NBER WORKING PAPER SERIES

WHEN DO CENTRAL BANK INTERVENTIONS

INFLUENCE INTRA-DAILY AND LONGER-TERM
EXCHANGE RATE MOVEMENTS?

Kathryn M. E. Dominguez

Working Paper 9875

http://www.nber.org/papers/w9875

NATIONAL BUREAU OF ECONOMIC RESEARCH

1050 Massachusetts Avenue
Cambridge, MA 02138
July 2003

I am grateful to Chang-ching Lin for outstanding research assistance, Olsen and Associates for providing the
FXFX intradaily indicative quotes and the Reuters news tape, Tim Bollerslev and Michael Melvin for sharing
their flexible fourier form regression programs and Alain Chaboud and Steve Weinberg for providing the
daily integrated volatility series. I thank Philippe Bacchetta, Paul DeGrauwe, Charles Goodhart, Rich Lyons,
Richard Olsen, Carol Osler, Mark Taylor and Jonathan Wright for helpful comments and suggestions. The
views expressed herein are those of the author and not necessarily those of the National Bureau of Economic
Research

©2003 by Kathryn M. E. Dominguez. All rights reserved. Short sections of text, not to exceed two
paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given
to the source.
When Do Central Bank Interventions Influence Intra-Daily and Longer-Term
Exchange Rate Movements?
Kathryn M. E. Dominguez
NBER Working Paper No. 9875
July 2003
JEL No. F31, G14, G15, E58

ABSTRACT

This paper examines dollar interventions by the G3 governments since 1989, and the reasons
that trader reactions to these interventions might differ over time and across central banks. Market
microstructure theory provides a framework for understanding the process by which sterilized

central bank interventions are observed and interpreted by traders, and how this process, in turn,
might influence exchange rates. Using intra-daily and daily exchange rate and intervention data, the

paper analyzes the influence of interventions on exchange rate volatility, finding evidence of both
within day and daily impact effects, but little evidence that interventions increase longer-term
volatility.

Kathryn M. E. Dominguez
Gerald R. Ford School of Public Policy
University of Michigan
Lorch Hall
611 Tappan Street
Ann Arbor, MI 48109
and NBER
kathrynd@umich.edu
I. Introduction

On May 31, 1995 the U.S. government purchased a total of $500 million against

marks and $500 million against yen on three occasions between the hours of 1:45pm and

2:26pm (Eastern Standard Time), resulting in a 2% increase in the value of the dollar

against both the mark and yen over the course of the day. 1 On other occasions when the

U.S. government intervened in the dollar exchange rate market, however, the dollar either

moved in the opposite direction to that expected, or did not move at all. This paper

examines dollar interventions by the G3 governments since 1989, and the reasons that

market reactions to these interve ntions might differ over time and across central banks.

Standard models of exchange rate determination identify at least two channels

through which interventions might be expected to influence exchange rates: the portfolio

balance channel and the signaling channel. However, neither of these channels is easily

reconciled with the empirical evidence, which suggests that sometimes intervention

works and sometimes it does not. Of course, standard exchange rate determination

models have a difficult time expla ining (often the lack of) exchange rate reactions to all

kinds of purportedly fundamental information, suggesting that it may be worth

reexamining standard models before drawing conclusions regarding the efficacy of

intervention.

One approach to exchange rate modeling that has gone some distance toward

reconciling observed short- term currency movements and economic theory is the market

microstructure approach. In the context of intervention, market microstructure provides a

1
During New York trading hours on May 31, 1995 the dem-usd rate opened at 1.385 and closed at 1.4135
and the yen-usd rate opened at 82.70 and closed at 84.40. The Bundesbank and the BOJ coordinated their
interventions with the Fed on this day. Reuters reports indicate that the Bundesbank purchased $395.6
million against the mark on two occasions (starting just before the Fed was in the market), and the BOJ
purchased $767.4 million against yen on one occasion (just before the last Fed operation).

1
framework for understanding the process by which central bank interventions are

observed and interpreted by traders, and how this process, in turn, might result in

exchange rate changes. Even if interventions are informative, for example if they reveal

information that is considered pric e relevant, exchange rates might not react immediately

if traders learn about interventions at different times. Likewise, if interventions are

largely non- informative (and are simply a central bank’s attempt to target exchange rates

away from fundamentals), interventions might still impact prices and volatility in the

very short-run if traders misinterpret the (lack of) information content of the

interventions.

Recent advances in market microstructure theory, new sources of data on

exchange rates and central bank interventions, and in particular, the availability of high

frequency data, offer new tools with which to shed light on the old question of when

central bank interventions are likely to influence exchange rates. 2 Section II offers a

simple version of a market microstructure model and introduces a role for intervention.

Section III describes the G3 intervention and exchange rate data. Section IV provides an

empirical examination of the intra-day and daily dynamics of interventions and exchange

rate volatility. Section 5 is the conclusion.

II. Market Microstructure and Intervention

The exchange rate microstructure model developed by Bacchetta and van

Wincoop (2003) provides a way to think about why trader heterogeneity (based on

differences in information or the interpretation of information) might lead to short-run price

2
See Dominguez and Frankel (1993ab) and Humpage (1999). Sarno and Taylor (2001) and Edison (1993)
provide excellent surveys of the intervention literature. Also, see Dominguez (2003b), Ito (2002), De
Grauwe and Grimaldi (2003) and Taylor (2003) for recent contributions.

2
and volatility effects in reaction to information revelation. 3 Interventions can easily be

included in the model with the potential to provide price-relevant information. In this

model, both information-based trades and non- informative trades can move exchange rates

in the short run depending on aggregate market ability to differentiate noise from

fundamentals. The model assumes that longer-term exchange rate behavior is well

explained by fundamentals.

Consider a standard asset pricing model of exchange rates in which the current

exchange rate, et, is the discounted present value of expected macro fundamental

differentials, Mt+k , and a risk premium, rpt+k , associated with non-fundamentals trade. 4

Assume traders have higher order expectations, Etk (such that the exchange rate at time t

depends on the fundamental at time t, and the average expectation of the fundamental in all

future periods, as a consequence of which the expectation of other investors’ expectations

matters). Exchange rates may then be expressed as:

k
1 ∞
 β  k
(1.1) et =
1+ β
∑   E t ( M t +k − rpt + k ),
k=0  1+ β 

where β can be interpreted as the interest semi-elasticity of money demand. If market

participants receive information (or signals) at time t about future fundamentals, Mt+k , but

this information is not common knowledge (either because people receive difference bits

of information or because they interpret the information differently), then there is a

common average signal among traders (assuming there are large numbers of market

3
See Lyons (2001) for a thorough discussion of market microstructure in foreign exchange markets as well
as Evans and Lyons (2002ab). For a more general treatment of market microstructure see O’Hara (1995).
4
In Bacchetta and van Wincoop (2003) the risk premium arises because some proportion of investors have
a noisy expected excess return on foreign bonds and the net supply of foreign bonds resulting from the
expectational error is non-observable (so that the irrational agents do not know their expectational error).

3
participants) but heterogeneity across individuals (and traders will expect their own

expectation next period to differ from that of others). Bacchetta and van Wincoop (2003)

show that this sort of information heterogeneity leads both to magnification and to

endoge nous persistence of the impact of non- fundamentals trade on the exchange rate.

Government exchange market interventions in this kind of setting may provide

information to market participants that allows them to distinguish more accurately

between fundamental and non- fundamental information. Define an intervention

operation known to individual i at time t, I ti , as providing information about future

fundamentals with probability ρ ti , and providing non- fundamentals information with

probability (1- ρ ti ):

(1.2) I ti = ρti ( M t + k ) + (1 − ρti )( rpt+ k ) + ηti

where ∑ ηti is the error term that goes to zero if interventions are common knowledge. If

the value of ρ ti is only learned over time, then intervention operations will in the short-

run (before ρ is known) potentially add to the weight of non-fundamentals driven

exchange rate movements

The model implies that, if interventions are “informative” ( ∑ ρti =1) and are

common knowledge ( ∑ ηti =0), then they should help reduce the rational confusion

(between fundamental and non-fundamental induced exchange rate movements) that arises

in the Bacchetta and van Wincoop (2003) model. Likewise, if interventions are not

common knowledge ( ∑ ηti ≠ 0), or if interventions are uninformative ( ∑ ρti <1), then

interventions will increase short-term non-fundamentals induced exchange rate

movements. Over time interventions should become common knowledge and the market

4
will learn the value of ρ . Therefore, only those interventions that are informative should

be expected to have long-term effects on exchange rates. This suggests that the very

short-run influence of interventions may differ from its longer run effects.

Previous empirical tests of how microstructure issues might influence the

relationship between intervention and exchange rates include Beattie and Fillion (1999),

Cai, Cheung, Lee and Melvin (2001), Chang and Taylo r (1998), Dominguez (2003a),

Evans and Lyons (2001), Fischer and Zurlinden (1999), Goodhart and Hesse (1993),

Neely (2002), Pasquariello (2001, 2002), Payne and Vitale (forthcoming) and Peiers

(1997). Each of these studies focuses on the effects of different central banks over

different sample periods, often using different data sets, making cross study comparisons

difficult. As a general matter, and in contrast to papers that study the longer-term effects

of sterilized central bank interventions, studies examining intra-daily effect of

intervention find strong evidence of impact effects. 5

LeBaron’s (1999) finding that intervention days are the source of unusual profits

for traders using technical analysis is noteworthy in this context. He finds that simple

moving average trading rule profits are significant in daily forex data if intervention days

are included in the sample -- when interventions are excluded, profits go to zero. Using

more finely timed data, Neely (2002) however, finds that interventions are unlikely to have

“caused” the increase in trading rule profits, but instead that interventions tend to arise

5
For example, Peiers (1997) examines how interactions between informed (defined to be indications
provided by Deutsche Bank (DB)) and uninformed foreign exchange traders (indications given by all other
banks) give rise to short-term price leadership during periods of central bank intervention. She finds that,
during the period October 1992 to September 1993, volatility increases five minutes prior to Bundesbank
interventions, and that there is evidence of DB price leadership from 60 to 25 minutes prior to Reuters
reports.

5
during periods when exchange rates are trending in a manner that would likely lead to

technical trading rule profits.

III. G3 Intervention and Excha nge Rate Data

The intra-daily exchange rate data used in this paper are the Reuter’s FXFX series

tick-by-tick indicative quotes on Fed intervention days as well as a control sample of 25

days with no interventions. 6 A limitation of the FXFX data is that because they are quotes

and not transactions they do not provide volume information, so it is not possible to

examine the joint dynamics of volume (or order flow) and price. 7 Another disadvantage of

the data set is that, because it includes only intervent ion days, it is not possible to measure

persistent effects of interventions.

The FXFX data used in the paper cover 69 days over the period August 1989 to

August 1995 when the Fed intervened in the dem- usd market, and 66 days when the Fed

intervened in the yen-usd market. 8 The propensity to intervene on a given day varied

across the sample period. The U.S, Japanese and German governments all intervened

actively in the early part of the sample, while only the Bank of Japan (BOJ) continued to

actively interve ne after 1992. Figures 1 and 2 show daily U.S. dollar intervention

6
The data are collected by Olsen and Associates (Research Institute for Applied Economics, Zurich
Switzerland) using O&A proprietary real-time data collection software and are filtered as recommended by
Dacorogna et al. (1993). The control dates were selected to provide a representative sample of non-
intervention days over the period when the intervention operations take place. These data are used to create
the volatility seasonal used in the empirical tests to follow.
7
Goodhart et al. (1996) and Danielsson and Payne (2002) find that the basic characteristics of 5-minute
FXFX returns closely match those calculated for transactions prices but find that quote frequency and bid-
ask spreads in the FXFX data are not good proxies to transaction volume or spreads.
8
Two additional Fed intervention operations have occurred since August 1995. On June 17, 1998 the Fed
sold $833 million against the yen in cooperation with the BOJ and on September 22, 2000 the Fed
purchased a total of 1.5 billion euros against the dollar in cooperation with the ECB, the BOJ, the Bank of
Canada and the Bank of England.

6
operations in the mark and yen markets along with the corresponding bilateral exchange

rates over the sample period. 9

The G3 central banks release historical daily intervention data. Unfortunately,

they do not provide the exact timing of interventions, nor do they disclose how many

operations occurred over the course of the day. 10 The only available source of timing

information for G3 interventions comes from Reuters reports of intervent ions (which is

also the most likely source of information for those traders in the market that are not

directly involved in the intervention transaction).

The Reuters news reports used in this study are from the Reuters AAMM Page

News (Money Market Headline News). Along with reports of central bank intervention,

the Reuters data include announcements of various macroeconomic statistics, statements by

central bank and government officials and reports of major economic events. In order to

control for the impact of other news on exchange rates, these Reuters news reports are also

included in the empirical work. In particular, dummy variables indicate the timing of all

major macroeconomic announcements and statements regarding exchange rate policy by

officia ls of the G-3 central banks on the intervention sample days. Table 1 lists each of the

dummy variables created from the Reuters reports and the day-of-week and average time

(GMT) when the announcements are made.

9
In the United States the U.S. Treasury and the Federal Reserve have independent legal authority to
intervene in foreign exchange markets. In practice, the U.S. Treasury and the Fed typically act jointly and
split the costs of intervention equally against their separate accounts. The New York Fed implements
intervention policy for the United States and for this reason I follow the convention of associating U.S.
intervention operations with the Fed in the paper. Similarly, in Japan intervention decisions are made by
the Ministry of Finance and implemented by the Bank of Japan (BOJ). The Bundesbank had sole
jurisdiction over German intervention decisions and implemented intervention operations prior to 1999.
10
The Swiss National Bank is an exception. The SNB provides exact timing and transaction prices for
interventions (see Fischer and Zurlinden (1999), Payne and Vitale (forthcoming) and Pasquarliello (2001)
for studies using these data).

7
 The Reuters reports indicate that central banks typically intervene during business

hours in their respective markets. 11 Frequency distributions of the times of G3

intervention suggest that the BOJ is most likely to intervene at 3:56:36 GMT (or around

1pm in Tokyo). The Bundesbank is most likely to intervene at 11:31:16 GMT (or at

12:30pm in Frankfurt). And, the Fed is most likely to intervene at 14:57:10 GMT (or

10am EST). Table 2 shows the relative timing of the Tokyo, Frankfurt and New York

markets using the GMT scale and indicates the times when Reuters reports that each

central bank is most likely to be in the market. It is worth noting that Tokyo business

hours end just as the Frankfurt market opens and the New York market overlaps the

Frankfurt market for two hours. The New York market closes two hours before the

Tokyo financial market opens.

Getting the timing of interventions right is critical to measuring the short-term

influence of interventions on foreign exchange markets. Evidence in Andersen,

Bollerslev, Diebold and Vega (2003) and Almeida, Goodhart and Payne (1998) suggests

that conditional mean adjustments of exchange rates to macro news occur quickly

(though they also find that conditional variance adjustments are more gradual). Tables 3

and 4 provide descriptive information about the central bank interventions that are

examined in the next section. The Fed intervened on 268 occasions over 104 days (in

either the dem- usd or yen-usd markets) over the sample period August 1989 through

August 1995. Many of the Fed’s intra-daily interventions were clustered in the same hour

of the day. Most Fed interventions occurred during the overlap in New York and

European trading hours. And 31% of Fed interventions were coordinated with the

11
Neely (2000) provides detailed information about the practice of central bank intervention based on
survey data.

8
Bundesbank in the dem- usd market, while 54% were coordinated with the BOJ in the

yen-usd market. 12

Before formally testing for a relationship between exchange rates, intervention

and macro announcements, it is interesting to examine the 25 largest returns over the

sample period together with the Reuters time-stamped events that surround these

unusually large returns. Tables 4 and 5 present this information for dem- usd and yen- usd

returns, respectively. The timing of large returns and the timing of macro announcements

tends to be very closely aligned. For examp le, many of the large returns are timed within

10 seconds of a (scheduled) macroeconomic announcement. In contrast, some of the

large returns are only loosely synchronized with interventions. The Reuters time-stamp

typically lags the large returns, sometimes by as much as two hours. Of course, it is

possible that the “cause” of the large return is unrelated to any news event reported by

Reuters.

IV. Measuring the Intra-daily influences of central bank interventions

A fundamental property of high frequency data is that observations can occur at

varying time intervals resulting in irregular spacing of quotes. Standard econometric

techniques require regularly spaced data. The approach to irregularly spaced data used in

this paper is to create from these data a regularly spaced time series over a discrete time

interval. Defining the tick-by-tick price (P) as the average of the bid and ask:

[log Pt , h + log Pt , h ]
ask bid

(1.3) Pt, h ≡
2

12
In the intra-day context a “coordinated” Fed intervention is defined as an intervention that occurs within
2 hours of a BOJ or Bundesbank intervention.

9
where t,h is the sequence of tick recording times which is irregularly spaced, then the

regular-space price is defined as:

[log Pt , n + log Pt , n ]
ask bid

(1.4) Pt, n ≡
2

where t,n is the sequence of the regular-spaced in time data and n is the time interval. 13

Equivalently, the nth return (R) on day t is defined as:

(1.5) Rt, n = Pt, n − Pt , n−1

and volatility, Vt,n , is measured as the absolute value of the 5-minute returns.

A number of previous studies have documented a strong seasonal pattern in the

average 5-minute exchange rate volatility (see, for example, Bollerslev and Domowitz

(1993), Dacorogna et al. (1993) and Guillaume et al. (1997)). This seasonality is also

readily apparent in both the sample of Fed intervention days and the control sample days.

Failure to take account of these intra-daily seasonals is likely to result in misleading

statistical analyses. In this paper de-seasonalization of the volatility series is achieved using

the Anderson and Bollerslev (1997ab, 1998) version of Gallant’s (1981) flexible fourier

form regression method.

Figures 3 and 4 shows average absolute dem-usd and yen- usd returns, respectively,

for each 5-minute interval across both the control sample days and the Fed intervention

days, along with estimated intra-day seasonal. The seasonal is calculated using the

Andersen and Bollerslev (1997a) flexible fourier form framework, which decomposes the

13
In practice the 5-minute price series used in this paper is formed by averaging the two immediately
adjacent bid and ask observations to the round 5-minute mark with weights proportional to the distance from
the end of the interval.

10
demeaned 5-minute returns into a daily volatility factor, σt , a periodic component for the

nth intraday interval, st,n and an i.i.d. mean zero unit variance innovation term, Zt,n .

σ t st , n Zt ,n
(1.6) Rt, n − E ( Rt , n ) = 1
N2

The daily volatility component σt in practice is estimated from a MA(1)-FIGARCH(1,d,1)

model for dem- usd and yen-usd fitted over 1752 daily returns from January 1989 though

December 1995. And the seasonal component is estimated using a flexible fourier form

(FFF) regression. Following Andersen and Bollerslev (1997a), and defining xt,n from

equation (1.6) as:

(1.7) xt, n ≡ 2log[ Rt, n − E ( Rt ,n ) ] − log σ t2 + log N =log st2,n +log Zt2, n

the approach is then based on a non-linear regression in the intraday time interval, n, and

the daily vola tility factor, σt :

(1.8) xt, n = f (θ ;σ t , n ) + log Z t2, n − E(log Z t2, n )

In the actual implementation the non- linear regression function is approximated by the

following parametric function:

(1.9)

J
n n2 D P
2π p 2π p
f (θ ; σ t , n) = ∑ σ tj[ µ 0 j + µ1 j + µ2 j + ∑ λij In = di + ∑ (γ pj cos n + δ pj sin n)]
j =0 N1 N 2 i=1 i =1 N N

where N1 =(N+1)/2 and N2 =(N+1)(N+2)/6 are normalized constants, N refers to the number

of return intervals per day (N=288), the tuning parameter P (P=8) determines the order of

the expansion14, the J flexible fourier forms are parameterized by quadratic components

14
Andersen and Bollerslev (1997a, 1998) and Cai, Cheung, Lee and Melvin (2001) find that P=6 fits the
dem-usd FXFX data over the sample 92-93 and the yen-usd FXFX data in 1998. Experimentation with

11
(terms with µ-coefficients), a number of sinusoids (the γ and δ coefficients) and time-

specific dummies (the λ coefficients). In practice estimation involved a two-step

procedure where $x t, n was replaced by the sample mean of the 5- minute returns and treated

as the dependent variable in the regression defined by equations (1.6) and (1.7). Defining

µf as the resulting estimate for the right hand side of equation (1.7), then the intraday
t, n

periodic seasonal component for interval n on day t is

$s t, n = T gexp( µf t, n /2)
(1.10)
∑ t=1 ∑ n=1 exp(µf t,n /2)
T /N N

An “event study” approach is used to examine the influence of central bank

intervention (and other macro announcements) on exchange rate volatility. 15 The general

regression specification is:

(1.11) Vt ,n = α 0 + ∑ k ∑ iα 1,ki Dtk, n+i + α 2 $s t, n + ε t ,n ,

where Dk denotes the (time-stamped to the nearest 5-minute) intervention and other

announcement dummy variables and $s t, n is the FFF volatility seasonal estimated over the

control sample days. 16 Using this general regression specification it is possible to test for

the impact and intra-day effects of intervention (and other macro news) by examining

whether the Dk s are statistically significant.

P=4,6 and 8 using both the control sample and Fed intervention day samples (over the years 1989-1995),
indicate that P=8 offers the best fit with these data.
15
See Dominguez (2003a) for a similar “event study” approach using returns rather than volatility.
16
Estimates of the intra -daily seasonal using the Fed intervention days produced very similar results.
Control sample days were used under the assumption that volatility on intervention days may differ from
non-intervention days (indeed figures 3 and 4 suggest that especially for the yen-usd market Fed
intervention days are more volatile than the control sample days), and while it is necessary to control for
intra-day cycles, it is also important not to inadvertently explain away what is unusual about intervention
days by only using intervention days to calculate the seasonal. I am grateful to Michael Melvin for
suggesting I use the control sample days for this purpose.

12
 Interventions in the event study by all three central banks take the value 1 if they

involve a purchase or sale of dollars and 0 otherwise. Interventions are included as (1,0)

dummy variables both because the dollar magnitudes are generally only available at a

daily (not intra-daily) frequency, and because there is some evidence that the sizes of

interventions depend on market reactions to initial trades suggesting that including

magnitudes might engender simultaneity bias.

Table 7 presents the results of the volatility event-study regression using the FXFX

five- minute dem-usd data. Significant one-hour leads were found for intervention

operations by all three central banks, suggesting that some traders know about these

operations well before the Reuters new release.17 Bundesbank interventions in the sample

had the largest (positive) influence on volatility by a factor of two relative either to Fed or

BOJ interventions. Fed interventions continued to influence dem-usd volatility for one hour

after the Reuters report, while Bundesbank interventions had effects for 25 minutes after

the report, and BOJ interventions had effects for just 10 minutes after the report. Only

announcements by the Fed significantly influenced volatility. Seven of the twelve U.S. and

German macro announcements are also found to be significant, with the significant lags

varying from impact to fifteen minutes after the Reuters time-stamp. The announcement

with the largest average influence on dem-usd variability is U.S. GNP. The intra-day

seasonal is highly significant in the regression. The interventions, announcements, macro

controls and seasonal together explain just under 25% of intra-day dem-usd volatility.

17
Various regression specifications were attempted, including imposing a polynomial distributed lag (pdl)
structure on the leads and lags of the intervention variables. Tests of the pdl restrictions suggested that the data
do not conform to this specification. Experimentation with various lead and lag combinations indicated that a
[-1hr,+2hr] window for the intervention variables and a [0,1hr] window for the macroeconomic
announcements was appropriate.

13
 The results of the yen-usd volatility event-study regression are presented in Table 8.

In contrast to the analogous dem-usd regression, Bundesbank interventions are not

statistically significant. There is evidence of one-hour Reuters announcement lags for both

the Fed and the BOJ. Fed interventions continue to have effects for an hour and a half after

the Reuters report, and BOJ interventions continue to have effects for one hour after the

report. Both Fed and BOJ central bank announcements also influence volality. Six of

twelve U.S. and Japanese macro announcements are significant. U.S. GNP again has the

largest effect.

Tables 7 and 8 include results from three alternative regression specifications. The

first of these alternative hypotheses asks whether the relationship between interventions

and volatility is related to the volume of trade. In particular, the regressions test whether

Fed interventions that occurred during the overlap in New York and European trading,

when volume is generally highest, had different effects than those that occurred during

other time periods. The results for both the dem-usd and yen-usd suggest that, regardless

of volume, Fed interventions have statistically significant effects over the 36 (5-min) leads

and lags. In both the high and low trade volume times, Fed interventions continued to have

one-hour lead effects and roughly one-hour lag effects in the dem-usd market, and one-

hour and twenty minute lag effects in the yen-usd market. Results for the remaining

variables in the regression were little changed by the inclusion of the interactive trade

volume dummy. The relative size of the coefficients on Fed intervention in low and high

volume trade volume suggests Fed interventions during high volume periods in the yen-usd

market had a slightly larger overall effect, and it is worth noting that 61% of Fed

interventions occurred during high trade volume times.

14
 The second alternative specification serves as a test of whether interventions that

are timed close to a (scheduled) macro announcement have different effects than those that

are not. The dummy variable distinguishing those interventions that occurred within a two-

hour window of a macro announcement are significant in both the dem-usd and yen-usd

volatility regressions. The relative size of the coefficients on the interactive dummy

suggests that these interventions have larger effects on volatility than interventions that are

not timed close to other announcements (although these continue to be significant in the

regressions). One possible explanation for this result is that traders are more sensitive to

news (including intervention news) at times when other major announcements are

released.18

The final set of alternative specifications examines the extent to which coordination

matters. Interventions are defined as being “coordinated” if at least one other of the G3

central banks intervened within a two hour window (and in the same direction). In the case

of the dem-usd market, 31% of all Fed interventions over this period were coordinated with

the Bundesbank. The results suggest that those interventions that were coordinated have a

slightly larger influence on volatility than unilateral interventions. Interestingly, the lag

effects for coordinated interventions last a good hour beyond those for unilateral

interventions. In the yen-usd market, 54% of all Fed interventions over this period were

coordinated with the BOJ. Results again suggest that both coordinated and unilateral Fed

interventions influence volatility, though the size of the effect is actually larger for

unilateral interventions. As in the dem-usd market, coordinated interventions had

18
Evans and Lyons (2002b) also find that currency trades have greater price impact if they are closely
timed with macro announcements.

15
significant lag effects on yen-usd volatility for an hour and a half after the Reuters report,

while the influence of unilateral interventions lasted for about forty- five minutes.

The event study results indicate that G3 interventions systematically influenced

intra-day exchange rate volatility over the sample period examined, August 1989 through

August 1995. The coefficient estimates indicate that Reuters reports generally lag

interventions by one hour, and intervention continues to influence volatility up to one and a

half hours after the Reuters report release, suggesting that intervention is neither common

knowledge, nor perfectly informative. It is also worth noting that all the coefficients on

intervention in tables 7 and 8 are positive, indicating that in the very short run interventions

are always associated with increases in volatility.

The next set of tests, reported in Tables 9, 10 and 11, examine the influence of G3

interventions on daily volatility19. Recall from the model in section II that trader

heterogeneity can make even informative intervention increase very short run exchange

rate volatility, though over time interventions should become common knowledge, and

traders should sort out any misinterpretation of intervention’s information content. Daily

realized dem-usd and yen-usd volatility is measured using the intra-daily returns data.

Following Anderson and Bollerslev (1998) we sum the squared 5-minute FXFX indicative

quote returns over each day (through GMT22) 20, such that:

264
(1.12) σ tFXFX = ∑R
n =1
2
t, n

19
A number of papers have examined the influence of intervention on daily exchange rate volatility. See,
for example, Bonser-Neal and Tanner (1996), Dominguez (1998), Chaboud and LeBaron (2001), Galati,
Melick and Micu (2002), Frenkel, Pierdzioch, and Stadtmann (2003) and Beine (2003).
20
The daily integrated volatilities were created by Steve Weinberg. The daily cutoff is GMT22 when
volatility is generally very low. Weekends are excluded and the volatilities are expressed as annualized
standard deviations.

16
to create a daily integrated volatility series. This measure better captures current volatility

relative to standard models such as GARCH because it is able to exploit

contemporaneous intraday information. As Anderson, Bollerslev, Diebold and Labys

(2003) explain,

“Suppose, for example, that the true volatility has been low for many days,
t=1,...T-1, so that both realized and GARCH volatilities are presently low as well.
Now suppose that the true volatility increases sharply on day T and that the effect
is highly persistent as is typical. Realized volatility for day, T, which makes
effective use of the day-T information, will increase sharply as well, as is
appropriate. GARCH or RiskMetrics volatility, in contrast, will not change at all
on day T, as they depend only on squared returns from days T-1, T-2,..., and they
will increase only gradually on subsequent days, as they approximate volatility
via a long and slowly decaying exponentially weighted moving average.”

In the context of measuring the impact of interventions on volatility – it is particularly

important that our measure of volatility reflect current (and not necessarily past) market

conditions. Figures 5 and 6 show the integrated volatility series as well as a MA(1)-

GARCH(1,1) volatility measure for the dem- usd and yen- usd, respectively.

Unsurprisingly the GARCH series are much smoother than the integrated volatility series,

though both sets of volatilities tend to move together.

Table 9 presents statistics on the distribution of the logarithm of realized volatility

for dem- usd and yen-usd, over the full sample period as well as on intervention days. For

the full sample (and non-intervention days sample) the measures of skewness and

kurtosis suggest the series are approximately Gaussian and the Ljung-Box statistics

indicate strong serial correlation. 21 Further, the estimates of the degree of fractional

21
Five observations in the dem-usd realized volatility series were significant outliers and are excluded
from the sample. Realized volatility on September 14, 1992 is two times higher than the second highest
day and realized volatilities on Septemb er 3-4, 1991and May 12-13, 1994 are three times smaller than the
next lowest day. We replaced these outliers with values from the next highest (or lowest) realized
volatilities in the sample distribution. Replacing these outliers does not significantly change the regression
results in Table 10, but does influence the skewness and kurtosis statistics for the full and non-intervention

17
integration, d, are significantly greater than zero and less than .5, which indicates

evidence of long-run dependence in the logarithmic volatilities. It is interesting to note

that on “all” interventions days, and particularly on coordinated intervention days, the

mean realized volatility is significantly larger than on non- intervention days. Realized

volatility on the day after an intervention remains higher for the dem- usd but not for yen-

usd, and volatility on the day before an intervention is no higher, on average, than on

other non- intervention days.

The statistics in table 9 suggest that the long-memory dynamics of the realized

logarithmic volatilities are best modeled using a fractional integration or ARFIMA

(autoregressive, fractionally integrated, moving average) model. Experimentation with

the two realized volatility series suggests that for the dem- usd an ARFIMA(1,d,0) best

describes the data, while for the yen-usd we use an ARFIMA(0,d,0) model. In order to

test whether daily interventions are correlated with realized volatility we include a

holiday dummy variable (H) to control for holiday and market closure effects as well as

daily contemporaneous and lagged G3 intervention indicators (the Dk s) as explanatory

variables:

(1 − L ) d (1 − φ L)( vt − µt ) = ε t ,
(1.13) µt = α 0 + α1k ' Dtk + α 2 Ht ,
ε t : N (0, ω )

where d is the fractional parameter, L denotes the lag operator, φ denotes the AR

parameter (set to 0 in the case of yen-usd), v t denotes the log of the integrated volatility

samples in Table 9. There were no apparent outliers in the yen-usd realized volatility series.

18
( σ FXFX ) at time t, α 0 is an intercept, α1k is the intervention parameter vector, and α 2 is the

holiday parameter. 22

The results in table 10 indicate that coordinated interventions continue to

systematically influence daily contemporaneous dem-usd and yen-usd volatility, though

there is no evidence that the influence of coordinated interventions extends beyond the day

of the operations. Similarly, contemporaneous unilateral Fed, Bundesbank and BOJ

operations all influence daily realized volatility in both markets. In the yen-usd market

unilateral Fed interventions are marginally significant after 3 days. The relative magnitudes

of the coefficients in the daily regressions suggest that coordinated interventions have a

substantially larger impact on daily volatility than do unilateral interventions. Coordinated

operations were also significant in the intra-day tests, though the relative size of the

coefficients on coordinated and unilateral operations were similar. This suggests that over

the course of the day the influence of coordinated operations (which may have a lower

probability of containing non- fundamental information) increases. Overall, the results in

tables 7-10 indicate that interventions by all three central banks positively influenced intra-

day and daily (contemporaneous) exchange rate volatility. There is little evidence that

interventions systematically influence volatility beyond the day of the operations. 23

The ARFIMA model specification in 1.13 implicitly assumes that shocks on

intervention days are as persistent as shocks on non-intervention days. A final set of

22
Estimation is based on the numerical quasi-maximum likelihood (QML) algorithm, where the likelihood
function is based on the Wold representation of the ARFIMA(1,d,0) processes with a Gaussian assumption.
The t-statistics reported in the tables are calculated using the corresponding heteroskedastic-consistent
standard errors based on finite difference approximation.
23
One day leads of coordinated and unilateral Fed interventions were also included in the regression
specification but lead coefficients were never statistically significant and are not reported in table 10.
These results are available from the author upon request.

19
regressions tests whether this assumption is valid by allowing the persistence of volatility

to differ across intervention and non-intervention days 24, specifically:

ν t = α0n + ν tn + ν tI + α 2 DtI + α 3 Ht + εt
ν tn = α1n (1 − DtI−1 )εt −1 + β nν tn−1
(1.14)
ν tI = DtI−1 (α 0I + α1I εt −1) + β Iν tI−1
ε t ~ N (0, ω )

where ν tn and ν tI are daily realized volatility on non- intervention and intervention days,

respectively. Note that if α 0I = 0 , α1n = α1I , and β n = β I 1.14 approximately reduces to

an ARMA(1,1) model. Regression estimates of this model are given in Table 11 and

indicate that the persistence of shocks on intervention and non- intervention days is not

statistically different. The likelihood ratio statistics suggest that we cannot reject the joint

hypothesis that α 0I = 0 , α1n = α1I and β n = β I . Further, the estimates of β I (the coefficient

on lagged volatility on intervention days) suggest that the effects of shocks to volatility

on intervention days fall by half in 1.5 days for dem-usd and 3 days for yen-usd.

V. Conclusions

This paper identifies circumstances in which central bank interventions influence

exchange rates. Microstructure theory suggests that trader heterogeneity can cause

exchange rates to move away from fundamentals in the short-run. Introducing intervention

operations into an already confused market, in turn, can increase the influence of non-

fundamentals trading on exchange rate movements. Over time, however, interventions are

common knowledge and traders should be able to distinguish informative from non-

24
Jones, Lamont and Lumsdaine (1998) estimate a similar model to measure the influence of
macroeconomic information on the persistence of bond market volatility.

20
informative interventions. This suggests that the influence of interventions on exchange

rates may well differ over the very short and longer runs.

The empirical tests in this paper examine the influence of G3 interventions on dem-

usd and yen-usd intra-daily (5- minute) indicative quote volatility as well as a measure of

realized daily volatility. Results suggest that intervention operations, especially those that

were coordinated, were consistently associated with increases in intra-day and daily

volatility, while there is little evidence that interventions influenced longer-term volatility.

The fact that interventions did not lead to longer-term increases in volatility may help

explain why governments, who presumably prefer not to increase market volatility,

continue to rely on interventions to influence currency values.

21
 References

Almeida, A., Goodhart, C., Payne, R., 1998, The Effects of Macroeconomic News on High
Frequency Exchange Rate Behavior, Journal of Financial and Quantitative Analysis, 33 (3), 383-
408.

Andersen, T. and Bollerslev, T., 1997a. Intraday Periodicity and Volatility Persistence in
Financial Markets, Journal of Empirical Finance, 4, 115-158.

Andersen, T. and Bollerslev, T. 1997b Heterogeneous Information Arrivals and Return Volatility
Dynamics: Uncovering the Long-Run in High Frequency Returns, Journal of Finance, 975-1005.

Andersen, T. and Bollerslev, T., 1998. Deutsche Mark-Dollar Volatility: Intraday Activity
Patterns, Macroeconomic Announcements, and Longer Run Dependencies, Journal of Finance,
53(1) 219-65.

Andersen, T., Bollersleb, T., Diebold, F. and Labys, P., 2003, Modeling and Forecasting Realized
Volatility, Econometrica, 71, 529-626.

Andersen, T., Bollerslev, T. Diebold, F. and Vega, C., 2003 Micro Effects of Macro
Announcements: Real Time Price Discovery in Foreign Exchange, American Economic Review,
93(1), 38-62.

Bacchetta, P. and van Wincoop, E., 2003, Can Information Heterogeneity Explain the Exchange
Rate Determination Puzzle?, NBER Working Paper #9498, February.

Beine, M., 2003, “Volatility Expectations and the Asymmetric Effects of Direct Interventions in
the FX Market, Journal of the Japanese and International Economies, 17, 55-80.

Beattie, N., Fillion, J., 1999. An Intraday Analysis of the Effectiveness of Foreign
Exchange Intervention. Bank of Canada Working Paper 99-4.

Bollerslev, T. and Domowitz, I., 1993 Trading Patterns and Prices in the Interbank Foreign
Exchange Market. Journal of Finance, 48, 1421-1443.

Bonser-Neal, C. and Tanner, G., 1996 Central Bank Intervention and the Volatility of Foreign
Exchange Rates: Evidence from the Options Market. Journal of International Money and Finance
15/6, 853-878.

Cai, J., Cheung, Y., Lee, R., Melvin, M. 2001 ‘Once-in-a-Generation’ Yen Volatility in
1998: Fundamentals, Intervention and Order Flow. Journal of International Money and
Finance, 20, 327-347.

Chaboud, A., LeBaron, B., 2001. Foreign Exchange Market Trading Volume and Federal
Reserve Intervention. Journal of Futures Markets, 21, 851-860.

Chang, Y., Taylor, S., 1998. Intraday Effects of Foreign Exchange Intervention by the
Bank of Japan. Journal of International Money and Finance 18, 191-210.

22
Dacorogna, M.M., U.A. Muller, R.J. Nagler, R.B. Olsen, and O.V. Pictet, 1993. A
Geographical Model for the Daily and Weekly Seasonal Volatility in the FX Market.
Journal of International Money and Finance 12, 413-438.

Danielsson, J. and Payne, R., 2002 Real Trading Patterns and Prices in Spot Foreign
Exchange Markets: An Empirical Investigation, Journal of International Money and
Finance, 21, 203-222.

De Grauwe, P. and Grimaldi, M., 2003, Intervention in the Foreign Exchange Market in a
Model with Noise Traders, mimeo, May.

Dominguez, K., 1998. Central Bank Intervention and Exchange Rate Volatility, Journal of
International Money and Finance 18, 161-190.

Dominguez, K., 2003a. The Market Microstructure of Central Bank Intervention. Journal
of International Economics 59, 25-45.

Dominguez, K., 2003b Foreign Exchange Intervention: Did it Work in the 1990s? in Dollar
Overvaluation and the World Economy, edited by C.F. Bergsten and J. Williamson,
Institute for International Economics, Special Report 16, 217-245.

Dominguez, K., Frankel, F., 1993a. Does Foreign Exchange Intervention Matter ? The
Portfolio Effect. American Economic Review 83, 1356-69.

Dominguez, K., Frankel, F., 1993b. Does Foreign Exchange Intervention Work? Institute
for International Economics, Washington, D.C.

Edison, H., 1993 The Effectiveness of Central Bank Intervention: A Survey of Post-1982
Literature. Essays in International Finance, Princeton University Press, Princeton.

Evans, M,, Lyons, R., 2001. Portfolio Balance, Price Impact and Secret Intervention.
National Bureau of Economic Research Working Paper 8356.

Evans, M., Lyons, R., 2002 Time-Varying Liquidity in Foreign Exchange, Journal of
Monetary Economics, 49(5) 1025-1051.

Evans, M., Lyons, R., 2002 Order Flow and Exchange Rate Dynamics, Journal of Political
Economy, 170-180.

Fischer, A.M., Zurlinden, M., 1999. Exchange Rate Effects of Central Bank Interventions:
An Analysis of Transaction Prices. Economic Journal 109, 662-676.

Frenkel, M. Pierdzioch, C. and Stadtmann, G. 2003 The Effects of Japanese Foreign

Exchange Market Interventions on the Yen/US Dollar Exchange Rate Volatility, Kie l
Working Paper No. 1165.

23
Galati, G., Melick, W. Micu, M., 2002, Foreign Exchange Market Intervention and
Expectations: An Empirical Study of the Dollar/Yen Exchange Rate, August mimeo.

Gallant, R. 1981 On the Bias in Flexible Functional Forms and an Essentially Unbiased
Form: The Fourier Flexible Form, Journal of Econometrics, 15(2), 211-45.

Goodhart, C., Hesse, T., 1993. Central Bank Forex Intervention Assessed in Continuous
Time. Journal of International Money and Finance 12, 368-389.

Goodhart, C., Ito, T., Payne, R., 1996. One Day in June 1993: A Study of the Working of
the Reuters 2000-2 Electronic Foreign Exchange Trading System in: Frankel, J., Galli, G.,
Giovannini, A., (Eds.), The Microstructure of Foreign Exchange Markets, The University
of Chicago Press for NBER, pp.107-179.

Guillaume, D., Dacorogna, M., Dave, R., Muller, U., Olsen, R., Pictet, O., 1997. From the
Bird's Eye to the Microscope: A Survey of New Stylized Facts on the Intra-Daily Foreign
Exchange Markets. Finance and Stochastics 1(2), 95-129.

Humpage, O., 1999, U.S. Intervention: Assessing the Probability of Success. Journal of
Money, Credit and Banking, vol31, No. 4, November, 731-747.

Ito, T., forthcoming, "Is Foreign Exchange Intervention Effective?: The Japanese
experiences in the 1990s" in Paul Mizen (ed.), Monetary History, Exchange Rates and
Financial Markets, Essays in Honour of Charles Goodhart, Volume 2, Cheltenham U.K.;
Edward Elgar Publishers and NBER working paper no. 8914 (April 2002).

Jones, C., Lamont, O., and Lumsdaine, R., 1998, Macroeconomic News and Bond Market
Volatility, Journal of Financial Economics, 47, 3, 315-337.

LeBaron, B., 1999. Technical Trading Rule Profitability and Foreign Exchange
Intervention. Journal of International Economics 49, 1, 125-143.

Lyons, R., 2001. The Microstructure Approach to Exchange Rates, MIT Press.

Neely, C., 2000. The Practice of Central Bank Intervention: Looking Under the Hood.
Central Banking Vol. XI, No.2, 24-37.

Neely, C., 2002, The Temporal Pattern of Trading Rule Returns and Central Bank
Intervention: Intervention Does Not Generate Technical Trading Rule Profits. Journal of
International Economics, 58, 211-232.

O'Hara, M., 1995. Market Microstructure Theory, Blackwell Business.

Pasquariello, P., 2001 Central Bank Intervention and the Intraday Process of Price
Formation in the Currency Markets, Stern School of Business, NYU.

24
Pasquariello, P., 2002 Informative Trading or Just Noise? An Analysis of Currency
Returns, Market Liquidity, and Transaction Costs in Proximity of Central Bank
Interventions, Stern School of Business, NYU, November.

Payne, R., Vitale, P., forthcoming, A Transaction Level Study of the Effects of Central
Bank Intervention on Exchange Rates. Journal of International Economics.

Peiers, B., 1997. Informed Traders, Intervention and Price Leadership: A Deeper View of
the Microstructure of the Foreign Exchange Market. Journal of Finance 52(4), 1589-1614.

Sarno, L., Taylor, M., 2001 Official Intervention in the Foreign Exchange Market: Is it
Effective and, If So, How Does it Work? Journal of Economic Literature, vol. XXXIX,
September, 839-868.

Taylor, M. 2003 Is Official Exchange Rate Intervention Effective? CEPR Discussion

Paper No. 3758.

25
 Table 1 Timing of Selected Reuters Announcements (1989-1995)

News Event Day-of-week Average time

(GMT)

Fed Intervention various 14:57:10

US Employment Friday 12:30(DST);13:30
US CPI/PPI Friday 12:30(DST);13:30
US M1 Thursday 20:30(DST);21:30
US Trade various 12:30(DST);13:30
US Consumer Credit various 19:30(DST)20:30
US Retail Sales various 12:30(DST);13:30
US Leading Indicators Wed or Fri 12:30(DST);13:30
US GNP various 12:30(DST);13:30
FOMC news Friday 20:30(DST);21:30
comments by US officials various 14:48:12

Bundesbank Intervention various 11.31:16

Bundesbank Meeting Thursday 11:30(DST):12:30
German M3 various 6:30(DST);7:30
comments by German officials various 10:27:13

BOJ Intervention various 3:56:36

Japanese Current Account various 6:30(DST);7:30
comments by Japanese officials various 6:51:07
DST denotes daylight savings time; otherwise, times are GMT.

Table 2 Typical timing of G3 interventions during the 24 hour GMT clock

GMT24(t-1) GMT3 GMT8 GMT11 GMT14 GMT17 GMT22

Tokyo 9am 12pm 5pm
BOJ interventions
Frankfurt 9am 12pm 3pm 6pm
Bundesbank Interventions
New York 9am 12pm 5pm
Fed Interventions
GMT is Greenwich Mean Time.

26
Table 3 G3 Central Bank Intervention Descriptive Statistics 1989-1995

Fed Buba BOJ Any

CB
Number of intervention days 104 39 63 206
Total number of intra-day interventions 268 83 145 496
nd
Probability of a 2 intervention within 1 hour of 0.56 0.29 0.26 0.49
prior one
Number of interventions followed by another one 156 24 37 244
within 1 hour
Total number of intervention clusters 44 25 16 83
Average duration of intervention clusters, 0:44:49 0:45:30 0:35:08 0:42:48
each 1 hour apart
Average number of interventions in a cluster 5.023 2.52 2.5 4.048
Note: An intervention cluster is defined as at least two interventions that occur within the
same hour.

Table 4 Timing of G3 Central Bank Interventions in the dem- usd and yen- usd markets

Dem-usd yen-usd
Percent of Interventions closely timed with macro announcements
1. all interventions (any bank) 7% 10%
2. Fed interventions 7% 13%
3. Buba interventions 6% 15%
4. BOJ interventions 9% 4.5%
Percent of Fed Interventions
1. occurring before 12:00 est 42.6% 57%
2. occurring after 12:00 est 13.4% 16%
3. unknown 0.004% 3%
4. occurring in both the am and pm 44% 24%
Percent of Fed Interventions occurring during European Trading 61% 61%
Percent of Fed Interventions Coordinated
1. with Buba 31%
2. with BOJ 54%
Note: “closely timed” is defined as occurring within 2 hours. Likewise, “coordinated”
Fed interventions are defined as interventions that occur within 2 hours of a BOJ or
Bundesbank intervention.

27
 Table 5 The 25 Largest 5- min dem-usd Returns (1989-95)

DATE TIME RETURN EVENT* REUTERS TIME-STAMP

9/15/89 12:35:00 0.009458 CPI;Trade 12:36;12:36

9/15/89 18:20:00 0.005188 Fed 13:56
9/15/89 18:25:00 0.011564 BB 14:28
10/5/89 13:05:00 0.005004 BB 12:09 and 13:11
10/6/89 12:35:00 0.005968 Employment 12:34
1/10/90 15:15:00 0.006123 Fed 19:05
3/19/91 13:10:00 0.00525 Fed;BB 16:00-16:13
3/19/91 13:40:00 0.00514 CPI;Housing Starts 13:33
3/19/91 14:35:00 0.005161 Fed;BB 16:00-16:13
3/19/91 14:40:00 0.005647 Fed;BB 16:00-16:13
3/19/91 14:45:00 0.008245 Fed;BB 16:00-16:13
5/17/91 14:30:00 0.007094 Fed;BB 19:15-20:53
5/17/91 14:35:00 0.005674 Fed;BB 19:15-20:53
5/17/91 15:05:00 0.005492 Fed;BB 19:15-20:53
7/12/91 13:25:00 0.005715 Fed;BB 13:27
7/20/92 14:20:00 0.005532 Fed;BB 14:30
7/20/92 14:30:00 0.006298 Fed;BB 14:30
8/11/92 12:20:00 0.005553 Fed;BB 12:30
8/21/92 13:25:00 0.006027 Fed;BB 13:35
4/29/94 14:30:00 0.006037 Fed 14:40
11/2/94 16:05:00 0.007848 Fed 16:13-18:53
5/31/95 12:40:00 0.010738 Fed 12:49
5/31/95 12:45:00 0.005854 BB 12:45
5/31/95 12:55:00 0.006077 Fed 12:51-12:54
8/15/95 12:20:00 0.008128 BB;Fed;BOJ 12:49-12:58

NOTE: At least one Fed interventions occurred on each of the days in the sample.

* Events are defined as any macro announcement or central bank intervention that
occurred within a [-1hr,5hr] window of the large return.

28
 Table 6 The 25 Largest 5- min yen-usd Returns (1989-95)

DATE TIME RETURN EVENT* REUTERS TIME-STAMP

9/15/89 12:35:00 0.005344 CPI;Trade;Fed;BB 12:36;12:36;13:56;14:28

9/15/89 18:20:00 0.007393 Fed;BB 13:56;14:28
9/15/89 18:25:00 0.005687 Fed;BB 13:56;14:28
9/15/89 18:35:00 0.006791 Fed;BB 13:56;14:28
10/6/89 12:35:00 0.005976 Employment;Fed 12:34;13:42
1/17/92 18:35:00 0.005997 Fed 18:38
1/17/92 19:10:00 0.005314 BOJ 19:14
1/17/92 19:15:00 0.004613 BOJ 19:14
4/27/93 12:55:00 0.004626 Fed 13:05-14:15
8/19/93 12:35:00 0.005207 Trade 12:30
8/19/93 14:30:00 0.005635 Fed 14:39
8/19/93 15:05:00 0.005217 Fed 15:07
4/29/94 13:10:00 0.005025 Fed 14:38-16:08
5/4/94 12:30:00 0.005113 Fed;BB 12:33;12:34
6/24/94 13:15:00 0.005839 Fed;BB;BOJ 13:45-16:59
6/24/94 13:35:00 0.005378 Fed;BB;BOJ 13:45-16:59
11/2/94 16:05:00 0.007266 Fed 16:13-18:58
5/31/95 12:40:00 0.008712 Fed 12:49
5/31/95 12:45:00 0.008348 BB 12:45
5/31/95 12:55:00 0.007176 Fed 12:51-12:54
5/31/95 14:30:00 0.004466 Fed 13:47
8/2/95 13:10:00 0.005012 Fed;BOJ 13:13;13:30
8/15/95 12:20:00 0.005948 BB;Fed;BOJ 12:49;12:58
8/15/95 23:25:00 0.004811 Fed;BB;US Retail Sales 14:28;14:24;19:06
8/15/95 24:00:00 0.004664 Fed;BB;US Retail Sales 14:28;14:24;19:06

NOTE: At least one Fed intervention occurred on each of the days in the sample.

* Events are defined as any macro announcement or central bank intervention that
occurred within a [-1hr,5hr] window of the large return.

29
 Table 7 Influence of Interventions on Intra-Day dem- usd Volatility
Vt ,n = α 0 + ∑ k ∑ iα 1,ki Dtk, n+i + α 2 $s t, n + ε t ,n
where V is 5- min dem-usd volatility (measured as the absolute value of the 5- min
returns); the Dk s include intervention, official central bank announcements and macro
announcements; i=-1 to +2hrs for the G-3 intervention variables and official
announcements and i=0 to +1hr for the macro announcements; t,n is the sequence of the
regular-spaced (every 5 minutes) intra-daily data for all the days on which the Fed
interve ned against the mark from 1989 to 1995 (69 days and a total of 151 reports of Fed
operations). The reported coefficients are multiplied by 100.

Do All G-3 Does Trade Does Does

Interventions Volume Proximity to Coordination
Matter? Matter?1 Macro News Matter?3
Matter?2
Independent Variable Coeff sum4 Coeff sum Coeff sum Coeff sum
Fed Intervention 0.121**
Buba Intervention 0.213** 0.212** 0.214** Na
BOJ Intervention 0.103** 0.103** 0.104** Na
Official Announcements Yes Yes Yes Yes
Macro Controls Yes Yes Yes Yes
Fed Interventions
During High Trade 0.122**
Volume
Fed Interventions
During Low Trade 0.121**
Volume
Fed Interventions Close 0.138**
to Macro News
Isolated 0.116**
Fed Interventions
Coordinated 0.139**
G-3 Interventions
Unilateral 0.122*
Fed Interventions
FFF Seasonal 0.169** 0.169** 0.169** 0.173**
R2 0.237 0.237 0.237 0.236
D.W. 1.99 1.99 1.99 1.99

Number of observations= 19,833. 1 High trade volume is defined as the overlap in US

and European trading hours. 2 Interventions that occur within 2 hours of a macro news
announcement are defined as “close”. 3 Coordinated interventions are defined as Fed
interventions that occur on the same day as at least one other of the G-3 central banks.
4
The coefficient is the sum of the 36 lead and lag coefficients on each of the intervention
variables. **, * denote statistical significance (of the 36 leads and lags) at the 1%, and
5% levels, respectively, using robust standard errors.

30
 Table 8 Influence of Interventions on Intra-Day yen-usd Volatility
Vt ,n = α 0 + ∑ k ∑ iα 1,ki Dtk, n+i + α 2 $s t, n + ε t ,n
where V is 5- min yen-usd volatility; Dk s include intervention, official central bank
announcements and macro announcements; i=-1 to +2hrs for the G-3 intervention
variables and official announcements and i=0 to +1hr for the macro announcements; t,n is
the sequence of the regular-spaced (every 5 minutes) intra-daily data for all the days on
which the Fed intervened against the yen from 1989 to 1995 (66 days and a total of 192
reports of Fed operations). The reported coefficients are multiplied by 100.

Do All G-3 Does Trade Does Does

Interventions Volume Proximity to Coordination
Matter? Matter?1 Macro News Matter?3
Matter?2
Independent Variable Sum coeff4 Sum coeff Sum coeff Sum coeff
Fed Intervention 0.175**
Buba Intervention 0.063 0.058 0.066 na
BOJ Intervention 0.121** 0.121** 0.123** na
Official Announcements Yes Yes Yes yes
Macro Controls Yes Yes Yes yes
Fed Interventions
During High Trade 0.183**
Volume
Fed Interventions During 0.164**
Low Trade Volume
Fed Interventions Close 0.192**
to Macro News
Isolated 0.148**
Fed Interventions
Coordinated 0.149**
G-3 Interventions
Unilateral 0.271*
Fed Interventions
FFF Seasonal 0.168** 0.166** 0.168** 0.164**
R2 0.188 0.188 0.188 0.189
D.W. 2.00 2.00 2.00 2.00

Number of observations= 18,969. 1 High trade volume is defined as the overlap in US and
European trading hours. 2 Interventions that occur within 2 hours of a macro news
announcement are defined as “close”. 3 Coordinated interventions are defined as Fed
interventions that occur on the same day as at least one other of the G-3 central banks.
4
The coefficient is the sum of the 36 lead and lag coefficients on each of the intervention
variables. **, and * denote statistical significance (of the 36 leads and lags) at the 1% and
5% levels, respectively, using robust standard errors.

31
 Table 9 Summary Statistics for Daily dem- usd and yen-usd Log Realized Volatility
on Intervention and Non-Intervention Days (1989-1995)

I. Full Sample a
Mean Std- max min Skew- Kurt- Q(20) d AR Obs
dev ness osis
Dem-usd -2.224 0.425 -0.967 -5.776 -2.565 20.336 1811.0 0.273 0.203 1564
Yen-usd -2.250 0.396 -0.703 -5.133 -1.204 9.137 3318.8 0.424 0.046 1564

II. Non-Intervention Daysa

Dem-usd -2.234 0.425 -0.967 -5.776 -2.727 21.254 1644.2 0.267 0.204 1477
Yen-usd -2.288 0.392 -1.075 -5.133 -1.475 10.624 2378.9 0.413 0.059 1313

III. All Intervention Daysb

Dem-usd -2.067 0.391 -1.067 -2.750 0.494 -0.540 123.2 0.395 na 87
Yen-usd -2.050 0.352 -0.703 -2.945 0.259 0.334 454.8 0.400 na 251

IV. Coordinated Intervention Days c

Dem-usd -1.935 0.386 -1.169 -2.750 0.025 -0.614 26.6 0.280 na 33
Yen-usd -2.052 0.423 -1.127 -2.945 0.115 -0.666 72.9 0.477 na 59

V. Day Before (All) Intervention Days d

Dem-usd -2.100 0.365 -1.261 -2.703 0.470 -0.726 17.1 na na 44
Yen-usd -2.204 0.346 -1.075 -2.891 0.819 1.193 58.7 0.331 na 84

VI. Day After (All) Intervention Daysd

Dem-usd -2.176 0.337 -1.284 -3.056 -0.014 0.617 29.2 na na 44
Yen-usd -2.271 0.441 -1.075 -4.920 -2.085 15.220 35.7 0.323 na 84

Notes: The sample covers the period August 15, 1989 through August 15, 1995. The daily
realized volatilities are constructed from sums of 5-minute squared returns, and are expressed as
annualized standard deviations. The statistics refer to the distribution of logarithmic realized
standard deviations. The column labeled Q(20) contains Ljung-Box test statistics for up to the
twentieth order serial correlation. The column labeled “d” gives the regression estimate of the
fractional integration parameter, d, from an ARFIMA(1,d,0) model. The column labeled “AR” is
the regression estimate of the autoregressive parameter from the ARFIMA(1,d,0) model.
a
Five observations in the dem-usd realized volatility series were significant outliers and are
excluded from the full and non-intervention samples. Realized volatility on September 14, 1992
is two times higher than the second highest day and realized volatilities on September 3-4,
1991and May 12-13, 1994 are three times smaller than the next lowest day. We replaced these
outliers with values from the next highest (or lowest) realized volatilities in the sample
distribution. There were no apparent outliers in the yen-usd realized volatility series.
b
“All” intervention days include days of unilateral and coordinated Fed, Bundesbank and BOJ
intervention operations.
c
Coordinated intervention days are defined as days when the Fed intervened with either the
Bundesbank or the BOJ (or both).
d
The number of days before and after an intervention excludes intervention days that follow or
precede other intervention days.

32
Table 10 Influence of Interventions on Realized Daily dem- usd and yen-usd Volatility

(1 − L ) d (1 − φ L)( vt − µt ) = ε t ,
µt = α 0 + α1k ' Dtk + α 2 Ht
ε t : N (0, ω )
where vt is the log of the sum of intra-day squared 5- minute returns (through 22GMT)
excluding weekends, Ht is a dummy variable indicating the day after a holiday or market
closure and the Dks are dummy variables that denote daily unilateral and coordinated
intervention operations.

Independent Variables DEM-USD Yen-USD

ARFIMA parameters Coeff t-stat Coeff t-stat
α0 -2.257** -37.452 -2.374** -18.122
D 0.270** 5.951 0.455** 7.019
φ 0.223* 2.325
ω 0.119** 7.052 0.083** 8.470
Holiday dummy 0.083 1.507 0.006 0.134
Fed Unilateral Intervention 0.148* 2.245 0.354 1.379
Lag t-1 -0.030 -0.543 0.216 1.601
Lag t-2 -0.025 -0.448 0.139 0.978
Lag t-3 -0.033 -0.575 0.197† 1.811
Lag t-4 0.083 1.442 0.223 1.250
Lag t-5 0.100 1.522 -0.034 -0.329
Fed Coordinated Intervention 0.296** 4.487 0.317** 5.951
Lag t-1 -0.030 -0.543 0.042 0.902
Lag t-2 -0.025 -0.448 0.004 0.098
Lag t-3 -0.033 -0.575 0.045 0.882
Lag t-4 0.083 1.442 0.046 0.967
Lag t-5 0.100 1.522 -0.033 -0.687
Buba/ BOJ Unilateral Intervention 0.152† 1.885 0.083** 8.470
Log Likelihood -552.749 -276.532
No of daily observations 1564 1564

Note: An ARFIMA(1,d,0) was used to model dem- usd integrated volatility and an
ARFIMA(0,d,0) was used to model yen-usd integrated volatility. Results for the
variables of interest (the Dk s) were robust to alternative ARFIMA specifications. **, *
and † denote statistical significance at the 1%, 5% and 10% levels, respectively, using robust
standard errors.

33
Table 11 Influence of Shocks on Realized Volatility on Non-Intervention
and Intervention Days
ν t = α0n + ν tn + ν tI + α 2 DtI + α 3 Ht + εt
ν tn = α1n (1 − DtI−1 )εt −1 + β nν tn−1
ν tI = DtI−1 (α 0I + α1I εt −1) + β Iν tI−1
ε t ~ N (0, ω )

where ν tn and ν tI are the log of the sum of intra-day squared 5- minute returns (through
22GMT) excluding weekends on non- intervention and intervention days, respectively. Ht
is a dummy variable indicating the day after a holiday or market closure and the DI is a
dummy variable that denotes G3 intervention operations.

DEM-USD Yen-USD
Parameter Coeff t-stat Coeff t-stat
α 0n -2.240 -89.848** -2.311 -60.965**
α 0I 0.012 0.239 0.030 1.615
α1n 0.513 4.244** 0.451 5.346**
α1I 0.647 2.364* 0.479 6.706**
α2 0.203 3.857** 0.204 7.536**
α3 0.081 1.258 0.005 0.131
βn 0.633 4.236** 0.849 11.306**
β I
0.736 10.775** 0.838 5.703**
ω 0.123 7.418** 0.085 9.599**
Log Likelihood -576.48 -293.319
Likelihood Ratio Statistic 2.371 4.304

Note: ** and * denote statistical significance at the 1% and 5% levels, respectively, using robust
standard errors. The likelihood ratio statistics suggest that we cannot reject the hypothesis that
α 0I = 0 , α1n = α1I and β n = β I , or in other words, that the persistence of shocks on
intervention and non- intervention days is not statistically different. The β I estimates
suggest that the effects of shocks on intervention days fall by half in 1.5 days for dem-usd
and 3 days for yen- usd.

34
 Figure 1
Dem-USD Exchange Rate and Central Bank Intervention, August 1989-1995

2.25 1500

1000

Intervention ($ Million)
500
dem-usd

dem-usd
1.75 Buba Intervention
Fed Intervention

-500

1.25 -1000
Aug-89 Aug-90 Aug-91 Aug-92 Aug-93 Aug-94

Figure 2
Yen-USD Exchange Rate and Central Bank Intervent ion, August 1989-1995

180 9000

7000

160

5000

140 3000
Intervention ($ Million)

1000
yen-usd

yen-usd
120 BOJ Intervention

-1000 Fed Intervention

100 -3000

-5000

80

-7000

60 -9000
Aug-89 Aug-90 Aug-91 Aug-92 Aug-93 Aug-94

35
 Figure 3
Dem-USD Intra-day Volatility

1.60E-01

1.40E-01

1.20E-01
Average Volatility

1.00E-01
fed sample volatility
8.00E-02 control sample volatility
seasonal volatility
6.00E-02

4.00E-02

2.00E-02

0.00E+00
1 13 25 37 49 61 73 85 97 109 121 133 145 157 169 181 193 205 217 229 241 253 265 277
5 Min time

Figure 4
Yen-USD Intra-Day Volatility

1.20E-01

1.00E-01

8.00E-02
Average Volatility

fed sample volatility

6.00E-02 control sample volatility
seasonal volatility

4.00E-02

2.00E-02

0.00E+00
1 13 25 37 49 61 73 85 97 109 121 133 145 157 169 181 193 205 217 229 241 253 265 277
5 min Time

36
 Figure 5
DEM-USD Daily Realized Volatility and MA(1)-GARCH(1,1) Volatility
2.5

1.5

Realized Volatility
MA(1)-GARCH(1,1)

0.5

0
Jan-89 Nov-89 Sep-90 Jul-91 May-92 Mar-93 Jan-94 Nov-94 Sep-95

Figure 6
YEN-USD Daily Realized Volatility and MA(1)-GARCH(1,1) Volatility

2.5

Realized Volatility
1.5
MA(1)-GARCH(1,1)

0.5

0
Jan-89 Nov-89 Sep-90 Jul-91 May-92 Mar-93 Jan-94 Nov-94 Sep-95

37