panel C. The share of each sector in Panel C, which consists of the remaining 308 non-speculator players, is similar to the share presented in Table 1. For example, of the 308 non-speculator players there were 42.9% foreign residents, compared to their total share (45%, in Table 1) in the entire sample. Lastly, the percentage of Pro and Con players, who changed their behavior on outlying appreciation and depreciation days, among potential speculators is larger than the respective figure among non-speculative players. This result is

17 Page 18 of 39

reasonable as active management of NBP with regard to outlying appreciation and depreciation days is expected among speculators rather than among non-speculators. 4.1

Forecast of the potential speculators' BNP and P&L

Dividing the players from the combined list into the four types of potential speculators

ip t

described above enabled us to examine the influence of each speculator type on the ER, especially on outlying days, and to forecast outlying days by tracking their behavior. The above

cr

results indicate the possibility of predicting the initiation of a trend in the FX market. This can

us

be done by tracking potential speculators, especially Pro players who substantially change their NBPs around outlying days and are the main beneficiaries of that behavior. The underlying

an

assumption is that most players will adhere to their behavioral patterns, e.g., Pro players will usually buy (sell) foreign currency on outlying depreciation (appreciation) days. To examine

M

this hypothesis, the type of player was determined based on 2008–09 while his behavior examined on outlying days during the out-of-sample period of 2010. Figure 2 presents the

ed

behavior (i.e., mean NBP) of 12 Pro, 16 Con, 12 Sell, and 18 Buy players in the vicinity of 15 outlying days during 2010, which are evenly divided between appreciation and depreciation

ce pt

days (7 appreciation and 8 depreciation days).

[Enter Figure 2 here]

As can be seen from the upper part of the figure (outlying appreciation days), Pro players

Ac

aggressively sold foreign currency on appreciation days while their behavior before and after the outlying days was ambiguous. Con and Sell players bought foreign currency on these outlying days even though the ER appreciated and Sell players used to sell foreign currency. The 58 non-speculators, who served as a control group for the 58 potential speculators, were randomly selected from a list of 308 (366 – 58) non-speculative players. This group did not change its NBP around outlying appreciation days, as expected. The lower part of Figure 2 presents the various NBPs of the four types of speculators as well as the control group around

18 Page 19 of 39

outlying depreciation days. Pro players were the only group who consistently bought foreign currency before and on outlying depreciation days. They also did it aggressively (mean of USD 2.73 million per day). Buy players bought foreign currency before the outlying depreciation days but sold foreign currency (negative NBP) on the outlying days themselves, while Con,

ip t

Sell, and Control groups exhibited an ambiguous trading pattern. The trading pattern of Pro players can be perceived as speculative behavior because their NBP

cr

was strongly related to the ER around and on outlying days. Additionally, both Pro and Buy

us

players changed their NBPs aggressively compared to other groups. Finally, the control group behavior reflects insensitivity to the ER on outlying days, as expected. The various groups'

an

trading patterns raise the interest in their respective trading P&L; this is depicted in Figure 3. [Enter Figure 3 here]

M

It can be seen that Pro players are the most profitable players around outlying days; then Buy players, and then other groups. Moreover, Pro players earned profits before and on outlying

ed

depreciation days consistently, contrary to all other groups. This result is consistent with their BNP which changed with the direction of the trend – appreciation or depreciation. Other

ce pt

players, however, could not achieve such performance though it is not easy to determine whether the P&L of the various speculator groups are significantly different from that of the control group. This and more detailed P&L's characteristics of the above groups, is presented in

Ac

Table 5.

[Enter Table 5 here]

The table shows the P&L of the four speculator groups versus the P&L of a control group on outlying days and on non-outlying (normal) days during the entire sample period. It is interesting to see major differences between the P&L of both various groups and sub periods. Consistently with their respective NBPs of Figure 3, Pro players gained the most from outlying days. For instance, the sum of profits of Pro players on 39 outlying days was USD 2.257 million

19 Page 20 of 39

while on non-outlying days (681 days) the respective figure was only USD 0.842 million. In contrast, Con players lost a sum of USD 0.987 million on outlying days compared to a total gain of USD 1.097 million on non-outlying days. Although there were 17 outlying appreciation and 22 depreciation days during the sample period, the gains of Pro players and the losses of

ip t

Con players were more substantial during outlying appreciation days. This is partially explained by the larger mean appreciation rates (in absolute terms) than depreciation rates on

cr

outlying days (-1.59% versus 1.48%, respectively). Similar to Pro players, Buy players gained

us

USD 1.470 million on outlying depreciation days but lost USD 0.632 million on non-outlying days. The P&L of Sell players is a mirror image to Buy players' P&L i.e., a gain on outlying

an

appreciation days and a loss on outlying depreciation days. These results are typical of players who stick to their trading patterns regardless of ER movements. Finally, in almost all cases, the

M

gap between each player's P&L and that of a control group found to be significant by a two-

is not randomly achieved.

ed

tailed t-test with unequal variances. This means that the P&L of the various speculator groups

This analysis is limited, however, to the relations between a particular speculator's behavior and

ce pt

the ER. To examine the future P&L of five types of speculators: Pro, Con, Buy, Sell, and All players in a grouping framework, the following fixed effect pooled regressions (with crosssection weights and White heteroskedasticity-consistent standard errors and covariance), were

Ac

estimated as follows:

(2) P & Li ,t 1  Ci  i NBPi ,t  i Appi ,t 1   i Depi ,t 1  1ERt   2 gap3mt  3TA100t   i ,t 1 where Ci is the cross-section (specific) parameter representing the particular speculator type i {i=1..5}, App and Dep are dummy variables for outlying appreciation and depreciation days, respectively, gap3m is the three-month interest rate differential between Israel and the US, TA100 is the change rate of the Tel- Aviv 100 Stock index, t is the current day, and  is the error term. Table 6 presents the results of the pooled regressions broken down by years.

20 Page 21 of 39

[Enter Table 6 here] As can be seen, in some years the common coefficients differently influenced the P&L of the various groups. For instance, ERt negatively and significantly influenced P&Lt+1in 2008 while positively and significantly in 2010. Additionally, the NBPt of both Pro and Con players

ip t

negatively influenced their respective P&Lt+1 while the NBPt of Buy players positively influenced their P&Lt+1. These relationships were found to be significant across the years. As

cr

the focus of the paper is the behavior of the potential speculators on outlying days, an important

us

piece of evidence is the influence of outlying appreciation (App) and depreciation (Dep) days on the speculators' P&L. It can be seen that in all cases there are positive relations between Pro

an

players' P&L and the dummies App and Dep and negative relations between Con players' P&L and App and Dep. This means that on both outlying appreciation and depreciation days Pro

M

(Con) gained (lost) money across the sample years, particularly during 2010 which is an out-ofsample period. Such strong relations across the years were found in the case of Buy and Sell

ed

players on outlying depreciation days but not on outlying appreciation days. Finally, the influence of both dummies App and Dep on the P&L of All Speculators was much smaller than

ce pt

the respective influences of these two dummies on the P&L of particular speculators. This evidence is explained by the fact that various speculators cancel each other out, e.g., Pro player

Ac

versus Con player.

5. Policy implications and robustness checks The different behavior of speculators during normal and outlying days is consistent with Osler (2006; 2008). However, it is in contrast with sectoral analysis using the 'from-whom-towhom' (sector) approach (Shrestha and Mink, 2011), as most of the potential speculators belong to the financial sector, while their behavior is different, such that there are Pro versus Con players or Buy versus Sell players within the same financial sector. As the categorization

21 Page 22 of 39

of players into four speculator types is feasible only where the trade database is rich enough, the above results may also have some consequences on the required reporting system. At present, international bodies such as the IMF, OECD, and BIS recommend (see Shrestha and Mink, 2011) a sectoral from-whom-to-whom approach while analyzing capital flows and

ip t

financial stability. This approach appears suboptimal in analyzing and tracking potential speculators because there are some sectors without speculators (mainly nonfinancial), while

cr

others, particularly the financial sector, consist of both speculator and non-speculator players.

us

Thus, following the current international bodies' recommendations and considering the sectoral level may not result in a robust speculative behavior. In contrast, the suggested method groups

an

speculators from all sectors into four types of speculators—Pro, Con, Sell, and Buy—based only on their NBP and their respective P&L. Therefore, it allows distinguishing between four

M

types of behavioral patterns on outlying days, which are the more sensitive days for speculative activity. Notice, however, that tracking potential speculators requires continual efforts from the

ed

authorities side as players can change their behavior, especially Con players who bear losses on outlying days so, their trading pattern is not sustainable.

ce pt

The results indicate that Pro players aggressively sell (buy) foreign currency on outlying appreciation (depreciation) days and mostly benefit from it. This evidence is unique to Pro players and holds across years, on appreciation and depreciation days and by several estimation

Ac

types.14 Based on the above results, both regulators and market makers may track speculators, especially Pro players, to identify or even forecast outlying days. Once the regulators recognize an extreme accumulation or unusual depletion of foreign currency by a Pro player, they can react either by using administrative measures or by direct intervention in the FX market.

5.1 Robustness checks

14

Buy players also gained from outlying days—but only from depreciation days, as expected.

22 Page 23 of 39

In order to assess the sensitivity of the results to the definition of an outlying day, an alternative definition - the Quantile method was examined. However, in order to control for the number of outlying days, the thresholds (percentiles) were calculated such that the total number of outlying days will be 39 exactly as the Hampel method indicated. Thus, the lower threshold

ip t

was the percentile 0.015 of the ER and the upper threshold was 0.985 using a 90-day window size (see a description of these two methods in the appendix). Table 7 presents the P&L of the

cr

various speculator types using the Quantile method rather than the Hampel method.

us

[Enter Table 7 here]

It can be seen that the figures in Table 7 are quite similar to those of Table 5 except for the

an

mean loss of Con players on outlying depreciation days that was much larger under the Quantile method than the respective figure under the Hampel method (USD 30.8 million versus

M

USD2.5 million). In addition to the selected method decreasing the number of outlying days from 39 to 30 or increasing to 50 did not change the results, qualitatively. The regressions of

ed

equations (1a) and (1b) were also checked by using GMM or 2SLS rather than 3SLS. The results in all cases did not change qualitatively thus, the results are quite robust for the

6. Summary

ce pt

definition of outlying days and for the regression techniques.

Ac

As speculative activity can cause sharp movements in exchange rates especially in small open economies, it is in the interest of regulators and market makers to identify and track a certain number of potential speculators, particularly on outlying days. This paper suggests a practical methodology for identifying potential speculators in FX markets by examining both the speculative characteristics of a player and the relationship between the ER and his or her net buying pressure (NBP). A player is identified as a potential speculator only if his speculative characteristics are extreme in comparison with those of other players and if his

23 Page 24 of 39

influence on exchange rates is significant on outlying days using the 3SLS regressions. As these two conditions are independent, considering a player as a speculator is a non-random result; a conjecture that was tested and confirmed in this paper. Implementing the proposed methodology on Israel's FX market identified 58 potential

ip t

speculators – almost all of them nonresident entities, local banks, and financial companies. Examining their activity between 2008–09 revealed four types of potential speculators: (a) 12

cr

Pro players who usually initiate a trend (sell foreign currency before or on outlying

us

appreciation days and buy foreign currency before or on outlying depreciation days), (b) 16 Con players who act as contrarians (buy foreign currency on outlying appreciation days and sell

an

foreign currency on outlying depreciation days), (c) 12 Buy players who usually buy foreign currency regardless of appreciation or depreciation days, and (d) 18 Sell players who usually

M

sell foreign currency regardless of appreciation or depreciation days. Based on these four potential speculator types that were found in 2008–09, it was possible to identify similar

ed

behavioral patterns before and on outlying days during 2010, which served as an out-of-sample period. Pro players purchased foreign currency before and on outlying depreciation days and

ce pt

sold foreign currency before and on outlying appreciation days. Moreover, Pro players and, to a lesser extent, Buy players were the greatest beneficiaries from the trading on outlying days as their NBP on outlying days was found to be relatively large. The results are robust for the

Ac

definition of outlying days and for the regression techniques. Regulators as well as market makers may track speculators, especially Pro players, to identify or even forecast outlying days as Pro players aggressively sell (buy) foreign currency on outlying appreciation (depreciation) days and mostly benefit from it.

24 Page 25 of 39

References Bjonnes, G.H, S. Holden, D. Rime, and H.O. Solheim (2009), ‗Large‘ vs. ‗small‘ players: A closer look at the dynamics of speculative attacks, CESifo Working Paper No. 2518. Burnside, C., M. Eichenbaum, and S. Rebelo (2007), The returns to currency speculation in emerging markets, American Economic Review, Vol. 97, 333-338.

ip t

Chang, C.C., P.F. Hsieh, and H.N. Lai (2009), Do informed option investors predict stock returns? Evidence from the Taiwan stock exchange, Journal of Banking and Finance 33, 757-764.

cr

Galai, D. and B.Z. Schreiber (2013), Bid-Ask Spreads and Implied Volatilities of key players in a FX options market, Journal of Futures Markets, 774-794.

us

Graham, B. and D. Dodd (1951), Security analysis: Principles and Technique, McGraw-Hill. Mehta, C.R. and P. Senchaudhuri (2003), Conditional versus unconditional Exact tests for

an

comparing two binomials, http://www.cytel.com/papers/twobinomials.pdf . Melvin, M. and M.P. Taylor (2009), The crisis in the foreign exchange market, CESifo

M

Working Paper No. 2707.

Menkhoff, L. and M. Schmeling (2010), Whose trades convey information? Evidence from a cross-section of traders, Journal of Financial Markets 13, 101–128.

ed

Menkhoff, L., L. Sarno, M. Schmeling, A. Schrimpf (2013), Information flows in dark markets: Dissecting customer currency trades, BIS Working Papers, No. 405.

ce pt

Moore, M.J. and R. Payne (2011), On the sources of private information in FX markets, Journal of Banking and Finance 35, 1250-1262. Osler, C.L. (2006), Macro lessons from microstructure, International Journal of Finance and Economics 11, 55-80.

Osler, C.L. (2008), Foreign exchange microstructure: A survey of the empirical literature,

Ac

Encyclopedia of Complexity and System Science, Springer. Shreshta, M. and R. Mink (2011), An Integrated Framework for Financial Flows and Positions on a From-Whom-to-Whom Basis, presented at the conference on strengthening sectoral position and flows data in the macroeconomic accounts, jointly organized by the IMF and OECD, February 28–March 2, Washington, DC.

25 Page 26 of 39

Appendix: Hampel versus Quantile method Hampel's method is a robust technique of identifying outliers, based on robust estimators for the location and the scale of the distribution, i.e., median and MAD (Mean Absolute Deviations) rather than the mean and the standard deviation. Hampel's thresholds for outlying

ip t

appreciation and depreciation days are: Limit  Median (dER)  z1 / 2 MAD(dER) where

MAD  1.4826  MediandER  Median (dER)  is a scale estimator, z1 / 2 is the 1-α/2

cr

percentile of the z distribution, and dER is the exchange rate changes. When the sample is

us

normally distributed, the MAD is a consistent estimator of the standard deviation and α is the type 1 error for detecting an outlier. As a result MAD is robust for jumps. The chosen moving

an

window for the MAD calculation was 90 trading days and an outlier is detected if MADt is

M

outside Limit.

ed

The Quantile method defines outliers by their extreme values vis-à-vis the last 90 trading days, as follows: POt (NOt) is a positive (negative) outlier if dERt > threshold_H (dERt<

ce pt

threshold_L) and threshold_H was set to the 98.5th percentile while threshold_L was set to the 1.5th percentile based on the last 90 trading days (likewise the moving window of the Hampel's method). This method detects the same number of outliers at both sides of the

only.

Ac

distribution regardless whether the period is tranquil or turmoil so it is used as a benchmark

26 Page 27 of 39

Table 1 Basic statistics of sectors' trading activity in the Israeli FX OTC market: 2008–09

Financial sector

Institutional Export Import investors firms firms Households Others 8 50 43 10 16 2% 14% 12% 3% 4% 1,382 6,381 5,555 1,304 2,004 2% 10% 8% 2% 3%

Total 366 100% 65,646 100%

ip t

# players % of total # active days % of total

Commercial sector

Domestic Financial Foreign banks firms residents 24 52 163 7% 14% 45% 8,711 10,904 29,405 13% 17% 45%

-0.13 285 -631 13

-1.03 133 -150 19

-2.83 125 -105 8

4.53 376 -175 13

0.07 20 -28 3

0.56 58 -100 6

0.00 376 -631 13

b) Position (POS), USD million Mean -3.11 Max 300 Min -296 Std. 43

-1.76 227 -506 26

2.57 646 -584 26

-85.81 351 -652 156

-22.60 194 -1031 48

24.90 427 -103 41

3.10 68 -18 9

2.09 153 -147 22

-1.17 646 -1031 44

9.22 369 0 22

4.84 301 0 17

5.90 859 0 21

20.95 2002 0 70

20.82 1529 0 51

24.89 584 0 48

18.77 253 0 32

18.25 278 0 31

10.17 2002 0 31

-0.05 1 -1 0.61

0.00 1 -1 0.81

-0.03 1 -1 0.86

0.07 1 -1 0.91

-0.64 1 -1 0.72

0.57 1 -1 0.74

0.00 1 -1 0.81

0.06 1 -1 0.93

-0.03 1 -1 0.84

0.12 1 0 0.32

0.08 1 0 0.27

0.07 1 0 0.26

0.02 1 0 0.14

0.01 1 0 0.08

0.01 1 0 0.12

0.06 1 0 0.23

0.03 1 0 0.17

0.06 1 0 0.23

0.76 373 0 5.50

0.75 2400 0 14.31

0.16 21 0 0.87

0.14 500 0 3.71

0.20 401 0 3.89

0.45 200 0 5.02

0.24 100 0 1.84

0.56 2400 0 9.87

e) Changing Position (CHG) Mean Max Min Std.

Ac

f) Leverage (LEV) Mean Max Min Std.

1.26 700 0 11.70

us

an

M

ed

d) Heterogeneity (HET) Mean Max Min Std.

ce pt

c) Days To Expiration (DTE) Mean Max Min Std.

cr

a) Net Buying Pressure (NBP), USD million Mean -0.52 0.13 Max 257 151 Min -281 -239 Std. 20 11

This table presents some activities of the main sectors. Net Buying Pressure (panel a) is all buying minus all selling. Position (panel b) includes all instruments (spot, TOM, same day, and futures and options in delta values) except swaps. Heterogeneity is calculated (panel d) as (buy - sell)/(buy + sell) so the maximal and minimal values are +1 and -1, respectively. Changing position (panel e) counts the number of times that a player changed his position from long to short or vice versa (change = 1, no change = 0). Leverage (panel f) is calculated as NBP/POS in absolute terms.

27 Page 28 of 39

Table 2 Sectors' trading activity: Means of 'normal' versus 'outlying' days, 2008–09

a) Net Buying Pressure (NBP), USD million Normal days -0.58 0.08 Outlying days 0.32 0.56 Gap 284 86

-1.04 -0.90 15

-2.79 -3.12 10

-1.86 -0.33 457**

2.54 3.01 16*

-86.99 -69.34 25**

-22.77 -20.33 12**

4.75 5.62 15*

5.84 6.56 11*

20.92 21.22 1

20.60 22.77 10

d) Heterogeneity (HET) Normal days -0.05 Outlying days -0.03 Gap 90

0.00 0.05 105**

-0.04 -0.03 40

0.07 0.12 43

e) Changing Position (CHG) Normal days 0.12 Outlying days 0.12 Gap 1

0.07 0.11 31**

0.07 0.09 21**

f) Leverage (LEV) Normal days Outlying days Gap

0.75 0.88 14*

0.74 0.86 14

Ac

1.21 2.08 42*

0.52 0.97 47

-0.03 0.26 111*

3.05 3.72 18*

2.06 2.50 18

-1.27 0.16 910**

24.41 29.12 16**

18.87 17.81 6

18.16 19.05 5

10.02 11.71 14**

-0.63 -0.65 3

0.57 0.55 5

-0.02 0.10 115*

0.06 0.10 43

-0.03 -0.01 148**

0.02 0.04 45*

0.01 0.01 33*

0.01 0.02 39**

0.05 0.10 43**

0.03 0.05 48**

0.05 0.07 24**

0.15 0.35 57**

0.14 0.18 25

0.20 0.28 29*

0.42 0.82 50

0.24 0.32 27*

0.55 0.70 22**

24.84 25.70 3

an

M

ce pt

c) Days To Expiration (DTE) Normal days 9.12 Outlying days 10.42 Gap 12*

ed

b) Position (POS), USD million Normal days -3.25 Outlying days -1.14 Gap 185*

4.51 4.74 5

0.01 0.59 98**

us

-0.14 -0.05 180

Total 366 100%

ip t

# players % of total

Commercial sector Institutional Export Import Household investors firms firms s Others 8 50 43 10 16 2% 14% 12% 3% 4%

cr

Financial sector Domestic Financial Foreign banks firms residents 24 52 163 7% 14% 45%

The table presents the differences between the main sector activities on normal days and on outlying days. Gap is calculated as 100*(outlying days - normal days)/outlying days in absolute terms. Outlying days are days when ILS/USD changes were exceptional using Hampel's method (see the Appendix). Net Buying Pressure (panel a) is all buying minus all selling. Position (panel b) includes all instruments (spot, TOM, same day, and futures and options in delta values) except swaps. Heterogeneity is calculated (panel d) as (buy - sell)/(buy + sell) so, the maximal and minimal values are +1 and -1, respectively. Changing position (panel e) counts the number of times that a player changed his position from long to short or vice versa (change = 1, no change = 0). Leverage (panel f) is calculated as NBP/POS in absolute terms. * - indicates that the figures on outlying days are larger than the respective figures on normal days at a significance level of 0.9 or more using Welch t-test (unequal variances), ** - indicates a significance level of 0.95 or more.

28 Page 29 of 39

Table 3 3SLS regression results of the four types of speculators, in-sample period (2008–09) Con

(buy on depreciation (sell on depreciation days & sell on days & buy on appreciation days) appreciation days) 12 16

# Players

Buy

Sell

(buy on both depreciation and appreciation days) 12

(sell on both depreciation and appreciation days) 18

-0.005 -0.18 0.268 2.51 -0.246 -3.52 -0.012 -0.26

-0.006 0.06 -0.007 -0.10 -0.117 -0.50 0.002 -0.03

0.011 0.22 0.023 0.07 0.012 0.24 0.004 -0.02

0.15 1.90

0.14 1.93

0.05 1.88

1.564 0.35 9.243 0.97 5.309 0.80 2.664 0.84 -0.577 -0.05 8.984 0.10

-0.059 -0.20 -3.658 -0.52 -2.705 -0.39 -0.123 0.15 -0.019 -0.23 -0.485 0.04

0.640 0.06 1.756 0.12 0.269 0.07 0.402 0.15 -0.120 -0.06 1.663 0.01

0.650 0.68 4.749 1.00 -1.999 -0.17 0.055 0.27 -0.051 0.03 0.998 0.00

0.09 2.03

0.03 2.04

0.04 2.03

0.05 2.07

-0.003 0.38 -0.313 -3.03 0.103 2.55 0.022 0.08

0.13 1.94

0.14 1.94

0.14 1.96

-0.078 -0.16 -9.681 -1.29 4.774 0.79 -0.442 -0.24 0.029 0.14 -0.451 -0.15

1.270 0.29 10.811 1.26 -3.545 -0.50 -0.070 -0.07 -0.004 -0.01 0.175 0.02

ce pt

2

Adj. R D.W.

ed

2 T-Stat. 2 T-Stat. 2 T-Stat. 2 T-Stat. 2 T-Stat. 2 T-Stat.

M

(2) Endogenous variable: NBPt

0.01 1.94

us

-0.013 0.27 -0.192 -2.96 -0.365 -2.11 -0.004 0.04

an

Adj. R2 D.W.

-0.002 -0.17 0.136 2.73 0.186 3.15 0.012 0.10

0.03 2.09

58

cr

(1) Endogenous variable: ERt 1 T-Stat. 1 T-Stat. 1 T-Stat. 1 T-Stat.

All Speculators Non-Speculators

ip t

Pro

13

This table presents the results (means) of the following two equation system run for each player, separately:

(1)

ERt  1  1 NBPt  1 Appt  NBPt  1Dept  NBPt  1t

Ac

(2) NBPt   2   2 ERt   2 Appt   2 Dept  2 gapt 1  2 IVt 1   2t

where ER is the exchange rate changes, NBP is the net buying pressure, App and Dep are dummy variables for outlying appreciation and depreciation days, respectively, gap is the interest rate differential (3 month) between the local and the US markets, IV is the implied volatility derived from one month FX options, and  is the noise term. Bold T-Stat. indicate a significance level of 0.95 or more.

29 Page 30 of 39

Table 4 Contingency tables and sectoral analysis of the four types of potential speculators and other players, 2008–09

# all coefficients & all players (a) # robust coefs. & speculative characteristics (b) # robust coefs. & non-speculative characteristics (c) # non robust coefs. & speculative characteristics (d) # non robust coefs. & non-speculative characteristics Fisher's Exact test (Prob.) Barnard's Exact test (Prob.)

Con (sell on depreciation days & buy on appreciation days)

Buy (buy on both depreciation and appreciation days)

Sell (sell on both depreciation and appreciation days)

All players

68 12 1 22 33

85 16 1 25 43

108 12 0 40 56

105 18 4 29 54

366 58 6 116 186

0.001 0.001

0.000 0.001

0.000 0.001

0.000 0.001

0.000 0.001

3.4% 17.2% 0.0% 0.0% 20.7%

3.4% 20.7% 3.4% 0.0% 27.6%

5.2% 5.2% 6.9% 3.4% 20.7%

0.8% 9.6% 2.9% 3.0% 1.2% 0.6% 3.6% 0.7% 22.4%

1.1% 13.4% 4.0% 4.1% 1.6% 0.8% 5.1% 1.0% 31.2%

10.3% 10.3% 10.3% 0.0% 31.0%

22.4% 53.4% 20.7% 3.4% 100%

1.0% 12.1% 3.7% 3.8% 1.5% 0.7% 4.6% 0.9% 28.2%

3.6% 42.9% 13.0% 13.3% 5.2% 2.6% 16.2% 3.2% 100.0%

ed

M

0.6% 7.8% 2.4% 2.4% 0.9% 0.5% 3.0% 0.6% 18.2%

Domestic banks Foreign residents Financial firms Import firms Others Institutional investors Export firms Households All 308 non-speculative players

an

Panel C: Sectoral analysis of the 308 non-speculative players

us

Domestic banks Foreign residents Financial firms Import firms All 58 speculative players

cr

Panel B: Sectoral analysis of the 58 speculative players

ip t

Panel A: contingency tables

Pro (buy on depreciation days & sell on appreciation days)

Panel A presents contingency tables using Fisher and Barnard tests. These tests estimate the exactness of 2X2 dependency table of the four speculator types. Small probability

ce pt

indicates a rejection of the null that the entries in the table are randomly selected. Fisher test is: where, (a), (b), (c), and (d) are the four entries (possibilities) for each player's type and n is the sum of the four entries. Barnard's test which is an iterative non-parametric test, has a higher power than Fisher's exact test, however, it is less popular due to computational requirements. It generates a distribution of Wald statistics.

Ac

Panel B depicts the share of the sectors in the speculator list (58 players) while Panel C presents the sector's share in the non-speculators list (308 players). A speculator is defined as a player whose speculative characteristics are extreme compared to others (see tables 1 and 2) and his regression coefficients for App and Dep are robust (see Table 3).

30 Page 31 of 39

Table 5 P&L (USD, '000) of the four speculator groups and a control group - Hampel method

Pro

Con

Buy

All Speculators

Sell

Control group

ER changes (%)

# Players

12

16

12

18

58

58

2946 4.1 42 417 -366

-372 -0.5 32 164 -376

2086 2.9 50 667 -260

271 0.4 38 174 -434

870 1.2 15 131 -125

1164 1.6 25 248 -198

2.5 0.0000

-2.1 0.0000

1.3 0.0000

-1.2 0.0000

2257 57.9 116 417 -323

-987 -25.3 96 163 -376

838 21.5 143 667 -260

195 5.0 108 174 -434

59.3 0.0000

-23.9 0.0000

22.9 0.0000

6.4 0.0554

11.5 0.0000

-932 -54.8 117 163 -376

-632 -37.2 113 187 -260

610 35.9 101 174 -159

25 1.5 39 107 -61

-31.7 0.0009

-14.1 0.0823

59.0 0.0000

24.6 0.0000

-55 -2.5 70 94 -218

1470 66.8 150 667 -78

-415 -18.9 109 174 -434

366 16.6 48 131 -125

28.0 0.0003

-17.8 0.0006

51.5 0.0000

-34.2 0.0000

1.3 0.7211

(e) Non outlying days (681 days) Sum 842 Mean 1.2 Std 42 Max 174 Min -565

1097 1.6 32 205 -263

718 1.1 50 363 -435

-601 -0.9 38 223 -415

524 0.8 15 66 -90

2.7 0.0000

2.2 0.0000

0.2 0.0004

1.9 0.0000

99.8 0.0000

ce pt

Gap (Speculator - Control) Prob.(Gap = 0)

Gap (Speculator - Control) Prob.(Gap = 0)

Gap (Speculator - Control) Prob.(Gap = 0)

Ac

(d) Outlying depreciation days (22 days) Sum 953 Mean 43.3 Std 126 Max 302 Min -323

2.3 0.0000

M

(c) Outlying appreciation days (17 days) Sum 1305 Mean 76.7 Std 104 Max 417 Min -13

ed

Sum Mean Std Max Min Gap (Speculator - Control) Prob.(Gap = 0)

-0.4 0.0000

391 10.0 44 131 -125

an

(b) Outlying days (39 days)

-0.02 0.72 3.23 -2.44

us

Sum Mean Std Max Min Gap (Speculator - Control) Prob.(Gap = 0)

cr

(a) All sample (720 days)

ip t

(buy on depreciation (sell on depreciation (buy on both (sell on both days & sell on days & buy on depreciation and depreciation and appreciation days) appreciation days) appreciation days) appreciation days)

-56 -1.4 67 164 -198

0.14 1.68 3.23 -2.44

-393 -23.1 63 64 -198

-1.59 0.61 -0.64 -2.44

337 15.3 66 164 -124

1.48 0.73 3.23 0.76

-742 -1.1 17 120 -131

-0.02 0.63 2.62 -2.18

This table shows the impact of the P&L on outlying days on the total profitability of various speculator groups. Outlying days were determined using the Hampel method and 90 days window size. The control group contains players who were randomly selected from a list of non-speculative players. Buy players and Pro players gained from trading on outlying days the most while contrarians (Con) lost money on these outlying days. The probability of Gap = 0 was calculated as a two-tailed T test with unequal variances.

31 Page 32 of 39

Table 6

Pooled (fixed effect) regression results of the P&L of various speculators by year

Endogenous variables: P&Lt+1 In-sample 2009

Out-of-sample 2010

Exogenous variables - Common coefficients -0.018 -0.013 -0.850*

-0.039** 0.000 0.133

0.016** 0.026** 0.033** -0.011* 0.018**

-0.001* -0.007** -0.002** -0.016** -0.001**

-0.004** -0.001** 0.005** -0.005** -0.005**

-0.002** 0.000** 0.006** -0.005** 0.005**

1.859** -2.543** -1.671** -0.006 -0.484**

1.589** -0.935** -2.214** 1.195** -0.078**

0.101** -0.500** -0.002 -0.989** -0.516**

1.304** -1.475** -1.306** 0.069** -0.372**

0.689** -0.064** 1.725** -0.178** 0.460**

0.780** -0.654** 1.467** -0.602** -0.069**

0.322** -0.149** 0.008** 0.108** 0.055**

0.560** -0.215** 0.923** -0.156** 0.173**

-0.005** -0.004** 0.011** -0.001* 0.017**

ce pt

Appt+1 - Pro players ( 1) Appt+1 - Con players ( 2) Appt+1 - Buy players ( 3) Appt+1 - Sell players ( 4) Appt+1 - All Speculators ( 5)

0.010 0.018 0.013 0.010 0.017

ed

NBP t - Pro players (1) NBP t - Con players (2) NBP t - Buy players (3) NBP t - Sell players (4) NBP t - All Speculators (5)

Adj. R2 D.W.

Ac

Dept+1 - Pro players ( 1) Dept+1 - Con players ( 2) Dept+1 - Buy players ( 3) Dept+1 - Sell players ( 4) Dept+1 - All Speculators ( 5)

-0.019* -0.009 -0.349

0.100** 0.028** 0.021** -0.004** 0.030**

an

-0.015 0.058** 0.043** -0.028 0.022

M

C - Pro players C - Con players C - Buy players C - Sell players C - All Speculators

us

Exogenous variables - Cross-section specific coefficients

0.024** -0.005 1.653**

cr

ERt ( 1) gap3mt ( 2) TA100t ( 3)

All sample period 2008-10

ip t

In-sample 2008

0.15 2.12

0.38 1.97

0.21 1.98

0.11 2.00

This table presents the results of a pooled (fixed effect) estimation:

P & Li ,t 1  Ci   i NBPi ,t   i App i ,t 1   i Dep i ,t 1   1 ERt   2 gap 3mt   3 TA100 t   i ,t 1

where C is the cross-section (specific) intercept, NBP is the net buying pressure, App and Dep are dummy variables for outlying appreciation and depreciation days, ER is exchange rate changes, gap3m is an interest rate differential between domestic and US treasury bills of 3 month, TA100 is changes in the Tel Aviv 100 stock index, and  is the error term. The index i {i = 1..5} reflects the means of the five speculator types: Pro, Con, Buy, Sell, and All speculators. * - indicates a significance level of 0.95 or more, ** - indicates a significance level of 0.99 or more.

32 Page 33 of 39

Table 7 P&L (USD, '000) of the four speculator groups and a control group - Quantile method

Pro Con (buy on depreciation (sell on depreciation days & sell on days & buy on appreciation days) appreciation days)

Sell (sell on both depreciation and appreciation days)

All Speculators Control group

16

12

18

58

2339 3.2 35 389 -248

-619 -0.9 38 331 -293

1840 2.6 46 627 -286

2106 2.9 43 291 -499

1074 1.5 16 140 -116

5.7 0.0000

1.6 0.0000

5.0 0.0000

5.3 0.0000

1842 47.2 96 389 -248

-1202 -30.8 82 145 -289

1032 26.5 126 627 -286

194 5.0 112 263 -469

345 8.8 38 140 -82

63.4 0.0000

-14.7 0.0000

42.6 0.0000

21.1 0.0000

25.0 0.0000

-178 -11.1 96 175 -286

750 46.9 107 263 -173

170 10.6 33 83 -35

-14.6 0.0634

5.1 0.4797

63.0 0.0000

26.8 0.0000

-709 -30.8 67 122 -191

1210 52.6 138 627 -74

-556 -24.2 107 117 -469

175 7.6 41 140 -82

-14.7 0.0013

68.7 0.0000

-8.1 0.1505

23.7 0.0000

842 1.2 42 174 -565

1097 1.6 32 205 -263

718 1.1 50 363 -435

-601 -0.9 38 223 -415

524 0.8 15 66 -90

2.3 0.0000

2.7 0.0000

2.2 0.0000

0.2 0.0004

1.9 0.0000

Sum Mean Std Max Min Gap (Speculator - Control) Prob.(Gap = 0)

78.9 0.0000

ce pt

Gap (Speculator - Control) Prob.(Gap = 0)

Ac

(d) Outlying depreciation days (23 days) Sum 839 Mean 36.5 Std 95 Max 273 Min -248 Gap (Speculator - Control) Prob.(Gap = 0)

-493 -30.8 102 145 -289

52.6 0.0000

M

(c) Outlying appreciation days (16 days) Sum 1003 Mean 62.7 Std 97 Max 389 Min -8

ed

Sum Mean Std Max Min Gap (Speculator - Control) Prob.(Gap = 0)

an

(b) Outlying days (39 days)

-1740 -2.4 30 175 -298

-0.02 0.72 3.23 -2.44

-614 -16.1 60 175 -157

0.28 1.60 3.23 -2.44

-259 -16.2 57 175 -79

-1.44 0.63 -0.63 -2.44

-355 -16.1 63 128 -157

1.47 0.73 3.23 0.57

-742 -1.1 17 120 -131

-0.02 0.63 2.62 -2.18

3.9 0.0000

us

(a) All sample (720 days)

58

ip t

12

ER (%)

cr

# Players

Buy (buy on both depreciation and appreciation days)

(e) Non outlying days (681 days) Sum Mean Std Max Min Gap (Speculator - Control) Prob.(Gap = 0)

Similar to Table 5, this table shows the impact of the P&L on outlying days on the total profitability of various speculator groups. Outlying days were determined in this table using the Quantile method (thresholds at percentiles 0.015 and 0.985) and 90 days window size. The control group contains players who were randomly selected from the list of non-speculative players. The probability of Gap = 0 was calculated as a two-tailed T test with unequal variances. It can be seen that the results are quite similar to those in Table 5.

33 Page 34 of 39

ip t cr us

Figure 1

Hampel vs. Quantile method: Outlying days and exchange rate (ER) changes

an

5%

In-sample period 4%

M

3%

d

2%

-2%

-3%

-4%

Ac c

-1%

ep te

1%

0%

Out-of-sample period

Hampel

ER

Quantile

The figure presents two alternative methods to determine whether a trading day is an outlying day. Hampel method is based on MAD (Median Absolute Deviation) while the Quantile method is based on a high and low threshold of 0.985 and 0.015, respectively. The number of outlying days during the entire period by both methods was 39 (5.4% of all days) and 28 out of 39 outlying days were overlapped. Both methods use 90 days window size.

Page 35 of 39

Figure 2 Net Buying Pressure (NBP, USD million) of the four speculator types and a control group around outlying days, 2010 Outlying appreciation days, Out-of-sample period 5 4 3

ip t

2 1

Pro

Con

0

Sell

cr

Buy -1

Control

us

-2 -3

-4

day -3

day -2

day -1

an

-5 Outlying Day

day +1

day +2

day +3

M

Outlying depreciation days, Out-of-sample period

ed

5

3

Pro 1

ce pt

Con Buy Sell Control

-1

-5

Ac

-3

day -3

day -2

day -1

Outlying Day

day +1

day +2

day +3

The figure presents NBPs of the four speculator types and a control group, starting three days before the outlying days and ending three days after, during 2010. The control group, which is composed of 58 randomly selected non-speculative players, exhibits a relatively stable NBP regardless of outlying days.

Page 36 of 39

Figure 3 P&L (USD, '000) of the four speculator types and a control group around outlying days, 2010 Outlying appreciation days, Out-of-sample period 50

40

Pro

ip t

30

20

Con Buy

cr

10

Sell Control

us

0

-10

day -3

day -2

an

-20 day -1

Outlying Day

day +1

day +2

day +3

Outlying depreciation days, Out-sf-sample period

M

50

40

ed

30

20

Pro Con

10

Sell

0

Control -10

-20

Ac

-30

ce pt

Buy

day -3

day -2

day -1

Outlying Day

day +1

day +2

day +3

The figure presents the P&L of the four speculator types and a control group, starting three days before the outlying days and ending three days after. The examined period was 2010, which is an out-of-sample period. There were 7 outlying appreciation days and 8 outlying depreciation days in 2010. The control group, which is composed of 58 randomly selected non-speculative players, exhibits small P&L regardless of outlying days.

36 Page 37 of 39

*Research Highlights



This paper suggests a methodology for identifying speculators in FX markets



A player is identified as a speculator only if his speculative characteristics are extreme and his influence on exchange rates on outlying days is significant.



Implementing the proposed methodology on Israel's FX market identified one group of speculators that joined or initiated the trend before the outlying days. It was possible to identify similar behavior before and on outlying days during

Ac

ce pt

ed

M

an

us

cr

2010, which was defined as an out-of-sample period.

ip t



Page 38 of 39

figure 1 Hampel vs. Quantile method: Outlying days and exchange rate (ER) changes In-Sample Period Out-Of-sample period

cr

ip t

Figure 2 Net Buying Pressure (NBP, USD million) of the four speculator types and a control group around outlying days, 2010 Outlying appreciation days, Out-of-sample period Outlying depreciation days, Out-of-sample period

Ac ce p

te

d

M

an

us

Figure 3 P&L (USD, '000) of the four speculator types and a control group around outlying days, 2010 Outlying appreciation days, Out-of-sample period Outlying depreciation days, Out-of-sample period

Page 39 of 39