You are on page 1of 33

F

o
r

P
e
e
r

R
e
v
i
e
w
Macroeconomic announcements, communication and order flow on the
Hungarian foreign exchange market
Journal: International Journal of Finance and Economics
Manuscript ID: IJFE-08-0140
Wiley - Manuscript type: Research Article
Keywords:
microstructure, foreign exchange, order flow, news, central bank
communication, Hungary, transition economy
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
F
o
r

P
e
e
r

R
e
v
i
e
w
Macroeconomic announcements, communication and order flow on
the Hungarian foreign exchange market
August 2008.
Abstract
We investigate the relation between the intradaily HUF/EUR exchange rate on the one hand
and news announcements and order flow on the other hand. We extend the existing literature
on foreign exchange market microstructure by considering a small open transition economy.
We find that the intradaily exchange rate depends on both news announcements and order
flow. We conclude that news on the HUF/EUR market are transmitted directly via immediate
reactions to news announcements as well as indirectly via order flow. We decompose the
news’ total effect on exchange rate and find that order flow accounts for approximately three
quarters, compared to one quarter for direct news impact. Although the HUF is pegged to the
EUR, the exchange rate reacts very qualitatively very similarly to exchange rates of major
currencies as reported in the literature, whereas it quantiatively differs: the importance of
indeirect news transmission is remarkably higher on the HUF/EUR market. Furthermore, we
extend the commonly used news to communication of central bankers significantly improving
the explanatory power of the estimates. Central bank communication is an important
determinant for the HUF/EUR rate.
JEL: F31, G14, G15
Keywords: microstructure, order flow, exchange rate, macroeconomic news, central bank
communication, Hungary
Page 1 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
1
Macroeconomic announcements, communication and order flow on
the Hungarian foreign exchange market*
1. Introduction
During recent years there has been a growing interest in the microstructure of financial mar-
kets and the relations between exchange rates, news and order flow. The microstructure ap-
proach deals with the purest form of financial intermediation, i.e. pure trading without any
transformation and has been pushed by the increasing availability of high-frequency data, that
typically ranges from tick-to-tick data to frequencies of about 20 minutes. The microstructure
approach allows additional insights in the price formation on financial markets. Whereas tra-
ditional macroeconomic models assume that news are directly transmitted to prices, the mi-
crostructure approach explicitly assigns an important role to the process of trading.
The key concept is the order flow, showing a close empirical relation to the exchange rate (see
e.g. Lyons 2001, Evans 2002, Payne 2003). The order flow is defined as the difference be-
tween trades initiated by the buyer, and trades initiated by the seller of an asset. Therefore
order flow can be described as buying (or selling) pressure on the market. Order flow signifi-
cantly differs from the turnover on the market. There are situations where the order flow is
almost zero, if trades are symmetrically initiated by buyers and sellers, although the turnover
is high and vice versa.
Whereas in traditional asset market models information is assumed to be common, the micro-
structure approach claims that information exists that is not shared by all market participants
and/or is interpreted differently by them (see e.g. the models by Kyle 1985, Glosten and Mil-
grom 1985, Evans 1995, Evans and Lyons 2006b). Thus, the trading mechanism itself be-
comes an important feature. However, the microstructure approach does not insist that it is not
the fundamentals which drive prices, but it focuses on the way how potentially dispersed in-
formation about fundamentals is transformed to prices. Although some standard assumptions
of perfect markets are relaxed within this framework (i. e. perfect information, perfect
competition), the market may still be regarded as efficient in the sense, that prices reflect all
known information. It implies that within both frameworks the market reaction to public news
announcements is expected to occur within very short time.
As dealers may revise their quotes based on news as well as on the quotes they receive from
other dealers, one may distinguish three different relations between news, order flow and the
exchange rate (see inter alia Evans and Lyons 2006a, Sager and Taylor 2006). First, the in-
formation content of announcements, that are publicly available and evaluated in a similar
manner by at least most of the traders (the ‘common-knowledge’ part of macro news), affects
prices directly and immediately. This channel directly corresponds to what one would call a
Page 2 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
2
traditional exchange rate model. Second, there may be news that are not unambiguous and
difficult to interpret or not publicly available, so there can be a component of news that is not
common-knowledge. In this case, private information is dispersed and transmitted to prices
via order flow (see also Evans and Lyons 2002a, Osler 2006). We refer to this channel as the
indirect effect of macro news intermediated by the induced order flow. Third, order flow un-
related to news announcements due to changing hedging or liquidity demands, or changing
risk tolerances (Evans and Lyons 2002a) may also affect the exchange rate.
The relations between the different sources of exchange rate variation are illustrated in Figure
1. The main difference between the microstructure approach and traditional asset price models
is therefore that trading is considered as a substantial part rather than an auxiliary activity to
price formation and hence deserves particular attention.
The recent empirical literature shows that particularly the first two channels of price impact
are present on foreign exchange markets: There is a vast empirical literature showing a sig-
nificant relationship between high-frequency exchange rates and news (Almeida et al. 1998,
Andersen and Bollerslev 1998a, Melvin and Yin 2000, Andersen et al. 2003, Bauwens et al.
2005) and between exchange rates and order flow (Lyons 1995, Yao 1998, Payne 2003,
Bjønnes and Rime 2005). More recent papers simultaneously analyze the relation between the
exchange rate, news and order flow.
Love and Payne (2008) conclude for major exchange rates that news affects both, the ex-
change rates as well as the order flow within a very short time, only one third of the news
content immediately being impounded to prices, a result that is roughly confirmed by Evans
and Lyons (2006a) for the DEM/USD rate. Both studies also find a significant increase of the
impact of order flow on prices around news announcements.
Evans and Lyons (2005) also find for the USD/EUR that the impact of news on the order flow
is stronger than on exchange rate returns. Furthermore, they identify delays in the direct im-
pact of news on the exchange rate and even longer ones for the impact on order flow. In con-
trast, Berger et al. (2005) find a very quick reaction of the order flow on news, within one
minute.
Cai et al. (2001) turn to the volatility of the JPY. They conclude that in 1998 order flow
played a more important role than news announcements and mainly contributed to the ex-
traordinarily high volatility of the yen. Dominguez and Panthaki (2005) come to the conclu-
sion that both, scheduled and unscheduled news affect the volatility of the USD/EUR and the
USD/GBP rates. Frömmel et al. (2008) find that both, the relation between order flow, news
and the volatility of the EUR/USD rate depend on the counterpart of one particular FX dealer.
While most of the existing literature focuses on exchange rates between the ‘big four’ USD,
EUR (DEM), JPY and GBP, only few papers deal with smaller developed markets, such as
the market for the Swedish (Lindahl and Rime 2006) or Norwegian krona (Rime 2001), but
Page 3 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
3
there are only very few papers on the microstructure of financial markets in Central and East-
ern European Countries (CEEC) (Hanousek and Podpiera 2003 on the Czech stock market;
Derviz 2003 and Scalia 2008 on the efficiency of FX interventions, Menkhoff and Schmeling
2008 on the Russian FX market). However, the small FX market of a transition economy may
show different characteristics than the markets for the major currencies. Besides the well-
known facts that transition economies are characterized by large capital flows and frequent
interventions by the central banks there are two main characteristics that are directly linked to
market microstructure: First, markets in transition economies are much smaller, thus less liq-
uid. Second, there may be more private information in the market. This may be due to the
lower news coverage, the low number of market participants, the more dynamic economic
environment or simply structural deficits in an evolving market. Both factors, liquidity and
private information, are key determinants of a markets microstructure.
Another strand of literature focuses on the impact of central bank interventions and communi-
cation on the exchange rate. A basic insight of the research on central bank interventions (for
surveys see e.g. Sarno and Taylor 2001, Vitale 2007) is that interventions are able to move the
exchange rate. They affect the first two moments of the exchange rate (Scalia 2008) and the
impact is usually stronger in emerging countries than in developed countries (Canales-
Kriljenko 2003). This may be due to less sterilization, the market's size and organization.
Besides direct interventions central bank communication may also be seen as a form of inter-
vention, that is – although less obvious at first sight – able to affect the exchange rate as well
(Ehrmann and Fratzscher 2007). There is a huge amount of research providing some ambigu-
ous empirical evidence on an impact of central bank communication on the exchange rate (see
the survey in Blinder et al. 2008). The impact of verbal interventions or communication stems
from their role in anchoring expectations on future monetary policy, i.e. the signaling or ex-
pectation channel of monetary policy (Sarno and Taylor 2001), but also by functioning as a
coordination advice for market participants (Reitz and Taylor 2008). Thus, communication
may complement intervention or substitute it (Fratzscher 2008).
Due to the more dynamic economic environment in a transition economy verbal interventions
may be more effective than in developed markets. However, again most of the work deals
with developed markets, mostly for the FOMC, the ECB, the Bank of England and the Bank
of Japan, and there are only few papers that focus on transition economies: Rozkrut et al.
(2007) find for the Czech Republic, Hungary and Poland, that speeches about monetary policy
affect the exchange rate. Égert (2007) finds influence of central bank communication for
Hungary, but not for other CEEC. He also concludes that the Hungarian National Bank
(MNB) used actual interventions very rarely, but mainly relies on verbal interventions.
The contribution of our paper to the literature is threefold:
Page 4 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
4
First, we focus on a small, less liquid market in a transition economy. Our aim is to analyze,
how much the results differ from those for the major foreign exchange markets. One could
expect a higher importance of order flow due to a higher share of private information and a
less stable and predictable economic environment.
Second, we use a broader data set of news, covering not only quantitative macroeconomic
announcements, but also communication variables. Again, the less stable economic environ-
ment, including uncertainty about monetary policy may lead some importance of central bank
communication, as it may anchor expectations. We also depart from most of earlier studies in
that we analyze the effects of the Hungarian quantitative news series individually and thus are
able to compare the importance of different news announcements.
Third, we had access to a database of outstanding quality containing very detailed data from
the Reuters D2000-2 trading system. It is substantially longer than most of those used in the
relevant literature on order flow (e.g. 8-month-long period of Love and Payne 2008, 4-month-
long periods of Evans and Lyons 2006a or Frömmel et al. 2008) and covers a period of two
years. We will use data with a frequency of one and ten minutes as well.
The outline of the paper is as follows: In section 1 we gave a brief overview of the role of
order flow in the trading process and recent empirical work. After describing the data in sec-
tion 2, we present the empirical results in section 3. The empirical analysis comprises four
subsections: First, we address the question whether scheduled macroeconomic news releases
have an immediate impact on exchange rate and on order flows. Second, we investigate the
exchange rate-order flow linkage and its sensitivity to news announcements. In section 3.3 we
decompose the news impact into a direct part and an indirect part transmitted by order flow.
Finally, we analyze the longer-term relationship between news, order flow and exchange rates
in section 3.4. Section 4 summarizes and concludes.
2. Data set
The data set used in this study covers the whole years of 2003 and 2004 and contains data on
the exchange rate, order flow and macroeconomic news release. Our whole data set consists
of three parts: the exchange rate data including the order flow, macroeconomic
announcements (quantitative data) and communication data (qualitative data). We will
describe them briefly in the subsequent subsections.
2.1 Order flow and exchange rate data
The data for the exchange rate and the order flow stems from the Reuters D2000-2 electronic
trading system
1
. As the competing electronic trading system EBS provides services on the
1
For a full description of characteristics and tendencies of spot FX market segments and the features of the
available data see Lyons (2001).
Page 5 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
5
HUF/EUR market only since February 2006 and the direct bilateral interdealer trading is
steadily loosing ground in the FX market (Rime 2003), our data set can be assumed to cover
most of the interdealer HUF/EUR market
2
.
This part of our data contains data on all the posted quotes (time-stamp, entry type, price,
quantity entered, quantity traded, etc.) and the executed trades (time-stamp, trade price,
quantity), but no information about the identity of market participants.
A number of additional advantages should be noted. First, with the help of „entry type”
variable, which can be regarded as a buy/sell indicator, the initiator of the trade can be
determined quite easily without having to make uncertain assumptions. Second, unlike in
many other papers (Lyons 1996, Love and Payne 2008) the size of the trades is known,
allowing us to calculate the order flow more accurately, without any distortion.
Following the literature (Andersen et al. 2002, Bauwens et al. 2005, Berger et al. 2008) we
exclude weekends and public holidays from the data set due to the lack of trades and quotes
during these days. For the same reason we drop the periods from 17:30 CET to 08:00 CET
next morning each day. There are few short time periods when the Reuters data feed was
broken, or other data problems occurred. We have removed these periods as well.
These data, which are accurate to the second decimal of the second, are then aggregated to
two different frequencies: We use 1-minute frequency to analyze (1) the impact of five
specific scheduled Hungarian quantitative news items on exchange rate returns and order
flow, (2) the linkage between order flow and exchange rate movements and (3) the joint
dynamics of exchange rate, order flow and pre-scheduled news. We use a 10-minute
frequency to explore the relationship between an extended set of news, order flows, exchange
rate returns and volatility. This distinction is due to the characteristics of the news variables
and will be discussed in detail later. From these aggregated quotes series we then calculate the
returns as 100 times the difference of the log prices.
The order flow is obtained by subtracting the seller-initiated traded quantities from buyer-
initiated traded quantities that occurred within the given interval of respectively 1 and 10
minutes. In the Reuters D2000-2 a trade may occur, because two limit orders are paired by the
system, or because a market order hits an existing limit order (which is the case in more than
75 per cent of all cases). In the first case the last limit order, the one that causes the trade, is
regarded as the initiating order, whereas in the latter case the market order is the initiating
one. This is in line with the microstructure literature (Berger et al. 2008).
2
For an extended description of the Reuters D2000-2 database, including a detailed description of data and a
comparison to other datasources, see Gereben and Kiss M. (2006), where the same database as here is used.
Page 6 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
6
In this way we get 293,550 observations for the 1-minute analysis and 29,355 for the 10-
minute analysis. Summary statistics of the return and order flow series are presented in Table
1.
2.2 Macroeconomic Announcements
This part of the news database consists of 5 quantitative news items containing the exact time,
the announced value and the market expectations for the included fundamentals. According to
previous examinations, to the results of the existing literature and to our intuition, four
Hungarian macroecomomic variables are selected (CPI, GDP growth, budget deficit and
current account deficit), whose effects on the HUF/EUR market can be regarded as relevant.
We also include the Monetary Council meetings in this subset, as base rate decision are
regarded as one of the most important sources information among fundamentals (Kiss M.
2004, Pintér and Wenhardt 2004).
3
For a brief description of the variables see Table 2.
We define the „news” component of these macro variables as the difference between the
actual values and the expectations, that are extracted from the regular monthly survey
conducted by Reuters Hungary among 15-25 market analysts. The only exception is the
Monetary Council’s rate decision. This question has become standard in the survey only
during 2005, so for our sample period no expectation data are available. So we use the change
in the short-term (3-months) government bond benchmark yield as the measure of surprise,
which is in line with the expectations hypothesis.
We use this surprise or news component of the macroeconomic announcements in absolute
terms as well as in a standardized manner, that enables us to compare the relative impact of
announcement categories. This approach has turned out to be standard in the empirical
literature (Balduzzi et al. 2001, Andersen et al. 2003, Love and Payne 2008):
[ ]

, ,
,
t i t i
t i
A E A
S
÷
= (1)
where
t i
A
,
is the announced value of data i, [ ]
t i
A E
,
is the expected value of data i, and oˆ is
the sample standard deviation of [ ]
t i t i
A E A
, ,
÷ .
4
3
This variables also includes meetings when the interest rate was unchanged. However, we excluded from the
examination the decisions of the extraordinary meetings, as they were announced unexpectedly under turbulent
market circumstances, therefore the market may be expected to react in a very different way than to scheduled
interest rate changes. In addition, we excluded the so-called non-ratesetting meetings, as well, because according
to the communicated policy the Monetary Council is not expected to make an interest rate decision at these
meetings.
4
In the empirical part of the paper we report results for both the absolute and the standardized measures of
surprise, where it is meaningful and has additional information content.
Page 7 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
7
An announcement is regarded as „positive/good” (positively signed) if it implies the
appreciation of the forint (i.e. lower-than-expected inflation or deficit figures, restrictive base
rate decisions) and is regarded as „negative/bad” analogously.
5
As all these mentioned news items are pre-scheduled, i.e. market participants know the hour
and minute of their release in advance and may react within a very short time, an analysis at
the high frequency of one-minute intervals seems to be most appropriate to capture their effcts
on the exchange rate. Our methodology is based on the approach of Love and Payne (2008),
but with some necessary modifications.
2.3 Communication variables
The second part of our news database includes qualitative and (mainly) non-scheduled news
items concerning central bank communication. We use news headlines reported on Reuters
screen and all kinds of written communication by the MNB to create the variables describing
central bank communication. We include only those items in the database containing some
kind of guidance about the future stance of monetary policy, and each statement counted only
one time, when it is mentioned first. Three main (monetary policy related) topics determine
the content of the MNB’s communication: exchange rate (for example, hints on the preferred
exchange rate level, intervention), policy rate (hints on the short-term interest rate path), eco-
nomic outlook.
Similarly as for the quantitative news in subsection 2.2, we transform the communication
dataset into 3 dummy variables, according to the "sign" of the news. Positive and negative
news sum up to approximately 73 per cent of the selected statements, while we consider the
rest of the communication as being neutral.
We take into account that the same statement can carry different meanings at different points
in time. We code the pieces of news from the market participants’ point of view, so the sign
of statements is determined in relation to market expectations. The comparison to market ex-
pectations is most important in case of interest rate communication, as a rising/declining in-
terest rate path was often present in expectations. Accordingly, we consider hints on a smaller
cut than the anticipated interest rate cut as tightening
6
.
We have performed the encoding of statements as follows: Four people compiled the data set,
and then two of them again checked the coding. Several people independently reviewed
5
Different theoretical exchange rate determination models differ in their predictions concerning the expected
direction of the exchange rate response of news surprises. For example, central bank reaction function models
predict an appreciation after higher-than-expected inflation data, while the same surprise is expected to cause
depreciation according to the portfolio-balance approach (for more details see Hoffman and Schlagenhauf 1985).
We used an event-study analysis for the classification, where the expected direction of the exchange rate
response was not unambiguous. Our results are consistent mainly with the portfolio-balance model.
6
As for the macroeconomic announcements, the difference between the three-month benchmark and the policy
rate is considered as the anticipated interest rate cut.
Page 8 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
8
doubtful cases to reduce the subjectivity of the coding as far as possible. This procedure al-
ready hints at this news category being a natural candidate for indirect price impact via order
flow as suggested by Evans and Lyons (2006a), due to its narrative and sometimes-
controversial character.
Only part of the communication (some pieces of written communication) is pre-scheduled.
Most of the communication news items appear without announcement on the Reuters news-
wire screen. This makes it difficult to determine the exact time when these pieces of informa-
tion arising from the central bank reach the market. For this reason, we conduct the part of the
analysis including communication variables at a lower, ten-minute frequency.
3. Empirical analysis
Due to the aforementioned attributes of our news variables the empirical analysis embraces
two different time aggregation frequencies – 1-minute and 10-minute intervals –, and
according to characteristics of the specified approaches we used different methodological
approaches – event study, VAR and conventional regression analysis – to analyze how news,
order flow and exchange rate depend on each other. We proceed in five steps:
First, as a preliminary exercise we analyze the direct impact of news announcements on the
HUF/EUR rate, without considering the role of order flow.
Second, we investigate the relation between the exchange rate and order flow, and whether
this relation changes during news release times.
Third, we decompose the impact of news on the exchange rate into a direct and an indirect
effect by means of a VAR. The first three steps are performed in line with Love and Payne
(2008) and applied to data at the frequency of one-minute intervals.
Fourth, we add the communication variables to our system and turn to a lower frequency of
ten-minute intervals. This approach covers effects on exchange rate returns as well as on the
volatility of exchange rates and is in line with, inter alia, Cai et al. (2001), Dominguez and
Panthaki (2006) and Frömmel et al. (2008).
3.1 The immediate impact of macroeconomic news on exchange rate and order flows
Following the approach by Love and Payne (2008), as a first step, we analyze the impact of
the surprise content of macroeconomic announcements on exchange rate returns and on order
flows separately. Therefore, we begin our empirical analysis by testing two hypotheses,
namely whether the surprise component of the included announcements has significant
influence on the HUF/EUR exchange rate (equation 2) and on order flow (equation 3). We
follow the commonly used approach by Andersen et al. (2003) and Berger et al. (2008) and
separately estimate the following equations for exchange rate returns and order flow:
Page 9 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
9
n t m n i t
l
k i
p
m i
p
n t
N p
, , , , ,
c | o + + = A
÷
=
¯
(2)

n t m n i t
l
k i
of
m i
of
n t
N OF
, , , , ,
n | o + + =
÷
=
¯
(3)
where
n t
p
,
A is 100 times the logarithmic change in the exchange rate from time t-1 to time t
on day n,
n t
OF
,
is order flow in the same interval, in million euro, and
m n t
N
, ,
is the surprise
component of news category m,
n t ,
c and
n t ,
n are the error terms.
7
We show results for each news category (CPI, GDP growth, C/A, budget deficit, base rate
decision) and for absolute as well as standardized news measures separately for a short event
window with k = 3, and with l = -1 minutes
8
.
In Table 3 we report estimates for specifications with the absolute news measures. At first
glance, the overall picture is that most of the macro news have a significant immediate impact
on returns as well as on the order flow. The comparatively low R
2
imply that macroeconomic
information accounts for only a small proportion of exchange rate variation. However, this is
partly due to the small number of news releases relative to the number of observations and in
line with the literature (Love and Payne 2008, Andersen et al. 2003).
Four out of the five announcements significantly affect both, the exchange rate as well as the
trading pattern. In the majority of cases the coefficients are significant and well-signed in the
minute(s) right after the announcement. The only exception are GDP releases, for which –
potentially due to the only eight GDP data releases in our sample – our data do not give
evidence of impact of GDP news neither on returns nor on order flow within short time. The
effect of GDP news, if there is any, must be slight, disperse and slowly impounded.
For the other news categories (CPI, budget deficit, C/A and monetary council meetings) we
find similar patterns: there is an immediate impact on the return as well as the order flow,
with the expected sign on the coefficients. For instance, for CPI surprises we find that 1
percent higher than expected inflation causes almost half a percent depreciation of the
Hungarian currency, and euro purchases/forint sales worth 17,2 million euro on average in the
7
We use OLS for our estimations and correct the coefficient variance/covariance matrix using Newey-West
standard errors to compensate for autocorrelation and heteroscedasticity. This does, however, not substantially
change the overall picture.
8
Actually, we estimated the equations for k = 60 and l = -10. There were, however, few significant coefficients
and no clear pattern. Furthermore we only show the results for the absolute measure as those for the relative
measure are basically the same. We only refer to them in the text if necessary. All results not shown here are
available from the authors on request.
Page 10 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
10
minute immediately following the release. For CPI, C/A and the Monetary Council meetings
we also find that the effects of a surprise last for several minutes.
9
The standardized measures indicate that CPI surprises have the strongest impact: One „unit”
of CPI surprise implies about more than 4 million euro net one-sided trading. Furthermore, it
causes an almost three times higher change in the exchange rate, than the second powerful
one, the base-rate decision surprise. This latter even has double influence on exchange rate
than budget deficit and current account deficit surprises, which have approximately the same
effect on the exchange rate in terms of the size of the impact.
3.2 The sensitivity of the exchange rate-order flow relationship to news release times
In the preceding analysis we showed that news about fundamentals do not only influence
exchange rate determination, but also trigger net buying or selling pressure on the currency in
accordance with the sign of the surprise. Thus, one should expect that order flow is correlated
with exchange rate returns, at least around news announcements. Therefore, we turn to the
examination of the potential connection between order flow and exchange rate.
There are theoretical justifications for both, a weaker as well as a stronger connection between
the exchange rate and order flow when news are published:
If the new information is unambiguous and is interpreted similarly by most of the market
participants, then trades convey less information about the fundamental value of the asset in
the presence of news. Market makers may then update their beliefs and change their quotes
according to this unambiguous information instead of depending upon the noisy trading. This
attitude can cause a temporary disconnection between exchange rate and order flow. In
addition, as liquidity and trading activity are assumably higher around news releases (see
Chaboud et al. 2004, Fleming and Remonola 1999), due to the larger volumes placed in the
order book, a unity of trade might cause less price movement.
As the effect of order flow on exchange rate originates in conveying private or dispersed
information, the association between order flow and exchange rate may be quite different in
the immediate aftermath of macro news releases. On the other hand, if traders disagree about
the interpretation of a given piece of news, well-informed or „smart” market participants are
believed to trade more intensively around news releases. In this case dealers pay more
attention to trading activity, and increase the probability of private information content of
one-sided flows. Under these circumstances a positive (negative) order flow more probably
indicates that the informed traders consider the asset more valuable (less valuable) than its
9
In few cases, even a lead is significant, which may indicate position taking in advance (Monetary Council
meetings) or just relate to the early release time of 8:30 a.m. (C/A) when the trade intensity is low (Gereben and
Kiss M., 2006).
Page 11 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
11
actual market price. Moreover, the previously demonstrated results from subsection 3.1
suggest a stronger connection as well.
Thus we estimate the following equation:
n t m t n t
k
k i
m i n t m n t
i I OF OF p
, , , , , ,
) ( n ¸ | o + · · + · + = A
¯
÷ =
(4)
where
m t
i I
,
) ( is a dummy variable indicating that a news release concerning macroeconomic
variable m occured in period t. We adopt an event-study approach and estimate equation (4)
only on announcement days, broken down by announcement type.
10
The approach follows
Love and Payne (2008), but we introduce additional dummies for different news categories.
We then compare the results with a benchmark equation without the interaction between order
flow und news releases.
Regarding the results of the benchmark, we can now compare, whether the explanatory power
of the estimation can be improved by letting the coefficients of the order flow differ from its
usual level around news releases, and differ by announcement types. We set k to 5; more
specifically under the expression „around news releases” we mean the five minutes before and
after the announcements. Our estimation results for the benchmark regression and the five
announcements separately are presented in Table 4.
11
First, order flow plays a significant role in exchange rate determination in the Hungarian FX
market and corroborates previous results on the HUF/EUR market obtained from daily data
(Gereben et al. 2006) as well as the literature on other exchange rates (see inter alia Evans and
Lyons 2002a). The parameter of order flow is positive and highly significant, the R
2
is 16%.
12
A net excess of forint buyer-initiated trades is worth 1 million (in the base currency of euro)
and leads to about 0.01 percent appreciation of the forint (Table 4).
Second, as expected, the order flow parameter significantly changes at announcement times. It
is significantly positive for all news categories, and interacts with the presence of news, also
for all news categories (Table 4). This means that order flow conveys additional private
information about the fundamentals in the minute, when scheduled public information reaches
10
We also estimated equation (4) only for announcement days, broken down by announcement type (Berger et
al. 2008) apply this event-study approach similarly, but do not use dummy variables and estimate the regressions
only in the minutes after the releases.
11
As well as in the previous section, due to convenience reasons we only report the contemporaneous results.
Leads and lags are hardly significant. The detailled results are available on request.
12
However, it should be noted that the statistical power of the benchmark regression is lower to some extent
(R
2
=16%), than is commonly found in the empirical literature. This is probably due to the 1-minute frequency
and the relatively low trading activity in the Hungarian FX market compared to the big markets (EUR/USD,
JPY/USD etc.), as on average there is only one trade every three minutes. This view is confirmed by the fact that
the R
2
was particularly low during the first year of our sample, when trading activity was much lower on the
HUF/EUR market (Gereben and Kiss M. 2006) and speculative attacks against the band took place.
Page 12 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
12
the market. The substantially increased R
2
show a close relation between order flow and the
exchange rate.
This result, together with the previously found conclusion, inidcates that part of the total
impact of new public information is transmitted to exchange rates partially via order flow. In
the following we further explore this issue.
3.3 Decomposing the channels of the news’ impact – A multivariate VAR analysis
Now we carry on with the decomposition of exchange rate responses to news into direct and
indirect (intermediated) parts, and try to measure the contribution of order flows to total
exchange rate responses to news shocks. For analysing this issue and testing the order flow
intermediating role we estimate a VAR model for exchange rate changes and order flows, as
follows.

A
t
t
OF
P
=
t
k
j
j t l j l
l
l
i i t
i t
i t
N
OF
P
OF c
|
o + O +

A
I + ·

+
¯ ¯ ¯
=
÷
= = ÷
÷
0
, ,
6
1 1
0
(5)
In this model, we allow contemporaneous order flows to have an effect on the price
quotations, but the flows are not affected by the contemporaneous returns. This assumption
seems to be a reasonable economic restriction, as the market is less liquid than the markets for
major currencies, and within the 1-minute frequency we are analysing the number of trades is
very limited. So we do not expect almost simultaneous feedback trading, which would allow
the order flow to respond to price movements at a frequency of less than a minute. We further
assume, that the structural form innovations, (
1,t
,
2,t
) are independent. This allows us to
estimate the equation separately by OLS.
Table 5 summarizes the results and for brevity we only report the results for the first five lags,
the contemporaneous order flow and the news.
13
There is a relatively strong and persistent
high-frequency negative autocorrelation in returns, while the high-frequency order flow
exhibits positive autocorrelation. The contemporaneous order flow affects the returns as
expected: a net forint purchase of 1 million euros goes along with an appreciation of the forint
by 0.94 basis points. Lagged flows have a similar but smaller effect. The order flow exhibits
positive dependency on recent returns up to 8 lags: forint appreciation leeds to forint
purchases and vice versa, which can be interpreted as an indication of positive feedback
trading in the market. News affect the order flow significantly: all types but the GDP
announcement have a significant contemporaneous effect, and several types of news affect the
flows for several minutes. The coefficients are in line with intuition: positive surprises (good
news) lead to an increase in net forint purchases, and with one exception the lagged effects
have the same sign. The direct effect of news on the exchange rate is less pronounced: only
13
The lag lengths were selected according to the Schwartz information criterion.
Page 13 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
13
the surprise content of the CPI announcement has a significant effect. But in this case the
effect is strong: a one-standard deviation positive surprise causes the HUF/EUR rate to
appreciate by 27.34 basis points.
To highlight the particular effect of CPI surprises as the obviously most effectful news
category, and the role of the order flow, Figure 2 shows the (cumulative) impulse responses of
the exchange rate and the order flow to a CPI announcement. We follow Love and Payne
(2008) and compare the actual IRF of the exchange rate returns to a hypothetical IRF, which
would prevail if the news had no affect on the order flow.
14
The hypothetical impulse
response shows the direct effect of the news shock on the exchange rate returns, and the
difference between the actual and the hypothetical impulse responses the indirect impact via
the order flow.
The share of the direct and indirect components is given in Table 6. At the time of the
announcement the direct channel dominates, 93.7% of the change in the exchange rate can be
attributed to the news announcement directly. The share of the order flow has in the
impounding the news into the prices increases rapidly over the time: 5 minutes following the
announcement it amounts to almost 60%, and stabilizes around 57%, indicating that order
flow indeed plays a substantial part in processing news, even in the presence of a strong direct
effect.
To the best of knowledge we are the first who do this kind of decomposition for an emerging
market, so the comparison of our results is possible only with developed markets. Evans and
Lyons (2006a) found that around two-thirds of news’ total effect can be linked to trading
activity in the DEM/USD market. They rejected the test for whether the direct channel is
more important than the indirect channel, concluding order flow accounts for at least half of
the effect of macro news on exchange rate. Love and Payne (2008) examined three big FX
markets and showed that on average one third of the new information is transmitted via order
flow. In general, our result is in line with the findings of the existing literature, although the
share that order flow has in incorporating news into prices is even higher in our case than it
was found concerning major FX markets.
3.4 Longer-term interdependence of news, order flow and exchange rate
As the inclusion of communication variables requires a ten-minute frequency we move on to
explore the longer-term relationship between exchange rate, order flow and fundamentals.
3.4.1 Methodology
14
In our framework this means computing IRF from a model where the coefficient in the order flow equation
is restricted to 0. This hypothetical model still allows for the order flow to influence the exchange rate, but
assumes that neither the size of the order flow nor its effect on the returns changes following a news
announcement.
Page 14 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
14
We measure volatility following the approach by Andersen and Bollerslev (1998a), which has
become standard in high-frequency analysis and is commonly used in the recent literature (see
inter alia Cai et al. 2001, Dominguez and Panthaki 2006, Frömmel et al. 2008) and calculate
the volatility of ten-minute returns. The approach is based on the model
t n , t n n , t
z p · · = ÷ o o µ A (6)
where µ is the unconditional mean of the return process, o
n
is the contribution of the daily
volatility level to the return volatility and o
t,n
consists of high-frequency contributions, in par-
ticular an intraday pattern, day of the week effects, which are reported for exchange rates in
previous empirical work (see inter alia Andersen and Bollerslev 1998, Cai et al. 2001), and
the impact of various kinds of news.
After the equation is squared and expressed in logs it evolves to
n , t n , t
4
1 k
k k
n
n , t
n , t
u X day ) t ( p c
p
log 2 h + · + · + + =
÷
· =
¯
=
o ¸
o
µ A
(7)
where p(t) is the above mentioned intraday volatility pattern, day
k
are dummies capturing po-
tential day-of-the-week effects and X
t,n
is a set of further determinants, including news, cen-
tral bank activities and order flow, which will be specified later.
The daily volatility level o
n
is usually achieved by estimating a GARCH model based on daily
data
15
. While commonly a standard GARCH(1,1) model with normally distributed conditional
variances is used, we apply a model with t-distributed errors. We do this because it is known
(see e.g. Hughes Hallett and Anthony 1997, Frömmel and Menkhoff 2001) that managed ex-
change rates as the HUF/EUR show a much higher kurtosis than the flexible DEM/USD rate
analyzed by Andersen and Bollerslev (1998a). In this case the use of a conditionally normally
distributed GARCH-process may lead to the underestimation of the “true” volatility. The re-
sults of the daily volatility and the intraday volatility estimations are illustrated in Figure 3.
The estimated conditional volatility shows a huge increase in January and again in June 2003,
when the market was characterized by turmoil due to speculative pressure on the forint.
The intraday pattern is simpler than the one usually found for the USD/EUR or USD/DEM
market
16
. However, the figure shows a clear pattern. One can easily distinguish the start from
a very low activity level in the morning, the peak between 9:00 and 9:20 a.m., when Hun-
15
Andersen and Bollerslev (1998b) show that even in the case of a continuous-time price process the GARCH-
model provides accurate forecasts. As there are only daily observations necessary, overnight breaks are no prob-
lem.
16
For the intraday pattern we use a HP filter, which describes the pattern sufficiently well. We did, however, fit
several possible trend functions and concluded that no one figures out the data better. As Melvin and Yin (2000)
point out, if the pattern is simpler than the one in, for instance, Anderson and Bollerslev (1998) there is a priori
no need to work with the more complex flexible Fourier form, and may consider using a simpler pattern.
Page 15 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
15
gary’s main macroeconomic news are usually published and trading starts getting more active,
the subsequent slowdown during lunchtime and the repeated increase in the afternoon when
the US markets open.
For analyzing the effects of news on volatility we estimate the following system of equations,
taking into account the relation between order flow (OF) and volatility (h):
n t
h
K
k
m
m l
n l t k l k
i
n i t i n h t h n t
n t
h
K
k
m
m l
n l t k l k
j
j j
i
n i t i n h t h n t n t
N h OF OF
N day h OF OF t p h
,
2
1 1
, , ,
2
1
, , 2 ,
,
2
1 1
, , ,
4
1
2
1
, , , 0 1 ,
) (
n o ¸ | e
c o . ¸ | | o e
¯ ¯¯ ¯
¯ ¯¯ ¯ ¯
= = ÷ =
+
=
÷ ÷
= = ÷ =
÷
= =
÷ ÷
+ + + + =
+ + + + + + + =
(8)
where the variables N
k,t,n
cover, as in the previous section, various types (k) of news as well as
lags. The news variables may affect order flow as well as volatility. Again, the volatility de-
pends contemporaneously on order flow, whereas order flow itself may react to volatility with
a delay of one interval only. Note that we use absolute order flow in the volatility equation, as
volatility is a one-sided measure.
The equation looks similar for the conditional mean. There are, however, some small differ-
ences. First, we refer to the signed order flow instead of the absolute one, as the return shows
a distinctive direction and so does the signed order flow (positive values of the order flow
mean that market participants seek to buy forint and sell euro). Second, there is no need to
include day of the week effects and an intraday pattern. Therefore the equations (8) evolve to:
n , t
2
1 h
K
1 k
m
m l
n , l t , k l , k
2
1 i
n , i t i n , h t h 2 n , t
n , t
2
1 h
K
1 k
m
m l
n , l t , k l , k
2
1 i
n , i t i n , h t h n , t 0 1 n , t
N p OF OF
N p OF OF p
n o A ¸ | e
c o A ¸ | | e A
¯ ¯ ¯ ¯
¯ ¯ ¯ ¯
= = ÷ =
+
=
÷ ÷
= = ÷ =
÷
=
÷ ÷
+ + + + =
+ + + + + =
(9)
In the case of the return analysis, including lagged returns to the order flow equation is even
more appealing as it explicitly allows for the existence of feedback trading, i.e. market par-
ticipants react on recent price changes for exploiting (potentially small and short) trends.
3.4.2 Results
The estimation results for the equations (8) are given in Table 6. It is obvious that in the order
flow equation there is a strong negative relation between present absolute order flow and
lagged volatility, and at the same time there is positive autocorrelation of the absolute order
flow. Similarly, in the volatility equation positive autocorrelation of the volatility and some
negative relation is present between the volatility and lagged absolute order flow. Further-
Page 16 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
16
more, a highly significant intraday pattern can be observed in volatility. We do not find a day-
of-the-week effect.
These results are in line with the empirical literature. Intraday volatility patterns are a well-
known feature of FX markets, as well as positive autocorrelation observable in financial
variables, such as volatility, order flow, etc. (Andersen et al. 2003, Bauwens et al. 2005,
Frömmel et al. 2008). The negative relation between (present/lagged) order flow and
(lagged/present) volatility is reasonable, as well. Both variables can be regarded as a measure
of market uncertainty about an asset’s fair value. If the volatility is increasing showing higher
uncertainty, then trades will be more probably symmetrically initiated by sellers and buyers,
and the chances of one-directional pressure on the asset (which is shown by absolute order
flow) should be less in the following period(s). Similarly, the higher selling/buying pressure
occurs in the market, that means, the more traders (the "smarters" or who has private
information) agree on the overvalued/undervalued price of the asset, which may cause a
decrease in the uncertainty about an asset’s value.
Monetary policy meetings, especially when the base rate is changed, apparently significantly
affect both, absolute order flow and volatility. Prior to the meetings, the absolute value of
order flow, i.e. the one-sided trading activity, and volatility increase and decrease after the
meeting. We interpret this as position taking in expectation of the meeting’s decision.
Whereas the variable meeting does not distinguish, whether there was an interest rate deci-
sion, this is captured by the following variables. Accordingly, a change in the base rate causes
the buying or selling pressure to be higher after the meeting, which reflects the position taking
due to the new interest rate settings. In contrast, there is only little impact on order flow be-
fore the interest rate decision. Interestingly, volatility tends to be low prior to interest rate
cuts.
Macro news announcements do not imply aggressive one-sided trading. The only exception is
the interval 20 minutes after positive news has been published, when the one-sided trading
seems to be more intense than during other times. This phenomenon can be due to the lower
frequency, as we have seen previously that the impact of news releases on order flow emerges
within very short time. Thus, some of the effects may have already been offset. In contrast,
there is a measurable direct effect of news on volatility. The impact is especially strong in the
case of negative news items: while volatility decreases before the releases, as everybody ex-
pects the figures, volatility increases considerably contemporaneously and after the an-
nouncements, as the surprise triggers heavy trading. The reaction on positive news is less un-
ambiguous, but still consistent with our previous results: there is an increase in volatility only
10 minutes after the publication.
It seems that there is no clear-cut tendency concerning the effects of central bank communica-
tion. The significance and the sign of the parameters vary along the relative time and the type
of the communication. Moreover, as central bank communication is not pre-scheduled in most
Page 17 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
17
of the cases, it is not expected to have significant impact on the market prior to the events.
Thus, we can draw just a cautious conclusion: Four significant and relevant parameters indi-
cate that central bank communication (essentially negative communication) might decrease
the uncertainty in the market.
The intensity of trading, measured with the number of deals conducted, is positively related
both to the absolute order flow and volatility. The explanatory power of the order flow regres-
sion is very impressive with a R
2
higher than 65%, that means two thirds of the market pres-
sure from asymmetrically initiated trades can be explained by fundamentals, such as macro
news, monetary policy decisions and communication, and lagged/contemporaneous values of
volatility, absolute order flow and trades variables. Meanwhile in the case of volatility regres-
sion the R
2
shows that the mentioned variables account for 16% of price variability, which
can be regarded as relatively high, as well.
The estimation results for the returns are presented in Table 7. The order flow equation shows
significant negative autocorrelation of the order flow, and a strong negative relation between
present order flow and lagged returns. This may be caused to some extent by negative feed-
back trading present on the Hungarian FX market.
17
On the other hand, returns are negatively
autocorrelated, while there is significant positive (negative) relation between the returns and
the contemporaneous and first (second) lag order flow.
Monetary policy meetings can coincide with either appreciation or depreciation after the
meetings depending on the relative time, and around the meetings net forint purchases are
observable. One has, however, to keep in mind that the variable meeting captures very differ-
ent meetings, including those with and without decision, which may be surprising or not.
Again, a look at the interest rate decisions seems to be more informative, as interest rate
changes significantly affect both the returns and the flows.
Interest rate cuts mainly coincide with depreciation of the forint, the only exception is 10-
minute lead prior to the decision. The impact of rate cuts on order flows is not fully consistent
over time: both buying and selling pressure can be observed. In contrast, the impact of interest
rate hikes on order flow is unambiguous: hikes cause aggressive one-sided trading on the
HUF buy side. In accordance with this, if the Monetary Council increases the base rate, the
forint tends to appreciate. The reaction on interest rate changes is therefore almost fully con-
sistent with theory, and shows a close relation between the news and both, order flow and
volatility.
The results on macroeconomic announcements are also mainly consistent with theory: The
strongest reaction follows positive news on the Hungarian economy, i.e. news that are ex-
pected to cause an appreciation of the forint. They are linked, with a lag of about ten minutes,
17
Csávás and Varga (2006) find that depending on the sample period both negative and positive feedback
trading is observable in the Hungarian FX market.
Page 18 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
18
to buying pressure and accordingly with appreciation of the forint. Together with results from
the previous section it is shown that news do not show only a very short-lived effect on vola-
tility during the most recent minutes after publication of the news, but also have an ongoing
impact on the exchange rate after publication of the news. The impact of negative news is also
visible, but less strong. Whereas the order flow shows significant sales after the news, there is
only a slight, albeit not significant depreciation of the forint. Therefore we find a strong
asymmetric behavior of market participants.
Central bank communication seems to affect the trading process more intensively, than the
exchange rate movements. Though, as it could be seen earlier, it is also true in this equation
system, that the significance and the sign of the parameters differ depending on the relative
time and the type of the communication. This is in line with the in general more ambiguous
character of this “news” category and the fact that they are not scheduled
18
.
Positive communication, however, appears to imply forint purchases and appreciation. There
seems to be some kind of overshooting, as part of the initial purchases are offset within the
next ten minutes. Neutral communication implies some negative order flow and depreciation,
while negative communication goes along with forint sales. The reaction of the exchange rate
is only slightly significant, but wrong-signed.
The results on the communication variables demonstrate the importance of central bank
communication for the HUF/EUR exchange rate and are in line with Rozkrut et al. (2007) and
Égert (2007).
We experience a relatively low explanatory power of the order flow equation, but on the con-
trary the fit of the exchange rate return equation may be considered rather good (R
2
=22%).
Together with the previously found consequences these results imply that macroeconomic and
microstucture variables together can explain a considerable part of exchange rate movements
and the characteristics of the trading activity on the HUF/EUR market.
4. Conclusions
During recent years the analysis of order flow has become an important issue, being exten-
sively debated in academia as well as by policy makers.
Our study adds to this discussion by complementing earlier papers. While most of the earlier
studies focus on exchange rates between the world’s major currencies, we consider the ex-
change rate of a small open economy, characterized by a less liquid market and comparatively
high capital flows, and raise the question, whether the role of order flow in this market is the
18
The latter point makes it even difficult to explain the significance of leads, but this may also be due to the
difficult timing of communication events.
Page 19 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
19
same as for the major ones. To make the results comparable we conduct our analysis in the
standard framework of order flow analysis.
Our results regarding the exchange rate, macroeconomic news and order flow indicate that the
HUF/EUR market behaves qualitatively in a very similar way as major and developed mar-
kets in the sense that we find a strong reaction to news, but also the strong relation between
the exchange rate and order flow as reported in the literature. Order flow itself can explain a
significant part of exchange rate variation. In addition, the order flow – exchange rate relation
changes significantly around news releases, namely order flow has additional information
content and stronger impact on the exchange rate (i.e. the same magnitude buying/selling ex-
cess causes larger appreciation/depreciation), when scheduled public information reaches the
market.
Overall, we find basically the same results as reported in the literature for developed foreign
exchange markets. We find a relation between order flow, news and the exchange rate that
does not differ in its basic structure. This is even more remarkable as the HUF at this time
was pegged to the EUR and our sample covers a period of strong pressure on the forint.
However, although the results are qualitatively the same, they are not comparable in
quantitative terms. Not surprisingly the importance of order flow seems to be higher than it
was found in major currency markets. Decomposing exchange rate responses to news into
direct and indirect (intermediated) parts we quantify the contribution of order flows to total
exchange rate responses to news. Our results suggest that three quarters of total price reaction
are intermediated by order flow, whereas the share for major markets is usually estimated as
30-50 per cent.
The differences are even more striking for the communication variables. In contrast to most
other currencies they seem to play an important role in the HUF/EUR market, reflecting the
experience that the MNB relied on verbal intervention rather than on actual interventons in
most times (Egert 2007).
Both differences of the HUF/EUR market – higher importance of order flow and relevance of
communication – may be attributed to the presence of private information due to the factors
mentioned in the introduction.
Page 20 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
20
References:
Almeida, Alvaro, Charles Goodhart and Richard Payne (1998). The Effects of Macroeco-
nomic News on High Frequency Exchange rate Behavior, Journal of Financial
and Quantitative Analysis 33:3, 383-408.
Andersen, Torben and Tim Bollerslev (1997). Intraday Periodicity and Volatility Persistence
in Financial Markets, Journal of Empirical Finance 4, 115–158.
Andersen, Torben and Tim Bollerslev (1998a). Deutsche Mark-Dollar Volatility: Intraday
Activity Patterns, Macroeconomic Announcements and Longer Run Dependen-
cies, Journal of Finance 53:1, 219-265.
Andersen, Torben and Tim Bollerslev (1998b). Answering the Sceptics: Yes, Standard Vola-
tility Models Do Provide Accurate Forecasts, International Economic Review
39:4, 885-905
Andersen, Torben, Tim Bollerslev, Francis X. Diebold and Clara Vega (2003). Micro Effects
of Macro Announcements: Real-Time price Discovery in Foreign Exchange,
American Economic Review 93:1, 38-62.
Balduzzi, Pierluigi, Edwin J. Elton and T. Clifton Green (2001). Economic News and Bond
Prices: Evidence from the U.S. Treasury Market, Journal f Financial and Quanti-
tative Analysis, 36, pp. 523-543.
Bandi, Federico M. and Jeffrey R. Russel (2003). Microstructure Noise, Realized Volatility
and Optimal Sampling, unpublished manuscript, University of Chicago.
Bauwens, Luc, Walid Ben Omrane and Pierre Giot (2005). News Announcements, Market
Activity, and Volatility in the Euro/Dollar Foreign Exchange Market, Journal of
International Money and Finance 24, 1108-1125.
Berger, David W., Alan Chaboud, Sergey Chernenko, Edward Howorka, Raj S.K. Iyer, David
Liu and Jonathan H. Wright (2008). Order Flow and Exchange Rate Dynamics in
Electronic Brokerage System Data, Journal of International Economics, forthcom-
ing.
Bjønnes, Geir Høidal and Dagfinn Rime (2005). Dealer Behavior and Trading Systems in
Foreign Exchange Markets, Journal of Financial Economics 75(3), 571-605.
Blinder, Alan S., Michael Ehrmann, Jakob de Haan, Marcel Fratzscher and David-Jan Jansen
(2008). Central Bank Communication and Monetary Policy: a Survey of Theory
and Evidence, ECB Working Paper no. 898.
Cai, Jun, Yan-Leung Cheung, Raymond S.K. Lee and Michael Melvin (2001). ‘Once in a
Generation’ Yen Volatility in 1998: Fundamentals, Intervention, and Order Flow,
Journal of International Money and Finance 20, 327-347.
Canales-Kriljenko (2003). Foreign Exchange Intervention in developing and Transition
Economies: Results of a Survey, IMF Working Paper 03/95.
Csávás, Csaba and Lóránt Varga (2006). Main characteristics of non-residents’ trading on the
foreign exchange and government bond markets, MNB Bulletin volume 1., issue 1.
Derviz, Alexis (2003). FOREX Microstructure, Invisible Price Determinants, and the Central
Bank’s Understanding of Exchange Rate Formation, CNB Working Paper 6/2003
Dominguez, Kathryn M.E. and Freyan Panthaki (2006). What defines ´News` in the Foreign
Exchange Markets?, Journal of International Money and Finance 25, 168-198.
Page 21 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
21
Égert, Balázs (2007). Central Bank Interventions, Communication and Interest Rate Policy in
Emerging European Economies, Journal of Comparative Economics 35(2) 387-
413.
Ehrmann, Michael, and Marcel Fratzscher (2007). Communication by Central Bank
Committee Members: Different Strategies, Same Effectiveness?, Journal of
Money, Credit, and Banking, 39(2-3): 509-41.
Evans, Martin D.D. (2002). FX Trading and Exchange rate Dynamics, Journal of Finance 57,
2405-2448.
Evans, Martin D.D. and Richard K. Lyons (2002a). Order Flow and Exchange Rate Dynam-
ics, Journal of Political Economy 110, 170-180.
Evans, Martin D.D. and Richard K. Lyons (2005). Do Currency Markets Absorb News
Quickly? Journal of International Money and Finance 24, 197-217.
Evans, Martin D.D. and Richard K. Lyons (2006a). How is Macro News Transmitted to Ex-
change Rates?, NBER Working Paper No. 9433.
Evans, Martin D.D. and Richard K. Lyons (2006b). Understanding order flow
International Journal of Finance & Economics 11:1, 3-23.
Evans, Martin DD, Lyons, Richard (2002b). Time varying liquidity in foreign exchange,
Journal of Monetary Economics, 49, 1025-1051.
Fleming, Michael J. and Eli M. Remonola (1999). Price Formation and Liquidity in the U.S.
Treasury Market: The Response to Public Information, Journal of Finance 54, 1901-
1915.
Fratzscher, Marcel (2008). Oral Interventions Versus Actual Interventions in Fx Markets – An
Event-Study Approach, Economic Journal 118:530, 1079-1106.
Frömmel, Michael and Lukas Menkhoff (2001). Risk Reduction in the EMS? Evidence from
Trends in Exchange Rate Properties, Journal of Common Market Studies, 39, 261-
281.
Frömmel, Michael, Alexander Mende and Lukas Menkhoff (2008). Order Flows, News, and
Exchange Rate Volatility, Journal of International Money and Finance 27:6, 994-
1012.
Gábriel, Péter and Pintér, Klara (2006). The effect of the MNB’s communication on financial
markets, MNB Working Papers 2006/9
Gereben, Áron, and Norbert Kiss M. (2006). A brief overview of the characteristics of inter-
bank forint/euro trading, MNB Bulletin volume 1., issue 2.
Gereben, Áron, György Gyomai and Norbert Kiss M. (2006). Customer order flow, informa-
tion and liquidity on the Hungarian foreign exchange market, MNB Working Papers
2006/8
Hanousek, Jan and Richard Podpiera (2003). Informed Trading and the Bid-Ask Spread: Evi-
dence from an Emerging Market, Journal of Comparative Economics 31, 275-296.
Hoffmann, Dennis and Don Edward Schlagenhauf (1985). The Impact of News and Alterna-
tive Theories of Exchange Rate Determination, Journal of Money, Credit and Bank-
ing, 17, 328-346.
Hughes Hallett, Andrew and Myrvin L. Anthony (1997). Exchange Rate Behaviour under the
EMS Regime: Was there any Systematic Change?, Journal of International Money
and Finance, 16, 537-560.
Page 22 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
22
Kiss M., Norbert (2004). The Effects of Macroeconomic News on Money Markets, MNB Oc-
casional Papers 2004/30
Kyle, Albert S. (1985). Continuous Auctions and Insider Trading, Econometrica 53:6. 1315-
1335.
Lindahl, Anna and Dagfinn Rime (2006). Forecasting the Foreign Exchange Market Using
Private Information, unpublished manuscript.
Love, Ryan and Richard Payne (2008). Macroeconomic News, Order Flows and Exchange
Rates, Journal of Financial and Quantitative Analysis , forthcoming.
Lyons, Richard K. (1995). Tests of microstructural hypotheses in the foreign exchange mar-
ket. Journal of Financial Economics 39, 321-351.
Lyons, Richard K. (2001). The Microstructure Approach to Exchange Rates, MIT press.
Melvin, Michael and Xixi Yin (2000). Public Information Arrival, Exchange Rate Volatility,
and Quote Frequency, The Economic Journal 110, 644–661.
Menkhoff, Lukas and Maik Schmeling (2008). Whose Trades Convey Information? Evidence
from a Cross-Section of Traders, Journal of Financial Markets, forthcoming.
Osler, Carol L. (2006). Macro lessons from microstructure. International Journal of Finance
& Economics 11:1, 55-80.
Payne, Richard (2003). Informed Trade in Spot Foreign Exchange Markets: An Empirical
Analysis, Journal of International Economics 61, 307-329.
Pintér, Klára and Wenhardt Tamás (2004). A jegybanki kamatdöntések elrejelezhetsége és
hatása a hozamokra (only in Hungarian), MNB Occasional Papers 2004/31
Reitz, Stefan and Mark P. Taylor (2008). The Coordination Channel of Foreign Exchange
Intervention: A Nonlinear Microstructural Analysis, European Economic Review
52:1, 55-76.
Rime, Dagfinn (2001). Private or Public Information in Foreign Exchange Markets? An Em-
pirical Analysis, SSRN Working Paper No. 14/2000.
Rime, Dagfinn (2003). New electronic trading systems in foreign exchange markets. In Derek
C. Jones (ed.) New Economy Handbook, Elsevier Science.
Rozkrut, Marek, Krzystof Rybiski, Lucyna Sztaba and Radosław Szwaja (2007). Quest for
central Bank Communication. Does it Pay to be Talkative?, European Journal of
Political Economy 23(1)
Sager, Michael J. And Mark P. Taylor (2006). Under the microscope: the structure of the
foreign exchange market, International Journal of Finance & Economics 11:1, 81-
95.
Sarno, Lucio and Taylor, Mark P. (2001). Official Intervention in the Foreign Exchange
Market: Is it Effective, and if so, How does it Work?, Journal of Economic
Literature 39, 839-868.
Scalia, Antonio (2008). Is Foreign Exchange Intervention Effective? Some Microanalytical
Evidence from the Czech Republic, Journal of International Money and Finance,
27, 529-546.
Vitale, Paolo (2007). A Guided Tour of the Market Microstructure Approach to Exchange rate
Determination, Journal of Economic Surveys 21, 903-934.
Page 23 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
23
Yao, Jian (1998). Market Making in the Interbank Foreign Exchange Market, New York Uni-
versity Salomon Center Working Paper #S-98-3
Page 24 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
24
Tables:

TABLE 1. Order flow and exchange rate returns summary statistics
Horizon h=1 min h= 10 minutes
Returns Order flow Returns Order flow
Number of observations 293 550 293 550 29 355 29 355
Mean 0.00004 0.005720 0.00046 0.0574
Standard Deviation 0.31 1.15 0.21 5.18
Skewness -0.50 0.21 -0.59 6.78
Kurtosis 63563 308 3347 369
JB normality test 4.94e+13 1.14e+09 1.37e+10 1.64e+08
Proportion positive 0.53 0.51 0.51 0.50
Autocorrelation of order 1 -0.308 0.218 -0.409 0.137
Autocorrelation of order 2 -0.184 0.129 -0.007 0.074
Autocorrelation of order 3 0.000 0.089 -0.002 0.053
TABLE 2. Description of news announcements
Observations
1
Source
2
Release time
3
Sign
4
Consumer price index 24 CSO 9:00 -1
GDP growth 8 CSO 9:00 +1
Current account balance 24 MNB 8:30 +1
Public sector balance 24 Ministry of
Finance
Varies:
17:00 or 10:00
+1
MPC meetings – base
rate decisions
42 MNB 14:00 +1
Notes:
1. Total number of observations in the sample period
2. CSO: Central Statistical Office, MNB: Magyar Nemzeti Bank (the central bank of Hungary)
3. Central European Time
4. +1 (-1) if greater than expected news is assumed to coincide with appreciation (depreciation) of
Hungarian forint
Page 25 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
25
TABLE 3. The effects of macroeconomic announcements on exchange rate and order flow
(absolute measure of surprises)
News
variable
(unit of
surprise)
Dependent
variable
Constant
1 min
lead
Release
minute
1 min
lag
2 min lag 3 min lag R
2
(adj)
returns
0.001
(1,70)
0.006
(1.19)
-0.490***
(-5.40)
0.044
(0.43)
0.047**
(2.10)
0.026*
(1.88)
0.13
CPI (1%)
order flows
0.06
(1.17)
0.27
(1.16)
-17.2***
(-8.60)
-3.06
(-1.06)
0.26
(0.42)
0.59
(0.61)
0.09
returns
-0.001
(-0.84)
0.000
(0.01)
0.019
(1.04)
-0.009
(-1.11)
0.034
(0.98)
0.025
(0.76)
0
GDP (1%)
order flows
0.016
(0.11)
-0.000
(-0.01)
-0.528
(-0.58)
-1.279
(-1.22)
0.450
(0.95)
2.37
(1.52)
0
returns
-0.004
(-0.05)
0.010
(0.84)
0.021**
(2.06)
-0.006
(-0.44)
0.011
(1.14)
0.001
(0.45)
0.01
BD (100
billion HUF)
order flows
2,246
(0.42)
0.094
(0.25)
1.453**
(2.06)
0.051
(0.10)
0.449
(0.72)
0.392
(1.31)
0.01
returns
-0.047
(-0.18)
-0.008***
(-8.17)
0.005
(1.04)
0.002*
(1.76)
0.001**
(2.20)
0.001**
(2.35)
0.01
CA (100
million EUR)
order flows
-3.598
(-0.82)
-0.227***
(-9.02)
0.167
(1.02)
0.152***
(3.72)
0.207***
(11.45)
0.158***
(3.87)
0.02
returns
0.001***
(3.26)
0.020
(1.31)
0.014**
(1.96)
-0.006
(-0.48)
0.001
(0.21)
0.004
(1.58)
0.04
MC meeting
(10 bp)
order flows
0.101*
(3.97)
0.760*
(1.89)
1.449***
(3.50)
0.902*
(1.80)
0.238**
(2.27)
0.653***
(2.82)
0.10
Asterisks refer to level of significance, ***: 1 per cent, **: 5 per cent, *: 10 per cent
TABLE 4. The sensitivity of order flow parameter to news releases
constant OF
OF*I(0)
Release time
R
2
(adj)
Whole sample 0.000 (1.42) 0.010*** (34.63) 0.16
2003 0.000 (0.55) 0.010*** (20.31) 0.12
Benchmark
equation
2004 0.000 (1.54) 0.010*** (32.00) 0.26
Announcement
category
CPI
-0.000
(-0.40)
0.008***
(21.38)
0.012***
(5.44)
0.26
GDP -0.002 (1.38)
0.009***
(6.49)
0.006**
(2.46)
0.20
BD
-0.000
(-0.42)
0.008***
(5.36)
0.013***
(6.94)
0.23
CA 0.000 (0.03)
0.010***
(2.73)
0.023***
(4.50)
0.16
Eq (5)
MC meeting 0.000 (0.61)
0.011**
(7.60
0.028***
(5.55)
0.28
Asterisks refer to level of significance, ***: 1 per cent, **: 5 per cent, *: 10 per cent
Page 26 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
26
Table 5: VAR results
Dependent variable OF Return
return
Lag -1min -0.020** -0.627***
Lag -2min -0.028*** -0.670***
Lag -3min -0.031*** -0.596***
Lag -4min -0.031** -0.565***
Lag -5min -0.026* -0.525***
OF -0.009***
Lag -1min 0.189*** -0.005***
Lag -2min 0.069*** -0.005***
Lag -3min 0.036*** -0.005***
Lag -4min 0.030*** -0.004***
Lag -5min 0.012*** -0.004***
Surprises
Budget deficit -1.447*** 0.006
Lag -2min -0.640**
Current account -0.872*** -0.019
Lag -1min -0.660***
Lag -2min -0.749***
Lag -3min -0.540***
CPI -3.826*** 0.273***
Lag -2min 0.448*
GDP 0.204 0.017
Lag -3min -1.160***
MT -1.097*** 0.041
Lag -1min -0.609***
Lag -2min 0.320*
Lag -3min -0.828***
Lag -4min -1.618***
Lag -8min 0.898***
Lag -9min -0.404**
Trade -18.200*** -0.065
Lag -2min -14.882***
Lag -3min 4.985*
D_MT(1) -0.380**
R
2
(adj.) 0.061 0.320
DW 2.003 2.004
Asterisks refer to level of significance, ***: 1 per cent, **: 5 per cent, *: 10 per cent,
Coefficients for leads and lags are only shown if significant at least at the ten per cent level
Page 27 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
27
TABLE 6. Regression on volatility and absolute order flow
Dependent variable |OF
t,n
| h
t,n
const 0.211*** -0.463***
Intraday pattern 0.908***
Volatility
Lag -10min -0.010** 0.094***
Lag -20min -0.009** 0.033***
|OF| 0.064***
Lag -10min 0.060*** -0.004
Lag -20min 0.015** -0.026***
Monday 0.010
Tuesday -0.014
Thursday -0.022
Friday 0.033
Monetary policy
Meeting -2.010*** 0.076
10 min before event 0.734* 1.220**
10 min after event -1.148*** 0.512
20 min after event -0.654* 0.285
Interest rate down -0.217*** -0.022**
10 min before event -0.009 -0.054***
20 min before event 0.070*** -0.022***
10 min after event 0.015** -0.026***
20 min after event 0.021*** -0.010
Interest rate up -0.016*** 0.008
10 min after event 0.011* -0.045***
20 min after event 0.038*** -0.001
Macro news Hungary
Negative news
(1)
0.000 1.533***
10 min before event -0.141 -0.931**
20 min before event 0.068 -0.562
10 min after event 0.363 1.255**
20 min after event 0.499 0.927**
Positive news
(1)
-0.373 0.343
10 min after event -0.501 1.138***
20 min after event 0.907** -0.275
Communication
Positive communication
(1)
0.412 0.792
20 min before event 0.841* -0.108
10 min after event -0.721* 0.206
Neutral communication
(1)
-0.685* 0.211
10 min before event 0.801** 0.121
20 min after event 0.752* -0.472
Negative communication
(1)
-0.528 -0.173
10 min after event 0.437 -1.046**
20 min after event 1.153*** -1.060**
Trades 0.690*** 0.160***
Lag -10min -0.088*** -0.006
R
2
(adj.) 0.652 0.156
DW 2.003 2.003
Asterisks refer to level of significance, ***: 1 per cent, **: 5 per cent, *: 10 per cent,
Coefficients for leads and lags are only shown if significant at least at the ten per cent level
(1)
Negative news and communication events are those which are assumed to imply de-
preciation of the forint, and vice versa.
Page 28 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
TABLE 7. Regression on returns and order flow
Dependent variable OF Return
const 0.063** -0.0001
return
Lag -10min -4.192*** -0.146***
Lag -20min -0.981** -0.033***
OF 0.007***
Lag -10min -0.126*** 0.0002*
Lag -20min -0.031*** -0.0002*
Monetary policy
Meeting -0.941 -0.056***
10 min before event 0.412 0.015
20 min before event 2.060*** 0.003
10 min after event 1.876** 0.035***
20 min after event 0.899 0.031***
Interest rate down 0.039*** -0.004***
10 min before event -0.251*** 0.001***
20 min before event 0.234*** -0.001***
10 min after event 0.126*** -0.001***
20 min after event -0.060*** -0.000
Interest rate up 0.024*** 0.002***
10 min before event 0.021** 0.000**
20 min before event -0.012 0.000
10 min after event 0.152*** -0.001***
20 min after event 0.042*** 0.001***
Macro news Hungary
Negative news
(1)
-0.033 -0.003
10 min before event -0.114 0.002
20 min before event 0.310 0.007
10 min after event -1.679*** -0.012
20 min after event -0.910 0.001
Positive news
(1)
-0.778 -0.010
10 min before event -0.013 0.017
20 min before event -0.053 -0.028**
10 min after event 2.588*** 0.067***
20 min after event 1.523* 0.037***
Communication
Positive communication
(1)
0.424 -0.008
10 min before event 3.556*** 0.040***
20 min before event -0.783 0.002
10 min after event -1.684** 0.006
20 min after event 2.409*** -0.019*
Neutral communication
(1)
-0.034 -0.008
10 min before event -2.703*** -0.020*
20 min before event 0.190 -0.016
10 min after event -0.905 0.014
20 min after event -0.036 -0.010
Negative communication
(1)
0.029 0.017*
10 min before event -2.207*** 0.002
20 min before event 0.796 0.014
10 min after event 0.765 0.014
20 min after event -1.995*** 0.011
R
2
(adj.) 0.065 0.222
DW 2.005 2.003
Asterisks refer to level of significance, ***: 1 per cent, **: 5 per cent,
*: 10 per cent
(1)
Negative news and communication events are those which are assumed to
imply depreciation of the forint, and vice versa.
Page 29 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
Page 30 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
Page 31 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
F
o
r

P
e
e
r

R
e
v
i
e
w
Page 32 of 32
http://mc.manuscriptcentral.com/ijfe
International Journal of Finance & Economics
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60

Page 1 of 32

International Journal of Finance & Economics

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Macroeconomic announcements, communication and order flow on the Hungarian foreign exchange market
August 2008. Abstract We investigate the relation between the intradaily HUF/EUR exchange rate on the one hand and news announcements and order flow on the other hand. We extend the existing literature on foreign exchange market microstructure by considering a small open transition economy. We find that the intradaily exchange rate depends on both news announcements and order flow. We conclude that news on the HUF/EUR market are transmitted directly via immediate reactions to news announcements as well as indirectly via order flow. We decompose the news’ total effect on exchange rate and find that order flow accounts for approximately three quarters, compared to one quarter for direct news impact. Although the HUF is pegged to the EUR, the exchange rate reacts very qualitatively very similarly to exchange rates of major currencies as reported in the literature, whereas it quantiatively differs: the importance of indeirect news transmission is remarkably higher on the HUF/EUR market. Furthermore, we extend the commonly used news to communication of central bankers significantly improving the explanatory power of the estimates. Central bank communication is an important determinant for the HUF/EUR rate. JEL: F31, G14, G15

Keywords: microstructure, order flow, exchange rate, macroeconomic news, central bank communication, Hungary

rP Fo

http://mc.manuscriptcentral.com/ijfe

ee

rR

ev

iew

International Journal of Finance & Economics

Page 2 of 32

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Macroeconomic announcements, communication and order flow on the Hungarian foreign exchange market* 1. Introduction
During recent years there has been a growing interest in the microstructure of financial markets and the relations between exchange rates, news and order flow. The microstructure approach deals with the purest form of financial intermediation, i.e. pure trading without any transformation and has been pushed by the increasing availability of high-frequency data, that typically ranges from tick-to-tick data to frequencies of about 20 minutes. The microstructure approach allows additional insights in the price formation on financial markets. Whereas traditional macroeconomic models assume that news are directly transmitted to prices, the microstructure approach explicitly assigns an important role to the process of trading. The key concept is the order flow, showing a close empirical relation to the exchange rate (see e.g. Lyons 2001, Evans 2002, Payne 2003). The order flow is defined as the difference between trades initiated by the buyer, and trades initiated by the seller of an asset. Therefore order flow can be described as buying (or selling) pressure on the market. Order flow significantly differs from the turnover on the market. There are situations where the order flow is almost zero, if trades are symmetrically initiated by buyers and sellers, although the turnover is high and vice versa. Whereas in traditional asset market models information is assumed to be common, the microstructure approach claims that information exists that is not shared by all market participants and/or is interpreted differently by them (see e.g. the models by Kyle 1985, Glosten and Milgrom 1985, Evans 1995, Evans and Lyons 2006b). Thus, the trading mechanism itself becomes an important feature. However, the microstructure approach does not insist that it is not the fundamentals which drive prices, but it focuses on the way how potentially dispersed information about fundamentals is transformed to prices. Although some standard assumptions of perfect markets are relaxed within this framework (i. e. perfect information, perfect competition), the market may still be regarded as efficient in the sense, that prices reflect all known information. It implies that within both frameworks the market reaction to public news announcements is expected to occur within very short time. As dealers may revise their quotes based on news as well as on the quotes they receive from other dealers, one may distinguish three different relations between news, order flow and the exchange rate (see inter alia Evans and Lyons 2006a, Sager and Taylor 2006). First, the information content of announcements, that are publicly available and evaluated in a similar manner by at least most of the traders (the ‘common-knowledge’ part of macro news), affects prices directly and immediately. This channel directly corresponds to what one would call a 1
http://mc.manuscriptcentral.com/ijfe

rP Fo

ee

rR

ev

iew

Cai et al. Dominguez and Panthaki (2005) come to the conclusion that both. only few papers deal with smaller developed markets. Bjønnes and Rime 2005). 2005) and between exchange rates and order flow (Lyons 1995. While most of the existing literature focuses on exchange rates between the ‘big four’ USD. private information is dispersed and transmitted to prices via order flow (see also Evans and Lyons 2002a. Love and Payne (2008) conclude for major exchange rates that news affects both. order flow unrelated to news announcements due to changing hedging or liquidity demands.manuscriptcentral. The main difference between the microstructure approach and traditional asset price models is therefore that trading is considered as a substantial part rather than an auxiliary activity to price formation and hence deserves particular attention. The recent empirical literature shows that particularly the first two channels of price impact are present on foreign exchange markets: There is a vast empirical literature showing a significant relationship between high-frequency exchange rates and news (Almeida et al. Furthermore. so there can be a component of news that is not common-knowledge. Yao 1998. Andersen and Bollerslev 1998a. 2003. Berger et al. the exchange rates as well as the order flow within a very short time. Bauwens et al. Melvin and Yin 2000. the relation between order flow.Page 3 of 32 International Journal of Finance & Economics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 traditional exchange rate model. 1998. within one minute. They conclude that in 1998 order flow played a more important role than news announcements and mainly contributed to the extraordinarily high volatility of the yen. (2001) turn to the volatility of the JPY. We refer to this channel as the indirect effect of macro news intermediated by the induced order flow. Frömmel et al. Second. or changing risk tolerances (Evans and Lyons 2002a) may also affect the exchange rate. but rP Fo http://mc. there may be news that are not unambiguous and difficult to interpret or not publicly available. only one third of the news content immediately being impounded to prices. Evans and Lyons (2005) also find for the USD/EUR that the impact of news on the order flow is stronger than on exchange rate returns. such as the market for the Swedish (Lindahl and Rime 2006) or Norwegian krona (Rime 2001). JPY and GBP. The relations between the different sources of exchange rate variation are illustrated in Figure 1. EUR (DEM). Andersen et al. (2008) find that both. Both studies also find a significant increase of the impact of order flow on prices around news announcements. Payne 2003. Osler 2006). news and the volatility of the EUR/USD rate depend on the counterpart of one particular FX dealer. In contrast. Third. More recent papers simultaneously analyze the relation between the exchange rate.com/ijfe ee rR 2 ev iew . (2005) find a very quick reaction of the order flow on news. scheduled and unscheduled news affect the volatility of the USD/EUR and the USD/GBP rates. they identify delays in the direct impact of news on the exchange rate and even longer ones for the impact on order flow. In this case. news and order flow. a result that is roughly confirmed by Evans and Lyons (2006a) for the DEM/USD rate.

thus less liquid. Second. Égert (2007) finds influence of central bank communication for Hungary. the low number of market participants. The contribution of our paper to the literature is threefold: rP Fo http://mc. but also by functioning as a coordination advice for market participants (Reitz and Taylor 2008). mostly for the FOMC. the ECB. Hungary and Poland. liquidity and private information. again most of the work deals with developed markets. Another strand of literature focuses on the impact of central bank interventions and communication on the exchange rate. the more dynamic economic environment or simply structural deficits in an evolving market. the signaling or expectation channel of monetary policy (Sarno and Taylor 2001). Thus. there may be more private information in the market. markets in transition economies are much smaller. The impact of verbal interventions or communication stems from their role in anchoring expectations on future monetary policy. Derviz 2003 and Scalia 2008 on the efficiency of FX interventions.g. i. Menkhoff and Schmeling 2008 on the Russian FX market). Both factors.e. but not for other CEEC.manuscriptcentral. There is a huge amount of research providing some ambiguous empirical evidence on an impact of central bank communication on the exchange rate (see the survey in Blinder et al. Sarno and Taylor 2001. 2008). Besides the wellknown facts that transition economies are characterized by large capital flows and frequent interventions by the central banks there are two main characteristics that are directly linked to market microstructure: First.International Journal of Finance & Economics Page 4 of 32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 there are only very few papers on the microstructure of financial markets in Central and Eastern European Countries (CEEC) (Hanousek and Podpiera 2003 on the Czech stock market. are key determinants of a markets microstructure. He also concludes that the Hungarian National Bank (MNB) used actual interventions very rarely. and there are only few papers that focus on transition economies: Rozkrut et al. However. However. They affect the first two moments of the exchange rate (Scalia 2008) and the impact is usually stronger in emerging countries than in developed countries (CanalesKriljenko 2003). Vitale 2007) is that interventions are able to move the exchange rate.com/ijfe ee rR 3 ev iew . the Bank of England and the Bank of Japan. but mainly relies on verbal interventions. A basic insight of the research on central bank interventions (for surveys see e. the small FX market of a transition economy may show different characteristics than the markets for the major currencies. the market's size and organization. (2007) find for the Czech Republic. that is – although less obvious at first sight – able to affect the exchange rate as well (Ehrmann and Fratzscher 2007). This may be due to less sterilization. communication may complement intervention or substitute it (Fratzscher 2008). Due to the more dynamic economic environment in a transition economy verbal interventions may be more effective than in developed markets. This may be due to the lower news coverage. that speeches about monetary policy affect the exchange rate. Besides direct interventions central bank communication may also be seen as a form of intervention.

Second. 4-monthlong periods of Evans and Lyons 2006a or Frömmel et al. how much the results differ from those for the major foreign exchange markets.1 Order flow and exchange rate data The data for the exchange rate and the order flow stems from the Reuters D2000-2 electronic trading system1. The outline of the paper is as follows: In section 1 we gave a brief overview of the role of order flow in the trading process and recent empirical work. macroeconomic announcements (quantitative data) and communication data (qualitative data). as it may anchor expectations. order flow and exchange rates in section 3. It is substantially longer than most of those used in the relevant literature on order flow (e. including uncertainty about monetary policy may lead some importance of central bank communication.com/ijfe . Finally. 2008) and covers a period of two years. we analyze the longer-term relationship between news. We will describe them briefly in the subsequent subsections. order flow and macroeconomic news release. We will use data with a frequency of one and ten minutes as well. we use a broader data set of news. less liquid market in a transition economy. Again. we had access to a database of outstanding quality containing very detailed data from the Reuters D2000-2 trading system. The empirical analysis comprises four subsections: First. Second. We also depart from most of earlier studies in that we analyze the effects of the Hungarian quantitative news series individually and thus are able to compare the importance of different news announcements.4. 2. we investigate the exchange rate-order flow linkage and its sensitivity to news announcements. 8-month-long period of Love and Payne 2008.manuscriptcentral. Third. we address the question whether scheduled macroeconomic news releases have an immediate impact on exchange rate and on order flows. After describing the data in section 2. Section 4 summarizes and concludes. Our aim is to analyze.g.3 we decompose the news impact into a direct part and an indirect part transmitted by order flow. In section 3. covering not only quantitative macroeconomic announcements. we present the empirical results in section 3. Our whole data set consists of three parts: the exchange rate data including the order flow. Data set The data set used in this study covers the whole years of 2003 and 2004 and contains data on the exchange rate. One could expect a higher importance of order flow due to a higher share of private information and a less stable and predictable economic environment. we focus on a small. the less stable economic environment. rP Fo ee rR ev iew 2. 4 http://mc.Page 5 of 32 International Journal of Finance & Economics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 First. but also communication variables. As the competing electronic trading system EBS provides services on the 1 For a full description of characteristics and tendencies of spot FX market segments and the features of the available data see Lyons (2001).

Love and Payne 2008) the size of the trades is known. quantity traded. with the help of „entry type” variable. 2002. without any distortion. Following the literature (Andersen et al. We use a 10-minute frequency to explore the relationship between an extended set of news. This part of our data contains data on all the posted quotes (time-stamp. the one that causes the trade. A number of additional advantages should be noted. For the same reason we drop the periods from 17:30 CET to 08:00 CET next morning each day. In the Reuters D2000-2 a trade may occur. quantity entered. whereas in the latter case the market order is the initiating one. From these aggregated quotes series we then calculate the returns as 100 times the difference of the log prices. quantity). are then aggregated to two different frequencies: We use 1-minute frequency to analyze (1) the impact of five specific scheduled Hungarian quantitative news items on exchange rate returns and order flow.manuscriptcentral. exchange rate returns and volatility. 2005.com/ijfe . which are accurate to the second decimal of the second. In the first case the last limit order. This distinction is due to the characteristics of the news variables and will be discussed in detail later. The order flow is obtained by subtracting the seller-initiated traded quantities from buyerinitiated traded quantities that occurred within the given interval of respectively 1 and 10 minutes. 2 5 http://mc. We have removed these periods as well. These data. etc. where the same database as here is used. or because a market order hits an existing limit order (which is the case in more than 75 per cent of all cases). because two limit orders are paired by the system. trade price. Bauwens et al. First. or other data problems occurred. (2006). see Gereben and Kiss M. Berger et al.) and the executed trades (time-stamp. is regarded as the initiating order. including a detailed description of data and a comparison to other datasources. Second. This is in line with the microstructure literature (Berger et al. rP Fo ee rR ev iew For an extended description of the Reuters D2000-2 database. order flows. price. There are few short time periods when the Reuters data feed was broken. which can be regarded as a buy/sell indicator. order flow and pre-scheduled news. (2) the linkage between order flow and exchange rate movements and (3) the joint dynamics of exchange rate. entry type. 2008). the initiator of the trade can be determined quite easily without having to make uncertain assumptions. allowing us to calculate the order flow more accurately. our data set can be assumed to cover most of the interdealer HUF/EUR market2. unlike in many other papers (Lyons 1996. 2008) we exclude weekends and public holidays from the data set due to the lack of trades and quotes during these days. but no information about the identity of market participants.International Journal of Finance & Economics Page 6 of 32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 HUF/EUR market only since February 2006 and the direct bilateral interdealer trading is steadily loosing ground in the FX market (Rime 2003).

E [Ai . 2004. therefore the market may be expected to react in a very different way than to scheduled interest rate changes.t rR ev E [Ai .t E [Ai .t ] . the announced value and the market expectations for the included fundamentals. However. where it is meaningful and has additional information content. Summary statistics of the return and order flow series are presented in Table 1. This question has become standard in the survey only during 2005.com/ijfe .t = Ai . 4 6 http://mc. as well. We use this surprise or news component of the macroeconomic announcements in absolute terms as well as in a standardized manner. 2003.2 Macroeconomic Announcements This part of the news database consists of 5 quantitative news items containing the exact time. This approach has turned out to be standard in the empirical literature (Balduzzi et al.t is the announced value of data i. Pintér and Wenhardt 2004). which is in line with the expectations hypothesis.manuscriptcentral.t ] ˆ iew (1) 3 This variables also includes meetings when the interest rate was unchanged. Love and Payne 2008): where Ai . We define the „news” component of these macro variables as the difference between the actual values and the expectations. and ˆ is the sample standard deviation of Ai . 2. to the results of the existing literature and to our intuition. as they were announced unexpectedly under turbulent market circumstances. whose effects on the HUF/EUR market can be regarded as relevant. So we use the change in the short-term (3-months) government bond benchmark yield as the measure of surprise.355 for the 10minute analysis.t ] is the expected value of data i. In addition. we excluded the so-called non-ratesetting meetings. because according to the communicated policy the Monetary Council is not expected to make an interest rate decision at these meetings. we excluded from the examination the decisions of the extraordinary meetings. as base rate decision are regarded as one of the most important sources information among fundamentals (Kiss M.3 For a brief description of the variables see Table 2. Andersen et al.Page 7 of 32 International Journal of Finance & Economics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 In this way we get 293. According to previous examinations. We also include the Monetary Council meetings in this subset. so for our sample period no expectation data are available. In the empirical part of the paper we report results for both the absolute and the standardized measures of surprise. that are extracted from the regular monthly survey conducted by Reuters Hungary among 15-25 market analysts. The only exception is the Monetary Council’s rate decision.4 rP Fo ee S i . 2001. four Hungarian macroecomomic variables are selected (CPI.550 observations for the 1-minute analysis and 29. that enables us to compare the relative impact of announcement categories. budget deficit and current account deficit). GDP growth.

Positive and negative news sum up to approximately 73 per cent of the selected statements. We take into account that the same statement can carry different meanings at different points in time.e. The comparison to market expectations is most important in case of interest rate communication.com/ijfe ee rR 7 ev iew . Several people independently reviewed Different theoretical exchange rate determination models differ in their predictions concerning the expected direction of the exchange rate response of news surprises.e. 6 5 As for the macroeconomic announcements. we transform the communication dataset into 3 dummy variables. lower-than-expected inflation or deficit figures. policy rate (hints on the short-term interest rate path). Our results are consistent mainly with the portfolio-balance model. hints on the preferred exchange rate level. Accordingly. We use news headlines reported on Reuters screen and all kinds of written communication by the MNB to create the variables describing central bank communication. and each statement counted only one time. We have performed the encoding of statements as follows: Four people compiled the data set. market participants know the hour and minute of their release in advance and may react within a very short time.2. when it is mentioned first. according to the "sign" of the news. We code the pieces of news from the market participants’ point of view. intervention). where the expected direction of the exchange rate response was not unambiguous. and then two of them again checked the coding. Three main (monetary policy related) topics determine the content of the MNB’s communication: exchange rate (for example.International Journal of Finance & Economics Page 8 of 32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 An announcement is regarded as „positive/good” (positively signed) if it implies the appreciation of the forint (i. as a rising/declining interest rate path was often present in expectations.manuscriptcentral. but with some necessary modifications. the difference between the three-month benchmark and the policy rate is considered as the anticipated interest rate cut. i. while we consider the rest of the communication as being neutral. 2. For example.5 As all these mentioned news items are pre-scheduled. Our methodology is based on the approach of Love and Payne (2008).3 Communication variables The second part of our news database includes qualitative and (mainly) non-scheduled news items concerning central bank communication. central bank reaction function models predict an appreciation after higher-than-expected inflation data. while the same surprise is expected to cause depreciation according to the portfolio-balance approach (for more details see Hoffman and Schlagenhauf 1985). We used an event-study analysis for the classification. Similarly as for the quantitative news in subsection 2. we consider hints on a smaller cut than the anticipated interest rate cut as tightening6. an analysis at the high frequency of one-minute intervals seems to be most appropriate to capture their effcts on the exchange rate. rP Fo http://mc. so the sign of statements is determined in relation to market expectations. We include only those items in the database containing some kind of guidance about the future stance of monetary policy. restrictive base rate decisions) and is regarded as „negative/bad” analogously. economic outlook.

we decompose the impact of news on the exchange rate into a direct and an indirect effect by means of a VAR. Empirical analysis Due to the aforementioned attributes of our news variables the empirical analysis embraces two different time aggregation frequencies – 1-minute and 10-minute intervals –. Cai et al.manuscriptcentral. Third. Dominguez and Panthaki (2006) and Frömmel et al. Therefore. and according to characteristics of the specified approaches we used different methodological approaches – event study.1 The immediate impact of macroeconomic news on exchange rate and order flows Following the approach by Love and Payne (2008). We follow the commonly used approach by Andersen et al. Fourth. This approach covers effects on exchange rate returns as well as on the volatility of exchange rates and is in line with. (2001). ten-minute frequency. VAR and conventional regression analysis – to analyze how news. (2008). (2008) and separately estimate the following equations for exchange rate returns and order flow: rP Fo http://mc. inter alia. and whether this relation changes during news release times. namely whether the surprise component of the included announcements has significant influence on the HUF/EUR exchange rate (equation 2) and on order flow (equation 3). We proceed in five steps: First. we add the communication variables to our system and turn to a lower frequency of ten-minute intervals. Second. without considering the role of order flow. due to its narrative and sometimescontroversial character. Only part of the communication (some pieces of written communication) is pre-scheduled. as a preliminary exercise we analyze the direct impact of news announcements on the HUF/EUR rate. (2003) and Berger et al. For this reason. we conduct the part of the analysis including communication variables at a lower. we begin our empirical analysis by testing two hypotheses. we analyze the impact of the surprise content of macroeconomic announcements on exchange rate returns and on order flows separately.Page 9 of 32 International Journal of Finance & Economics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 doubtful cases to reduce the subjectivity of the coding as far as possible. we investigate the relation between the exchange rate and order flow. Most of the communication news items appear without announcement on the Reuters newswire screen. This procedure already hints at this news category being a natural candidate for indirect price impact via order flow as suggested by Evans and Lyons (2006a). The first three steps are performed in line with Love and Payne (2008) and applied to data at the frequency of one-minute intervals. 3. 3. as a first step. This makes it difficult to determine the exact time when these pieces of information arising from the central bank reach the market. order flow and exchange rate depend on each other.com/ijfe ee rR 8 ev iew .

n are the error terms. C/A. not substantially change the overall picture. We only refer to them in the text if necessary.n and t. C/A and monetary council meetings) we find similar patterns: there is an immediate impact on the return as well as the order flow.n (2) of + Nt i .n = p + l i=k l i=k of i. 8 Actually.n = OFt . if there is any.n . Andersen et al. budget deficit. this is partly due to the small number of news releases relative to the number of observations and in line with the literature (Love and Payne 2008.n (3) where p t . 9 http://mc. OFt .n . 2003). must be slight. and N t .m Nt i . the overall picture is that most of the macro news have a significant immediate impact on returns as well as on the order flow. The effect of GDP news.m is the surprise component of news category m. disperse and slowly impounded. with the expected sign on the coefficients.n is order flow in the same interval. and with l = -1 minutes8. the exchange rate as well as the trading pattern. In the majority of cases the coefficients are significant and well-signed in the minute(s) right after the announcement. for which – potentially due to the only eight GDP data releases in our sample – our data do not give evidence of impact of GDP news neither on returns nor on order flow within short time. GDP growth. For the other news categories (CPI. The only exception are GDP releases. base rate decision) and for absolute as well as standardized news measures separately for a short event window with k = 3.com/ijfe . t . The comparatively low R2 imply that macroeconomic information accounts for only a small proportion of exchange rate variation. This does. however. Furthermore we only show the results for the absolute measure as those for the relative measure are basically the same. Four out of the five announcements significantly affect both. n is 100 times the logarithmic change in the exchange rate from time t-1 to time t on day n.m + t .7 We show results for each news category (CPI. There were.International Journal of Finance & Economics Page 10 of 32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 pt . However. in million euro. budget deficit. however. and euro purchases/forint sales worth 17.2 million euro on average in the rP Fo ee rR ev iew 7 We use OLS for our estimations and correct the coefficient variance/covariance matrix using Newey-West standard errors to compensate for autocorrelation and heteroscedasticity. few significant coefficients and no clear pattern.m p i .m + t .manuscriptcentral. All results not shown here are available from the authors on request.n. For instance. we estimated the equations for k = 60 and l = -10. In Table 3 we report estimates for specifications with the absolute news measures. At first glance. for CPI surprises we find that 1 percent higher than expected inflation causes almost half a percent depreciation of the Hungarian currency.

which may indicate position taking in advance (Monetary Council meetings) or just relate to the early release time of 8:30 a. as liquidity and trading activity are assumably higher around news releases (see Chaboud et al. well-informed or „smart” market participants are believed to trade more intensively around news releases. the base-rate decision surprise. 2004. For CPI. (C/A) when the trade intensity is low (Gereben and Kiss M. the association between order flow and exchange rate may be quite different in the immediate aftermath of macro news releases. than the second powerful one. 3. This attitude can cause a temporary disconnection between exchange rate and order flow. Market makers may then update their beliefs and change their quotes according to this unambiguous information instead of depending upon the noisy trading.9 The standardized measures indicate that CPI surprises have the strongest impact: One „unit” of CPI surprise implies about more than 4 million euro net one-sided trading. 10 http://mc. Under these circumstances a positive (negative) order flow more probably indicates that the informed traders consider the asset more valuable (less valuable) than its rP Fo ee rR ev iew 9 In few cases.2 The sensitivity of the exchange rate-order flow relationship to news release times In the preceding analysis we showed that news about fundamentals do not only influence exchange rate determination. This latter even has double influence on exchange rate than budget deficit and current account deficit surprises. even a lead is significant. In this case dealers pay more attention to trading activity. then trades convey less information about the fundamental value of the asset in the presence of news. but also trigger net buying or selling pressure on the currency in accordance with the sign of the surprise. it causes an almost three times higher change in the exchange rate. if traders disagree about the interpretation of a given piece of news. C/A and the Monetary Council meetings we also find that the effects of a surprise last for several minutes. a weaker as well as a stronger connection between the exchange rate and order flow when news are published: If the new information is unambiguous and is interpreted similarly by most of the market participants. and increase the probability of private information content of one-sided flows. Thus.Page 11 of 32 International Journal of Finance & Economics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 minute immediately following the release. 2006). Furthermore. Fleming and Remonola 1999).com/ijfe . In addition.manuscriptcentral. Therefore. we turn to the examination of the potential connection between order flow and exchange rate. On the other hand. As the effect of order flow on exchange rate originates in conveying private or dispersed information.. due to the larger volumes placed in the order book. which have approximately the same effect on the exchange rate in terms of the size of the impact. a unity of trade might cause less price movement. There are theoretical justifications for both.m. at least around news announcements. one should expect that order flow is correlated with exchange rate returns.

International Journal of Finance & Economics Page 12 of 32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 actual market price. we can now compare. Regarding the results of the benchmark. as expected. and interacts with the presence of news.n = + OFt .12 A net excess of forint buyer-initiated trades is worth 1 million (in the base currency of euro) and leads to about 0. and differ by announcement types. 11 However. The parameter of order flow is positive and highly significant. The detailled results are available on request. This is probably due to the 1-minute frequency and the relatively low trading activity in the Hungarian FX market compared to the big markets (EUR/USD.m is a dummy variable indicating that a news release concerning macroeconomic variable m occured in period t.manuscriptcentral. when trading activity was much lower on the HUF/EUR market (Gereben and Kiss M. than is commonly found in the empirical literature. It is significantly positive for all news categories. also for all news categories (Table 4). broken down by announcement type (Berger et al. 12 11 http://mc.01 percent appreciation of the forint (Table 4). 2006) as well as the literature on other exchange rates (see inter alia Evans and Lyons 2002a). m + t . more specifically under the expression „around news releases” we mean the five minutes before and after the announcements.n + k i= k i . broken down by announcement type. whether the explanatory power of the estimation can be improved by letting the coefficients of the order flow differ from its usual level around news releases. Leads and lags are hardly significant.11 First. but we introduce additional dummies for different news categories. the order flow parameter significantly changes at announcement times. due to convenience reasons we only report the contemporaneous results. We adopt an event-study approach and estimate equation (4) only on announcement days. Second.com/ijfe . the previously demonstrated results from subsection 3. Moreover. order flow plays a significant role in exchange rate determination in the Hungarian FX market and corroborates previous results on the HUF/EUR market obtained from daily data (Gereben et al. when scheduled public information reaches rP Fo ee rR ev iew 10 We also estimated equation (4) only for announcement days.m m OFt . This means that order flow conveys additional private information about the fundamentals in the minute. the R2 is 16%. 2006) and speculative attacks against the band took place. JPY/USD etc. As well as in the previous section. as on average there is only one trade every three minutes. We then compare the results with a benchmark equation without the interaction between order flow und news releases. n I (i ) t . We set k to 5. Our estimation results for the benchmark regression and the five announcements separately are presented in Table 4.n (4) where I (i ) t . Thus we estimate the following equation: pt .10 The approach follows Love and Payne (2008).). 2008) apply this event-study approach similarly. This view is confirmed by the fact that the R2 was particularly low during the first year of our sample. but do not use dummy variables and estimate the regressions only in the minutes after the releases. it should be noted that the statistical power of the benchmark regression is lower to some extent (R2=16%).1 suggest a stronger connection as well.

com/ijfe . 12 http://mc. The order flow exhibits positive dependency on recent returns up to 8 lags: forint appreciation leeds to forint purchases and vice versa. In the following we further explore this issue. The direct effect of news on the exchange rate is less pronounced: only rP Fo l i =1 i Pt OFt i i + 6 l =1 k j =0 l. inidcates that part of the total impact of new public information is transmitted to exchange rates partially via order flow. which can be interpreted as an indication of positive feedback trading in the market. This result. together with the previously found conclusion. This allows us to estimate the equation separately by OLS. that the structural form innovations. which would allow the order flow to respond to price movements at a frequency of less than a minute.manuscriptcentral.94 basis points. News affect the order flow significantly: all types but the GDP announcement have a significant contemporaneous effect. we allow contemporaneous order flows to have an effect on the price quotations.t) are independent.3 Decomposing the channels of the news’ impact – A multivariate VAR analysis Now we carry on with the decomposition of exchange rate responses to news into direct and indirect (intermediated) parts.13 There is a relatively strong and persistent high-frequency negative autocorrelation in returns. The coefficients are in line with intuition: positive surprises (good news) lead to an increase in net forint purchases. Table 5 summarizes the results and for brevity we only report the results for the first five lags. 2. The contemporaneous order flow affects the returns as expected: a net forint purchase of 1 million euros goes along with an appreciation of the forint by 0. We further assume.Page 13 of 32 International Journal of Finance & Economics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 the market. This assumption seems to be a reasonable economic restriction. and try to measure the contribution of order flows to total exchange rate responses to news shocks. The substantially increased R2 show a close relation between order flow and the exchange rate. Lagged flows have a similar but smaller effect. as the market is less liquid than the markets for major currencies. and several types of news affect the flows for several minutes. ( 1. j N l .t j + t (5) ee rR ev iew 13 The lag lengths were selected according to the Schwartz information criterion.t. and with one exception the lagged effects have the same sign. but the flows are not affected by the contemporaneous returns. For analysing this issue and testing the order flow intermediating role we estimate a VAR model for exchange rate changes and order flows. 3. and within the 1-minute frequency we are analysing the number of trades is very limited. Pt = + OFt + OFt 0 In this model. as follows. while the high-frequency order flow exhibits positive autocorrelation. So we do not expect almost simultaneous feedback trading. the contemporaneous order flow and the news.

indicating that order flow indeed plays a substantial part in processing news. rP Fo http://mc. 3. The share of the order flow has in the impounding the news into the prices increases rapidly over the time: 5 minutes following the announcement it amounts to almost 60%.1 Methodology 14 In our framework this means computing IRF from a model where the coefficient in the order flow equation is restricted to 0. To the best of knowledge we are the first who do this kind of decomposition for an emerging market. and stabilizes around 57%.4 Longer-term interdependence of news.34 basis points. order flow and exchange rate As the inclusion of communication variables requires a ten-minute frequency we move on to explore the longer-term relationship between exchange rate.14 The hypothetical impulse response shows the direct effect of the news shock on the exchange rate returns. They rejected the test for whether the direct channel is more important than the indirect channel. even in the presence of a strong direct effect. We follow Love and Payne (2008) and compare the actual IRF of the exchange rate returns to a hypothetical IRF. Love and Payne (2008) examined three big FX markets and showed that on average one third of the new information is transmitted via order flow. To highlight the particular effect of CPI surprises as the obviously most effectful news category. and the role of the order flow. 3.4. This hypothetical model still allows for the order flow to influence the exchange rate.com/ijfe ee rR 13 ev iew . Figure 2 shows the (cumulative) impulse responses of the exchange rate and the order flow to a CPI announcement.International Journal of Finance & Economics Page 14 of 32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 the surprise content of the CPI announcement has a significant effect.manuscriptcentral. order flow and fundamentals. which would prevail if the news had no affect on the order flow. our result is in line with the findings of the existing literature. 93. although the share that order flow has in incorporating news into prices is even higher in our case than it was found concerning major FX markets. The share of the direct and indirect components is given in Table 6. but assumes that neither the size of the order flow nor its effect on the returns changes following a news announcement. and the difference between the actual and the hypothetical impulse responses the indirect impact via the order flow. Evans and Lyons (2006a) found that around two-thirds of news’ total effect can be linked to trading activity in the DEM/USD market.7% of the change in the exchange rate can be attributed to the news announcement directly. concluding order flow accounts for at least half of the effect of macro news on exchange rate. At the time of the announcement the direct channel dominates. But in this case the effect is strong: a one-standard deviation positive surprise causes the HUF/EUR rate to appreciate by 27. so the comparison of our results is possible only with developed markets. In general.

Hughes Hallett and Anthony 1997. 2008) and calculate the volatility of ten-minute returns. when the market was characterized by turmoil due to speculative pressure on the forint. The estimated conditional volatility shows a huge increase in January and again in June 2003. dayk are dummies capturing potential day-of-the-week effects and Xt. and the impact of various kinds of news. 2001). for instance. we apply a model with t-distributed errors. 16 14 http://mc. including news. As there are only daily observations necessary.n (7) n n is usually achieved by estimating a GARCH model based on daily ee rR ev iew 15 Andersen and Bollerslev (1998b) show that even in the case of a continuous-time price process the GARCHmodel provides accurate forecasts. For the intraday pattern we use a HP filter. which will be specified later. As Melvin and Yin (2000) point out.n + ut . The approach is based on the model p t . The daily volatility level 15 data . Cai et al. fit several possible trend functions and concluded that no one figures out the data better. Dominguez and Panthaki 2006. volatility level to the return volatility and t.n zt n (6) is the contribution of the daily where µ is the unconditional mean of the return process. the peak between 9:00 and 9:20 a.1) model with normally distributed conditional variances is used.g.Page 15 of 32 International Journal of Finance & Economics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 We measure volatility following the approach by Andersen and Bollerslev (1998a).n consists of high-frequency contributions. overnight breaks are no problem. In this case the use of a conditionally normally distributed GARCH-process may lead to the underestimation of the “true” volatility.. when Hun- rP Fo p t . We did. which are reported for exchange rates in previous empirical work (see inter alia Andersen and Bollerslev 1998.n µ = c + p( t ) + 4 k =1 k day k + X t . which describes the pattern sufficiently well. Anderson and Bollerslev (1998) there is a priori no need to work with the more complex flexible Fourier form.m. Frömmel and Menkhoff 2001) that managed exchange rates as the HUF/EUR show a much higher kurtosis than the flexible DEM/USD rate analyzed by Andersen and Bollerslev (1998a).n = 2 log where p(t) is the above mentioned intraday volatility pattern. day of the week effects.n µ= n t . 2001. While commonly a standard GARCH(1. Frömmel et al.manuscriptcentral. the figure shows a clear pattern. which has become standard in high-frequency analysis and is commonly used in the recent literature (see inter alia Cai et al. and may consider using a simpler pattern. After the equation is squared and expressed in logs it evolves to ht . if the pattern is simpler than the one in. One can easily distinguish the start from a very low activity level in the morning. The results of the daily volatility and the intraday volatility estimations are illustrated in Figure 3. However.n is a set of further determinants. We do this because it is known (see e. in par- ticular an intraday pattern.com/ijfe . central bank activities and order flow. however. The intraday pattern is simpler than the one usually found for the USD/EUR or USD/DEM market16.

e. The news variables may affect order flow as well as volatility. as in the previous section.l N k . market participants react on recent price changes for exploiting (potentially small and short) trends.t + l .t +l .n + K (9) + t .n where the variables Nk. The equation looks similar for the conditional mean.n 2 + + 2 i =1 i pt i .2 Results The estimation results for the equations (8) are given in Table 6.n + k = 1l = m t .t l . ev k = 1l = m k .t l .n = 1 + p(t ) + + 2 h=1 h 0 OFt .com/ijfe . various types (k) of news as well as lags. and at the same time there is positive autocorrelation of the absolute order flow.n rR 2 p t . Again. whereas order flow itself may react to volatility with a delay of one interval only.n K + m 2 i =1 i t i .n + t .t.l N k . the volatility depends contemporaneously on order flow. as volatility is a one-sided measure. there is no need to include day of the week effects and an intraday pattern. Similarly. Therefore the equations (8) evolve to: rP Fo 0 OFt . Further- 15 http://mc.n (8) 2 OFt h. in the volatility equation positive autocorrelation of the volatility and some negative relation is present between the volatility and lagged absolute order flow. some small differences.n + + 2 h=1 2 i =1 i t i.n h + k =1 l = m k .l k =1 l = m N k . For analyzing the effects of news on volatility we estimate the following system of equations. the subsequent slowdown during lunchtime and the repeated increase in the afternoon when the US markets open.l N k . taking into account the relation between order flow (OF) and volatility (h): ht .n + t .manuscriptcentral. i.n = OFt .n In the case of the return analysis.4. First. as the return shows a distinctive direction and so does the signed order flow (positive values of the order flow mean that market participants seek to buy forint and sell euro). There are.n + 2 h =1 h OFt h . Note that we use absolute order flow in the volatility equation.n = 1+ + i =1 i pt i .n h OFt h. however.n cover.International Journal of Finance & Economics Page 16 of 32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 gary’s main macroeconomic news are usually published and trading starts getting more active.n m + K m k . including lagged returns to the order flow equation is even more appealing as it explicitly allows for the existence of feedback trading.n iew 3. Second. It is obvious that in the order flow equation there is a strong negative relation between present absolute order flow and lagged volatility.n h + 4 j =1 j dayj + K m k . we refer to the signed order flow instead of the absolute one.n 2 h =1 ee h OFt h .n = OFt .

com/ijfe rP Fo ee rR ev iew . this is captured by the following variables. a change in the base rate causes the buying or selling pressure to be higher after the meeting. 2005. and the chances of one-directional pressure on the asset (which is shown by absolute order flow) should be less in the following period(s). It seems that there is no clear-cut tendency concerning the effects of central bank communication. the more traders (the "smarters" or who has private information) agree on the overvalued/undervalued price of the asset. 2003. there is only little impact on order flow before the interest rate decision.e. Prior to the meetings. This phenomenon can be due to the lower frequency. Monetary policy meetings. Moreover. (Andersen et al. These results are in line with the empirical literature. absolute order flow and volatility. as everybody expects the figures. apparently significantly affect both. The negative relation between (present/lagged) order flow and (lagged/present) volatility is reasonable. We do not find a dayof-the-week effect. In contrast. volatility increases considerably contemporaneously and after the announcements. Thus. The reaction on positive news is less unambiguous.Page 17 of 32 International Journal of Finance & Economics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 more. Similarly. which reflects the position taking due to the new interest rate settings. In contrast. whether there was an interest rate decision. the higher selling/buying pressure occurs in the market. Whereas the variable meeting does not distinguish. a highly significant intraday pattern can be observed in volatility. order flow. etc. the absolute value of order flow. as we have seen previously that the impact of news releases on order flow emerges within very short time. We interpret this as position taking in expectation of the meeting’s decision. The significance and the sign of the parameters vary along the relative time and the type of the communication. The impact is especially strong in the case of negative news items: while volatility decreases before the releases. Intraday volatility patterns are a wellknown feature of FX markets. volatility tends to be low prior to interest rate cuts. as central bank communication is not pre-scheduled in most 16 http://mc. but still consistent with our previous results: there is an increase in volatility only 10 minutes after the publication. the one-sided trading activity. Accordingly. i. Macro news announcements do not imply aggressive one-sided trading. as well. Interestingly.manuscriptcentral. as well as positive autocorrelation observable in financial variables. If the volatility is increasing showing higher uncertainty. 2008). and volatility increase and decrease after the meeting. such as volatility. Both variables can be regarded as a measure of market uncertainty about an asset’s fair value. as the surprise triggers heavy trading. some of the effects may have already been offset. which may cause a decrease in the uncertainty about an asset’s value. Frömmel et al. The only exception is the interval 20 minutes after positive news has been published. when the one-sided trading seems to be more intense than during other times. especially when the base rate is changed. there is a measurable direct effect of news on volatility. then trades will be more probably symmetrically initiated by sellers and buyers. Bauwens et al. that means.

if the Monetary Council increases the base rate. order flow and volatility. a look at the interest rate decisions seems to be more informative. measured with the number of deals conducted. is positively related both to the absolute order flow and volatility. as interest rate changes significantly affect both the returns and the flows. This may be caused to some extent by negative feedback trading present on the Hungarian FX market. The order flow equation shows significant negative autocorrelation of the order flow. In contrast. The explanatory power of the order flow regression is very impressive with a R2 higher than 65%.manuscriptcentral. including those with and without decision. to keep in mind that the variable meeting captures very different meetings. Again. and around the meetings net forint purchases are observable.com/ijfe ee rR 17 ev iew . it is not expected to have significant impact on the market prior to the events. as well. They are linked. returns are negatively autocorrelated. however. with a lag of about ten minutes. In accordance with this. monetary policy decisions and communication. 17 Csávás and Varga (2006) find that depending on the sample period both negative and positive feedback trading is observable in the Hungarian FX market. One has. i. and shows a close relation between the news and both. Monetary policy meetings can coincide with either appreciation or depreciation after the meetings depending on the relative time. Thus. The intensity of trading. that means two thirds of the market pressure from asymmetrically initiated trades can be explained by fundamentals.17 On the other hand.International Journal of Finance & Economics Page 18 of 32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 of the cases. news that are expected to cause an appreciation of the forint. Meanwhile in the case of volatility regression the R2 shows that the mentioned variables account for 16% of price variability. the impact of interest rate hikes on order flow is unambiguous: hikes cause aggressive one-sided trading on the HUF buy side. The estimation results for the returns are presented in Table 7. The impact of rate cuts on order flows is not fully consistent over time: both buying and selling pressure can be observed. the forint tends to appreciate. the only exception is 10minute lead prior to the decision. absolute order flow and trades variables. we can draw just a cautious conclusion: Four significant and relevant parameters indicate that central bank communication (essentially negative communication) might decrease the uncertainty in the market. and a strong negative relation between present order flow and lagged returns. Interest rate cuts mainly coincide with depreciation of the forint. while there is significant positive (negative) relation between the returns and the contemporaneous and first (second) lag order flow. The reaction on interest rate changes is therefore almost fully consistent with theory. The results on macroeconomic announcements are also mainly consistent with theory: The strongest reaction follows positive news on the Hungarian economy. which can be regarded as relatively high. rP Fo http://mc. and lagged/contemporaneous values of volatility. which may be surprising or not.e. such as macro news.

(2007) and Égert (2007). This is in line with the in general more ambiguous character of this “news” category and the fact that they are not scheduled18. whether the role of order flow in this market is the 18 The latter point makes it even difficult to explain the significance of leads. The results on the communication variables demonstrate the importance of central bank communication for the HUF/EUR exchange rate and are in line with Rozkrut et al. and raise the question. but on the contrary the fit of the exchange rate return equation may be considered rather good (R2=22%).Page 19 of 32 International Journal of Finance & Economics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 to buying pressure and accordingly with appreciation of the forint. we consider the exchange rate of a small open economy. but also have an ongoing impact on the exchange rate after publication of the news.com/ijfe . Though. There seems to be some kind of overshooting. Neutral communication implies some negative order flow and depreciation. as part of the initial purchases are offset within the next ten minutes. Together with the previously found consequences these results imply that macroeconomic and microstucture variables together can explain a considerable part of exchange rate movements and the characteristics of the trading activity on the HUF/EUR market. albeit not significant depreciation of the forint. 18 http://mc. it is also true in this equation system. rP Fo ee rR ev iew 4. but this may also be due to the difficult timing of communication events. appears to imply forint purchases and appreciation. while negative communication goes along with forint sales. We experience a relatively low explanatory power of the order flow equation. Therefore we find a strong asymmetric behavior of market participants.manuscriptcentral. than the exchange rate movements. Conclusions During recent years the analysis of order flow has become an important issue. Together with results from the previous section it is shown that news do not show only a very short-lived effect on volatility during the most recent minutes after publication of the news. Our study adds to this discussion by complementing earlier papers. there is only a slight. Positive communication. but less strong. While most of the earlier studies focus on exchange rates between the world’s major currencies. being extensively debated in academia as well as by policy makers. Whereas the order flow shows significant sales after the news. that the significance and the sign of the parameters differ depending on the relative time and the type of the communication. Central bank communication seems to affect the trading process more intensively. The reaction of the exchange rate is only slightly significant. The impact of negative news is also visible. however. but wrong-signed. characterized by a less liquid market and comparatively high capital flows. as it could be seen earlier.

Both differences of the HUF/EUR market – higher importance of order flow and relevance of communication – may be attributed to the presence of private information due to the factors mentioned in the introduction. namely order flow has additional information content and stronger impact on the exchange rate (i. they are not comparable in quantitative terms. Decomposing exchange rate responses to news into direct and indirect (intermediated) parts we quantify the contribution of order flows to total exchange rate responses to news. Our results regarding the exchange rate. We find a relation between order flow. news and the exchange rate that does not differ in its basic structure. Order flow itself can explain a significant part of exchange rate variation. However.com/ijfe ee rR 19 ev iew . we find basically the same results as reported in the literature for developed foreign exchange markets. This is even more remarkable as the HUF at this time was pegged to the EUR and our sample covers a period of strong pressure on the forint. although the results are qualitatively the same.e. macroeconomic news and order flow indicate that the HUF/EUR market behaves qualitatively in a very similar way as major and developed markets in the sense that we find a strong reaction to news. Our results suggest that three quarters of total price reaction are intermediated by order flow.International Journal of Finance & Economics Page 20 of 32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 same as for the major ones.manuscriptcentral. the same magnitude buying/selling excess causes larger appreciation/depreciation). Overall. but also the strong relation between the exchange rate and order flow as reported in the literature. In addition. the order flow – exchange rate relation changes significantly around news releases. rP Fo http://mc. when scheduled public information reaches the market. reflecting the experience that the MNB relied on verbal intervention rather than on actual interventons in most times (Egert 2007). To make the results comparable we conduct our analysis in the standard framework of order flow analysis. The differences are even more striking for the communication variables. whereas the share for major markets is usually estimated as 30-50 per cent. In contrast to most other currencies they seem to play an important role in the HUF/EUR market. Not surprisingly the importance of order flow seems to be higher than it was found in major currency markets.

Journal of Financial Economics 75(3). What defines ´News` in the Foreign Exchange Markets?. Market Activity. and the Central Bank’s Understanding of Exchange Rate Formation. Clifton Green (2001). Cai. Alan Chaboud. Intraday Periodicity and Volatility Persistence in Financial Markets.K. 38-62. Bauwens. Michael Ehrmann.K. pp. Intervention. Central Bank Communication and Monetary Policy: a Survey of Theory and Evidence. Iyer. Jakob de Haan. International Economic Review 39:4. Journal f Financial and Quantitative Analysis. and Freyan Panthaki (2006). 219-265. Alexis (2003). News Announcements. Torben. Realized Volatility and Optimal Sampling. Treasury Market. Canales-Kriljenko (2003). Andersen. issue 1. FOREX Microstructure. Journal of Empirical Finance 4.S. MNB Bulletin volume 1. Andersen. Torben and Tim Bollerslev (1998a). Russel (2003). 327-347. Marcel Fratzscher and David-Jan Jansen (2008).Page 21 of 32 International Journal of Finance & Economics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 References: Almeida. Journal of International Money and Finance 20. Francis X. Main characteristics of non-residents’ trading on the foreign exchange and government bond markets. David W. Luc. ‘Once in a Generation’ Yen Volatility in 1998: Fundamentals. Raj S. 571-605. Diebold and Clara Vega (2003). Journal of International Economics. Balduzzi. CNB Working Paper 6/2003 Dominguez. Economic News and Bond Prices: Evidence from the U. Edwin J. Alvaro. forthcoming. Micro Effects of Macro Announcements: Real-Time price Discovery in Foreign Exchange. Bandi. Torben and Tim Bollerslev (1998b). and Volatility in the Euro/Dollar Foreign Exchange Market. Journal of International Money and Finance 25. Order Flow and Exchange Rate Dynamics in Electronic Brokerage System Data. Journal of International Money and Finance 24. Lee and Michael Melvin (2001). Geir Høidal and Dagfinn Rime (2005). Blinder. Csaba and Lóránt Varga (2006). Bjønnes. Journal of Financial and Quantitative Analysis 33:3. Raymond S. University of Chicago. Charles Goodhart and Richard Payne (1998). The Effects of Macroeconomic News on High Frequency Exchange rate Behavior. Jun. David Liu and Jonathan H. Tim Bollerslev. Standard Volatility Models Do Provide Accurate Forecasts. Yan-Leung Cheung.. Answering the Sceptics: Yes. Csávás.E. Torben and Tim Bollerslev (1997). and Order Flow. Federico M. Derviz. IMF Working Paper 03/95. and Jeffrey R. Pierluigi. Sergey Chernenko. Wright (2008). Walid Ben Omrane and Pierre Giot (2005).. Elton and T. American Economic Review 93:1. Microstructure Noise. Andersen. 168-198. 898. 523-543. Alan S. Dealer Behavior and Trading Systems in Foreign Exchange Markets. Edward Howorka. ECB Working Paper no. 36. Invisible Price Determinants. 383-408. Berger.manuscriptcentral.com/ijfe ee rR 20 ev iew . 885-905 Andersen.. Journal of Finance 53:1. unpublished manuscript. Kathryn M. 115–158. rP Fo http://mc. Foreign Exchange Intervention in developing and Transition Economies: Results of a Survey. Macroeconomic Announcements and Longer Run Dependencies. 1108-1125. Deutsche Mark-Dollar Volatility: Intraday Activity Patterns.

MNB Bulletin volume 1. Péter and Pintér. Evans. (2006). Lyons (2006a). and Marcel Fratzscher (2007).D. Journal of Money. Informed Trading and the Bid-Ask Spread: Evidence from an Emerging Market. Do Currency Markets Absorb News Quickly? Journal of International Money and Finance 24. Áron. Communication and Interest Rate Policy in Emerging European Economies. Martin D. Hughes Hallett. Fratzscher. The Impact of News and Alternative Theories of Exchange Rate Determination. Michael. Martin D. Frömmel. MNB Working Papers 2006/9 Gereben. 275-296. 170-180. Risk Reduction in the EMS? Evidence from Trends in Exchange Rate Properties. Evans. Martin D. Andrew and Myrvin L. Lyons (2002a). 21 http://mc. Jan and Richard Podpiera (2003). Journal of International Money and Finance 27:6. 17. Dennis and Don Edward Schlagenhauf (1985). Evans. Evans. Journal of Finance 54.com/ijfe rP Fo ee rR ev iew . and Norbert Kiss M.D. Central Bank Interventions. Ehrmann. Michael. 197-217. Journal of Money. Same Effectiveness?. and Richard K. Journal of International Money and Finance.D. Lyons (2005). NBER Working Paper No. Áron. Treasury Market: The Response to Public Information. 261281. 9433. Martin DD. Order Flow and Exchange Rate Dynamics. 19011915. Hoffmann. 39(2-3): 509-41. Klara (2006). Understanding order flow International Journal of Finance & Economics 11:1. and Richard K. The effect of the MNB’s communication on financial markets. Alexander Mende and Lukas Menkhoff (2008). Richard (2002b). FX Trading and Exchange rate Dynamics.D. How is Macro News Transmitted to Exchange Rates?. 39. Price Formation and Liquidity in the U. Time varying liquidity in foreign exchange. 49. 328-346.International Journal of Finance & Economics Page 22 of 32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 Égert. and Richard K. Economic Journal 118:530. Balázs (2007). Journal of Comparative Economics 31. 537-560. Order Flows. Credit. 1025-1051. Communication by Central Bank Committee Members: Different Strategies. and Banking. and Eli M. Evans. News. Martin D. and Richard K. Customer order flow. (2006). Evans. Journal of Comparative Economics 35(2) 387413. information and liquidity on the Hungarian foreign exchange market. Credit and Banking. Remonola (1999). Oral Interventions Versus Actual Interventions in Fx Markets – An Event-Study Approach. Michael and Lukas Menkhoff (2001). György Gyomai and Norbert Kiss M.manuscriptcentral. Anthony (1997). MNB Working Papers 2006/8 Hanousek. 3-23.D. 16. Gábriel. Journal of Political Economy 110. (2002). 9941012. Fleming. Journal of Finance 57. and Exchange Rate Volatility. Lyons. 2405-2448. Martin D. Journal of Common Market Studies. Michael J.S. issue 2. Gereben. Frömmel. 1079-1106.. Journal of Monetary Economics. Marcel (2008). A brief overview of the characteristics of interbank forint/euro trading. Lyons (2006b). Exchange Rate Behaviour under the EMS Regime: Was there any Systematic Change?.

. Journal of Financial and Quantitative Analysis . Lucyna Sztaba and Radosław Szwaja (2007). Mark P. 307-329. (1985). Is Foreign Exchange Intervention Effective? Some Microanalytical Evidence from the Czech Republic. 55-80. Ryan and Richard Payne (2008). forthcoming. (1995). rP Fo http://mc. Michael and Xixi Yin (2000). International Journal of Finance & Economics 11:1. Sarno. Michael J. Antonio (2008). Taylor (2006). 55-76. (2001). Norbert (2004). forthcoming.com/ijfe ee rR 22 ev iew . Econometrica 53:6. 321-351. A jegybanki kamatdöntések el rejelezhet sége és hatása a hozamokra (only in Hungarian). Vitale. Osler. Elsevier Science. 13151335. Forecasting the Foreign Exchange Market Using Private Information. and Quote Frequency. Stefan and Mark P. (2001). Under the microscope: the structure of the foreign exchange market. Marek. The Effects of Macroeconomic News on Money Markets. Pintér. Klára and Wenhardt Tamás (2004). European Economic Review 52:1. International Journal of Finance & Economics 11:1. (2006). Lyons. How does it Work?. 14/2000. Lukas and Maik Schmeling (2008). A Guided Tour of the Market Microstructure Approach to Exchange rate Determination. Order Flows and Exchange Rates. 839-868. Journal of International Economics 61. MNB Occasional Papers 2004/31 Reitz. SSRN Working Paper No. And Mark P. The Microstructure Approach to Exchange Rates. Official Intervention in the Foreign Exchange Market: Is it Effective. The Economic Journal 110. Rozkrut. Love. Dagfinn (2001). Journal of International Money and Finance. Payne. Richard (2003). Dagfinn (2003). Tests of microstructural hypotheses in the foreign exchange market. Whose Trades Convey Information? Evidence from a Cross-Section of Traders. European Journal of Political Economy 23(1) Sager. Quest for central Bank Communication. Macro lessons from microstructure. Macroeconomic News. Does it Pay to be Talkative?. The Coordination Channel of Foreign Exchange Intervention: A Nonlinear Microstructural Analysis. Private or Public Information in Foreign Exchange Markets? An Empirical Analysis. Informed Trade in Spot Foreign Exchange Markets: An Empirical Analysis. Paolo (2007).manuscriptcentral. Richard K. 903-934. Lyons. Journal of Financial Economics 39. Jones (ed. Melvin. Rime. Continuous Auctions and Insider Trading. Taylor (2008). Carol L. Lindahl. Richard K. Journal of Financial Markets. Lucio and Taylor. 529-546. Journal of Economic Surveys 21. Krzystof Rybi ski. 8195. Scalia. and if so.Page 23 of 32 International Journal of Finance & Economics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 Kiss M. MNB Occasional Papers 2004/30 Kyle. Public Information Arrival. Journal of Economic Literature 39. 27. Albert S. Exchange Rate Volatility. New electronic trading systems in foreign exchange markets. Menkhoff. 644–661. Rime. In Derek C. unpublished manuscript. Anna and Dagfinn Rime (2006).) New Economy Handbook. MIT press.

New York University Salomon Center Working Paper #S-98-3 rP Fo http://mc. Jian (1998).International Journal of Finance & Economics Page 24 of 32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 Yao.manuscriptcentral.com/ijfe ee rR ev iew 23 . Market Making in the Interbank Foreign Exchange Market.

Page 25 of 32 International Journal of Finance & Economics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 Tables: TABLE 1.00004 0.50 0.409 -0. Description of news announcements Observations1 24 8 24 24 http://mc.31 -0.129 0.089 Returns 29 355 h= 10 minutes Order flow 29 355 0.184 0.51 -0.21 308 1.94e+13 0.00046 0.137 0.074 0.053 0.21 -0. Central European Time 4.com/ijfe ee Source2 CSO CSO MNB Ministry of Finance MNB Release time3 9:00 9:00 8:30 Varies: 17:00 or 10:00 14:00 Sign4 -1 +1 +1 +1 rR 24 ev iew .000 TABLE 2. +1 (-1) if greater than expected news is assumed to coincide with appreciation (depreciation) of Hungarian forint rP Fo -0. CSO: Central Statistical Office.0574 5. Order flow and exchange rate returns summary statistics Horizon Number of observations Mean Standard Deviation Skewness Kurtosis JB normality test Proportion positive Autocorrelation of order 1 Returns 293 550 0.59 3347 1.50 63563 4.18 6.15 0.14e+09 0.51 0.218 0.53 -0. Total number of observations in the sample period 2.308 h=1 min Order flow 293 550 0.002 Autocorrelation of order 2 Autocorrelation of order 3 Consumer price index GDP growth Current account balance Public sector balance MPC meetings – base 42 +1 rate decisions Notes: 1.005720 1.78 369 1.manuscriptcentral.37e+10 0. MNB: Magyar Nemzeti Bank (the central bank of Hungary) 3.64e+08 0.007 -0.

490*** (0.449 1.70) (1.00) ev iew 0.000 0.019 -0.001*** returns MC meeting (-0.61) 0.10) CPI (1%) 0.009*** GDP -0.02) (3.21) (1.89) (3.96) (3.36) 0.246 0.001 (0. *: 10 per cent rP Fo 0.40) (2.760* 1.55) 0.84) (0.02) (-0.023*** (4.016 -0.95) -0.2*** order flows (1.31) 0.450 order flows (0.58) 0.008*** returns CA (100 (1.16) (-1.01 0.06 0.20) million EUR) 0.82) (-9. *: 10 per cent 0.42) (-8.94) 0.008*** BD Eq (5) (-0.453** order flows (0.010*** (32.27 -3.227*** order flows (1.001 0.001** (2.18) (-8.000 (-0.051 0.002 (1.48) (0.76) (2.034 returns (-0.000 0.449*** 0.158*** (3.01 0.61) (7.094 0.54) http://mc.12 0.31) 0. **: 5 per cent.19) (-5.16 0.000 (1.17) (1.60 Asterisks refer to level of significance.008*** (21. ***: 1 per cent.84) (-0.06 0.31) (1.manuscriptcentral.38) (6.10 TABLE 4.025 (0.28 .42) (5.167 -3.55) 0.028*** (5.com/ijfe ee rR 25 OF OF*I(0) Release time R2 (adj) 0.021** returns BD (100 (-0. **: 5 per cent.42) 0.014** 0.03) (2.80) (2. The effects of macroeconomic announcements on exchange rate and order flow (absolute measure of surprises) News variable (unit of surprise) Dependent variable returns Constant 1 min lead Release minute 1 min lag 2 min lag 3 min lag 0.11) (-0.004 (1.58) (-1.88) 0.047** -0.528 -1.101* 0.010*** (20.001** -0.50) (1.006 0.76) 2.005 -0.01) (-0.50) 0.59 (0.06) (0.82) R2 (adj) 0.06) 0.44) 0.152*** 0.010 -0.06) billion HUF) 2.006 0.653*** (2.52) 0.46) 0.63) 0.60) -0.01) (1.279 0.11) (0.22) (0.207*** -0.02 0.002* 0.010*** CA 0.43) (1.98) GDP (1%) 0.598 0.73) 0.42) (0.000 (0.23 0.72) (2.04) (-1.011** MC meeting 0.006** (2.044 0.006 0.97) (1.27) Asterisks refer to level of significance.05) (0.37 (1.16 0.009 0.26 0.17) (1.09 0 0 0.26 -17.25) (0.238** order flows (3.000 (1.011 0.026* (1.38) 0.392 (1.000 -0.International Journal of Finance & Economics Page 26 of 32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 TABLE 3.44) (1.013*** (6.001 0. The sensitivity of order flow parameter to news releases constant Benchmark equation Whole sample 2003 2004 Announcement category CPI -0.04 0.000 (0.45) 0. ***: 1 per cent.012*** (5.01 0.49) -0.001 0.14) (2.35) 0.45) -0.020 0.40) 0.004 0.26) (10 bp) 0.010*** (34.13 0.047 0.26 0.902* 0.20 0.87) 0.000 (0.04) (-0.10) (0.72) (11.

627*** -0.005*** -0.manuscriptcentral.028*** -0.565*** -0. ***: 1 per cent.097*** -0.320 R2 (adj.828*** -1.005*** -0.898*** -0.009*** -0.061 0.826*** 0.189*** 0.004*** Surprises Budget deficit Lag -2min Current account Lag -1min Lag -2min Lag -3min CPI Lag -2min GDP Lag -3min MT Lag -1min Lag -2min Lag -3min Lag -4min Lag -8min Lag -9min Trade Lag -2min Lag -3min D_MT(1) 0.448* 0.026* Return -0.273*** 0.872*** -0.749*** -0.065 iew .882*** 4.447*** -0.com/ijfe rR ev -0.) DW 2.160*** -1. *: 10 per cent.985* -0.012*** OF -0.525*** -0.006 -0. **: 5 per cent.031** -0.004 Asterisks refer to level of significance.036*** 0.003 2.670*** -0.540*** -3.030*** 0.069*** 0.380** 0.019 0. Coefficients for leads and lags are only shown if significant at least at the ten per cent level rP Fo ee 26 -1.005*** -0.200*** -14.204 -1.020** -0.660*** -0.618*** 0.609*** 0.031*** -0.404** -18.041 http://mc.640** -0.Page 27 of 32 International Journal of Finance & Economics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 Table 5: VAR results Dependent variable return Lag -1min Lag -2min Lag -3min Lag -4min Lag -5min OF Lag Lag Lag Lag Lag -1min -2min -3min -4min -5min 0.320* -0.004*** -0.017 0.596*** -0.

026*** -0.528 0.841* -0.004 -0.054*** -0.275 0.026*** 0.652 0.211 0.010** -0.437 1. *: 10 per cent.512 0.211*** ht.373 -0.015** -0.138*** -0.752* -0.721* -0. **: 5 per cent.046** -1.010 0.931** -0.022 0.463*** 0.010 -0.033*** 0.070*** 0.021*** -0.285 -0. ***: 1 per cent.n -0.121 -0.412 0.000 -0.220** 0.045*** -0.907** 0.060*** 0.255** 0.033 0.022** -0.manuscriptcentral.217*** -0.908*** Interest rate down 10 min before event 20 min before event 10 min after event 20 min after event Interest rate up 10 min after event 20 min after event Macro news Hungary Negative news(1) 10 min before event 20 min before event 10 min after event 20 min after event Positive news (1) 10 min after event 20 min after event Communication Positive communication(1) 20 min before event 10 min after event Neutral communication(1) 10 min before event 20 min after event Negative communication(1) 10 min after event 20 min after event Trades Lag -10min 0.343 1.088*** 0.734* -1.363 0.com/ijfe rR ev -0.153*** iew .094*** 0.792 -0. and vice versa.927** 0.International Journal of Finance & Economics Page 28 of 32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 TABLE 6.801** 0.472 -0.076 1.533*** -0.014 -0.173 -1.156 R2 (adj.501 0.499 0.001 1.006 -0.064*** -0.009 0.160*** -0. Regression on volatility and absolute order flow Dependent variable const Intraday pattern Volatility Lag -10min Lag -20min |OF| Lag -10min Lag -20min Monday Tuesday Thursday Friday Monetary policy Meeting 10 min before event 10 min after event 20 min after event -2.011* 0.654* -0.108 0.010*** 0. rP Fo ee 27 http://mc.) DW 2.141 0.015** 0.008 -0.016*** 0.003 2.003 Asterisks refer to level of significance.038*** 0.060** |OFt.562 1.148*** -0. Coefficients for leads and lags are only shown if significant at least at the ten per cent level (1) Negative news and communication events are those which are assumed to imply depreciation of the forint.690*** 0.685* 0.009** 0.n| 0.068 0.022*** -0.206 0.

679*** 20 min after event -0. **: 5 per cent.063** return Lag -10min -4.409*** -0.146*** Lag -20min -0.002 10 min after event 0.020* 20 min before event 0. *: 10 per cent (1) Negative news and communication events are those which are assumed to imply depreciation of the forint.042*** 0.001*** 20 min before event 0.114 0.manuscriptcentral.008 Positive communication(1) 10 min before event 3.Page 29 of 32 International Journal of Finance & Economics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 TABLE 7.011 -1.002 -2.001*** 20 min after event 0.033 -0.412 0.126*** -0.556*** 0.053 -0.002 20 min before event 0.234*** -0.008 Neutral communication(1) 10 min before event -2.010 Positive news (1) 10 min before event -0.796 0. rP Fo http://mc.424 -0.028** 10 min after event 2.012 0.033*** OF 0.684** 20 min after event -0.876** 0.001*** 20 min after event -0.778 -0.031*** -0.060*** Interest rate up 0.588*** 0.0001 0.0002* Lag -20min -0.910 0. and vice versa.035*** 20 min after event 0.190 -0.126*** 0.003 Negative news(1) 10 min before event -0.017* 10 min before event 0.002*** 10 min before event 0.000** 20 min before event -0. Regression on returns and order flow Dependent variable OF Return const -0.021** 0.017 20 min before event -0.039*** -0.040*** 20 min before event -0.001*** Macro news Hungary -0.007*** Lag -10min -0.005 2.067*** 20 min after event 1.014 20 min after event 0.222 R2 (adj.703*** -0.015 20 min before event 0.060*** 10 min after event 1.037*** Communication 0.207*** 20 min before event 0.004*** 10 min before event -0.065 0.056*** 10 min before event 0.192*** -0.251*** 0.029 Negative communication(1) 0.031*** Interest rate down 0.003 2.006 -1.941 -0.523* 0.783 0.014 10 min after event 0.019* 2.034 -0.007 10 min after event -0.036 -0.com/ijfe ee rR ev iew .003 Asterisks refer to level of significance.013 0.899 0.010 0.001 -0.024*** 0.152*** -0.310 0.981** -0.765 0.995*** 0.014 20 min after event -0.012 -1.001*** 10 min after event 0.000 10 min after event 0.905 0. ***: 1 per cent.) DW 2.0002* Monetary policy Meeting -0.016 10 min after event -0.000 -0.

International Journal of Finance & Economics Page 30 of 32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 r Fo http://mc.com/ijfe Pe er Re vi ew .manuscriptcentral.

manuscriptcentral.Page 31 of 32 International Journal of Finance & Economics 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 r Fo http://mc.com/ijfe Pe er Re vi ew .

com/ijfe Pe er Re vi ew .manuscriptcentral.International Journal of Finance & Economics Page 32 of 32 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 r Fo http://mc.