Sterling currently exhibitssomething of a split personality.On the one hand, daily pricecharts of Sterling versus mostpopular currencies show amarked tendency to trending – acharacteristic that in many casesappears to coincide (approximately)with the introduction of the euro.On the other, Sterling’s intradaybehaviour is far less tractable,with a tendency to do virtuallynothing for extensive periods andthen abruptly take off in a violentdirectional move.

To some extent, this behaviour isdriven by the way many traders’risk profiles and time horizonshave changed over the past four

or five years. While traders mightpreviously have been happy totake and hold positions in Sterlingfor a month or more, this is nowfar less common. This is partlydue to a lower tolerance of volatilereturns, and one way to addressthis is to spend less time in themarket. As a result, many tradersnow prefer to ‘grab it and run’ inorder to minimise their exposure.

Ironically, this only serves toexaggerate Sterling’s little intradayfoibles, as an increasing numberof traders hover over theirkeyboards awaiting some form ofintraday breakout. If a significantpiece of news or a substantialflow appears, there is a frantic

scramble to climb aboard themove and benefit from what maybe the only P & L opportunity ofthe trading session.

Sterling – trading the intraday stop go

1

By Andy Webb

Jekyll and Hyde

Method oremotion?However, scrambling after theherd doesn’t guarantee that theinitial move out of the congestionarea won’t prove to be a falsebreakout that then develops intoanother sideways trading range.Some means of objectivelyquantifying the market’s

degree of congestion – andtherefore the likelihood of any breakout following throughinto a substantive move – isobviously desirable.

One possible solution is thecongestion count, which countsthe number of consecutive barswith price ranges that overlapthat of the current bar. In Figure 1the two red horizontal lines mark

the high and low of the currentbar – any consecutive precedingbars that intersect with theselines will be included in thecongestion count. The first, fifth and 10th bars included in the current congestion count of 10 have been marked (the congestion count alwaysincludes the current bar).

2

Figure 1

There are, in fact, two versions ofthe congestion count: the basic‘last’ version (see Figure 2bottom chart pane), which onlyrequires that the consecutivepreceding bars overlap thecurrent bar, and the ‘group’version, which requires thatthese bars must also overlapeach other in order to beincluded in the count. The basic

premise is that high congestioncount values often precede majorbreakouts. This certainly provesto be the case in Figure 2, wherea ‘last’ congestion count readingof 26 on a 60 minute Cable chart(highlighted by the vertical cursor)is followed by a 70 tick rally. (Togive an indication of the relativeimportance of a congestioncount reading of 26, the green

line overlaid on the congestioncount is its long term (1000 bar)moving average, which shows areading of only 6.78).

3

Figure 2

False premiseHowever, one carefully selectedchart on its own proves nothing –taken in isolation a highcongestion count only indicatesthat x number of bars areoverlapping. It is perfectlypossible to achieve a highcongestion count (particularlywith the ‘last’ version of the

indicator) simply because the last bar in a group has a widerange. For example, the 60minute EURGBP chart in Figure 3shows the ‘last’ version of thecongestion count suddenlyspiking to a value of 23 for justthis reason. It therefore has littlevalue as a trading signal, for, by the time the congestion countvalue is calculated (on the

close of the bar), the party isalready over.

4

Figure 3

The ‘group’ version of congestioncount goes some way towardsaddressing this issue. Figure 4shows the same data as Figure3, but with the ‘group’ version ofthe indicator. The requirementthat preceding bars must overlapeach other (as well as the currentone) in order to be included inthe count results in a reading ofonly 4, confirming that there was

in fact no appreciable build up ofcongestion. (‘Group’ congestioncount values are typically lowerthan those for the ‘last’ typecongestion count – for example,a ‘group’ congestion countreading above 15 in Cable wouldbe considered exceptional.)

5

Figure 4

Defining therange – 1One way of ensuring that a highcongestion count is not simplyflagging a price bar with a highrange is to filter possible entrysignals by specifying theacceptable size of the average truerange 1 of either a single bar orseries of bars.

Figure 5 shows a pair of short entrysignals on Cable. The rules forthese signals are:

1 The ‘last’ congestion count reading of the preceding bar must be 20 or greater

2 The 21 period average true range be less than 100 periodaverage true range

3 The current bar crosses belowthe low of the preceding bar minus 0.05% of the close of the preceding bar (the actual trade entry price is shown by the red horizontal dash overlaid on the entry bar.)

Figure 5

6

Although these are obviously tworather idealised examples, thebigger picture is also in positiveterritory, albeit by not a greatdeal. Figure 6 is the output tablefrom an evaluation of the shortentry signal outlined above,together with its long entrycounterpart. (Rules 1 and 2 forthe long entry are the same asfor the short signal, and rule 3 isthe reverse of rule 3 for the shortentry – ie. that the current barcrosses above the high of thepreceding bar plus 0.05% of the

close of the preceding bar.) Theentry signals have been testedon a portfolio of currency pairs(EURGBP, GBPCHF, GBPAUD,GBPCAD and GBPUSD) from 15 June 2002 to 20 June 2003using 60 minute price bars.

The top row of the table showsthe total gain in percentageterms (percentages are basedupon the entry price of eachtrade 2) at the close of each barout as far as six bars after entry.(For bars one to three,

performance as at the open ofeach bar is also shown.) It isimportant to note that these arenot complete trade systemresults, but only an assessmentof the quality of the long andshort entry signals, and howtrades triggered by those signalswould perform over the barsfollowing entry. The lower twothirds of the display show thefirst few of the 259 individualtrade entries that are included inthe summary statistics.

Figure 6

7

Figure 7 shows the relativecontribution of the variouscurrency pairs to the overallresults. Again, the picture is

reasonably encouraging with onlyone pair (EURGBP) showing aloss over the test period.

Figure 7

8

Defining therange – 2Though the basic signals havedelivered positive results (whichwere confirmed by the threepreceding years back as far asthe introduction of the euro) thereis still plenty of scope forimprovement. For example, thelow average result per trade (rowtwo of the table) is unlikely tocover much more than slippageand the bid/offer spread in a livetrading situation. One of theissues here is ensuring that the

trading range of the congestionperiod is tight enough to give anybreakout sufficient impetus. (Acommonly acceptedcharacteristic of successfulbreakouts is that they arepreceded by narrow, congestedtrading ranges.)

While requiring the short termaverage true range to be lessthan the long term average truerange (as in the previousexample) goes some waytowards addressing this, a betteralternative might be to use thecongestion count itself as an

input for precisely defining therange of the congestion period.For example, in Figure 8 thevertical cursor highlights acongestion count (‘groupversion’) value of 13, which has been used as the look backperiod for the maximum/minimumvalues on the chart (the maroonand blue/green lines boundingthe price bars). The next stepwould be to compare the size of the range of the congestionimmediately prior to the breakout with the size of previousranges that have the samecongestion count.

Figure 89

Figure 9

Figures 9 and 10 show the tradesignal evaluation results for sucha rule set based on the samedata as the first set of entrysignals. The additionalrequirement is that the size of thetrading range on the bar prior toentry has to be smaller than thelong-term average of all rangeswith the same congestion countvalue. As can be seen, the trade

count has dropped by more than50%, but there has also been adramatic improvement inprofitability – for example, theaverage performance of allentries after 10 bars is +0.16%,which for GBPUSD would equateto approximately 27 pips atcurrent levels. At first glance thismay not seem particularlyscintillating, but it should be

borne in mind that these resultsare just for raw entry signals –eg. there are no stop losses toweed out poor entry signalsalong the way. Figure 10 alsoreveals that although GBPCADwas by some way the strongestperformer, the other fourcurrency pairs also made solid contributions.

10

Figure 10

11

Figure 11

Closing the loopFigure 11 takes things a stepfurther by providing an indicationof congestion count potential asa tool for trading intraday Sterlingbreakouts. The long and shortentry signals shown in Figure 9have been used as the basis fora simple trading system andapplied to GBPAUD. (Theweakest market in the entry

signal evaluation performedabove.) Only two exit signalshave been used – a fixedpercentage stop loss and a fixedpercentage profit target, withneither of the exit percentagesbeing optimised. The results arebased on a lot size of £1m pertrade, so the AUD185300 profitequates to an annual return(before slippage and executioncosts) of approximately 7.6% atcurrent exchange rates. Other

trade statistics are alsoencouraging, such as a removeto neutral 3 of 19.23% and61.54% winning trades. Althoughthe total trade count of 26 is low,the system also proved profitablein the preceding years back tothe introduction of the euro (albeitonly marginally in 2001–2002)and has been particularly strongsince the end of the test periodshown below.

12

Footnotes1 Average true range is the moving average of true range, which is the greatest of:

• High for the price bar minus the low for the price bar

• High for the price bar less the close for the previous price bar

• Close for the previous price bar and the low for the current price bar

2 So for example, the first individual trade signal listed (on GBPUSD) that occurred on 17 June 2002 shows a loss of 0.16%

by the close of the third bar after entry. In this case the entry price was 1.4783 (not shown) so the trade would have been

underwater by approximately 24 pips by the close of the third bar. (By contrast, the third individual entry listed – GBPCAD,

dated 18 June 2002 – was showing a gain of 0.64% by the close of the sixth bar on an entry price of 2.2908 = +147 pips.)

3 Percentage remove to neutral is the percentage of winning trades that would have to be excluded from a trade system’s

results to reduce its total net profit to zero. A percentage remove to neutral of greater than 10 suggests that a trading

system is reasonably robust.

The views expressed in this paper are those of the author and not necessarily those of EBS.

Andy Webb:

Andy Webb is a freelance writer with 20+ years' experience, most recently specialising inderivatives, technology and trading methodologies. He has written regularly for a wide rangeof journals – including The Sunday Times, Sunday Business, Treasury & Risk Management,Global Finance, Derivatives Strategy and Wall Street & Technology. He is also activelyinvolved in developing and programming FX trading models for several hedge funds andproprietary trading desks.

ANDY WEBB

Practical realityWhile measuring the length andrange of congestion periods mayprove a useful method for copingwith Sterling’s intraday vagaries,its success is obviouslydependent upon whether there isa tight/stable enough bid/offerspread to actually execute tradesclose to the level indicated by the

entry signals. The mostvulnerable moments in thisrespect are during the transitionperiods between one time zone’strading session and the next.(Eleven of the 115 raw tradingsignals shown in Figure 9occurred during these periods.)Fortunately the situation here isimproving rapidly, with EBS’‘focus hours’ already having a

favourable impact – in the case of EURGBP pushing thebid/offer spread down to 2-3pips between 07.00 and 09.00London time. Under thosecircumstances, a congestionbreakout strategy with a modestpercentage profit target becomes viable.

February 2004

13