This article was downloaded by: [Linnaeus University]On: 09 October 2014, At: 20:33Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

International Economic JournalPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/riej20

The Effects of US MacroeconomicSurprises on the Intraday Movements ofForeign Exchange Rates: Cases of USD-EUR and USD-JPY Exchange RatesYoung Wook Han aa Department of Economics , Economic Research Institute, HallymUniversity , Chuncheon, Republic of KoreaPublished online: 21 Sep 2010.

To cite this article: Young Wook Han (2010) The Effects of US Macroeconomic Surprises on theIntraday Movements of Foreign Exchange Rates: Cases of USD-EUR and USD-JPY Exchange Rates,International Economic Journal, 24:3, 375-396, DOI: 10.1080/10168737.2010.504777

To link to this article: http://dx.doi.org/10.1080/10168737.2010.504777

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the“Content”) contained in the publications on our platform. However, Taylor & Francis,our agents, and our licensors make no representations or warranties whatsoever as tothe accuracy, completeness, or suitability for any purpose of the Content. Any opinionsand views expressed in this publication are the opinions and views of the authors,and are not the views of or endorsed by Taylor & Francis. The accuracy of the Contentshould not be relied upon and should be independently verified with primary sourcesof information. Taylor and Francis shall not be liable for any losses, actions, claims,proceedings, demands, costs, expenses, damages, and other liabilities whatsoever orhowsoever caused arising directly or indirectly in connection with, in relation to or arisingout of the use of the Content.

This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

International Economic JournalVol. 24, No. 3, 375–396, September 2010

The Effects of US MacroeconomicSurprises on the Intraday Movementsof Foreign Exchange Rates: Cases ofUSD-EUR and USD-JPY Exchange

Rates

YOUNG WOOK HAN

Department of Economics, Economic Research Institute, Hallym University, Chuncheon,Republic of Korea

(Received 13 July 2009; accepted 10 December 2009)

ABSTRACT This paper characterizes the intraday dynamics of the high frequency US Dollar(USD)–Euro (EUR) and US Dollar (USD)–Japanese Yen (JPY) foreign exchange rates thathave been subject to macroeconomic fundamentals. Even though the FIGARCH model witha normality assumption is found to be a good starting point, it appears to be inappropriateto represent the underlying movements of the high frequency returns due to the occurrencesof jumps. Hence, this paper relies on the FIGARCH model with the mixture distributionthat allows for the time-varying jumps that are determined by the US macroeconomic sur-prises. This paper generally finds that the US macroeconomic surprises are closely relatedto the intraday movements in the volatility process of the high frequency returns processthrough the jumps. In particular, the US macroeconomic surprises appear to affect the move-ments in the volatility process of the foreign exchange rates asymmetrically depending on thesigns of the surprises and spuriously increasing the long memory persistence in the volatilityprocess due to the jumps.

KEY WORDS: Intraday foreign exchange rates, time-varying jumps, FIGARCH, long memoryproperty, US macroeconomic surprises, mixture distributionJEL CLASSIFICATIONS: C22, F31, G15

Correspondence Address: Young Wook Han, Department of Economics, Hallym University,Hallymdaehakgil 39, Chuncheon, 200-702, Republic of Korea. Email: ywhan@hallym.ac.kr

1016-8737 Print/1743-517X Online/10/030375–22 © 2010 Korea International Economic AssociationDOI: 10.1080/10168737.2010.504777

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

376 Y.W. Han

1. Introduction

The link between news about economic fundamentals and the variation of assetprices is an important issue in financial economics. It has been one of the leastwell understood issues. Some influential empirical studies have gone so far tosuggest that for some assets, prices – like foreign exchange rates and fundamentals– are largely disconnected. See Frankel and Rose (1995) for a good survey ofthe empirical exchange rate literature. Only modest progress has been made inexplaining exchange rate movement with macroeconomic fundamentals in thelong run (Mark, 1995). In particular, econometric attempts to explain short runmovements in exchange rates have had limited success so far.

As stressed in the literature on market microstructure (see O’Hara, 1995), theinformation including both public and private components is a major determinantof foreign exchange rate movements. With respect to the public information, themost common type is the scheduled announcements of macroeconomic funda-mentals. This kind of public information is officially forbidden to be announceduntil the scheduled release time so that there is no prior leakage of the information.Thus, the new public information of macroeconomic fundamentals reaches allparticipants in foreign exchange markets at the same time and then the exchangerates will adjust to incorporate the information. Regarding private information,there are two types of private information. One is the payoff-related private infor-mation that can be accessed to yet unreleased public information by governments(Lyons, 2001) and the other is the unrelated payoff information that a dealer hasregarding interim states of the market by order flow, i.e. excess buyer-initiated orseller-initiated trading, which reflects a market’s information process mechanism.The second type of the private information seems to be most probable type inforeign exchange markets. However, as presented by Lyons (2001), Love and Payne(2008) and Evans and Lyons (2008), much of the private information is in factclosely related to the public information about macroeconomic fundamentals andit can be changed by public information.

Thus, public information about macroeconomic fundamentals can cause largemovements or jumps in foreign exchange rates given the specifications of the meanprocess and the volatility process, as pointed by Andersen et al. (2003). In lightof the evidence on the existence of jumps in the process of foreign exchangerates, this paper models the jumps and looks for an economic explanation fortheir existence. Since the jumps appear to be quite important in understandingthe dynamics of foreign exchange rates, modeling the jumps directly might giveinteresting additional insight into the mechanism of exchange rate dynamics.

While many studies have reasonably well explained the link between macroeco-nomic fundamentals and exchange rate movements in the long run, the empiricalattempts to explain a short run linkage between exchange rates and macroeco-nomic fundamentals have had very limited success due to econometric problemslike small sample bias and some factors that are difficult to capture in econo-metric models, such as irrationality of market participants, bubbles and herdbehavior (Ehrmann & Fratzscher, 2005). However, the general pattern in foreignexchange markets has recently been found to be the very short run exchange rateconditional mean adjustment, characterized by a jump immediately following the

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

The Effects of US Macroeconomic Surprises 377

announcements of macroeconomic news, and little movement thereafter, whilethe exchange rate volatilities adjust gradually (Andersen et al., 2003). Thus, theissue of how the public information of macroeconomic fundamentals and foreignexchange rates are incorporated in the short run has been important in financialeconomics even if it has also been one of the least well understood issues.

For the analysis, this paper uses two kinds of new data sets. One is the datasetconsisting of four years of real time 30-minute high frequency US Dollar (USD)–Euro (EUR) and US Dollar (USD)–Japanese Yen (JPY) exchange rates, which isprovided by Olsen and Associates and the other is the dataset of the macroe-conomic news containing the information of surveyed expectations and actualrealizations for the US macroeconomic fundamentals provided by the Interna-tional Money Market Service (MMS), which are similar to that used in Andersenet al. (2003), Galati and Ho (2003), Faust et al. (2004), and Ehrmann andFratzscher (2005). By exploring the survey data on foreign exchange market partic-ipants’ expectations of the US announcements for macroeconomic fundamentals,this paper extracts the US macroeconomic surprises components by computingthe difference between the expected surveys and the realizations of the US macroe-conomic announcements. Among many US macroeconomic indexes, this paperselects five major indexes such as Durable Goods, GDP, PPI, Unemployment andNAPM index.

Then this paper investigates how the jumps related to the US macroeconomicsurprises can explain the intraday movements of the foreign exchange rates. Forthis purpose, this paper adopts the Fractionally Integrated Generalized AutoRe-gressive Conditional Heteroskedasticity (FIGARCH) model combined with theBernoulli distribution, which allows for the possibility of time-varying jumpscaused by the US macroeconomic surprises. Furthermore, to highlight the signeffects of the macroeconomic surprises, this paper distinguishes positive surprisesfrom negative surprises. The general findings of this paper are that the FIGARCHmodel with the Bernoulli distribution performs quite well and is more appropriateto model the high frequency exchange returns than the basic FIGARCH model ofBaillie et al. (1996) and that the US macroeconomic surprises affect the intradaymovements in the volatility process of the foreign exchange rates asymmetrically,depending on the signs of the surprises through the jumps: the positive surprisesincrease the movements of the volatility process with the high jump probabilitywhile the negative surprises decrease the movements with the low jump prob-ability. In addition, the US macroeconomic surprises tend to increase the longmemory persistence in the volatility process of the high frequency returns seriesdue to the jumps. Thus, the US macroeconomic surprises would be an importantrole in explaining the intraday movements in the volatility process of the highfrequency exchange returns.

This paper attempts to make a contribution to the literature in two centralregards. First, it focuses on the intraday movements in the volatility processof foreign exchange rates, which are related to the surprises of macroeconomicfundamentals, and in particular on the question of whether the presence of asym-metries in the movements depends on the surprise signs. This issue could be oflarge relevance because, from a policy perspective, news about economic fun-damentals is important for overall exchange rate movement. Second, this paper

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

378 Y.W. Han

provides the overall importance of the intraday movements related to the timevarying jumps that affect the long memory property in the volatility process offoreign exchange rates at a high frequency level. Thus, this paper could be help-ful to improve our understanding of the relationship between news arrivals andintraday exchange rate movement by providing an empirical examination of pricediscovery in the context of foreign exchange.

The plan for the rest of this paper is as follows: section 2 presents a basicanalysis of the high frequency returns series for the USD-EUR and the USD-JPY exchange rates. For the analysis of the high frequency returns, this paperapplies the Flexible Fourier Form (FFF) proposed by Gallant (1981, 1982) toeliminate the intraday periodicity in the high frequency returns and then usesthe basic FIGARCH model of Baillie et al. (1996) with a normal distribution toinvestigate the volatility process of the high-frequency filtered returns. Section3 of the paper describes the normal mixture distribution model, the Bernoullijump process, in order to account for the time-varying jumps caused by the USmacroeconomic surprises in the high frequency returns. Thus, the FIGARCHmodel with the mixture distribution is presented to investigate the effects of thejumps on the intraday movements in the volatility process of the high frequencyreturns. Section 4 provides a conclusion.

2. Basic Properties of High-frequency Foreign Exchange Rates

This section is concerned with the set of the 30-minute high frequency USD-EUR and USD-JPY spot exchange rate data provided by Olsen & Associatesof Zurich, in which Reuter FXFX quotes are taken every 30-minutes for thecomplete calendar years of 1999 through 2002, in which the Euro currency wasfirst introduced in the world currency markets and the foreign exchange marketswere relatively been stable after the Asian crisis. The sample period is 00:00 GMT(Greenwich Mean Time), January 4, 1999 through to 00:00 GMT, January 1,2003. Each quotation consists of a bid and an ask price and is recorded in time tothe nearest second. Following the procedures of Baillie et al. (2000, 2004), the spotexchange rate for each 30-minute interval is obtained by linearly interpolating theaverage of the log bid and the log ask at the two closest ticks.

The nth 30-minute high frequency return for day t is,

Rt,n = 100 × [ln(St,n) − ln(St,n−1)] (1)

where St,n is the 30-minute spot foreign exchange rate, which is the average of thelog bid price and the log ask price. It has become fairly standard in this literatureto remove atypical data associated with slower trading patterns during weekends(Müller et al., 1990; Bollerslev & Domowitz, 1993). Hence, the returns duringthe weekends from Friday 21:30 GMT through Sunday 20:00 GMT are excluded.

In particular, this definition of the weekend was motivated by the ebb andflow in the daily FX activity patterns documented in Bollerslev and Domowitz(1993). They presented that the weekend data with much lower trading activitiesare excluded since they cannot provide any economic implications. However, thereturns for holidays occurring during the sample are retained in order to preserve

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

The Effects of US Macroeconomic Surprises 379

the number of returns associated with one week, as suggested in the Appendixof Andersen and Bollerslev (1997) and Baillie et al. (2000). The eventual sampleused in the subsequent analysis contains 1042 trading days, each with 48 intervalsof 30-minute duration, which realizes a total of 50,016 observations for the highfrequency returns.

The 30-minute high frequency returns of the USD-EUR and the USD-JPYexchange rates are presented in Figures 1(a) and (b) respectively. They are centeredon zero but there exists obvious volatility clustering in the two high frequencyreturn series. The sample means of the 30-minute exchange return are found to be−0.0002 and 0.0001 for the USD-EUR and the USD-JPY exchange rates, whichare very close to zero and indistinguishable at the standard significance level giventhe sample deviations of 0.01 and 0.1. However, both of the returns appear notto be normally distributed since the sample skewness and kurtosis are 0.30 and17.82 for the USD-EUR returns and 0.17 and 16.71 for the USD-JPY returns,which are all found to be statistically significant.1 In particular, the estimatedkurtosis statistics for the two high frequency returns are found to be relativelylarge, which implies the rejection of a Gaussian normal distribution assumption.The high excess kurtosis may be due to the occurrences of numerous jumps thathave taken place in the two high frequency returns. These jumps could lead to thelevel and volatility outliers that the normal distribution cannot take into account(Hotta & Tsay, 1998).

Figures 2(a) and (b) present the first 480 autocorrelation coefficients for thereturns, squared returns and absolute returns of the unadjusted (raw) 30-minuteUSD-EUR and USD-JPY exchange rates respectively. For the two high frequencyreturns, there are small, negative but very significant first-order autocorrelations.The weak negative correlations may be attributed to a combination of a smalltime varying risk premium, bid-ask bounce, and/or non-synchronous tradingphenomena while higher order autocorrelations are not significant at conven-tional levels; see Andersen and Bollerslev, (1997) and Goodhart and O’Hara (1997)for a description of this issue in high frequency currency markets. In particular,Andersen and Bollerslev (1997) presented that the small first autocorrelation pro-vides some support for the hypothesis that foreign exchange dealers positiontheir quotes asymmetrically relative to the perceived true market prices as a wayto manage their inventory positions, thus causing the midpoint of the quotedprices to move around in a fashion similar to the bid-ask bounce observed onorganized exchanges.

The autocorrelation functions of the squared and absolute returns of the highfrequency returns exhibit a pronounced U-shape pattern (intraday periodicity),associated with substantially slow decays of the autocorrelations. The generalpattern is consistent with the studies of Müller et al. (1990), Baillie and Boller-slev (1991), Dacorogna et al. (1993), Andersen and Bollerslev (1998), Baillieet al. (2000, 2004) and Andersen et al. (2003). The pattern is generally attributed tothe opening of the European, Asian and North American markets superimposedon each other as presented by Andersen and Bollerslev (1997).

1According to Jarque and Bera (1987), the standard errors of the sample skewness and the samplekurtosis in their corresponding normal distributions are (6/T)1/2 and (24/T)1/2.

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

380 Y.W. Han

(a) USD-EUR

(b) USD-JPY

Figure 1. 30 minute spot returns.

In order to remove the strong intraday periodicity, this study follows a similarapproach used by Andersen and Bollerslev (1997, 1998) and Baillie et al. (2000,2004), and adopts a two-step estimation method. First, the intraday periodicity isremoved by applying the FFF approach of Gallant (1981, 1982); see the Appendix

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

The Effects of US Macroeconomic Surprises 381

Figure 2. Correlograms of 30-minute raw (unfiltered) high frequency exchange returns.

of Andersen and Bollerslev (1997) for details. Without any economic theory forstimulating a particular parametric form for the intraday seasonal pattern, theflexible nonparametric procedure seems to be natural.2 One advantage of this

2There are several alternative procedures for the intraday seasonal adjustment, which are (i) adopt-ing an intraday timescale to deseasonalize the volatility (Dacorogna et al., 1993); (ii) using a set ofmultiplicative factors (Taylor & Xu, 1998; Chang & Taylor, 2003); (iii) dividing returns by their

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

382 Y.W. Han

approach is that it allows the shape of the periodic pattern in the market to dependon the overall level of the volatility (Andersen & Bollerslev, 1997). In particular,Baillie et al. (2000, 2004) show that the FFF method appears to be appropriate forrepresenting intraday periodicity without inducing any nonlinearity and obviousdeficiencies with the model. The high frequency USD-EUR and USD-JPY returnsare then filtered from the FFF method. The filtered high frequency returns aredefined as:

yt,n = Rt,n/pt,n (2)

where pt,n is the intraday seasonality series estimated from the FFF method.3Figures 3(a) and (b) present the first 480 autocorrelation coefficients for returns,

squared returns and absolute returns of the filtered 30-minute USD-EUR andUSD-JPY exchange rates respectively. While they reveal a familiar lack of auto-correlation in returns, the figures represent a dramatic reduction of the strongintraday periodicity and the more marked persistence in autocorrelations ofsquared and absolute returns. In particular, the autocorrelation functions for thesquared and absolute returns do not display the usual exponential decay associ-ated with the stationary and invertible process, but rather appear to be generatedby a long memory process with the hyperbolic decay.

A model that is consistent with these stylized facts is the MA(1)-FIGARCH(1,d, 1) process,

yt,n = μ + εt,n + θεt,n−1 (3)

εt,n = zt,nσt,n (4)

σ 2t,n = ω + βσ 2

t,n−1 + [1 − βL − (1 − ϕL)(1 − L)d]ε2t,n (5)

where zt,n is an i.i.d.(0,1) process, and the time indexes t = 1, . . . , 1042 days andn = 1, . . . , 48. The FIGARCH model in equation (5) is motivated by the standardGeneralized AutoRegressive Conditional heteroskedasticity (GARCH) model ofBollerslev (1986). The parameter (d) characterizes the long memory property ofhyperbolic decay in volatility because it allows for autocorrelations decaying ata slow hyperbolic rate. When d = 0, then equation (5) reduces to the standardGARCH (1,1) model, and when d = 1, then equation (5) becomes the IntegratedGARCH, or IGARCH (1,1) model implying the complete persistence of the con-ditional variance to a shock in squared returns. The attraction of the FIGARCHprocess is that for 0 < d < 1, it is sufficiently flexible to allow for intermedi-ate ranges of persistence, which represents the slow hyperbolic rates of decay inthe autocorrelations of the squared returns. Furthermore, the associated impulse

cross-sectional average (Melvin & Yin, 2000); and (iv) using the fractionally cyclically integratedmodel (Gray et al., 1989, 1994; Gil-Alana, 2001). But no one procedure has been found to be clearlysuperior.3The intra-day seasonality pt,n is estimated based on the combination of trigonometric functionsand polynomial terms since the approach could result in better approximation properties whenestimating regularly recurring patterns, as presented by Andersen and Bollerslev (1997).

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

The Effects of US Macroeconomic Surprises 383

Figure 3. Correlograms of 30-minute filtered high frequency exchange returns.

response weights also exhibit quite persistent hyperbolic decay. The FIGARCHprocess has impulse response weights, σ 2

t = ω/(1 − β) + λ(L)ε2t, where for

large lags k, λk ≈ kd−1, which is essentially the long memory property or ‘Hursteffect’ of hyperbolic decay. The FIGARCH process is strictly stationary andergodic for 0 ≤ d ≤ 1, and shocks will have no permanent effect. Further detailsconcerning the FIGARCH process can be found in Baillie et al. (1996).

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

384 Y.W. Han

The above equations (3) through (5) are estimated by using nonlinearoptimization procedures to maximize the Gaussian log likelihood function,

ln(ζ ; θ) = −(T/2) ln(2π)(1/2)�t=1,...,1042, n=1,...,48[ln(σ 2t,n + ε2

t,nσ−2t,n )] (6)

where θ is a vector containing the unknown parameters to be estimated. However,it has long been recognized that most asset returns are not well represented byassuming zt,n in equation (4) is normally distributed; for example, see McFarlandet al. (1982). Thus, the inference is based on the Quasi Maximum LikelihoodEstimation (QMLE) of Bollerslev and Wooldridge (1992), which is valid whenzt,n is non-Gaussian. Denoting the vector of parameter estimates obtained frommaximizing equation (6) using a sample of T observations on equations (3), (4)and (5) with zt,n being non-normal by θ̂T, then the limiting distribution of θ̂T is:

T1/2(θ̂T − θ) −→ N[0, A(θ0)−1B(θ0)A(θ0)

−1] (7)

where A(θ0) =[

∂2 ln(ξ)∂θ0∂θ0

]and B(θ0) =

[∂ ln(ξ)

∂θ0

]represent the Hessian and outer

product gradient respectively; and θ0 denotes the vector of true parameter values.Equation (7) is used to calculate all the robust standard errors that are reportedin the subsequent tables, with the Hessian and outer product gradient matricesbeing evaluated at the point θ̂T for practical implementation. In particular, toobtain maximum likelihood estimates and the second-order efficiency, an iterativeprocedure is called for. However, this paper uses the BHHH algorithm (Berndtet al., 1974) as suggested by Bollerslev (1986) because of the recursive problem inthe FIGARCH process.

Table 1 presents the estimation results of applying the above model to thefiltered high frequency USD-EUR and USD-JPY exchange returns. The long mem-ory volatility parameters (d) are estimated to be 0.19 and 0.27 for the USD-EURand the USD-JPY returns, and are found to be statistically significant. Thus, thehypotheses that d = 0 (stationary GARCH) and also d = 1 (integrated GARCH)can be comprehensively rejected for the returns using standard significance levels.Table 1 also reports the Robust Wald test statistics denoted by Wd=0 for testingthe null hypothesis of GARCH (1,1) versus a FIGARCH (1,d,1) data generat-ing process. Under the null, W will have an asymptotic χ2

1 distribution and theGARCH (1,1) model is rejected for the high frequency returns at standard signif-icance levels. It presents strong support for the hyperbolic decay and persistenceof the FIGARCH model as opposed to the conventional exponential decay associ-ated with the stable GARCH(1,1) model.4 Furthermore, a sequence of diagnosticportmanteau tests on the standardized residuals and squared standardized resid-uals failed to detect any need to further complicate the model.5 In general, the

4Beine et al. (2002) presented the cumulative impulse functions of FIGARCH models and comparedthem with those of GARCH and IGARCH models. The FIGARCH model seems to be betterrepresenting the hyperbolic decay of the impulse functions than GARCH and IGARCH models.5Tests of model diagnostics are performed by the application of the Box-Pierce portmanteau statisticon the standardized residuals. The standard portmanteau test statistic Qm = T�j=1,mr2

j , where rj isthe jth order sample autocorrelation from the residuals is known to have an asymptotic chi squared

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

The Effects of US Macroeconomic Surprises 385

Table 1. Estimated MA(1)-FIGARCH (1,d,1) model for 30-minute highfrequency filtered exchange rate returns

USD-EUR USD-JPY

μ 0.0002 0.0007∗∗(0.0004) (0.0004)

θ −0.0639∗∗∗ −0.0541∗∗∗(0.0062) (0.0060)

d 0.1949∗∗∗ 0.2701∗∗∗(0.0132) (0.0199)

ω 0.0002∗∗∗ 0.0005∗∗∗(0.0001) (0.0001)

β 0.9125∗∗∗ 0.7336∗∗∗(0.0232) (0.0339)

ϕ 0.8931∗∗∗ 0.6090∗∗∗(0.0290) (0.0408)

ln(L) 49417.502 48054.125Skewness −0.116 0.043Kurtosis 10.746 11.309Q(50) 88.101 76.119Q2(50) 58.587 49.852Wd=0 217.656∗∗∗ 184.766∗∗∗

Notes: (i) QMLE asymptotic standard errors are in parentheses below the corre-sponding parameter estimates. The asterisks (∗∗∗,∗∗∗, ∗) represents the significancelevel of 1%, 5% and 10%. (ii) The quantity ln(L) is the value of the maximized loglikelihood. (iii) The sample skewness and kurtosis refer to the standardized residuals.(iv)The Q(50) and Q2(50)statistics are the Ljung-Box test statistics for 50 degrees offreedom to test for serial correlation in the raw normalized and squared normalizedresiduals.

various diagnostic statistics all indicate that the FIGARCH model is superior tothe usual GARCH for modeling the long memory in the conditional variance pro-cess of the two high frequency returns series. Thus, this basic FIGARCH modelseems to match the dynamics of the high frequency exchange rate returns and isa satisfying starting point to study the nature of the underlying distributions.

However, the estimated excess kurtosis is found to be 10.75 and 11.31 forthe high frequency returns of the USD-EUR and the USD-JPY exchange ratesrespectively, which are still large, implying the rejection of the normal distributionassumption. As mentioned before, the jumps occurring in the high frequencyexchange returns appear to cause the high excess kurtosis. A potential sourceof the jumps in the high frequency returns may be important events in foreignexchange markets, such as news surprises from macroeconomic fundamentals orcentral bank interventions (Andersen et al., 2003; Beine & Laurent, 2003). Theseevents concerning expected future flows can result in price changes well abovenormal and might be better captured by jumps rather than normal innovations.

distribution with m − k degrees of freedom, where k is the number of parameters estimated inthe conditional mean. Similar degrees of freedom adjustment are used for the portmanteau teststatistic based on the squared standardized residuals when testing for omitted ARCH effects. Thisadjustment is in the spirit of the suggestions by Diebold (1988).

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

386 Y.W. Han

These jumps might lead to the level and volatility outliers that cannot be taken intoaccount by the simple normal distribution, as Hotta and Tsay (1998) presented.Thus, the basic FIGARCH models under the assumption of the usual normaldistribution seem to be inappropriate to represent the high frequency returnsseries properly.

Since the presence of the jumps is primarily responsible for the rejection ofthe usual normality assumption, it seems to call for the use of another model.One model to be considered in this paper is to introduce time varying jumpsthrough the use of a normal mixture distribution. These jumps may be relatedto the public information about macroeconomic fundamentals that influence theintraday movements of the high frequency exchange rates.

3. Intraday Movements and Time-varying Jumps with the US MacroeconomicSurprises

This section is concerned with modeling the jumps in the high frequency returnsseries of the USD-EUR and the USD-JPY exchange rates to investigate the intradaymovements of the foreign exchange rates. In particular, this paper employs thejump diffusion process proposed by Press (1967) in order to account for theconditional jumps in the 30-minute high frequency returns of the USD-EUR andthe USD-JPY exchange rates. Initially, Press (1967) proposed a jump diffusionmodel for stock prices under the assumption that the logarithm of the stock pricefollows a Brownian motion process on which i.i.d. normal distributed jumps areincluded. Jorion (1988) used a Press-type model to find some statistical evidenceof jumps in the USD-DM exchange rate for the post 1971 free floating period. Thisjump diffusion model has subsequently been widely employed to model featuresof the Exchange Rate Mechanism (ERM) of the European Monetary System(EMS) such as the jumps resulting from realignments of the ERM bands and highexcess kurtosis. See Vlaar and Palm (1993, 1997), Nieuwland et al. (1994), Neely(1999), Baillie and Han (2001) and others.

This paper relies on a jump diffusion process-FIGARCH model that assumesthat the high frequency returns are drawn from a mixture of normal distributionand a diffusion process combined with an additive jump process. In particular,this paper considers this model in the context of a Bernoulli-normal distribution.The Bernoulli distribution models the stochastic jumps in the 30-minute highfrequency exchange returns series. The main characteristic of the Bernoulli pro-cess is that, over a fixed time period, one relevant information arrives in foreignexchange markets and a jump occurs in the high frequency exchange rates withprobability (λ), which is drawn from a Bernoulli distribution and is forced in the(0,1) interval. The jump size is given by the random variable, which is assumed tobe NID(ν, δ2).

Since a constant jump probability model may not provide economic and finan-cial insights as presented by Beine and Laurent (2003), this paper adopts thetime-varying jump probability that is determined endogenously. Initially, thistime-varying jump probability model has been extensively used in papers suchas Nieuwland et al. (1994), Vlaar and Palm (1997) and Neely (1999) for theanalysis of the foreign exchange rates in European Monetary System (EMS) and

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

The Effects of US Macroeconomic Surprises 387

has recently been used to investigate the effects of central bank interventionson the movements of foreign exchange rates. Following the previous papers, thispaper considers the time-varying jump probability associated with the macroeco-nomic surprises that are obtained from the macroeconomic fundamentals. Themacroeconomic fundamentals have been considered to play an important rolein explaining the movements of foreign exchange rates. A voluminous empiricalliterature has provided evidence that the jumps in major foreign exchange rateseries appear to be linked with macroeconomic fundamentals and the jumps maycause asymmetry and high excess kurtosis in the process of foreign exchange rates(Goodhart et al., 1993; Ghosh, 1997; Almeida et al., 1998; Andersen et al., 2003).In fact, foreign exchange rates are the relative prices of two currencies so thatthe news from the two areas should affect the exchange rate at the same weight.But the news only from the US, not that from the other areas like Japan or EUregions, generally seems to have a statistically significant impact on the Dollar-DEM or the USD-EUR exchange rates (Edison, 1997; Andersen et al., 2003).6Thus, this paper focuses only on the US macroeconomic fundamentals and inves-tigates the effects of the US macroeconomic surprise on the intraday movementsof the foreign exchange rate returns.

In order to calculate the US macroeconomic surprise, this paper follows themethod proposed by Andersen et al. (2003) using the International Money MarketServices (MMS) real time data on expected and realized (announced) US macroe-conomic indicators sampled from January 1999 to December 2002. Thus, the USmacroeconomic surprise associated with US macroeconomic fundamental k is,

xkt,n = (Ak

t,n − Ekt,n)/σ

k (8)

where Akt,n is the announced (realized) value of a macroeconomic fundamental

(k) released at time (n) and day (t), Ekt,n is the market expected value of the

macroeconomic fundamental (k) as distilled in the MMS median forecast, andσ k is the sample standard deviation of (Ak

t,n − Ekt,n). In particular, for the MMS

expectations data for macroeconomic indicators, MMS has conducted a Fridaytelephone survey of about 40 money managers, collected forecasts of all indicatorsto be released during the next week and reported the median forecasts from thesurvey.7 For more detailed information about the MMS data, see the Appendix.

Among many US macroeconomic fundamentals, this paper selects five majorones, Durable goods, PPI, GDP, Unemployment and NAPM index and uses them

6Preliminary tests on several macroeconomic surprises from EMU and Japan indicate that the effectsof the EMU and Japanese surprises on the high frequency USD-EUR and Dollar-JPY exchange ratesappear to be generally statistically insignificant. The results are not reported here to save space, butthey are available upon request to the author. As pointed by Andersen et al. (2003), one possibleexplanation for this disparity is that the release day and time of the US macroeconomic news areknown exactly but not for EMU and Japan, in which only the release day is exactly known butthe release time varies. Uncertain release times may result in less market trading around the releasetimes, hence resulting in insignificantly smaller effects.7Balduzzi et al. (1998) and Andersen et al. (2003) presented that the MMS forecasts data containvaluable information about the forecasted variable and they are mostly unbiased.

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

388 Y.W. Han

as the explanatory variables of the jump probability.8 Thus, the time-varying jumpprobability (λt,n) is,

λt,n = 1 −⎡⎣1 + exp

⎛⎝λ0 +

∑k=1,5

λkxkt,n

⎞⎠

⎤⎦

−1

(9)

where the xkt,n are the five US macroeconomic surprises expected to be related to

the jump probability. Furthermore, to highlight the effects of the macroeconomicsurprises contained in the major macroeconomic fundamentals, this paper dis-tinguishes positive surprises from negative surprises and explores the effects ofthe macroeconomic surprise on the jumps. If the actual realization value is largerthan the surveyed forecast value, the surprise is classified as positive since it cancontribute to the growth of the economy. In addition, if the actual value is lessthan the expected value, it is regarded as negative since it may slow the economydown. This rule seems to be straightforward for the surprise from Durable goods,GDP and NAPM index. But the surprises from employment and PPI have reversedsigns so that the positive (negative) surprises (i.e. higher (lower) actual realiza-tions than expected surveyed values) is associated with lower (higher) economicactivity.

In order to investigate the effects of the US macroeconomic surprise effects onthe intraday movements of the high frequency returns series, this paper generalizesthe basic MA(1)-FIGARCH (1,d,1) model in section 2 by allowing the mixturedistribution, the Bernoulli normal distribution with the time-varying jump prob-ability. Then the QMLE estimation method is also used for the generalized model.Since the statistical and economic motivations for the jumps and the long mem-ory property are quite different, this work chooses the model specification thataccounts for the two features in high frequency exchange rate returns at thesame time. Hence, this paper investigates the 30-minute high frequency returnsof the USD-EUR and the USD-JPY exchange rates by employing the MA(1)-FIGARCH(1,d,1)-Bernoulli jump diffusion process with the time-varying jumpprobability associated with the either positive or the negative US macroeconomicsurprises in order to consider both the jumps and the long memory property in thehigh frequency exchange returns process. The inclusion of the jump process mayreduce the influence of the jumps on the MA(1) – FIGARCH (1,d,1) specification.

The combined MA(1)-FIGARCH (1,d,1) model with Bernoulli jump process is,

yt,n = μ + λt,nν + εt,n + θεt,n−1 (10)

εt,n ∼ (1 − λt,n)N(−λt,nν, σ 2t,n) + λt,nN((1 − λt,n)ν, σ 2

t,n + δ2) (11)

σ 2t,n = ω + βσ 2

t,n−1 + [1 − βL − (1 − ϕL)(1 − L)d]ε2t,n (12)

8For the explanatory variables of the jump probability, five major US macroeconomic variables areused after eliminating the macroeconomic variables which are announced more than one in the tworegions on the same day.

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

The Effects of US Macroeconomic Surprises 389

The 30-minute high frequency returns are still specified as following the MA(1)process, with the time-varying jump probability (λt,n) associated with theUS macroeconomic surprise, which is drawn from a Bernoulli distribution(0 < λt,n < 1) and ν is the mean of the jump distribution. Following Vlaar andPalm (1997) and Beine and Laurent (2003), the distribution of the disturbances εt,nis given by the mixture of two normal distributions conditional on the absence ofa jump, an event that has probability (1 − λt,n), εt,n is normally distributed withexpectation −λt,nν and variance σ 2

t,n, otherwise the expectation is increased bythe mean jump size ν and the variance is increased by δ2, which captures the vari-ance of the jump size, implying the additional volatility related to the jumps. And,the volatility process is the FIGARCH(1,d,1) model as developed in section 2. Thelog likelihood function for the combined model has the following form,

ln(ξ) = −(T/2) ln(2π) + �t=1,...,262, n=1,...,48 ln{[(1 − λt,n)/σt,n]× exp[−(εt,n + λt,nν)2/2σ 2

t,n] + [λt,n/(σ2t,n

+ δ2)1/2] × exp[−(εt,n − (1 − λt,n)ν)2/2(σ 2t,n + δ2)]} (13)

The form of the likelihood function for the Bernoulli-normal mixture distribu-tion is basically similar to that proposed by Vlaar and Palm (1997), which usedmacroeconomic variables such as the inflation differentials, and the trade balanceof two countries as the explanatory variable of the jump probability in a GARCHframework. In addition, the normalized residuals are used for the statistical infer-ence instead of the usual standardized residuals since the standardization maynot lead to i.i.d. residuals in the mixture distribution model with time dependentvariance as suggested by Beine and Laurent (2003).

Table 2 represents the maximum likelihood results from the FIGARCH modelwith the Bernoulli-normal mixture distribution for the high frequency returnsseries of the USD-EUR and the USD-JPY exchange rates. Generally, the resultsare quite similar for the two currencies. The LR test is performed to test thenull hypothesis of the FIGARCH model with the normal distribution versus theFIGARCH model with the mixture distribution. The large LR test statistics, 3250and 3370 of the high frequency USD-EUR and USD-JPY returns reject the nullhypothesis at standard significance levels confirming that the FIGARCH modelwith the mixture distribution outperforms the model with the normal distributionalone regardless the individual significance for each of the additional parameters.Thus, it seems relevant over such a period to introduce the possibility of jumps inthe high frequency exchange rate process.

For the two high frequency returns, the parameters (δ2) are all positive andare statistically significant in the cases where the jumps are associated with eitherthe positive or the negative US macroeconomic surprises while the parameters(υ) are all negative but are not significant at all in both cases. This estimationresults show that the additional volatility is associated with the jumps relatedto the macroeconomic surprises while the jumps do not affect the mean processitself. This implies that the jumps mostly affect the volatility process of the highfrequency returns rather than the mean process. Furthermore, the estimated val-ues of the parameters (δ2) for the positive surprises and the negative surprises are

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

390 Y.W. Han

Table 2. Estimated MA(1)-Bernoulli jump process – FIGARCH (1,d,1) model for the filtered30-minute high frequency exchange rate returns

USD-EUR USD-JPY

Positive surprises Negative surprises Positive surprises Negative surprises

μ 0.0011 0.0011 0.0014 0.0018∗∗(0.0009) (0.0009) (0.0009) (0.0009)

λ0 −2.0689∗∗∗ −2.0686∗∗∗ −2.2496∗∗∗ −2.2517∗∗∗(0.1225) (0.1218) (0.1107) (0.1092)

λD 0.0397∗∗ −0.0216∗∗∗ 0.0229∗∗∗ −0.0175∗∗∗(0.0198) (0.0052) (0.0086) (0.0068)

λP 0.0155 −0.0269∗∗∗ 0.0261∗∗∗ −0.0027(0.0099) (0.0102) (0.0152) (0.0066)

λG 0.0128∗∗∗ −0.0041 0.0064 −0.0199*(0.0047) (0.0135) (0.0053) (0.0110)

λU 0.0276∗∗ −0.0326∗∗∗ 0.0194∗∗∗ −0.0294∗∗∗(0.0112) (0.0122) (0.0058) (0.0072)

λN 0.0133 −0.0205∗∗ 0.0139 −0.0221∗∗∗(0.0094) (0.0106) (0.0138) (0.0063)

υ −0.0041 −0.0038 −0.0051 −0.0069(0.0037) (0.0037) (0.0043) (0.0042)

δ2 0.0304∗∗∗ 0.0305∗∗∗ 0.0373∗∗∗ 0.0373∗∗∗(0.0033) (0.0033) (0.0039) (0.0039)

θ −0.0761∗∗∗ −0.0762∗∗∗ −0.0740∗∗∗ −0.0742∗∗∗(0.0048) (0.0048) (0.0049) (0.0049)

d 0.1067∗∗∗ 0.1063∗∗∗ 0.1215∗∗∗ 0.1216∗∗∗(0.0065) (0.0065) (0.0059) (0.0059)

ω 0.0001∗∗∗ 0.0001∗∗∗ 0.0002∗∗∗ 0.0002∗∗∗(0.0000) (0.0000) (0.0000) (0.0000)

β 0.9047∗∗∗ 0.9052∗∗∗ 0.6009∗∗∗ 0.6037∗∗∗(0.0250) (0.0246) (0.0440) (0.0435)

ϕ 0.8989∗∗∗ 0.8995∗∗∗ 0.5630∗∗∗ 0.5659∗∗∗(0.0266) (0.0261) (0.0448) (0.0443)

ln(L) 52674.383 52677.947 51425.793 51433.362Skewness 1.022 1.022 1.025 1.025Kurtosis 1.060 1.060 1.067 1.067Q(50) 91.990 89.353 106.160 105.842Q2(50) 90.661 88.649 106.041 106.609

Notes: (i) The same as Table 1 except that a jump intensity of λt,n where 0 < λ < 1 and λt,n = [1 + exp(λ0 +∑k=1,5 λixk

t,n)]−1 and is specified by the Bernoulli process. And, the jump size (vt,n) is assumed to be NID(ν,δ2). (ii) The parameters (λD, λP, λG, λU and λN) represent the estimated values for Durable Goods, PPI, GDP,Unemployment and NAPM index respectively.

found to be quite similar where the values are 0.0304 and 0.0305 for the USD-EURreturns and 0.0373 and 0.0373 for the USD-JPY returns respectively. This suggeststhat the jumps associated with either the positive or the negative macroeconomicsurprises almost equally increase the intraday movements of the volatility processregardless of the signs of the macroeconomic surprises. Thus, the US macroeco-nomic surprises generally increase the probability of the big loss of the Euro andJapanese Yen currency against the US Dollar since the mean jump sizes (υ) are

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

The Effects of US Macroeconomic Surprises 391

all negative (although they are all insignificant) but the variances of jumps size(σ 2) are very large so that the volatility increases significantly with the surprises.The risk of the big loss may be at least partly compensated by the increase of theexpected excess returns due to the rise in the volatility.

Now, we turn to the values of the jump parameter estimates. The result con-firms that the US macroeconomic surprises are associated with the jumps observedin the movements of the high frequency exchange rates. For both high frequencyreturns, almost all US macroeconomic surprises are found to be statistically signif-icant and influence the intraday movements of the foreign exchange rates throughthe jump probability. Since the jumps are associated with the increase in the highfrequency exchange return volatility, this implies that the results of this paper arefully consistent with the many previous papers that reported the effects of macroe-conomic fundamentals on the exchange rate volatility (DeGennaro & Shrieves,1997; Andersen & Bollerslev, 1998; Melvin & Yin, 2000; Cai et al., 2001). The signeffect generally appears to be maintained. The signs of the estimated parameters(λk) for the explanatory variables of the jumps are mostly positive for the positivesurprises but they are all negative for the negative surprises. Since the parameters(λk) represent the effects of the US macroeconomic surprises on the jump prob-ability, the estimation results show that the movements of the volatility processof the foreign exchange rates reacts to the US macroeconomic surprise asymmet-rically depending on their signs (sign effect): the positive surprises increase themovements of the exchange rates with the high jump probability but the negativeones decrease the movements with the low probability. In particular, the positivesurprises of the Durable Goods variable and the negative surprises of the Unem-ployment variable appear to affect the movements of the foreign exchange ratesmost greatly.

The main contribution of the FIGARCH model with the mixture distributionlies in the significant decrease in the excess kurtosis and the long memory param-eter of the volatility process. For the two high frequency returns, the kurtosis ofthe residuals are all significantly decreased to around 1.0, regardless of the factthat the jump probability is associated with the positive or negative surprises.And, the long memory parameters (d) estimated in the FIGARCH model withthe mixture distribution are also decreased to around 0.1067 and 0.1063 for thepositive and negative surprises in the high frequency USD-EUR returns and about0.1215 and 0.1216 for the two surprises respectively in the high frequency USD-JPY returns and they are all very significant. The decrease in the long memoryproperty of the two high exchange returns suggests that the additional volatilitycaused by the jumps may spuriously induce the higher values of the long memorypersistence. This is because the jumps are fully accounted for in the mixture distri-bution model by the additional parameter (δ2), the volatility associated with thejumps. Thus, the long memory persistence of the volatility process in the high fre-quency exchange rates can decrease when jumps are accounted for appropriately.This result strongly supports the interaction between the jumps and the volatilitypersistence of high frequency exchange rates. This is quite consistent with thefindings of Diebold and Inoue (2001), Granger and Hyung (2004) and Beine andLaurent (2003) who presented that the long memory property can decrease whenstructural changes or jumps are properly modeled.

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

392 Y.W. Han

Hence, the estimation results seem to be quite understandable given that thejumps associated with US macroeconomic surprises are fully accounted for inthe mixture distribution since the jumps may spuriously cause the additionalvolatility. These results provide evidence that the jumps associated with the USmacroeconomic fundamentals increase the movements in the volatility processof the high frequency returns and that the US macroeconomic surprises asym-metrically influence the movements through the jump probability depending onthe signs of the surprises. Thus, the higher long memory property seems to berelated to the volatility adjustments to the additional movements, which react tothe US macroeconomic surprises asymmetrically, and the additional movementsrelated to the jumps seem to be one of the possible driving forces behind the longmemory property in the volatility process of the high frequency returns.9

4. Conclusions

This paper considers four years of high frequency returns data for the 30-minuteUSD-EUR and USD-JPY exchange rates and investigates the intraday movementsof the high frequency returns. Special attention is devoted to modeling the jumpsrelated to macroeconomic fundamentals appropriately in order to deepen ourunderstanding of the relations between foreign exchange rate movements andmacroeconomic fundamentals.

Initially, this paper uses the FIGARCH model with the normality assumption toconsider the high frequency returns series. However, the excess kurtosis estimatedfrom the FIGARCH model indicates that the normality assumption is found tobe inappropriate in representing the high frequency returns due to the occurrenceof the jumps in the process of the high frequency exchange returns. Thus, thispaper employs the mixture distribution model, the Bernoulli-normal distributionmodel, to account for the jumps in the high frequency returns. In particular, thispaper generalizes the FIGARCH model by combining the Bernoulli distributionwith the time-varying jump probability associated with the US macroeconomicsurprises (the differences between the announced values and the market expectedvalue of the US macroeconomic fundamentals).

The general results of this paper are (i) that the usual normality assumptionappears not to be appropriate for the high frequency exchange returns data,mainly due to the occurrences of jumps in the high frequency exchange ratereturns; (ii) that the major parts of the jumps seem to be closely related to theUS macroeconomic surprises in the foreign exchange markets; and (iii) that thenormal mixture distribution model, the Bernoulli-normal distribution with atime varying jump probability associated with the US macroeconomic surprise,is found to be quite successful in representing the intraday movements in thevolatility process of the high frequency returns. This paper also provides evidencethat macroeconomic surprises are an important driving force behind the exchangerate movements. The intraday movements of the volatility process of the foreignexchange rates appear to react asymmetrically depending on the signs of the

9The structural breaks may be another possible factor that can explain the long memory propertyin foreign exchange rates as presented by Diebold and Inoue (2001) and Granger and Hyung (2004).

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

The Effects of US Macroeconomic Surprises 393

macroeconomic surprises: the positive surprises increase the movements while thenegative surprises decrease them. This paper finds that the jumps associated withthe US macroeconomic surprises (either positive or negative) increase exchangerate volatility, which is in line with the findings in the empirical papers such asNieuwland et al. (1994), Vlarr and Palm (1997) and Neely (1999). Thus, the jumpsseem to spuriously induce the higher value of the long memory property in thevolatility process of the high frequency returns due to the additional volatilitycaused by the jumps as presented by Beine and Laurent (2003).

The mixture–normal-distribution model with the time-varying jump probabil-ity helps us to understand the importance of the movements in foreign exchangerates as a key determinant of volatility and the economic interpretation of thejumps. Such a model can be an alternative modeling for representing the dynam-ics of high frequency foreign exchange rates to other models with a normaldistribution or excluding jumps.

Acknowledgements

This work is supported by the Korea Research Foundation Grant (KRF-2006-332-B00054) and by the HallymUniversity Research Grant. The author is grateful to Olsen and Associates for making available their real timeexchange rates and to Money Market Services (MMS) International for their news data. And, the author wouldlike to thank the editor and the two anonymous referees for their valuable comments and suggestions.

References

Almeida, A., Goodhart, C. & Payne, R. (1998) The effects of macroeconomic ‘news’ on high frequency exchangerate behavior, Journal of Financial Quantitative Analysis, 33(3), pp. 383–408.

Andersen, T.G. & Bollerslev, T. (1997) Intraday periodicity and volatility persistence in financial markets, Journalof Empirical Finance, 4(2-3), pp. 115–158.

Andersen, T.G. & Bollerslev, T. (1998) Deutsche mark-dollar volatility: intraday activity, patterns, macroeco-nomic announcements and longer run dependence, Journal of Finance, 53(1), pp. 219–265.

Andersen, T.G., Bollerslev, T., Diebold, F.X. & Vega, C. (2003) Micro effects of macro announcements: Real-timeprice discovery in foreign exchange, American Economic Review, 93(1), pp. 38–62.

Baillie, R.T. & Bollerslev, T. (1991) Intra-day and inter-market volatility in foreign exchange rates, Review ofEconomic Studies, 58(3), pp. 565–585.

Baillie, R.T. & Han, Y.W. (2001) Testing target zone models using efficient methods of moments, Journal ofBusiness and Economic Statistics, 19(3), pp. 273–277.

Baillie, R.T., Bollerslev, T. & Mikkelsen, H.O. (1996) Fractionally integrated generalized autoregressiveconditional heteroskedasticity, Journal of Econometrics, 74(1), pp. 3–30.

Baillie, R.T., Cecen, A.A. & Han, Y.W. (2000) High frequency deutschemark - US dollar returns: FIGARCHrepresentations and non linearities, Multinational Finance Journal, 4(3-4), pp. 247–267.

Baillie, R.T., Cecen, A.A., Erkal, C. & Han, Y.W. (2004) Merging the stochastic with the nonlinear determin-istic: the case of high frequency European exchange rates, Journal of International Financial Markets,Institutions and Money, 14(5), pp. 401–418.

Balduzzi, P., Elton, E.J. & Green, C. (1998) Economic news and the yield curve: evidence from the U.S. treasurymarket, Salomon Center Working Paper S-98-5, Stern School of Business, New York University.

Beine, M. & Laurent, S. (2003) Central bank interventions and jumps in double long memory models of dailyexchange rates, Journal of Empirical Finance, 10(5), pp. 641–660.

Beine, M., Benassy-Quere, A. & Lecourt, C. (2002) Central bank intervention and foreign exchange rate: newevidence from FIGARCH estimations, Journal of International Money and Finance, 21(1), pp. 115–144.

Berndt, E.K., Hall, B.H., Hall, R.E. & Hausman, J.A. (1974) Estimation inference in nonlinear structuralmodels, Annals of Economics and Social Measurement, 4, pp. 653–665.

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

394 Y.W. Han

Bollerslev, T. (1986) Generalized autoregressive conditional heteroscedasticity, Journal of Econometrics, 31(3),pp. 307–327.

Bollerslev, T. & Domowitz, I. (1993) Trading patterns and prices in the interbank foreign exchange market,Journal of Finance, 48(4), pp. 1421–1443.

Bollerslev, T. & Wooldridge, J.M. (1992) Quasi-maximum likelihood estimation of dynamic models with timevarying covariances, Econometric Reviews, 11(2), pp. 143–172.

Cai, J., Cheung, Y.-L., Lee, R.S.K. & Melvin, M. (2001) Once-in-generation Yen volatility in 1998: fundamentals,interventions and order flow, Journal of International Money and Finance, 20(3), pp. 327–347.

Chang, Y.-C. & Taylor, S.J. (2003) Information arrivals and intraday exchange rate volatility, Journal ofInternational Financial Markets, Institutions and Money, 13, pp. 85–112.

Dacorogna, M.M., Müller, U.A., Nagler, R.J., Olsen, R.B. & Pictet, O.V. (1993) A geographical model for dailyand weekly seasonal volatility in the foreign exchange markets, Journal of International Money Finance,12(4), pp. 413–438.

DeGennaro, R.P. & Shrieves, R.E. (1997) Public information releases, private information and volatility in theforeign exchange market, Journal of Empirical Finance, 4(4), pp. 295–315.

Diebold, F.X. (1988) Modeling the persistence of conditional variances: a comment, Econometric Reviews, 5(1),pp. 51–56.

Diebold, F.X. & Inoue, A. (2001) Long memory and regime switching, Journal of Econometrics, 105, pp. 131–159.

Edison, H.J. (1997) The reaction of exchange rates and interest rates to news releases, International Journal ofFinance and Economics, 2(2), pp. 87–100.

Ehrmann, M. & Fratzscher, M. (2005) Exchange rates and fundamentals: new evidence from real-time data,Journal of International Money Finance, 24(2), pp. 317–341.

Evans, M.D.D. & Lyons, R.K. (2008) How is macro news transmitted to exchange rates? Journal of FinancialEconomics, 88(1), pp. 26–50.

Faust, J., Rogers, J.H., Wang, S.-Y. & Wright, J.H. (2004) The high frequency responses of exchange rates andinterest rates to macroeconomic announcements, Board of Governors of the Federal System, InternationalFinance Division Paper No. 784.

Frankel, J.A. & Rose, A.K. (1995) A survey of empirical research on nominal exchange rates, in: G. Grossman& K. Rogoff (Eds) Handbook of International Economics (Amsterdam: North-Holland).

Galati, G. & Ho, C. (2003) Macroeconomic news and the Euro/Dollar exchange rate, Economic Notes, 32(3),pp. 371–398.

Gallant, A.R. (1981) On the bias in flexible function forms and an essentially unbiased form, the Fourier flexibleform, Journal of Econometrics, 15(2), pp. 211–245.

Gallant, A.R. (1982) Unbiased determination of production technologies, Journal of Econometrics, 20(2),pp. 285–323.

Ghosh, D. (1997) Negative autocorrelation around large jumps in intra-day foreign exchange data, EconomicLetters, 56(2), pp. 235–241.

Gil-Alana, L.A. (2001) Testing of stochastic cycles in macroeconomic time series, Journal of Time Series Analysis,22, pp. 411–430.

Goodhart, C.A.E. & O’Hara, M. (1997) High frequency data in financial markets: issues and applications,Journal of Empirical Finance, 4(2-3), pp. 73–114.

Goodhart, C.A.E., Hall, S.G., Henry, B.S. & Pesaran, B. (1993) News effects in a high frequency model of theSterling-Dollar exchange rate, Journal of Applied Econometrics, 7(2), pp. 199–121.

Granger, C.W.J. & Hyung, N. (2004) Occasional structural breaks and long memory with an application to theS&P500 absolute stock returns, Journal of Empirical Finance, 11(3), pp. 399–421.

Gray, H.L., Yhang, N. & Woodward, W.A. (1989) On general fractional process, Journal of Time Series Analysis,10, pp. 233–257.

Gray, H.L., Yhang, N. & Woodward, W.A. (1994) On general fractional process: a correction, Journal of TimeSeries Analysis, 15, pp. 561–562.

Hotta, L.K. & Tsay, R.S. (1998) Outliers in GARCH process, unpublished manuscript, University of Chicago.Jarque C.M. & Bera, A.K. (1987) Test for normality of observations and regression residuals, International

Statistical Review, 55(2), pp. 163–172.Jorion, P. (1988) On jump processes in the foreign exchange and stock markets, Review of Financial Studies, 1,

pp. 427–445.

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

The Effects of US Macroeconomic Surprises 395

Love, R. & Payne, R.G. (2008) Macroeconomic news, order flow and exchange rates, Journal of Financial andQuantitative Analysis, 43(2), pp. 467–488.

Lyons, R.K. (2001) The Microstructure Approach to Exchange Rates (Cambridge, MA: MIT Press).Mark, N.C. (1995) Exchange rates and fundamentals: evidence on long horizon predictability, American

Economic Review, 85, pp. 201–218.McFarland, J.W., Pettit, R. & Sung, S.K. (1982) The distribution of foreign exchange price changes: trading day

effects and risk measurement, Journal of Finance, 37(3), pp. 693–715.Melvin, M. & Yin, X. (2000) Public information arrival, exchange rate volatility and quote frequency, Economic

Journal, 110(465), pp. 644–661.Müller, U.A., Dacorogna, M.M., Olsen, R.B., Pictet, O.V., Schworz, M. & Morgenegg, C. (1990) Statistical

study of foreign exchange rates, empirical evidence of price change law and intraday analysis, Journal ofBanking and Finance, 14(6), pp. 1189–1208.

Neely, C.J. (1999) Target zones and conditional volatility: the role of realignments, Journal of Empirical Finance,6(2), pp. 177–192.

Nieuwland, F., Verschoor, W. & Wolff, C.C.P. (1994) Stochastic trends and jumps in EMS exchange rates,Journal of International Money Finance, 13(6), pp. 699–727.

O’Hara, M. (1995) Market Microstructure (London: Blackwell).Press, J. (1967) A compound events model for security prices, Journal of Business, 40(3), pp. 317–335.Taylor, S.J. & Xu, X.X. (1998) The incremental volatility information in one million foreign exchange

quotations, Journal of Empirical Finance, 4, pp. 317–340.Vlaar, P.J.G. & Palm, F. (1993) The message in weekly exchange rates in the European Monetary System: mean

reversion, conditional heteroskedasticity and jumps, Journal of Business and Economic Statistics, 11(3),pp. 351–360.

Vlaar, P.J.G. & Palm, F. (1997) Interest rate differentials and excess returns in the European Monetary System,Journal of International Financial Markets, Institutions and Money, 7(1), pp. 1–20.

Appendix. The International Money Market Services (MMS) Data

1. Examples of MMS median expectations and actual realizations for five majorUS macroeconomic variables

Macroeconomic Median Date of Actual Date of Release timevariable expectation1 survey realization release (Eastern time)2

Durable goods −0.5 02-19-99 3.9 02-25-99 08:30am/10:00am3

GDP 4.10 06-18-99 4.30 06-25-99 08:30amPPI 0.10 02-12-99 0.5 02-18-99 08:30amNAPM 46.5 01-29-99 49.5 02-01-99 10:00amUnemployment 4.40 01-29-99 4.3 02-05-99 08:30amDurable goods −1.50 03-19-99 −5 03-24-99 08:30am/10:00amGDP 1.80 09-24-99 1.60 09-30-99 08:30amPPI −0.10 03-05-99 −0.4 03-12-99 08:30amNAPM 50 02-26-99 52.4 03-01-99 10:00amUnemployment 4.30 02-26-99 4.4 03-05-99 08:30am

Notes: 1. Median expectations data are conducted by telephone survey of about fortymoney managers, with collected forecasts of all indicators to be released during the nextweek and reported the median forecasts from the survey.2. The announcements of US macroeconomic variables are scheduled in advance.3. Whenever the Durable Goods and GDP variables are released on the same day, theDurable Goods are released at 10:00am.

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014

396 Y.W. Han

2. Basic descriptions of US macroeconomic surprises

Positive Surprises Negative Surprises

Macroeconomic Standard Standardvariable Mean deviation Mean deviation

Durable goods 64.99 73.66 −73.02 56.33GDP 87.67 70.65 −81.31 64.48PPI 112.93 57.83 −77.82 52.35NAPM 88.62 60.80 −68.57 50.63Unemployment 96.47 39.11 −110.95 47.45

Dow

nloa

ded

by [

Lin

naeu

s U

nive

rsity

] at

20:

33 0

9 O

ctob

er 2

014