You are on page 1of 20

Volume, Order Flow and the FX market:

Does it matter who you are?∗


Geir H. Bjønnes† Dagfinn Rime‡
Haakon O.Aa. Solheim§

February 2002
Comments welcome
Preliminary and incomplete. Do not quote

Abstract
We study the impact of order flow and volume using an unique
data set of daily trading in the Swedish krona (SEK) market. The
data set covers 95 per cent of worldwide SEK-trading, and is disag-
gregated on 13 reporting banks’ buying and selling with seven different
counterparties in five different instruments. The preliminary sample
covers the first six months of 1998, while the whole data set, which
we will receive during the spring 2002, begins in 1993.
A unique feature in our data is the ability to differentiate between
counterparties. We focus on three relationships: Between (i) volume
and volatility, (ii) order flow and exchange rate changes and (iii)
the dynamics of liquidity to the market. We find that while we can
restate previously reported findings from the literature, the findings
depend on the identity of the counterparty. We find that the volume
of the presumably least informed counterpart contributes positively
to volatility, indicating a noise-trader role for this group. Further-
more, customers’ order flow contribute significantly to exchange rate
changes, supporting the importance of these flows previously stated in

We thank Antti Koivisto for helpful discussions and assistance with collecting the data
set.

Stockholm Institute for Financial Research, email: geir.bjonnes@sifr.org.

Norges Bank (Central Bank of Norway) and Stockholm Institute for Financial Re-
search. email: dagfinn.rime@norges-bank.no.
§
Corresponding author. Norwegian School of Management (BI). email:
haakon.o.solheim@bi.no.

1
the literature. Finally, we find that customer order flow is a function
of past returns in the equity markets, changes in the interest rate dif-
ferential and exchange rate movements. Again this process will differ
between different groups of counterparties, indicating the importance
of heterogeneity.
Keywords: Order flow analysis, volume volatility relation, microstruc-
ture, exchange rates
JEL Classification: F31

1 Introduction
This paper revisits three previously studied topics in the financial literature;
the relationship between volume and volatility, between order flow and ex-
change rate changes and the topic of liquidity provision. In the FX-market
such research has until recently been difficult due to the lack of good trading
data. In this paper we use a unique data set provided by Sveriges Riksbank
(the Swedish central bank). The data is based on daily reporting from 13
primary dealers. Each primary dealer reports total purchases and sales with
six categories of counterparties: (1) Swedish primary dealer, (2) foreign pri-
mary dealer, (3) Swedish bank, (4) foreign bank, (5) Swedish customer, and
(6) foreign customer. For each counterparty-category total purchases and
sales are split into different instruments, (i) spot, (ii) outright forwards, (iii)
short swaps (tomorrow-next), (iv) FX swaps, and (v) options.1 The data
covers as much as 95 per cent of all currency trading in Swedish kroner.
This paper makes several contributions. First, we use transaction volume
data which covers almost the entire market for Swedish kroner to examine
the relationship between volatility and transaction volume. This is an im-
portant contribution considering the lack of transaction volume data from
other currency markets. As in many other studies from different market set-
tings (see Karpoff, 1987), we find a positive relationship between volatility
and volume. The results are also consistent with Galati (2000) who finds
a positive relationship between volatility and volume for five of the seven
currencies measured against dollar. Compared with the currency markets
studied by Galati (2000), the Swedish currency market is larger.
Second, with our detailed data we can for the first time study the under-
lying source of the relationship between volume and volatility in the foreign
exchange market. In particular, we examine the importance of heterogene-
ity in explaining volatility. This is possible since total transaction volume
can be split into different categories of counterparties. Studies from other
1
A short swap is a contract to be delivered within two days, e.g. before a spot contract.

2
market settings suggest that heterogeneity among the market players may
be important in understanding volatility (see e.g. Grinblatt and Keloharju,
2001). In asymmetric information models (e.g. Kyle, 1985; Admati and Pflei-
derer, 1988) more trade from informed investors increase volatility because
of the generation of private information. On the other hand, Shalen (1993)
argue that uninformed traders increase volatility because they cannot differ-
entiate liquidity demand from fundamental value change. Daigler and Wiley
(1999), studying futures markets, find that trade of presumable less informed
investors tend to be more correlated with volatility than trade of more in-
formed investors. Our results point in the same direction, as we we find a
positive correlation between volume from customers and volatility.
Third, we use our data to explain movements in foreign exchange rates.
In particular, we study the effects of customer trades. Recent literature by
Evans and Lyons (2002) and Fan and Lyons (2000) suggest that orders from
customers are important in explaining movements in foreign exchange rates
because customer orders are the ultimate driver of all interdealer trading. We
are able to test for the relationship between order flow and the exchange rate
differentiating between customer order flow, interbank flows and central bank
interventions. We find that customer order flow2 is much more important
than interbank flows for the determination of exchange rates.
Fourth, we test for exogeneity between order flow and the exchange rate.
Killeen, Lyons and Moore (2001) argue that order flow shall be both weakly
and strongly exogenous with regard to a flexible exchange rate. They find
that order flow is strongly exogenous before the EMS exchange rates were
fixed, and endogenous after the rates were fixed. However, we do not find
that order flow is strongly exogenous with regard to the exchange rate in our
sample. Rather the exchange rate seems to Granger cause changes in order
flow.
Last, we will look at the dynamics of liquidity provisions. Chordia, Sarkar
and Subrahmanyam (2001) investigate the determinants of bond and stock
market liquidity. They find that volume changes in bond and stock mar-
kets “are predictable to a considerable degree using lagged market returns,
lagged spreads, and lagged volume.”(Chordia et al., 2001). However, little
theoretical work has been done on movements in liquidity.
We identify determinants of order flow. We find that customer order flow
does depend on returns in the exchange market, as well as past returns in
the Swedish equity markets and changes in interest differentials. However,
the determinants of order flows differ between groups. This confirms our
2
Notice that our definition of customer order flow contain trades made by customers
as well as by non-reporting banks with reporting banks.

3
Figure 1: Swedish and German Interest rates and the SEK/EUR exchange
rate

1.3 5.2 8.9

1.2
4.8 8.8
1.1

4.4 8.7
1.0

0.9 Interest rate diff (left axis) 4.0 8.6


Swedish 3 month
0.8 German 3 month
3.6 8.5
0.7
SEK/EUR exchange rate
0.6 3.2 8.4
1/01 1/29 2/26 3/26 4/23 5/21 6/18 1/01 1/29 2/26 3/26 4/23 5/21 6/18

indication that foreigners and locals, customers and interbank markets are
different.
The paper is organized as follows. Section 2 gives a detailed presentation
of our data. In Section 3 we derive testable hypothesis and present the
results. Section 4 concludes.

2 Data
The sample used is daily observations for the first six months of 1998, making
a total of 129 observations.3 According to the market survey conducted by
BIS in 1998 this period was calm. Our dependent variable is the SEK/DEM
exchange rate.4 In the order flow regressions we use return (∆ log SEK/DEM),
while in the volatility regressions we use the absolute value of return. We
focus on SEK/DEM since this covers close to 100 of all interbank trading
and 80–90% of customer trading. We use the 3 month money market inter-
est rate from Sweden and Germany. These series, together with the interest
rate differential, is showed in figure 1.
The figure confirms that this was a calm period. Interest rates are very
stable throughout (the exception being a reduction of the Swedish repo rate
by 25 points June 10), with a mean interest rate differential around 100
3
The whole data set on transaction flow starts in 1993 and run up to present. We will
receive all the data during the spring.
4
Note however, that as the standard is changing to EUR, the exchange rate is indexed
to the EUR equivalent terms (SEK/DEM*1.95583).

4
points and a standard deviation of 14 point. The exchange rate is similarly
stable, with only 4.6% between top and bottom. Mean return is 0.004% with
0.3 standard deviation and maximum return being ca. 1%. Although the
fall of 1998 was very turbulent in financial markets we are apparently not
picking up anything from that period in this sample. This is assuring as we
do not want to pick extreme events in the current analysis.
The remainder here will be devoted to the description of the transaction
flows since they represent the most novel part of this paper. The data set is
provided to us from the Sveriges Riksbank (Central Bank of Sweden), and
will eventually contain daily observations starting in 1993 until today.
The data set is extremely detailed. The Riksbank receives daily reports
from 13 Swedish and foreign banks on their buying and selling of five different
instruments against seven different counterparts. The reported series is an
aggregate of Swedish krona trading against all other currencies, measured in
krona, and covers 90–95 of all worldwide trading in SEK exchange rates. As
mentioned above, the vast majority of the flows is SEK/DEM transactions.
The 13 reporting banks are anonymized. We know however that there are
five Swedish banks, five foreign banks, and three branches of foreign banks
located in Sweden. The reporters are the main liquidity providers in the
SEK-market. The five instruments are spot, forward, options, short swaps
(tomorrow/next) and long swaps. In this paper we focus on the spot and
forward flows.
Particularly important is the counterparty information. The seven coun-
terparties are Swedish and foreign customers, reporting banks, and other
banks, and in addition the Riksbank. This enables us to distinguish cus-
tomer order flow from interbank trading between the liquidity providers and
interbank trading with more speculative banks. This is to the best of our
knowledge an unique feature of the present data set.
Figure 2 shows the net selling of foreign currency to different counterparts
by the reporting banks in the spot market. A positive number indicates that
the counterpart group has accumulated currency from the reporters. We
immediately see that Swedish customers and other banks are accumulating
currency in the spot market (right panel). This is in line with the current ac-
count surplus of Sweden. Figure 3 shows the net purchase of foreign currency
spot and forward.
These data are unique in several aspects. (i) The series are longer and/or
of higher frequency than in many previously available data sets.5 Evans and
Lyons (2002) have daily observations from the interbank market but only for
5
The whole data set is long (nine years). The present sample however is of similar
length as Evans and Lyons (2002).

5
Figure 2: Net spot position of foreign and Swedish Customers+Other banks

10000 30000
Net SEK-sale (FG-group)
25000
0
20000

-10000 15000

10000
-20000

5000
-30000
0
Net SEK-sale (SG group)
-40000 -5000
1/01 1/29 2/26 3/26 4/23 5/21 6/18 1/01 1/29 2/26 3/26 4/23 5/21 6/18

[FG-group: foreign customers and other banks, SG-group: Swedish customers


and other banks.]

Figure 3: Net spot+forward position of foreign and Swedish Cus-


tomers+Other banks

10000 15000

Net spot+forward (FG-group) Net spot+forward (SG-group)


0 10000

-10000 5000

-20000 0

-30000 -5000

-40000 -10000
1/01 29/01 26/02 26/03 23/04 21/05 18/06 1/01 29/01 26/02 26/03 23/04 21/05 18/06

[FG-group: foreign customers and other banks, SG-group: Swedish customers


and other banks.]

6
79 days. Wei and Kim (1997), Cai, Cheung, Lee and Melvin (2001) and Rime
(2001) have long series, but only at weekly frequency. Fan and Lyons (see
Lyons, 2001) also have long series, but only for the monthly frequency. (ii)
We have the complete trading of several banks. The weekly series mentioned
above aggregate over the banks. (iii) Detailed counterparty information.
The data in Evans and Lyons (2002) is only interbank flow, but do not
differentiate between groups of banks, while the data in Wei and Kim (1997),
Cai et al. (2001) lump all flows, customers and interbank, together. The data
in Rime (2001) differentiate between local customers and foreigners which
primarily are banks. Information on customer flow is important because
this is the basic source of demand. Fan and Lyons can distinguish between
financial and non-financial customers, but have only series for one bank. On
the other hand, Evans and Lyons and Cai et al. have observations on the
main exchange rates.

3 Results
In this section we present some results. One should have in mind that these
estimations are only conducted on a set covering 129 trading days of data. A
longer set is necessary to confirm the trends reported below. However, this
is a problem experienced in much of the microstructure literature.
The large number of variables that can be extracted from our set do
however make for a rather complex system of variable names. To assure that
all tables can be understood properly table 1 provides variable names and
explanations.
In the first section we discuss the relationship between volume and volatil-
ity. In the second section we discuss contemporaneous correlation between
order flow and returns. In the last section we look at the dynamics of order
flow and test for Granger causality between order flow and exchange rates.
Notice that all estimations are done using daily observations.

3.1 Volume versus volatility


How to measure volatility is not given. As mentioned above, volatility is low
in the period under investigation. We find no evidence of an ARCH process
in in the exchange rate over this sample period.
The optimal way to measure volatility in a sample of this kind would
probably be to take the difference between highest and lowest observed value
intra day. However, for some reason such series do not seem to be available

7
Table 1: Variable explanations
Variable Explanation
lSEK The log of the price of ECU denominated in SEK
returns d(lSEK)
absreturns absd(lSEK)
rdif3 Swedish 3 month interbank rate - German 3 month interbank
rdif10 Swedish 10 year gov. bonds - German 10 years gov. bonds
lOMX Log of the OMX equity index (Stockholm exchange)
OFLGGL Aggregate order flow all customers and non-reporting banks, spot and forward
OFSGGL Agg. order flow from Swedish customers and non-reporting banks in the spot and forward market
OFFGGL Agg. order flow from foreign customers and non-reporting banks in the spot and forward market

8
OFLRGL Aggregate order flow in the interbank market, spot and forward
OFSSGL Aggregate order flow, Sveriges Riksbank
d() First differential
LLSLL Total volume transacted in the spot market
LLSLLU Unexpected volume
LLSLLE Expected volume
SGSLLU Total unexp. volume from Swedish customers and other banks in the spot market
FGSLLU Total unexp. volume from foreign customers and other banks in the spot market
LRSLLU Total unexp. volume, interbank, spot market
SSSLLU Total unepx. volume, Riksbanken, spot market
D1, D2, D3, D4 Dummies for Monday, Tuesday, Wednesday and Thursday respectively
in the SEK/DEM currency cross, only in SEK/USD. We choose to measure
volatility as the absolute value of return from close to close.6
Table 2 report the first four estimations on volatility. In estimation (1) we
explain volatility using a number of “fundamental variables”—e.g. absolute
change in the stock market index and short term interest rates, and with
daily dummies. As can be seen, the explanatory power is weak.
In estimation (2) we add the change in total volume in the spot market,
d(LLSLL). In line with a number of previous papers we find a significant
relationship between volume and volatility.
Hartmann (1999) report that “unexpected” flows increase spreads in the
exchange rate market, while expected flows reduce spreads. We differentiate
between expected and unexpected flows by using an AR(5) model. We esti-
mate an AR(5) for each variable, and use the fitted values as expected flows,
and the residuals as the unexpected flows. As can be seen from estimation
(3) unexpected flows seems to drive the volume-volatility relationship. Ex-
pected flows have no significant relationship with volatility. Estimation (4)
is conducted using only unexpected flows.
As pointed out, an interesting part of our data is the ability to differentiate
between counterparties. Daigler and Wiley (1999) are able to distinguish
between counterparties on the basis of their proximity to the market. They
find that the volume from customers far from the market seems to increase
volatility more than volume from presumably more informed traders. In our
sample we can distinguish between the Riksbank, reporting currency banks,
other banks and customers. We choose to pool customers and other banks
in two distinct groups: Swedish and foreign. We pool all inter-reporter trade
in one group.
“Reporters” comprise the main large banks trading in SEK. One should
therefore expect reporters to be well informed about the SEK-market. Like-
wise, the Riksbank should be expected to have superior knowledge of this
market. The least informed counterparties are expected to be customers and
other banks.
In table 3 we report a number of estimation where we differentiate be-
tween counterparties. In estimation (5) we include all the four groups. We
only include unexpected flows, as expected flows have little or no power. We
find that only the volume of foreign customers seem to have significant re-
lationship with volatility. One should however remark that our results are
not clear cut. The sign for both reporters and the Riksbank is also posi-
tive, indicating a positive relationship between volume and volatility. Only
6
The results for high-low in SEK/USD do not differ significantly from those reported
below. However, R2 is somewhat lower.

9
Table 2: Estimating |returns| (a)—01.01.1998 to 06.30.1998
|returns| (1) t-stat (2) t-stat (3) t-stat (4) t-stat
Constant 0.0021 4.83 ** 0.0004 0.76 0.0020 1.49 0.0024 14.82 **
Market spot volume, D(LLSLL) 0.0000 3.89 **
Unexpected volume, D(LLSLLU ) 7.69E-08 4.34 ** 7.30E-08 4.35 **
Expected volume, D(LLSLLE) 4.88E-09 0.10
Monday dummy, D1 -0.0001 -0.17 0.0004 0.82 0.0003 0.60
Tuesday dummy, D2 0.0000 0.05 0.0001 0.18 -0.0001 -0.20
Wednesday dummy, D3 -0.0002 -0.32 0.0000 -0.02 -0.0001 -0.22

10
Thursday dummy, D4 0.0001 0.14 0.0001 0.16 0.0000 0.03
Abs. stock index return, |d(lOM XC)| 0.0432 1.96 0.0263 1.24 0.0348 1.54
Abs. interest differential, |d(rdif )| -0.0028 -0.30 -0.0064 -0.72 -0.0053 -0.57
Adjusted R2 -0.02 0.09 0.11 0.13
DW 2.03 2.11 2.10 2.00
S.E. of regression 0.0019 0.0018 0.0018 0.0018
Note1: Estimated with OLS
Note2: * - 5 per cent, ** - 1 per cent
Note3: Coefficient sizes: All returns are measured such that a return of 1 per cent is 0.01. Volume is measured in
million SEK.
the sign for Swedish customers and other banks is negative. We probably
need a longer data set to be able to confirm whether information (or lack of
information) is really the driving force in the volume-volatility relationship.

3.2 Order flow versus returns


Order flow is here defined as change in net position over the day. We define
order flow in such a manner that a negative order flow indicates that the
counterparty sell foreign currency, and purchases SEK. According to standard
microstructure theory we should expect a positive correlation between order
flow and the value of foreign currency denominated in SEK.
One should point out that change in net position is not enough to really
establish that a flow is an ‘order flow’. Lyons (2001) define an order flow as
a signed transaction. To identify order flow we need to know who initiates
the trade. That information is not contained in our data.
However, all trades between customers and reporters or between other
banks and reporters are initiated outside the reporting bank. To our knowl-
edge it is also reasonable to expect that trades between customers and other
banks and reporters are initiated outside the reporting bank. We will there-
fore make the assumption that daily change in the net position of customers
and other banks can be considered as daily order flow.
Change in net positions between reporters is an interbank flow. It is not
given whom initiates these trades. Also trades conducted by Sveriges Riks-
bank is problematic seen from the order flow perspective. Rightly, Sweden
does in this period have a floating exchange rate. The money supply should
therefore be exogenous, and trades by the Riksbank should be expected to
be initiated by the Riksbank. However, it is not clear that the Riksbank
does not take e.g. liquidity considerations when doing transactions in the
currency market. Indeed we find a negative correlation between the change
in the Riksbank’s net positions and returns in the exchange rate market.
Table 4 reports a number of estimations on returns. In estimation (1) we
regress returns on the change in the three month interest rate differential to
German interest rates and the change in the the stock exchange. We find a
significant relationship between returns in the exchange rate and returns in
the stock market, indicating that the SEK strengthen when the stock market
rise. However, the explanatory power of this model is low.
In estimation (2) we include aggregate customer order flow—i.e. the
change in net positions over all customers and other banks in the spot and
forward markets.7 We find that this variable, as expected, is significant and
7
As it seems like the participants in our data set do manage spot and forward together,

11
Table 3: Estimating |returns| (b)—01.01.1998 to 06.30.1998
|returns| (1) t-stat (5) t-stat (6) t-stat (7) t-stat
Constant 0.002 4.825 ** 0.002 4.977 ** 0.002 15.209 ** 0.002 15.208 **
Unexp., spot, SGp, D(SGSLLU ) -1.42E-08 -0.241 -7.97E-09 -0.142
Unexp., spot, FG, D(F GSLLU ) 1.39E-07 3.946 ** 1.31E-07 3.803 ** 1.52E-07 5.134 **
Unexp., spot, rep., D(LRSLLU ) 8.23E-08 1.114 7.58E-08 1.040
Unexp., spot, Riksb., D(SSSLLU ) 1.02E-06 1.176 1.01E-06 1.175
Monday dummy, D1 0.000 -0.167 0.000 0.860
Tuesday dummy, D2 0.000 0.050 0.000 0.124

12
Wednesday dummy, D3 0.000 -0.318 0.000 0.098
Thursday dummy, D4 0.000 0.137 0.000 0.003
Abs. stock index ret., |d(lOM XC)| 0.043 1.963 0.035 1.675
Abs. interest dif., |d(rdif )| -0.003 -0.299 -0.005 -0.584
2
Adjusted R -0.02 0.17 0.17 0.17
DW 2.03 2.01 1.92 1.89
S.E. of regression 0.002 0.002 0.002 0.002
Note1: Estimated with OLS
Note2: * - 5 per cent, ** - 1 per cent
positive. The size of the parameter might be difficult to interpret. However,
notice that returns is measured in change in logarithms of the exchange rate.
Flows are measured in millions of SEK. Evans and Lyons (2002) report an
order flow effect of 0.5 per cent in the DEM/USD market for a trade of 1
billion USD. A trade of 1 billion USD would equal approximately a trade of
8000 million SEK. That would give an effect of 0.006 on the exchange rate,
or in other terms, an effect of 0.6 per cent. This is close to the findings of
Evans and Lyons.
In estimation (3) we differentiate between counterparties. Interestingly,
we find that Swedish and foreign customers and other banks behave as ex-
pected: the parameter is significant and the sign is positive.8 Inter-reporter
flows are not significant. Order flow from the Riksbank is significant, but
with a negative sign. This might indicate that we do not interpret the or-
der flow definition properly with regard to the central bank. The lack of
effect from interbank trade may also be due to noise introduced through our
definition of order flow.
A potential problem in estimations (1-3) is the question whether or not
order flows ar endogenous with regard to the exchange rate. The microstruc-
ture theory clearly predicts that order flow should be exogenous. However,
as the results in the next section will show, this is not clear in our data. One
possibility is that order flow is partly a function of feedback trading. This
could affect our results significantly.
To control for feedback trading we include 5 lags of the return in the
exchange rate in estimation (4). Only one of the five lags is significant—lag
4. However, as can be seen when comparing estimation (3) and (4), although
R2 rises considerably, the results for the order flow variables does not change
noticeably.

3.3 The dynamics of order flow


On volume vs. volatility there is a vaste theoretical literature. On order flow
vs. returns there is a growing literature. On the dynamics of order flow little
has been done in the theoretical area.
Some hypotheses have been put forward. Killeen et al. (2001) argue that
under a flexible exchange rate one should expect order flow and exchange
rates to be (i) cointegrated, (ii) the order flow should be weakly exogenous
with regard to the exchange rate, and (iii) the order flow should be strongly
we find it reasonable to include both spot and forward positions when calculating order
flow.
8
Notice that the size of the order flow parameter for these two groups is no in fact
exactly 0.5 per cent per 1 billion USD equivalent.

13
Table 4: Estimating returns—01.01.1998 to 06.30.1998
returns (1) t-stat (2) t-stat (3) t-stat (4) t-stat
Constant 0.000 0.576 0.000 0.984 0.001 2.776 ** 0.001 2.798 **
Tot. customer OF, D(OF LGGL) 7.72E-07 3.987 **
Swe. customer OF, D(OF SGGL) 6.61E-07 2.979 ** 6.24E-07 2.880 **
For. customer OF, D(OF F GGL) 6.11E-07 2.951 ** 7.06E-07 3.442 **
Reporter OF, D(OF LRGL) -1.94E-07 -0.637 -1.76E-07 -0.594
Reserves, D(OF SSGL) -3.63E-06 -3.026 ** -3.78E-06 -3.194 **

14
Interest dif., d(rdif ) 0.002 0.173 -0.002 -0.151 -0.006 -0.507 0.000 -0.035
Stock index return, d(lOM XC) -0.054 -2.472 * -0.051 -2.437 * -0.055 -2.695 ** -0.053 -2.689 **
Lagged return (4) 0.195 2.445 *
Adjusted R2 0.03 0.14 0.18 0.30
DW 2.09 2.02 2.05 2.15
S.E. of regression 0.003 0.003 0.003 0.001
Note1: Estimated with OLS
Note2: * - 5 per cent, ** - 1 per cent
: We include 5 lagged values of return in this regression. Only the fourth lag is significant.
Table 5: Estimating Granger causality—01.01.1998 to 06.30.1998
d(OFSGGL) does not Granger cause d(lSEK) F-stat 0.70 0.62
d(OFFGGL) does not Granger cause d(lSEK) F-stat 1.00 0.42
d(OFLRGGL) does not Granger cause d(lSEK) F-stat 1.20 0.32
d(lSEK) does not Granger cause d(OFSGGL) F-stat 2.56 0.03 *
d(lSEK) does not Granger cause d(OFFGGL) F-stat 3.91 0.00 **
d(lSEK) does not Granger cause d(OFLRGGL) F-stat 1.55 0.18
Note1: Estimated with bivariate VAR, 5 lags. Probability levels in the right-
most column.
Note2: * - 5 per cent, ** - 1 per cent

exogenous with regard to the exchange rate. Strong exogeneity is defined by


Killeen et al. (2001) as the observation that order flow should Granger cause
the exchange rate.
The sample is to short to conduct satisfactory cointegration analysis.9
However, we believe that we can with some degree of reason do a Granger
causality test. The results are reported in table 5. Evidence seems to be
that one can not reject that there is no relationship from order flows to the
exchange rate. This is the case for order flows from Swedish customers and
other banks, for foreign customers and other banks, and for changes in net
positions of reporters. However, one can reject a hypothesis of no relationship
from the exchange rate to order flow for Swedish and foreign customers and
other banks. This indicates that order flow in this sample seems to react to
past returns, not the other way around.
The Granger causality tests indicate that past returns in the currency
market might be one explanatory variable for order flow. Chordia et al.
(2001) find a relationship between the volume and past returns in the stock
and bond markets. To investigate this issue closer we conduct a process of
econometric modelling. 10 We use the general to specific modelling approach,
and begin with a model containing the same period and five lags of returns in
(i) the exchange rate and (ii) the Swedish stock index (OMX), (iii) the change
in the three month and (iv) 10 year interest differential to Germany (the last
indicating uncertainty about future inflation differentials), and order flows
for the the four groups (v) Swedish customers and other banks, (vi) foreign
customers and other banks, (vii) net changes in positions between reporters
and (viii) net changes in reserves (measured as net changes in the positions
of the Riksbank). We also include dummies for different week-days. We then
eliminate variables have no explanatory power, and test modelling progress
9
Some preliminary results do indicate the presence of a stable cointegration vector.
However, it seems like the exchange rate is weakly exogenous, not the order flow.
10
Since there is little theory in this area, we feel that this approach is justified.

15
using F-tests. We report only final modelling results.
The results for order flow from Swedish customers and other banks is
reported in table 6. We see that our model has reasonable explanatory
power—the R2 is above 0.5. Order flow from Swedish customers and other
banks seems to depend on simultaneous and past flows, as could be expected.
However, order flow also seems to depend on past returns in the exchange
rate, past returns in the OMX and past interest rate changes. To summarise
our findings:
• As expected, a same day depreciation leads to flow out of SEK. However
a depreciation leads to purchases of SEK two days hence. Assuming
that this group contains a number of importers and exporters, this seem
reasonable.
• A change in the short term interest rate on the same day the previous
week tends to affect flows.
• Past returns on the Stockholm stock exchange is positively correlated
with the purchase of SEK (remember that a purchase of SEK is a
negative number).
• An increase in the 10 year interest differential to Germany seems to
make Swedish customers and other banks depart from SEK. This might
indicate that an increase in the long term spread indicates increased
uncertainty about conditions in Sweden.
The results for the order flow of Swedish customers and other banks
become even more interesting when compared with similar results for foreign
customers and other banks, as reported in table 7. As can be seen, neither
short-term interest rates nor the stock market index is contained in this
regression. In fact the only lagged macro variable included is the long-term
interest rate differential. On the other hand, daily dummies are important
here. It seems like foreign flows depend on an other set of variables than
local flows.
With regard to inter-reporter flows, these do not depend on macro vari-
ables at all. However, one should notice that our ability to explain these
flows are much less than our ability to explain customer flows. It is a well
known from the banking literature that the amount of interbank flows are
very difficult to justify in standard models. We must conclude that it is sat-
isfying to find that macro variables do not explain changes in net positions
between reporting banks. However, our findings in this area provides even
more indications of the need for a good general equilibrium model of the
interbank market.

16
Table 6: Estimating d(OF SGGL)—01.01.1998 to 06.30.1998
t-stat
Constant 270.43 1.85
d(OF SGGL)3 0.13 2.01 *
d(OF F GGL) -0.63 -10.82 **
d(OF F GGL)4 -0.11 -1.88
d(OF LRGL) -0.35 -3.05 **
d(OF LRGL)1 0.23 1.95
d(OF SSGL)5 -1.11 -2.44 *
d(lSEK) 1023.20 3.27 **
d(lSEK)2 -907.48 -2.85 **
d(rdif 3)5 -126.03 -2.75 **
d(lOM X)2 -206.16 -2.37 *
d(lOM X)3 -223.92 -2.55 *
d(lOM X)5 -153.50 -1.92
d(rdif 10) 9739.16 2.78 **
d(rdif 10)1 4957.39 1.49
d(rdif 10)3 -7323.82 -2.29 *
D2 -268.04 -1.10
Adjusted R2 0.61
AR 1-4 test 0.367 (0.83)
ARCH 1-4 test 0.370 (0.83)
Note1: Estimated with OLS
Note2: * - 5 per cent, ** - 1 per cent

Table 7: Estimating d(OF F GGL)—01.01.1998 to 06.30.1998


t-stat
Constant 372.1383 2.229 *
d(OF SGGL) -0.72631 -10.615 **
d(lSEK) 1236.07 3.575 **
d(rdif 10)1 8208.599 2.364 *
d(rdif 10)4 -7736.53 -2.309 *
D2 -884.934 -3.063 **
D3 -842.137 -2.837 **
D4 -839.289 -2.873 **
2
Adjusted R 0.55
AR 1-4 test 0.973 (0.43)
ARCH 1-4 test 0.588 (0.67)
Note1: Estimated with OLS
Note2: * - 5 per cent, ** - 1 per cent

17
Table 8: Estimating d(OF LRGL)—01.01.1998 to 06.30.1998
t-stat
d(OF SGGL) -0.16 -3.41 **
d(OF LRGL)1 0.31 3.58 **
2
Adjusted R 0.15
AR 1-4 test 1.100 (0.36)
ARCH 1-4 test 0.918 (0.46)
Note1: Estimated with OLS
Note2: * - 5 per cent, ** - 1 per cent

4 Conclusions
Our starting point was: does it matter for the FX markets who you are? The
answer is that different counterparties clearly seems to affect the markets
differently. This opens a wide field of possible research questions.
We have used this paper to investigate a number of different questions
related to microstructure theory. As pointed out on several occasions above,
our results should at this point be interpreted with care. However, we believe
our results illustrate the potential in this kind of data. Such data can both
be used to test the number of existing (and in part competing) theories
that exists in the microstructure literature. It can also be used to point out
interesting avenues for new theoretical research.

References
Admati, A. R. and P. Pfleiderer (1988). “A theory of intraday patterns:
Volume and price variability.” Review of Financial Studies, 1(1), 3–40.
Cai, J., Y.-L. Cheung, R. S. K. Lee and M. Melvin (2001). “‘once-in-a-
generation’ yen volatility in 1998: Fundamentals, intervention, and order
flow.” Journal of International Money and Finance, 20(3), 327–47.
Chordia, T., A. Sarkar and A. Subrahmanyam (2001). “Common determi-
nants of bond and stock market liquidity: The impact of financial crises,
monetary policy and mutual funds.” Tech. Rep. 140, Federal Reserve Bank
of New York.
Clark, P. K. (1973). “A subordinated stochastic process model with finite
variance for speculative prices.” Econometrica, 41, 135–55.
Copeland, T. E. (1976). “A model of asset trading under the assumption of
sequential information arrival.” Journal of Finance, 31, 1149–68.

18
Copeland, T. E. (1977). “A probability model of asset trading.” Journal of
Financial and Quantitative Analysis, 12, 563–78.
Daigler, R. T. and M. K. Wiley (1999). “The impact of trader type on the
futures volatility-volume relation.” Journal of Finance, 54(6), 2297–2316.
Evans, M. D. and R. K. Lyons (2002). “Order flow and exchange rate dy-
namics.” Journal of Political Economy, 110, 170–180.
Fan, M. and R. K. Lyons (2000). “Customer-dealer trading in the foreign
exchange market.” typescript.
Galati, G. (2000). “Trading volumes, volatility and spreads in foreign ex-
change markets: Evidence from emerging market countries.” Working
paper 93, Bank for International Settlements.
Grinblatt, M. and M. Keloharju (2001). “How distance, language, and culture
influence stockholdings and trades.” Journal of Finance, 56(3), 1053–73.
Harris, M. A. and A. Raviv (1993). “Differences of opinion make a horse
race.” Review of Financial Studies, 6(3), 473–506.
Hartmann, P. (1999). “Trading volumes and transaction costs in the foreign
exchange market. evidence from daily dollar–yen spot data.” Journal of
Banking and Finance, 23, 801–24.
Isard, P. (1995). Exchange Rate Economics. Cambridge University Press.
Jorion, P. (1996). Risk and Turnover in the Foreign Exchange Market, pp.
19–37. ucp, Chicago.
Karpoff, J. M. (1987). “The relation between price changes and trading
volume: A survey.” Journal of Financial and Quantitative Analysis, 22(1),
109–26.
Killeen, W. P., R. K. Lyons and M. J. Moore (2001). “Fixed versus flexible:
Lessons from EMS order flow.” Tech. Rep. 8491, NBER.
Kyle, A. S. (1985). “Continuous auctions and insider trading.” Econometrica,
53(6), 1315–35.
Lyons, R. K. (2001). The Microstructure Approach to Exchange Rates. MIT
Press, Cambridge, Mass.
Melvin, M. and X. Yin (2000). “Public information arrival, exchange rate
volatility, and quote frequency.” Economic Journal, 110(465), 644–61.

19
Rime, D. (2001). “Private or public information in foreign exchange markets?
An empirical analysis.” University of Oslo.

Shalen, C. T. (1993). “Volume, volatility, and the dispersion of beliefs.”


Review of Financial Studies, 6(2), 405–34.

Wei, S.-J. and J. Kim (1997). “The big players in the foreign exchange
market: Do they trade on information or noise.” Working Paper 6256,
NBER.

20

You might also like