0.80

-0.20

0.20

0.00

0.60

0.401-

-0.40

i0 400 600 800

Number of Days since January 5,1993

1000

FIGURE 5.4 Daily difference between the last bid quote before 2 a.m. GMT and the last bid quote before 8 p.m. GMT. The early quote is systematically higher than the later one. Data sample: Quotes of the USD 3-month interbank money market rates published by Telerate. The analysis runs from January 5, 1993, to January 31, 1995.

0.5%). As explained in Section 2.3.1, the segmentation even caused negative JPY interest rates in the European and American markets.

5.3.2 Positive Impact of Official Interventions

One special type of trader is the central banks, as the time and the size of their interventions can be measured on an intradaily basis. Central banks may operate either directly through officially announced interventions or indirectly through unannounced interventions. Official interventions operate essentially as signals given to the markets and are therefore difficult to measure, see Edison (1993) for a review of the literature on central bank interventions. Some evidence is given in Goodhart and Hesse (1993) of the positive effects in the long run of official interventions, although they may result in short-term losses for the central bankers. One could, however, easily extend the analysis to any other long-term trader. A trader who can afford to keep a large open position for a long time will have some impact on the market through his reputation, even if he doesnt have a large share of the market. This is the case of some hedging funds, for example. Peiers (1997) shows the positive impact of unannounced interventions and interventions of a central bank, the Bundesbank, through the biggest player on the market, namely the Deutsche Bank.

5.3.3 Mixed Effect of News

News is a very broad concept covering a phone call of a customer who wants to make a large FX transaction (due to inventory imbalances, for instance), a conversation with a colleague, price forecasts and histories when used in technical analysis programs or the economic forecasts of the research department of a bank, general economic and political news, and major economic news announcements.

Intraday Interval Index (in hours GMT)

FIGURE 5.5 Intraday distribution of 15-min mean changes for the absolute returns (Equation 5.1 for the USD-DEM). Sudden peaks are darkened. The values are average: over all weeks of a sampling period of many years.

News is therefore difficult to quantify. Goodhart (1989) first tried to quantir news by looking at the "news" pages of Reuters. General economic and poh-ical news was displayed on the AAMM page (until February 1997). Goodhar (1989) found that "small" news does not have a significant effect on the behavior of the foreign exchange rates. Distinct and relatively large price movement: unrelated to any news are indeed apparent. The price formation process seem, to prevail, notwithstanding the presence or absence of news. In contrast, maic. economic news announcements such as trade, unemployment, budget deficit, gross domestic product growth have significant impact (Goodhart, 1989). Economic news announcements along with the market expectations and the effect the previous announcement were displayed in Reuters FXNB page. Effective news-that is, the difference between the market expectation and the actual figure that is released-increases the volatility as the dispersion of traders views on tkz impact of the effective news widens.

In Figure 5.5 we study the systematic effect of news release over a full da*. We plot as a function of daytime (in GMT) a quantity that reflects the chanp. of volatility by examining its relation to the neighboring values. This quantity i: defined as

hi = exp lnr, - -lnr,- ir,-+1

(5.1

where the index i represents the time of day (in steps of 15-min), and the / , are averaged over all working days of a long sampling period. The three right peaks in Figure 5.5 show the clear-cut effect of news release in New York and Japan. News is not released every working day, but when it is, this happens at the typical daytimes indicated by peaks. The two peaks for the United States are separated by 1 hr and reflect the change of daylight saving time, which does not exist in Japan. The first two peaks on the left correspond to the beginning of the Japanese trading session and to the time just after the Japanese lunch. Goodhart et al. (1993) further show that major economic news announcements, such as release of the U.S. trade figures or changes in the U.K. base interest rate, have a significant impact on the return process. This effect however extends over 3 or 4 days as markets eventually incorporate the effects of the news. Moreover, the direction of the effect on the level of the price is difficult to predict. This can be explained by the highly nonlinear dynamics of the FX rates (Guillaume, 1994).

An alternative way to quantify the impact of news is with the mixture of distribution hypothesis (Clark, 1973; Tauchen and Pitts, 1983; Andersen, 1996). In this framework, the clustering of the volatility results from the clustering of the news arrival process. Because the news arrival process is an unobserved variable, proxies for the market activity such as the volume of trade are used (volume is not available in the FX markets). Moreover, as shown in Jones et al. (1994), volume can be rather noisy. Therefore, empirical studies in the FX intradaily markets use the tick frequency or the spread as proxies for the level of activity. Although a certain correlation between these variables and the volatility is obvious from the simple inspection of Figure 5.12, severe limitations harm the use of these variable as noted earlier. Moreover, Dave (1993) shows that tick frequency can only be a good approximation of the volume when markets are analyzed as separate geographical entities. Thus, there is no overlap between markets and the data are not disaggregated by individual bank subsidiary. Goodhart (1989) also shows that tick frequency does not specifically rise when news is released. Therefore, empirical evidence in favor of this mixture of distribution hypothesis is only partial, (Demos and Goodhart, 1992; Bollerslev and Domowitz, 1993). A more recent paper by Melvin and Yin (2000) provides new support for the link between news arrival frequency and quote frequency.

In another study, Almeida et al. (1998) are able to quantify the effect of news to a short-lived response of 15 min on average, confirming the results of Figure 5.5. The peaks of that figure disappear if longer time intervals are examined. This is also confirmed in a study by Franke and Hess (1998) on other very liquid markets such as the U.S. treasury bond market and the German Bund futures market. By studying the effect of scheduled U.S. macroeconomic news releases, these authors were able to detect an increase of volatility of the U.S. Treasury bond futures contracts. This anomalous volatility would last from few minutes to a maximum of an hour. Moreover, they show that the futures Bund price reacts significantly to an American news announcements. They attribute this reaction to the increasing integration of the German bond market. It is only with the use of more sophisticated