News
Macro news announcements typically generate a quick surge in currency trading volume and volatility. As shown in Figures 2A and 2B, which are taken from Chaboud et al. (2004), volume initially surges within the first minute by an order of magnitude or more. Dealers assert that the bulk of the exchange-rate response to news is often complete within ten seconds (Cheung and Chinn 2001).
Carlson and Lo (2006) closely examine one macro announcement the timing of which was unanticipated. They show that in the first half-minute spreads widened and in the second half-minute trading surged and the price moved rapidly. Chaboud et al. (2004) shows that after the first minute volume drops back substantially but not completely in the next few minutes. The remaining extra volume then disappears slowly over the next hour. The response of returns to news is particularly intense after a period of high volatility or a series of big news surprises (Ehrmann and Fratzscher 2005, Dominguez and Panthaki 2006, 2007), conditions typically interpreted as heightened uncertainty.
The U.S. macro statistical releases of greatest importance are the GDP, the unemployment rate, payroll employment, initial unemployment claims, durable goods orders, retail sales, the NAPM index, consumer confidence, and the trade balance (Anderson et al. 2003). Strikingly, money supply releases have little or no effect on exchange rates (Cai et al. 2001, Anderson et al. 2003, Cheung and Chinn 2001, Evans and Lyons 2008), consistent with the observation above that aggregate money supply and demand seem irrelevant for short run exchange-rate dynamics.
Statistical releases bring a home-currency appreciation when they imply a strong home economy. A positive one-standard deviation surprise to U.S. employment, which is released quite soon after the actual employment is realized, appreciates the dollar by (0.98 percent). For GDP, which is released with a greater lag, a positive one-standard deviation surprise tends to appreciate the dollar by 0.54 percent (Andersen et al. 2002). Responses are driven by associated anticipations of monetary policy: anything that implies a stronger economy or higher inflation leads investors to expect higher short-term interest rates (Chaboud et al. 2006) and thus triggers a dollar appreciation and vice versa.
Federal Reserve announcements following FOMC meetings do not typically elicit sharp increases in trading volume and volatility (Chaboud et al. 2006). Instead, FOMC announcements bring only a small rise in trading volume (Figure 2C) and tend to reduce exchange-rate volatility (Chang and Taylor 2003). This suggests that Federal Reserve policy shifts are generally anticipated, which is encouraging since that institution prefers not to surprise markets. 16
Unanticipated changes in monetary policy do affect exchange rates. Fratscher (2007) finds that an unanticipated 25 basis-point rise in U.S. interest rates tends to appreciate the dollar by 4.2 percent. Kearns and Manners (2005), who analyze other Anglophone countries, find that a surprise 25 basis-point interest-rate rise tends to appreciate the home currency by only 38 basis points. Kearns and Manners also note a more subtle dimension of response: If the policy shift is expected merely to accelerate an already-anticipated interest-rate hike, the exchange-rate effect is smaller (only 23 basis points, on average) than if the shift is expected to bring consistently higher interest rates over the next few months (43 basis points on average).
Evidence presented in Evans and Lyons (2005) suggests that exchange rates overshoot in responses to news announcements. For some types of news between a tenth and a quarter of the initial response is typically reversed over the four consecutive days. The reversals are most pronounced for U.S. unemployment claims and the U.S. trade balance. This contrasts strikingly with the well-documented tendency for initial the stock-price response to earnings announcements to be amplified after the first day, a phenomenon known as “post-earnings announcement drift” (Kothari (2001) provides a survey). Nonetheless, over-reaction to fundamentals has been documented repeatedly for other financial assets (Shiller 1981; Campbell and Shiller 1988; Barberis and Thaler 2002).
Exchange-rate responses to a given macro news statistic can vary over time, as dealers are well aware (Cheung and Chinn 2001). During the early 1980s, for example, the dollar responded fairly strongly to money supply announcements but, as noted above, this is no longer the case. This shift appears to have been rational since it reflected public changes in Federal Reserve behavior: in the early 1980s the Fed claimed to be targeting money supply growth, a policy it has since dropped. The possibility that such shifts are not entirely rational is explored in Bachetta and van Wincoop (2004). Cheung and Chinn (2001) provide further discussion of how and why the market’s focus shifts over time. Using daily data, Evans and Lyons (2005) find little evidence of such shifting during the period 1993-1999. This could reflect the masking of such effects in their daily data or it could indicate that such shifting was modest during those years of consistent economic expansion and consistent monetary policy structure.
Information relevant to exchange rates comes from many more sources than macroeconomic statistical releases. Trading volume and volatility are triggered by official statements, changes in staffing for key government positions, news that demand for barrier options is rising or falling, reports of stop-loss trading, even rumors (Oberlechner 2004). Jansen and De Hahn (2005) show that ECB statements affect conditional volatility but not the level of exchange rates. As documented in Dominguez and Panthaki, much of the news that affects the market is non-fundamental.
Numerous asymmetries have been documented in these responses to news. The effects of U.S. macro announcements tend to be larger than the effect of non-U.S. news (Goodhart et al. 1993, Evans and Lyons 2005). Ehrmann and Fratzscher (2005) attrtibute this asymmetry, at least in part, to the tendency for non-U.S. macroeconomic statistical figures to be released at unscheduled times and with a greater lag. Ehrmann and Fratzscher also shows that exchange rates respond more to weak than strong European news, and Andersen et al. (2002) report a similar pattern with respect to U.S. announcements. The source of such asymmetries is not well understood.
Carlson and Lo (2006) show that many interdealer limit orders are not withdrawn upon the advent of unexpected macro news. This might seem surprising, since by leaving the orders dealers seem to expose themselves to picking-off risk. It may not be the dealers themselves,
17
however, that are thus exposed. The limit orders left in place may be intended to cover take-profit orders placed by customers, so the customer may be the one exposed to risk.
To be concrete: suppose a customer places a take-profit order to buy 5 at 140.50 when the market is at 140.60. The dealer can ensure that he fills the order at exactly the requested price by placing a limit order to buy 5 at 140.50 in the interdealer market. Suppose news is then released implying that the exchange rate should be 140.30. The dealer loses nothing by leaving the limit order in place: the customer still gets filled at the requested rate of 140.50.
This interpretation may appear to push the mystery back one step, because now the customer is buying currency at 140.50 when the market price of 140.30 would be more advantageous. Why wouldn’t customers change their orders upon the news release, or withdraw them beforehand? This could reflect a rational response of customers to the high costs of monitoring the market intraday. Indeed, as noted earlier it is to avoid those costs that customers place orders in the first place. The customers, choosing not to monitor the market, may not even be aware of the news.
D. Intervention
Numerous studies have examined the intraday effects of official (sterilized) foreign exchange intervention. This is motivated in part by inherent limitations in the study of intervention at longer horizons: since intervention may be precipitated by intraday market developments, simultaneity may bias the analysis at daily or lower horizons (Vitale 2007). It is also intriguing to examine intraday effects since markets clearly take intervention reports seriously (see below), even though the effects of intervention at longer horizons appear to be limited (Sarno and Taylor 2001).
Studies of intervention using intraday data are often hampered by the inability to pinpoint the exact time of intervention. Such times are only revealed by the Swiss, Canadian, and Danish central banks. When exact intervention times are not available the “event” of intervention is typically identified with reference to Reuters news reports. Indirect evidence has suggested that these appear to be released with a (presumably time-varying) delay of 30 minutes to one hour (see, for example, Chang and Taylor 1998; Dominguez 2006). However, a direct comparison of the Reuters reports of Swiss intervention with actual intervention times shows that “the standard deviation of the prediction error can be measured in hours and not in minutes and there is evidence of Reuters intervention reporting before the SNB (Swiss National Bank) intervention occurs” (Fisher 2006, p. 1239). Fisher (2006) implicitly questions whether newsier reports of intervention provide a reliable basis for evaluating the intraday effects of intervention.
This section briefly reviews basic results from this literature: additional results are discussed in later sections. Studies of Swiss, Canadian, and Danish intervention show that intervention affects exchange rates as one would expect: purchases (sales) of a foreign currency (typically dollars or euros) bring a stronger (weaker) currency (see Fischer and Zurlinden (1999) for Switzerland, Fatum and King (2005) for Canada, and Fatum and Pedersen (2007) for Denmark). The effect is immediate in the Swiss and Canadian data but takes roughly 30 minutes to reach full strength in the Danish data. The effect persists for at least a few hours in the Swiss data (Payne and Vitale 2003). However, the effect may not be permanent and, if it is inconsistent with the direction of monetary policy, it is economically negligible even if it does last beyond a few hours (Fatum and Pedersen 2007). In the Swiss case rates begin to move in the quarter hour prior to the intervention (Payne and Vitale 2003). This indicates that intervention can be anticipated by the market which presumably is related to the tendency of intervention events to
18
be clustered. The ability to predict intervention could also reflect the fact that it is often precipitated by market developments.
The Swiss intervention data indicate that intervention has a stronger effect when it is coordinated with other central banks, consistent with studies using data from other countries and daily (or longer) time horizons (Dominguez 2003, Vitale 2007). The Swiss data differ from studies of Japanese intervention, however, when they indicate that intervention is more effective when it reinforces an existing trend, rather than attempting to counter a trend (Ito 2007; Payne and Vitale 2003). The Danish data indicate that intervention is only effective when it is consistent with current monetary policy fundamentals (Fatum and Pedersen 2007).
Both Fatum and Pedersen (2007) and Dominguez (2003) find that the effects are strongest during high volatility. D’Souza (2002) finds that the effects of interventions are similar to the effects of other customer trades. Dominguez finds, using Reuters intervention reports for non-Swiss authorities, that the intraday effects of intervention are strongest when the intervention is closely timed (within two hours) with macro announcements.
The intraday data tend to indicate that intervention is accompanied by a quick but brief boost to volatility, though some studies find the contrary result. Studies finding a negative association between volatility and intervention include Beattie and Fillion (1999) and Fatum and King (2005), both of which study Canadian intervention. Studies that find a positive association include Chang and Taylor (1998), which finds the effect most clearly using five- and ten-minute data, and Dominguez (2003), which finds that the peak effect lasts only a quarter hour and the overall effect disappears within an hour. Given a volatility spike, standard theories (e.g., Ho and Stoll 1981) predict that intervention would also bring higher bid-ask spreads in the interdealer market, a hypothesis that gains qualified support from the evidence. Naranjo and Nimalendran (2000) provide evidence for this hypothesis based on daily intervention data, and Chari (2007)provides support on news reports of intervention.
Macro news announcements typically generate a quick surge in currency trading volume and volatility. As shown in Figures 2A and 2B, which are taken from Chaboud et al. (2004), volume initially surges within the first minute by an order of magnitude or more. Dealers assert that the bulk of the exchange-rate response to news is often complete within ten seconds (Cheung and Chinn 2001).
Carlson and Lo (2006) closely examine one macro announcement the timing of which was unanticipated. They show that in the first half-minute spreads widened and in the second half-minute trading surged and the price moved rapidly. Chaboud et al. (2004) shows that after the first minute volume drops back substantially but not completely in the next few minutes. The remaining extra volume then disappears slowly over the next hour. The response of returns to news is particularly intense after a period of high volatility or a series of big news surprises (Ehrmann and Fratzscher 2005, Dominguez and Panthaki 2006, 2007), conditions typically interpreted as heightened uncertainty.
The U.S. macro statistical releases of greatest importance are the GDP, the unemployment rate, payroll employment, initial unemployment claims, durable goods orders, retail sales, the NAPM index, consumer confidence, and the trade balance (Anderson et al. 2003). Strikingly, money supply releases have little or no effect on exchange rates (Cai et al. 2001, Anderson et al. 2003, Cheung and Chinn 2001, Evans and Lyons 2008), consistent with the observation above that aggregate money supply and demand seem irrelevant for short run exchange-rate dynamics.
Statistical releases bring a home-currency appreciation when they imply a strong home economy. A positive one-standard deviation surprise to U.S. employment, which is released quite soon after the actual employment is realized, appreciates the dollar by (0.98 percent). For GDP, which is released with a greater lag, a positive one-standard deviation surprise tends to appreciate the dollar by 0.54 percent (Andersen et al. 2002). Responses are driven by associated anticipations of monetary policy: anything that implies a stronger economy or higher inflation leads investors to expect higher short-term interest rates (Chaboud et al. 2006) and thus triggers a dollar appreciation and vice versa.
Federal Reserve announcements following FOMC meetings do not typically elicit sharp increases in trading volume and volatility (Chaboud et al. 2006). Instead, FOMC announcements bring only a small rise in trading volume (Figure 2C) and tend to reduce exchange-rate volatility (Chang and Taylor 2003). This suggests that Federal Reserve policy shifts are generally anticipated, which is encouraging since that institution prefers not to surprise markets. 16
Unanticipated changes in monetary policy do affect exchange rates. Fratscher (2007) finds that an unanticipated 25 basis-point rise in U.S. interest rates tends to appreciate the dollar by 4.2 percent. Kearns and Manners (2005), who analyze other Anglophone countries, find that a surprise 25 basis-point interest-rate rise tends to appreciate the home currency by only 38 basis points. Kearns and Manners also note a more subtle dimension of response: If the policy shift is expected merely to accelerate an already-anticipated interest-rate hike, the exchange-rate effect is smaller (only 23 basis points, on average) than if the shift is expected to bring consistently higher interest rates over the next few months (43 basis points on average).
Evidence presented in Evans and Lyons (2005) suggests that exchange rates overshoot in responses to news announcements. For some types of news between a tenth and a quarter of the initial response is typically reversed over the four consecutive days. The reversals are most pronounced for U.S. unemployment claims and the U.S. trade balance. This contrasts strikingly with the well-documented tendency for initial the stock-price response to earnings announcements to be amplified after the first day, a phenomenon known as “post-earnings announcement drift” (Kothari (2001) provides a survey). Nonetheless, over-reaction to fundamentals has been documented repeatedly for other financial assets (Shiller 1981; Campbell and Shiller 1988; Barberis and Thaler 2002).
Exchange-rate responses to a given macro news statistic can vary over time, as dealers are well aware (Cheung and Chinn 2001). During the early 1980s, for example, the dollar responded fairly strongly to money supply announcements but, as noted above, this is no longer the case. This shift appears to have been rational since it reflected public changes in Federal Reserve behavior: in the early 1980s the Fed claimed to be targeting money supply growth, a policy it has since dropped. The possibility that such shifts are not entirely rational is explored in Bachetta and van Wincoop (2004). Cheung and Chinn (2001) provide further discussion of how and why the market’s focus shifts over time. Using daily data, Evans and Lyons (2005) find little evidence of such shifting during the period 1993-1999. This could reflect the masking of such effects in their daily data or it could indicate that such shifting was modest during those years of consistent economic expansion and consistent monetary policy structure.
Information relevant to exchange rates comes from many more sources than macroeconomic statistical releases. Trading volume and volatility are triggered by official statements, changes in staffing for key government positions, news that demand for barrier options is rising or falling, reports of stop-loss trading, even rumors (Oberlechner 2004). Jansen and De Hahn (2005) show that ECB statements affect conditional volatility but not the level of exchange rates. As documented in Dominguez and Panthaki, much of the news that affects the market is non-fundamental.
Numerous asymmetries have been documented in these responses to news. The effects of U.S. macro announcements tend to be larger than the effect of non-U.S. news (Goodhart et al. 1993, Evans and Lyons 2005). Ehrmann and Fratzscher (2005) attrtibute this asymmetry, at least in part, to the tendency for non-U.S. macroeconomic statistical figures to be released at unscheduled times and with a greater lag. Ehrmann and Fratzscher also shows that exchange rates respond more to weak than strong European news, and Andersen et al. (2002) report a similar pattern with respect to U.S. announcements. The source of such asymmetries is not well understood.
Carlson and Lo (2006) show that many interdealer limit orders are not withdrawn upon the advent of unexpected macro news. This might seem surprising, since by leaving the orders dealers seem to expose themselves to picking-off risk. It may not be the dealers themselves,
17
however, that are thus exposed. The limit orders left in place may be intended to cover take-profit orders placed by customers, so the customer may be the one exposed to risk.
To be concrete: suppose a customer places a take-profit order to buy 5 at 140.50 when the market is at 140.60. The dealer can ensure that he fills the order at exactly the requested price by placing a limit order to buy 5 at 140.50 in the interdealer market. Suppose news is then released implying that the exchange rate should be 140.30. The dealer loses nothing by leaving the limit order in place: the customer still gets filled at the requested rate of 140.50.
This interpretation may appear to push the mystery back one step, because now the customer is buying currency at 140.50 when the market price of 140.30 would be more advantageous. Why wouldn’t customers change their orders upon the news release, or withdraw them beforehand? This could reflect a rational response of customers to the high costs of monitoring the market intraday. Indeed, as noted earlier it is to avoid those costs that customers place orders in the first place. The customers, choosing not to monitor the market, may not even be aware of the news.
D. Intervention
Numerous studies have examined the intraday effects of official (sterilized) foreign exchange intervention. This is motivated in part by inherent limitations in the study of intervention at longer horizons: since intervention may be precipitated by intraday market developments, simultaneity may bias the analysis at daily or lower horizons (Vitale 2007). It is also intriguing to examine intraday effects since markets clearly take intervention reports seriously (see below), even though the effects of intervention at longer horizons appear to be limited (Sarno and Taylor 2001).
Studies of intervention using intraday data are often hampered by the inability to pinpoint the exact time of intervention. Such times are only revealed by the Swiss, Canadian, and Danish central banks. When exact intervention times are not available the “event” of intervention is typically identified with reference to Reuters news reports. Indirect evidence has suggested that these appear to be released with a (presumably time-varying) delay of 30 minutes to one hour (see, for example, Chang and Taylor 1998; Dominguez 2006). However, a direct comparison of the Reuters reports of Swiss intervention with actual intervention times shows that “the standard deviation of the prediction error can be measured in hours and not in minutes and there is evidence of Reuters intervention reporting before the SNB (Swiss National Bank) intervention occurs” (Fisher 2006, p. 1239). Fisher (2006) implicitly questions whether newsier reports of intervention provide a reliable basis for evaluating the intraday effects of intervention.
This section briefly reviews basic results from this literature: additional results are discussed in later sections. Studies of Swiss, Canadian, and Danish intervention show that intervention affects exchange rates as one would expect: purchases (sales) of a foreign currency (typically dollars or euros) bring a stronger (weaker) currency (see Fischer and Zurlinden (1999) for Switzerland, Fatum and King (2005) for Canada, and Fatum and Pedersen (2007) for Denmark). The effect is immediate in the Swiss and Canadian data but takes roughly 30 minutes to reach full strength in the Danish data. The effect persists for at least a few hours in the Swiss data (Payne and Vitale 2003). However, the effect may not be permanent and, if it is inconsistent with the direction of monetary policy, it is economically negligible even if it does last beyond a few hours (Fatum and Pedersen 2007). In the Swiss case rates begin to move in the quarter hour prior to the intervention (Payne and Vitale 2003). This indicates that intervention can be anticipated by the market which presumably is related to the tendency of intervention events to
18
be clustered. The ability to predict intervention could also reflect the fact that it is often precipitated by market developments.
The Swiss intervention data indicate that intervention has a stronger effect when it is coordinated with other central banks, consistent with studies using data from other countries and daily (or longer) time horizons (Dominguez 2003, Vitale 2007). The Swiss data differ from studies of Japanese intervention, however, when they indicate that intervention is more effective when it reinforces an existing trend, rather than attempting to counter a trend (Ito 2007; Payne and Vitale 2003). The Danish data indicate that intervention is only effective when it is consistent with current monetary policy fundamentals (Fatum and Pedersen 2007).
Both Fatum and Pedersen (2007) and Dominguez (2003) find that the effects are strongest during high volatility. D’Souza (2002) finds that the effects of interventions are similar to the effects of other customer trades. Dominguez finds, using Reuters intervention reports for non-Swiss authorities, that the intraday effects of intervention are strongest when the intervention is closely timed (within two hours) with macro announcements.
The intraday data tend to indicate that intervention is accompanied by a quick but brief boost to volatility, though some studies find the contrary result. Studies finding a negative association between volatility and intervention include Beattie and Fillion (1999) and Fatum and King (2005), both of which study Canadian intervention. Studies that find a positive association include Chang and Taylor (1998), which finds the effect most clearly using five- and ten-minute data, and Dominguez (2003), which finds that the peak effect lasts only a quarter hour and the overall effect disappears within an hour. Given a volatility spike, standard theories (e.g., Ho and Stoll 1981) predict that intervention would also bring higher bid-ask spreads in the interdealer market, a hypothesis that gains qualified support from the evidence. Naranjo and Nimalendran (2000) provide evidence for this hypothesis based on daily intervention data, and Chari (2007)provides support on news reports of intervention.
AVT INVENIAM VIAM AVT FACIAM