Market Technician No40

web site: www.sta-uk.org The Journal of the STA

1

At the time of going to press, market sentiment on bothsides of the Atlantic is gloomy. Merger and acquisitionactivity has dropped off dramatically and the equity marketshave come cascading down. No one is sure exactly whenand where the bottom will be but it is interesting to notethat, in the face of this uncertainty, even economists arereferring to support levels.

Given the difficult market conditions, Nick Glydon and NicolaMerrell of JP Morgan are to be congratulated for coming topin Reuters’s survey of technical analysts. HSBC were secondand Collins Stewart came third. Nick Glydon pithily summedup the current situation as “a trend is a trend until it comes toa bend – wait for the bend”. The JP Morgan team’sassessment of some of the major markets is given on page 18.

For all technical analysts – no matter what system they use –data are the basic building blocks. The STA actively lobbiesthe market data industry to make improvements to dataquality. It would, however, be useful to have the views of themembership on this issue and we are therefore currentlyconducting a survey on data quality. Feedback is critical andwe would urge all members (and non-members) to take afew minutes to complete the short questionnaire which is

available on our website (www.sta-uk.org). To encourageparticipation, we are giving away a case of champagne to arespondent, picked at random, at the end of the surveyperiod. We also hope to run an article on data cleaning in thenext issue of the Journal.

There has been a very encouraging response to David Watts’software survey in the last issue. In this sort of exercise thereare inevitably some companies which inadvertently get leftout. TraderMade was one such example and we would like toapologise to them for this. A review of their services will beincluded in the next issue. If anyone has spotted any otheromissions then please let either David Watts or DeborahOwen know. We hope to repeat this survey annually andwant to make it as comprehensive as possible.

COPY DEADLINE FOR THE NEXT ISSUEMAY 31ST 2001

PUBLICATION OF THE NEXT ISSUEJULY 2001

April 2001 ISSUE No. 40

MARKET TECHNICIAN

FOR YOUR DIARY

IN THIS ISSUE

Monday 9th April 2001 Exam Revision Day Southbank University

Wednesday 11th April 2001 ‘Channels and Cycles’ Dr. Brian Millard, Qudos Publications Ltd

Friday 27th April 2001 STA Diploma Examination Southbank University

Wednesday 9th May 2001 Monthly Meeting

Wednesday 13th June 2001 Monthly Meeting

N.B. The monthly meetings will take place at the Institute of Marine Engineers, 80 Coleman Street, London EC2 at 6.00 p.m.

D. Watts Software review 3

N. Elliott &Y. Harada Ichimoku kinko hyo charting 4

M. Wignall Open interest, volume and priceanalysis: the special case of markets with inter-exchange mutual offsetting systems. 7

C. Osler Support for resistance:technical analysis and intradayexchange rates 10

JP Morgan The TMT massacre

Goldman Sachs IMM positions and currency trends 19

WHO TO CONTACT ON YOUR COMMITTEE

CHAIRMANAdam Sorab, Deutsche Bank Asset Management, 1 Appold Street, London EC2A 2UU

TREASURERVic Woodhouse. Tel: 020-8810 4500

PROGRAMME ORGANISATIONMark Tennyson d’Eyncourt. Tel: 020-8995 5998 (eves)

LIBRARY AND LIAISONMichael Feeny. Tel: 020-7786 1322The Barbican Library contains our collection. Michael buysnew books for it where appropriate, any suggestions for newbooks should be made to him.

EDUCATIONJohn Cameron. Tel: 01981-510210Clive Hale. Tel: 01628-471911George MacLean. Tel: 020-7312 7000

EXTERNAL RELATIONSAxel Rudolph. Tel: 020-7842 9494

IFTAAnne Whitby. Tel: 020-7636 6533

MARKETINGSimon Warren. Tel: 020-7656 2212Kevan Conlon. Tel: 020-7329 6333Tom Nagle. Tel: 020-7337 3787

MEMBERSHIPSimon Warren. Tel: 020-7656 2212Gerry Celaya. Tel: 020-7730 5316Barry Tarr. Tel: 020-7522 3626

REGIONAL CHAPTERSRobert Newgrosh. Tel: 0161-428 1069Murray Gunn. Tel: 0131-245 7885

SECRETARYMark Tennyson d’Eyncourt. Tel: 020-8995 5998 (eves)

STA JOURNALEditor, Deborah Owen, 108 Barnsbury Road, London N1 0ES

Please keep the articles coming in – the success of the Journaldepends on its authors, and we would like to thank all thosewho have supported us with their high standard of work. Theaim is to make the Journal a valuable showcase for members’research – as well as to inform and entertain readers.

The Society is not responsible for any material published inThe Market Technician and publication of any material orexpression of opinions does not necessarily imply that theSociety agrees with them. The Society is not authorised toconduct investment business and does not provideinvestment advice or recommendations.

Articles are published without responsibility on the part of theSociety, the editor or authors for loss occasioned by anyperson acting or refraining from action as a result of any viewexpressed therein.

11th April

9th May

13th June

4th July

12th September

10th October

14th November

5th December

* * * *

Speakers’ Panel

The STA is regularly approached by the financial medialooking for technical analysts who are prepared tocomment on the markets.

Anyone seeking to raise their profile in the marketsshould complete the questionnaire on our website.

* * * *

IFTA Conference, October 2002

Plans for the IFTA Conference in London next year areproceeding. We would like to put together an interestingand varied programme for the three days. If anyonehas any suggestions for speakers would they pleasecontact Anne Whitby.

2 MARKET TECHNICIAN Issue 40 – April 2001

Networking Dates for STA Monthly Meetings

ANY QUERIES

For any queries about joining the Society, attendingone of the STA courses on technical analysis ortaking the diploma examination, please contact:

STA Administration Services(Katie Abberton)

P.O. Box 2, Skipton BD23 6YH. Tel: 07000 710207 Fax: 07000 710208

www.sta-uk.org

For information about advertising in the journal,please contact:

Deborah Owen

108 Barnsbury Road, London N1 OES. Tel/Fax: 020-7278 7220

Issue 40 – April 2001 MARKET TECHNICIAN 3

Nature’s heartbeat starts to beat again.

Since Y2K, software users have faced numerous issues not least isthat many programs have been “retired”, as they are now no longerY2K compatible. So it is good to hear that one such time analysisprogram – Natures Pulse – has sprung back to life, no doubt due todemand by existing users. Its new owners Pulse Technologies havereleased version 4.1 with Y2K compatibility. Details are available viatheir web site www.Pulsetech.org.

* * * *Tradestation Pro.

For System junkies the roll out of the new web based TradestationPro continues.

Tradestation Pro (now under the ownership of TradestationTechnologies, formerly Omega Research) offers live streaming datainto Tradestation to trade or test your trading system. With bettersystem reports and the soon-to-be released Intelligent Direct-AccessOrder-Routing and online execution of trades, Tradestation Pro is amajor upgrade. No longer does it take hours to set up the data feedor maintain the database. Costs are $99 per month for existingcustomers, with a 10-day free trial.

See www.Tradestation.com

* * * *Linnux Software

For those using the Linnux operating system, TA software can be aproblem. Version 1.4 of the Market Analysis System (MAS) forLinnux systems is now available. MAS is a software application thatprovides tools for stock charting and analysis.

MAS is released as free software. You can read more about MAS anddownload it at:

http://eiffel-mas.sourceforge.net/

* * * *Stock Research Web Browser

Using the Internet for Stock Research? You may consider using thispowerful web browser designed completely for stock research. It isspecifically set up to save you searching for real-time quotes, charts,analysis, msg boards and more. The Browser can be customized forany market and any website. Powerful link wizards navigate directlyto the data and charts you need. No built in advertisements, a veryuseful tool.

www.speedresearch.com

* * * *Free PF charts online

StockCharts.com has US PF charts available free online, Plus a ChartSchool that teaches a basic course in technical analysis for stockcharts. The lessons available are listed as: Overview, Chart Analysis,Indicator Analysis, Market Analysis, and Trading Strategies.

See www.Stockcharts.com.

* * * *Free Price Data

Grain market research offers free ASCII data files for US Futures, Stocks,Bonds and Mutual Funds. See http://www.grainmarketresearch.com.Free offerings include:

End of Day FuturesIntra-day Futures

End of Day StocksEnd of Day Mutual Funds

http://www.grainmarketresearch.com

* * * *Another Metastock Data Downloader

The Sentinel Software end-of-day product suite comprises the mostcomprehensive set of utilities available for downloading MetaStockdata from the Prestel Citi Feed service.

http://www.sentinelsoftware.co.uk/

* * * *AIQ TradingExpert Pro

AIQ TradingExpert Pro version 5.1 is a program that interfaces withthe real-time Mytrack datafeed quote manager, which is a longfavourite of mine for quick time real-time quotes. AIQ TradingExpertPro is known for its Expert Trading System and its ability to handlevast amounts of stock data, sorted into Markets and Sectors.

Stock Market Sector Analysis is one of the great advantages of thisprogram. Real-time quotes and charts are available via the internetfeed from $79.00 per month plus exchange fees. For Mytrack seewww.mytrack.com or for AIQ see www.aiq.com or phone (775) 831 2999 for details. Both are excellent programs.

Bytes and PiecesDavid Watts

Saving grace?Jesus and Satan were having an ongoing argument aboutwho was better on his computer. They had been going at itfor days, and God was tired of hearing all of the bickering.Finally, God said. “Cool it. I am going to set up a test thatwill run two hours, and I will judge who does the better job.”

So Satan and Jesus sat down at the keyboards and typed away.They moused.They did spreadsheets.They wrote reports.They sent faxes.They sent e-mail.They sent e-mail with attachments.They downloaded files.They did some genealogy reports.They created labels and cards.They did every known job.

Jesus worked with heavenly efficiency, and Satan was fasterthan Hell. But 10 minutes before their time was up, lightningsuddenly flashed across the sky, thunder rolled, rain poured,and – of course – the electricity went off.

Satan stared at his blank screen and screamed every curseword known in the Underworld. Jesus just sighed.

The electricity finally came back on, and each of themrestarted their computers. Satan started searching frantically,screaming “It’s gone! It’s all GONE! I lost everything whenthe power went out!”

Meanwhile, Jesus quietly started printing out all of his filesfrom the past two hours. Satan observed this and becameirate.

“Wait! He cheated! How did he do it?”

God shrugged and said, “Jesus saves”.


The major problem in trying to learn about Ichimoku Kinko Hyocharting is that all the text books are in Japanese and there are, asyet, no English translations. However we hope this article will providea useful introduction.

The Kanji, the Chinese character for Ichimoku, means ‘at a glance’,Kinko means ‘balance’, while Hyo simply means ‘charts’. SanjinIchimoku was the pseudonym used by a Mr. Goichi Hosoda whiledevising the method in the 1960’s for looking at stock market moves.More recently the system gained in popularity when a Mr. Sasakiwrote a book called ‘The Study of Ichimoku Kinko’ in 1995, havingcorrectly predicted the Nikkei’s collapse in the early 1990’s and theYen’s move below 100.00 to the US dollar with this method. Thecentral concept is the trend of the market.

Candlestick chart

The first thing to place on the chart are candlesticks, daily ones. HereI am using JPY. For anyone who does not use these, candlesticks arebasically like daily bar charts, with most of the same rules. The ‘body’of the candle is the difference between the opening price and theclosing price. The candle is coloured in if the close is lower than theopening price.

The fact that daily candlesticks are used means that the system is formedium to long term trades and is not suitable for jobbers and daytraders.

With this we watch for reversal patterns which I find are especiallyclear when using candles. Some of these have super names, like‘hammer’, ‘shooting star’ and ‘hanging man’.

Note that I am not using current prices for USD/JPY. This isdeliberate and not a ‘cop-out’. The reason is that USD/JPY has beentrading broadly sideways all year. Ichimoku Kinko Hyo is a trendingsystem. Ergo, it is useless in current market conditions.

Candlestick + moving averages

Two simple moving averages are added to the daily candlestick chart: anine day, called Tenkan-sen, and a 26 day, called Kijun-sen. The choiceof days is based on the fact that in Japan they used to work a six dayweek. There are, I believe, 26 working days in the average month.

The moving average does NOT use the closing price, but is an averageof the day’s range: high plus the low divided by two. This is probably atruer indication of the day’s activity. This is particularly so for marketswhere the closing price is at some arbitrary time, like Foreign Exchange;also for thin markets where the closing price may be manipulated.

The crossover of the two lines gives buy and sell signals as withconventional moving averages.

Candlestick + averages + clouds

To this chart two lines are added that make up the clouds. The first,known as Senkou Span A is calculated by adding the Tenkan andKijun values and dividing by two. This line is then plotted 26 daysahead of the last complete day’s trading.

The second line, imaginatively called Senkou Span B, is calculated byfinding the highest price of the last 52 days, adding the lowest priceof the last 52 days, and dividing by two. This is also plotted 26 daysahead. So although the lines have the same name, their constructionis very different.

Candle above cloud

The clouds have a variety of uses and add a completely newdimension to the chart. Firstly, if the candle is above the cloud (as inthe chart of the candlestick averages clouds), the trend is for higherprices. Like moving averages, this system really does only work intrending markets. The top of the cloud is the first level of supportand the bottom of the cloud is the second level of support. Fromexperience I have seen that these do often work but one has to givethem a little leeway.

Ichimoku Kinko Hyo ChartingBy Nicole Elliott and Yuichiro Harada

ICHIMOKU KINKO HYO

IBJ The Industrial Bank of Japan, Ltd.

USD/JPY Candle


USD/JPY Candle + Moving Average


USD/JPY Candle + Moving Average + Cloud


JPY =, Bid (Candle) Daily

JPY =, Bid (Candle) (MA 9)(MA 26) Daily

JPY =, Bid (Candle), Bid (Ichimoku 9, 26, 52, 26) Daily


Candle below cloud

The opposite is the case when candles are below the cloud, with thisbecoming the area of resistance. Very often the market seems tomove through the first support/resistance level and fails somewherein the middle of the cloud. In this case the shape of the candlestickmust be watched closely to see if it gives a reversal signal. Cloudscan be very useful in adjusting a basic trading position. Partial profitscan be taken or tentative new positions can be entered into withoutwaiting for moving averages to cross. They are particularly helpful forpicking levels in complex option strategies.

The thickness of the cloud is also important. The thicker the cloud, theless likely it is that prices will manage a sustained break through it.The thinner the cloud, the greater the chance of a successful break.

Thin cloud ahead

You will note that the cloud changes colour. This is when the SenkouSpan A and B cross over. The change of colour is not important, butit is the fact that the cloud is terribly thin at this point that matters.This gives an idea of when the market is likely to change trend. Itgives dates (I’d say three or four days around the point) when thereis an increased chance of a successful move through the cloud area.

The distance between the cloud and the current price is notsignificant. This is interesting as other Japanese charting methods,like the Disparity Index and the Divergence index, do take this verymuch into account and are used in a similar way to the RelativeStrength Index we know.

Chikou span

The final line to be added is the Chikou Span. This is today’s closingprice plotted 26 days behind the latest close. This is used in combinationwith the latest candlestick, and is explained in detail below.

The above explanation shows how clouds can work as resistance orsupport areas. But, of course, the cloud is not the only chart point.You will also find that the other lines are equally important, such asKijun-sen, Tenkan-sen and Chikou-span.

When the Tenkan-sen breaks up through the Kijun-sen, we call thiscondition “Koten”, and is normally considered to indicate marketsentiment turning more bullish. On the other hand we call the oppositesituation “Gyakuten” which suggests a bearish turn to the market.

An additional feature of Kijun-sen is that the inclination of the angleof the line is also important. We have already established that Kijun-sen uses longer-term data than Tenkan-sen, so it therefore changesmore slowly. In Ichimoku Kinko Hyo, when the Kijun-sen has someangle it indicates that the market has a direction. In other words, ifKijun-sen is moving horizontally, the market has no direction.

Next we will look at Chikou-span. This line is generally regarded tobe more important than the other lines that we have so far covered.As previously explained, the latest closing price is plotted as if it isfollowing the actual candles, 26 days behind. When the Chikou-spanbreaks above the candle (the candle of 26 days ago) it is consideredthat the sentiment in today’s market is turning bullish; “Koten”. Inthe opposite situation, when the line breaks below the candles, itshows sentiment turning bearish; “Gyakuten”. So it works much likemoving averages crossing, but in this case it is the Chikou-span linecrossing the actual candles of 26 days ago that gives today’s marketits direction. Additionally, the relationship between Chikou-span andthe clouds also works in a similar way, with crossovers giving newbullish and bearish signals. Both candles and the clouds can also besupport and resistance levels for Chikou-span. Even though this isplotted behind current prices, and the candles being used are thoseof 26 days ago, you can project these levels to today’s priceindicating where the market is likely to stall or fail. Personally I amnot convinced that this is the most important line. There are so manysubjective decisions that one can never be sure. For example, thecandles can be very tall, in which case deciding when the Chikou-span has really crossed can be very difficult. (as shown below).

USD/JPY Candle + Ichimoku


Koten and Gyakuten


The Inclination of Kijun-sen


USD/JPY Candle + Ichimoku




Chikou-span’s Koten/Gyakuten



Having covered the fundamentals, you are now in a position to lookat the markets using this analysis. But it is important to bear in mindthat we have only covered the first part of Ichimoku Kinko Hyo.

Most Japanese traders just stick to the cloud formations for theirtrading but Ichimoku charting does incorporate more complextechniques.

After the clouds, Ichimoku Kinko Hyo consists of three main parts, orideas. These are “Wave Principle”, “Timespan Principle” and “PriceTarget Principle”.

The point is that each part affects the other. For instance, theconcept of time and price are inseparably linked in this theory. Let’sremember the cloud formation. Each of the two lines that make upthe cloud will be similar to a 50% retracement level. Senkou Span Ais a weighted average of the last 26 days, being made up of Tenkan-sen (9 day) and Kijun-sen (26 day) values and divided by two.Senkou Span B is the mid-point of the last 52 days. This one givesadded importance to major highs and lows, ignoring the smaller dailymoves. For this reason it can often move sideways for many days.This idea, of halves with time marching on, is the innermost secret ofIchimoku Kinko Hyo.

The Wave Principle shows us that all price movement is resolved intosmaller pieces of waves. these are categorized into several types. It isa very simple principle as the graph below shows.

Those types are ‘I’, ‘V’, ‘N’, ‘Y’ and ‘P’. The name of the wavecomes from its shape. ‘I’ wave is the simplest pattern; up or down.‘V’ wave is a simple reversal pattern and made by two ‘I’ waves;up-and-down or down-and-up. ‘N’ wave is constructed with threelegs; up-down-up or down-up-down. ‘Y’ wave is diverging priceaction; several waves in a broadening pattern. ‘P’ wave is a pricepattern with convergence; several waves in a triangle formation.This principle tells us any price movement should be resolved intothose types of wave.

Although it may seem a bit fussy to call the waves all these differentnames, when it comes to measuring the level, you have to knowwhat type of wave and how many waves have been made already.So the classification itself won’t tell us much but it is necessary inorder to measure the target level. Note that Mr. Ichimoku did notcare how many waves were needed to fulfil a long-term trend. This isobviously very different from the Elliott Wave Principle, which allowsonly 3 or 5 waves per trend.

The Timespan Principle gives us a basic means to measure the nexttarget in terms of time. It is a very similar concept to the one that wealready know where Fibonacci numbers are projected as numbers ofdays forward, and similar to cycle theory. Ichimoku Kinko Hyo adopts“Kihon Suchi” the day of a turn, which suggests when the marketturns around, or when the market movement accelerates upward ordownward. The key numbers are 9, 17, 26 and some biggernumbers. These numbers were found to be the most useful and themost reliable after analyzing masses and masses of examples, mainlystock prices. They are not ‘magic’ numbers in terms of arithmetic,such as prime numbers or Fibonacci numbers. These are the numbersof days forward when the market may stall or turn.

The “Simple” number in this table means numbers with independent

values and the “Compound” is the number made by compounding“Simple” numbers.

For instance the compound number 42 is the number sum of 17 and26 minus 1. 17 and 26 are the “Simple” numbers and 1 is deductedbecause it has one attachment part. So, an ‘I’ wave that took 9 daysto move higher, will drop as an ‘I’ wave, also in 9 days, as in Figure1. Or an ‘I’ wave that took 26 days to move higher can also move asa ‘V’ wave next, but it will also take 26 days, as in Figure 2. Notethat there are many combinations of different waves, and that theseare just two examples. Also, the number of days in the wave is notexact. You must allow a day or two either side of the number. Thereason why the calculation requires 9 days, 26 days and 52 days isbecause they are Ichimoku Kinko’s most basic building blocks forcalculating time targets.

My personal opinion is that these numbers could perhaps bereviewed and refined, adjusting them to suit different markets better.

The final concept, the Price Targeting principle, will be familiar toyou. It is very similar to targets generated by patterns in the charts,and similar to targets based on the height of waves in Elliott Wavetheory. But in Ichimoku there are only four ways to measure wavesand they will only measure the very next wave, not a series ofwaves. They are ‘V’, ‘N’, ‘E’ and ‘NT’.

Most use the height of the ‘A’ wave as the starting point, with thetarget usually being calculated starting at ‘C’.

Nicole Elliott and Yuichuro Harada are technical analysts with The Industrial Bank of Japan

Kihon Suchi (Basic Values)


Time Measurement


Types of Wave


Target Measurement



Part 2 Internal Dynamics Of MOS-ableMarkets1

MOS - A RECAPITULATION

The MOS being examined is that which exists between the ChicagoMercantile Exchange (CME) and the Singapore Exchange DerivativesTrading Ltd, (SGX-DT), which is the derivatives arm of the SingaporeExchange. The facility is available for clearing members on both theCME and SGX-DT. It allows new or liquidating trades to be executedon the SGX-DT by CME members [when their own market is closed]and then be transferred back to Chicago, should they wish to do so.The opposite is also catered for i.e. SGX-DT firms can have tradesexecuted on the CME [when their own market is closed] and thenhave them transferred back to Singapore, should they wish to do so.

This arrangement apparently defies the logic of traditional, stand-alone, or non-MOS markets, described earlier. For these markets ithas been shown that all the transactions are contained within themarket of interest, although the “transfer” of commitment insidethem is an illusion. However, it is significant that their internaldynamics are such that the open interest and volume figuresproduced at the end of the day have a direct relationship to theintra-day price changes that have occurred.

However, MOS-able markets involve an additional set of processeswhich allows them to achieve the seemingly impossible. These can beviewed as (a) the provision of a notional counter-party acting as anintermediary between the exchanges – the Inter-Exchange Matching(IEM) system and (b) the adoption of a special form of novation thatallows exchanges themselves to act as notional counter-parties toeach other – through “mirror” accounts. These two aspects takentogether represent the mechanism that allows commitment to betransferred between exchanges – the MOS.

PARTICIPANT INTERACTION AND COUNTER-PARTY FORMATION

The MOS process itself can best be explained by representing thetransfer logic via four steps, Figures 5, 6, 7 and 8, each of whichcontains appropriate sets of data flow diagrams (DFD).2 However,for all practical purposes, the process would be viewed as beingseamless in the real world. The first step is represented by the DFD inFigure 5. This shows the situation where Participant A has assumedLong commitment at the CME as a result of the normal orderexecution and novation (OE&N) process described3 earlier in Figure 3.The participant could be viewed as coming from the right hand[output] side of Figure 3, but now appears on the left hand [input]side of Figure 5, where it will, this time, undergo an additionaltransformation which is unique to the MOS operation. The IEM takespart in the MOS operation by allowing an order entered into theCME pit which is meant to be “exported” to the SGX to be matchedwith an order from the SGX that requires it to be “imported” there.In logical terms the IEM can be viewed as a notional participant thatcan operate in the CME and SGX, and can thus bridge the twoexchanges. It is incorporated into the DFD in Fig. 5 in that sense.

Participant A which is currently Long (CME) and which has itsnotional counter-party in the form of the CME Clearing House whichis Short i.e. the “NCP = CH(CME) (Short)” begins by “selling” itscommitment to the CME IEM and so becomes Flat. The CME IEM,which is Flat, “buys” the commitment and so becomes Long. Notethat it has no counter-party, as the IEM is a notional or conceptualparticipant, rather than a physical one and thus there is no directneed for novation.

The result is that the actual Long commitment which formerly existed

Figure 5 A Representation Of The Transfer Of Actual LongCommitment From The Chicago Pit To The Chicago Inter-ExchangeMatch System

Note: MOS = Mutual Offset System [operation], NCP = Notional Counter-Party,IEM = Inter-Exchange Match, CH = Clearing House

in terms of Participant A is removed from the CME market in which itoriginally came into existence, as described in Figure 3. As a resultthat market’s commitment is distorted [in a logical sense] because itcame into existence by virtue of the Short commitment thatappeared at the same time, within that market. Open interest isdefined as the number of contracts that are still open, and as acontract is a bi-lateral entity, Short commitment cannot remain inexistence without an equal and opposite amount of Longcommitment also existing. Thus if Long commitment is “removed”by the CME IEM process as part of the MOS activity it somehow hasto be replaced to retain the symmetry of the CME’s actual counter-party and Clearing House notional counter-party arrangements,associated with its “parenting” Short commitment that is still beingheld by CME Participant B.

THE COUNTER-PARTY OF COUNTER-PARTIES

This is achieved by yet another logical artifice – that of “MirrorAccounts”. Two can be viewed as existing, one at the CME and theother at the SGX. They can be considered to be complementaryhalves of an account that over-arches both exchanges. The MirrorAccounts allow a form of novation to be undertaken, not at theindividual participant/participant level, as described previously, as in aconventional stand-alone exchange – but at an exchange/exchangelevel. These accounts allow the CME to act as a notional counter-party to the SGX and the SGX to act as a notional counter-party tothe CME when the IEM facility is invoked at either the Chicago orSingapore exchanges. Logically, these Mirror Accounts can be viewedas notional “participants” in order to integrate their activities into theDFD representation. They play a key role in ensuring that openinterest is credited to the appropriate exchange during MOStransactions.

Their utilization can be represented via Figure 5 with the CME’sMirror Account starting Flat and “buying” the Long commitmentthat Participant A is “selling” as part of the MOS process. As a resultit becomes Long and adopts the notional counter-party thatParticipant A had i.e. the “NCP = CH (CME) (Short)”. This has theeffect of acting as a surrogate for Participant A, thus enabling theLong commitment that is to be transferred to Singapore to be“retained” in Chicago, but in a notional form. This “sleight of hand”

Open Interest, Volume and Price Analysis: The Special Case ofMarkets with Inter-Exchange Mutual Offsetting Systems (MOS)

By Dr Michael Wignall MSTA

Figure 5


preserves the symmetry stemming from the original transactioninvolving CME’s Participant A buying and B selling through theirOE&N [described in Fig. 3] to be conserved.

THE “TRANSFER” OF COMMITMENT TO ATHIRD PARTY AT THE OTHER EXCHANGE

The second step which can be viewed as an “export” activity fromChicago can be represented by Figure 6. This shows the CME IEMLong which was at the right hand [output] side of Figure 5 now atthe left hand [input] side of Figure 6 where it will undergo anothertransformation as part of the MOS operation. It “sells” to becomeFlat whilst the IEM representing Singapore “buys” to become Long.Again there are no notional counter-parties involved in thesetransactions, as they do not involve physical participants.

Figure 6 A Representation Of The Transfer Of The Chicago Inter-Exchange Match Long Commitment To The Singapore Inter-Exchange Match System

Note: MOS = Mutual Offset System [operation], IEM =Inter-Exchange Match,NCP = Notional Counter-Party, DCP = Direct Counter-Party

This third step, which can be viewed as an “import” to Singapore, isrepresented by Figure 7. This shows the SGX IEM Long which was atthe right hand [output] side of Figure 6 now at the left hand [input]side of Figure 7 where it will undergo another transformation as partof the MOS operation. It “sells” to become Flat whilst the SingaporeParticipant C [the third party, on whose behalf Participant A originallybought the Long position in Chicago], “buys” from the SGX IEM tobecome Long. The net result is that the SGX IEM becomes Flat whilstParticipant C becomes Long. Here, however, Participant C is real in aphysical sense and cannot logically have Long commitment inSingapore without Short commitment also having been created there.

Figure 7 A Representation Of The Transfer Of The Singapore Inter-Exchange Match Long Commitment To An Actual Singapore Third Party


As no actual transaction took place in the physical market exchange[as it was closed], the necessary Short commitment has to benotional to provide the necessary symmetry. This is achieved via the

SGX Mirror “participant” which starts Flat and “sells” to becomeShort. The outcome is that Participant C is provided with thenecessary Short counter-party i.e. “DCP = Mirror (SGX-DT) (Short)”,whilst the Mirror (Short) has Participant C as its direct counter-party.This is the opposite process of the Mirror account in the CME, but ithas the same effect – the preservation of the necessary symmetry ofthe various transactions, both real and notional.

The final step is represented by Figure 8 which involves the novationprocess acting to de-couple the Long Participant C from the Mirror“participant” and provide the necessary involvement of the ClearingHouse. It can be seen that the process involved is that describedearlier in Figure 3, which involved the de-coupling of the originalCME Participants A and B from each other. Consequently the MOSprocess ends with the Participant C having the SGX-DT ClearingHouse as its notional counter-party, as does the Mirror “participant”.

Figure 8 A Representation Of The Impact Of The Novation ProcessOn The Output Of The Singapore Inter-Exchange Match System


The end result of the MOS activity in terms of the state of theexchanges at Chicago and Singapore is summarised in tabular formfor clarity in Figure 9. The original CME Participant B remains in themarket with its Short commitment and with the CME Clearing House(Long) acting as its notional counter-party. Similarly the CMEClearing House has Long and Short commitment with direct counter-parties of Participant B and the Mirror “participant” respectively.There is no sign of the other original Participant A, because it has leftthe market.

Figure 9 A Summary Of The State Of The Chicago And SingaporeExchanges On Completion Of The MOS Operation

Figure 6Figure 8

Figure 9

Figure 7



At the Singapore end, Participant C is Long and has the SGX ClearingHouse which is Short as its notional counter-party. The Mirror“participant” is Short and has the SGX Clearing House which is Longas its notional counter-party. Finally, the SGX-DT Clearing House hasShort and Long commitment with direct counter-parties of ParticipantC and the Mirror “participant”, respectively. The net result is thatParticipant A at the CME has transferred its Long position toParticipant C at the SGX-DT.

It can be seen that there is symmetry of commitment between all theparticipants, both within each exchange and between eachexchange. In the case of the latter the CME Mirror “participant” isLong, whilst the SGX-DT Mirror “participant” is Short. With theprocess of transferring commitment from Chicago to Singaporehaving been examined, the resulting open interest and volumeimplications can be explored.

CONSEQUENTIAL IMPACT ON VOLUME ANDOPEN INTEREST

Unlike the situation involving a single or stand-alone exchange, theopen interest and volume situation that was described earlier4 inrelation to Figure 4, the operation of a MOS facility between twoexchanges has implications for both exchanges. If we assume, onceagain, that Participants A, B and C were the only active actualparticipants involved in the MOS trade, the open interest and volumechanges that occur this time are illustrated in Tables 5, 6 and 7.These reflect the inter-linking processes described in the three DFDdiagrams representing the Chicago – Singapore MOS process, but forsimplification do not show the role of the exchanges’ ClearingHouses, acting as counter-parties to each participant. Thepresentation in tabular form allows the changes in open interest andvolume to be followed, as opposed to the processes involved, asdepicted in the DFDs of Figs. 5-8. The changes are represented in anidealised sequence over 10 separate steps for reasons of clarity. In thereal world the process would not appear so structured.

The beginning of the process, at the CME, is represented by Table 5,containing steps 1-4, which illustrates the interaction of ParticipantsA and B, which trigger the MOS process to “export” commitment tothe SGX. As they both start Flat, the filling of their orders throughoffsetting against each other results in open interest increasing by 1,as does the transaction volume. The IEM mechanism, together withthe action of the CME Mirror Account are both notional transactionsand thus do not impact on the CME’s open interest or volumestatistics, in their own right. What is important to note is that in thetable the open interest of +1 has been assigned to the initialinteraction of Participants A and B and left there, even thoughParticipant A has departed the market by step 3. This is for simplicity.In reality the CME’s Mirror Account becomes Long in order tosubstitute for it and maintain a counter-party to Participant B whichremains in the market, as explained in Fig. 5.

Table 5 Commitment Formation At The CME

Notes: IEM = Inter-Exchange Match, M-CME = Mirror at CME, M-SGX = Mirror at SGX

The “no-mans’ land”, in logical terms, between the two exchanges,is represented by Table 6. This includes steps 6-7, the notional

transactions whereby the CME’s IEM transfers its commitment to theSGX’s IEM. There is no impact on either exchanges’ open interest orvolume statistics.

Table 6 Inter-Exchange Transfer Of Commitment

Notes: IEM = Inter-Exchange Match

The final aspect of the MOS process, the “import” of thecommitment to the SGX is represented by Table 7 and includes steps8-10. Here the open interest and volume is generated by the actionof Participant C offsetting its order to buy against the order to sellthat is available via the IEM. In order to maintain the symmetry ofthe filled buy order which is a physical transaction, the SGX’s MirrorAccount goes Short, thus legitimising the open interest of +1, whichis credited to that exchanges’ statistics – as is the associatedtransaction volume.

Table 7 Commitment Formation At The SGX

Notes: IEM = Inter-Exchange Match, M-SGX = Mirror at SGX, M-CME = Mirror at CME

What has been described via these three Tables is the impact on theMOS process of a Chicago based participant creating commitment inChicago and having it transferred to Singapore for retention there.This is only one of a number of permutations that can occur throughusing the facility, and the one chosen to provide a specific, detailedexample. The others are: a CME participant can initiate commitmentin the SGX and either have it retained there, or have it brought backto the CME. Or it can initiate commitment locally in the CME andhave it retained there. Similarly, an SGX participant can initiatecommitment in the CME and either have it retained there or broughtback to the SGX. Or it can initiate it locally in the SGX and eitherhave it retained there or have it transferred to the CME.

Each of these operations generates its own set of changes in theopen interest and transaction volume that are peculiar to theindividual activity. They include MOS inter-exchange transfers andnon-MOS activity. The latter represent situations where commitmentis retained in the exchange it was formed in – thus mimicking aconventional, stand-alone exchange. Comparing the two types ofactivity with each other can provide the key to identifying MOSinduced distortions that are capable of invalidating the conventionalapproach to open interest, volume and price analysis.

Notes (Part 2)

1. For Part 1 see the Market Technician, Nov. 2000, Issue No. 39

2. DFDs provide a means of representing “systems” in a structuredform. For an overview of their characteristics, see Part 1, Note 2.

3. For Figures 1 to 4 inclusive, see Part 1.

4. For Tables 1 to 4 inclusive, see Part 1.

Copyright ( Michael Wignall 2001)

MOS: CME

STAGE PARTICIPANT PRE – EXECUTION POST- IMPACT ON IMPACT ON# ACTIVITY COMMITMENT EXECUTION OPEN TRANS-

COMMITMENT INTEREST ACTIONVOLUME

1 A(B) buys 1 A = Flat A = Long 1 ) )) ∆ = +1 ) ∆ = +1) )) )

2 B(A) sells 1 B = Flat B = Short 1 ) )

3 A(IEM CME) A = Long 1 A = Flat NA NAsells 1

4 IEM (CME) IEM = Flat IEM (CME) = NA NAbuys 1 Long 1

5 M-CME M-CME M-CME NA NA(M-SGX) Buys 1 = Flat = Long 1

MOS: CME aASGX-DT



6 IEM (CME) IEM (CME) = IEM (CME) = NA NAsells 1 Long Flat

7 IEM (SGX) IEM (SGX) = IEM (SGX) = NA NAbuys 1 Flat Long 1

MOS: SGX-DT



8 IEM (SGX) IEM (SGX) = Long IEM (SGX) = Flat NA NAsells 1

9 M-SGX (M-CME) M-SGX = Flat IEM (SGX) = ) )sells 1 Short 1 ) ∆ = +1 ) ∆ = +1

) )

10 C (IEM SGX) C = Flat C = Long 1 ) )buys 1


• Among the technical trading signals supplied to customers byforeign exchange trading firms are “support” and “resistance”levels. These levels indicate points at which an exchange rate trendis likely to be interrupted or reversed.

• A rigorous test of the levels specified by six trading firms during the1996-98 period reveals that these signals were quite successful inpredicting intraday trend interruptions.

• Although all six firms were able to identify turning points inexchange rate trends, some firms performed markedly better thanothers. As a group, the firms predicted turning points in the dollar-yen and dollar-pound exchange rates more accurately than turningpoints in the dollar-mark exchange rate.

• In addition, the predictive power of the support and resistancelevels appeared to last at least five business days after they werefirst communicated to customers.

Early in the morning of each business day, the major foreignexchange trading firms send their customers lists of technical tradingsignals for that day. Timely technical signals are also supplied bymajor real-time information providers. These signals, which are basedprimarily on prior price and volume movements, are widely used byactive foreign exchange market participants for speculation and fortiming their nonspeculative currency transactions. In fact, 25 to 30percent of foreign exchange traders base most of their trades ontechnical trading signals (Cheung and Chinn 1999; Cheung andWong 1999). More broadly, technical analysis is used as either aprimary or secondary source of trading information by more than 90percent of foreign exchange market participants in London (Allenand Taylor 1992) and Hong Kong (Lui and Mole 1998).

The technical trading signals provided to customers vary over timeand across technical analysts, but the vast majority of the dailytechnical reports include “support” and “resistance” levels.According to technical analysts, support and resistance levels arepoints at which an exchange rate trend is likely to stop and may bereversed. For example, a firm publishing a support level of $1.50/£would claim that the dollar-pound exchange rate is likely to stopfalling if it reaches $1.50/£. If the firm also provided another supportlevel of $1.45/£, the firm would claim that if the exchange ratepasses through $1.50/£, it is likely to stop falling at $1.45/£.

Despite the almost universal use of support and resistance levels inshort-term exchange rate forecasting, the ability of these tradingsignals to predict intraday trend interruptions has never beenrigorously evaluated. This article undertakes such a test, using actualsupport and resistance levels published daily by six firms fromJanuary 1996 through March 1998. The firms include commercialbanks, investment banks, and real-time information providers basedin the United States and abroad. I examine the value of threecurrencies relative to the U.S. dollar: the German mark, the Japaneseyen, and the British pound. Support and resistance levels for theseexchange rates are tested against indicative exchange rate quotessampled at one-minute intervals between 9 a.m. and 4 p.m. NewYork time.

These tests strongly support the claim that support and resistancelevels help predict intraday trend interruptions for exchange rates. Allsix of the firms studied were able to identify points where intradaytrends were likely to end. However, some firms were better thanothers at identifying such points.

For most firms, the predictive power of support and resistance levelslasted at least five business days beyond the levels’ publication date.Despite their overall success at identifying points of trendinterruptions, none of the firms correctly assessed the relativelikelihood of trend interruptions at the different levels. These results

are consistent across firms and are sustained over a number ofsensitivity analyses.

The statistical tests are based on the bootstrap technique (Efron1979, 1982), a nonparametric method frequently used to evaluatetechnical trading strategies (Brock et al. 1992; Levich and Thomas1993). To implement the tests, I compare the behaviour of exchangerates upon reaching published support and resistance levels with thebehaviour upon reaching 10,000 sets of arbitrarily chosen supportand resistance levels. If the outcome associated with the actual levelsexceeds the average outcome for the arbitrary levels in a highproportion of months, I conclude that the published levels havesignificant predictive power.

To complement the analysis of these signals’ predictive power, I alsoanalyze the signals themselves. I show that support and resistancelevels provided by individual firms tend to be fairly stable from day today. Their range varies very little over time. Firms do not agreeextensively with each other on the relevant signals.

The specific conclusion that exchange rates tend to stop trending atsupport and resistance levels has no precedent in the academicliterature. The closest point of comparison is a study by Lo et al.(2000), which finds that the conditional distribution of financial pricesis sensitive to the presence of a broad variety of technical tradingsignals, consistent with the results presented here.

The finding that support and resistance levels are able to predicttrend interruptions is consistent with other studies of the usefulnessof technical trading rules when applied to currencies. Filter rules werefound to be profitable as early as 1984 (Dooley and Shafer 1984),less than a decade into the floating rate period, and this finding hasbeen confirmed repeatedly (Sweeney 1986; Levich and Thomas1993). Moving-average crossover rules have also been testedfrequently on exchange rates, with similar results (Levich and Thomas1993; Menkhoff and Schlumberger 1995). More recently, Chang andOsler (1998) find that a trading strategy based on the head-and-shoulders chart pattern is profitable for dollar exchange rates vis-à-visthe mark and the yen, although not for four other dollar exchangerates.

This study differs from those earlier studies in four notable ways.First, the technical trading signals used here are intended toanticipate trend reversals, rather than trend continuations. Second,this study uses a type of trading signal that is actively used by marketparticipants. Third, it uses trading signals that were produced bymarket participants. Other academic studies of technical analysishave typically constructed technical trading signals of their own.Finally, this study uses data sampled at one-minute intervalsthroughout the New York trading day, while most earlier studies haveused data sampled at daily or lower frequencies.

The two existing studies of support and resistance levels— Curcio etal. (1997) and Brock et al. (1992)—test the hypothesis that pricestend to move rapidly once the levels are breached. Curcio et al. findthat the hypothesis is not true on average for currencies, but mayhold true during periods of strong trending. Brock et al. find that thehypothesis is true for daily movements of the S&P 500 stock index,but the profits may not be sufficient to offset transaction costs. Thehypothesis that prices will trend once a trading signal is breached isnot unique to support and resistance levels and is not examined here.

TECHNICAL ANALYSIS

Technical analysts claim that they can predict financial pricemovements using an information set limited to a few variables, suchas past prices. Many of the major technical indicators were describedas early as 1930 by Shabacker, who based his conclusions on

Support for Resistance: Technical Analysis andIntraday Exchange Rates

By Carol Osler


observations of U.S. stock prices. By now, technical indicators arewidely used in major financial markets around the world, includingforeign exchange and futures markets. There are two magazinesdevoted exclusively to the topic, each of which has more than40,000 subscribers. To learn about technical analysis, one can consulta myriad of manuals, software, and on-line sources. Alternatively,one can take courses in technical analysis.

Casual observation and conversations with market participantsindicate that support and resistance levels are the most widely usedtechnical indicators in the foreign exchange market. This conclusion isalso suggested by the fact that support and resistance levels are theonly indicators provided by all six of the technical services covered inthis research. In fact, some services provide no technical indicators atall other than support and resistance levels.

Support and Resistance Levels Defined

Before delving further into the analysis, it is important to explore thedefinition of support and resistance levels provided by technicalanalysts themselves. According to one major technical analysismanual, “support is a level or area on the chart under the marketwhere buying interest is sufficiently strong to overcome sellingpressure. As a result, a decline is halted and prices turn back again…Resistance is the opposite of support” (Murphy 1986, p. 59).

A review of technical analysis manuals reveals that there is littledisagreement among analysts on this definition (Arnold 1993;Edwards and Magee 1997; Hardy 1978; Kaufman 1978; Murphy1986; Pring 1991; Sklarew 1980). For example, Pring states:“Support and resistance represent a concentration of demand andsupply sufficient to halt a price move at least temporarily” (p. 199).Likewise, Arnold (1993) observes: “A support level is a price level atwhich sufficient demand exists to at least temporarily halt adownward movement in prices” (p. 67).1

To identify the support and resistance levels relevant for the comingday, practicing technical analysts consult a variety of informationinputs. These include visual assessments of recent price performance,simple numerical rules based on recent price performance, inferencebased on knowledge about order flow, and market psychology.

The simplest approach to visual assessment is to look at recentminima and maxima: “Usually, a support level is identifiedbeforehand by a previous reaction low,” and “a resistance level isidentified by a previous peak” (Murphy 1986, p. 59). According toPring (1991), one could also identify support and resistance levels bydrawing a trendline, or “channel,” in which recent peaks areconnected by one line and recent troughs are connected by another:“A good trendline represents an important support and resistancezone” (p. 105).

One numerical rule used to infer support and resistance levels is the“50 percent rule,” which asserts that a major market move will oftenbe reversed by about 50 percent in the first major correction (Pring,p. 187). Fibonacci series, which are widely used, suggest that 38.2percent and 61.8 percent retracements of recent rises or declines arecommon.

Market insiders sometimes identify support and resistance levels usingprivate information or information circulated informally in the marketabout certain market participants. For example, if a technical analystlearned in conversation that Japanese exporters are selling at 100, heor she would report a resistance level at ¥100.00/$. Similarly, if atrader knew that his or her own firm had a large order at DM1.50/$,he or she might expect unusual price behaviour at that point.

Simple market psychology is also used to help identify support andresistance levels. According to Murphy (1986): “Traders tend to thinkin terms of important round numbers… as price objectives and actaccordingly. These round numbers, therefore, will often act aspsychological support or resistance levels.” As I will demonstratelater, published support and resistance levels are round numbers thatend in 0 or 5 much more often than they would if they were chosenat random.

Some of the firms in the sample provided explanations for theirchosen support and resistance levels. All of the approaches listedabove are well represented among those explanations.

Properties of the Support and ResistanceDatabase

The data examined here include support and resistance levels for themark, yen, and pound in relation to the U.S. dollar, published daily bysix firms from January 1996 through March 1998. Two of the firms didnot report support and resistance levels for the pound. In total, thereare approximately 23,700 support and resistance values (combined) forthe mark, 22,800 for the yen, and 17,700 for the pound.

The six providers of technical analysis include commercial banks,investment banks, and news services. Some operate in the UnitedStates and others operate abroad. The commercial and investmentbanks provide the information free of charge to their customers,hoping that customers will be encouraged to direct more businesstoward them. Some news services charge for the information. Sinceall the providers hope that the usefulness of their signals willgenerate additional business, they have every incentive to maximizeaccuracy. In the analysis that follows, the firms are assigned numbersto preserve anonymity.

TABLE 1

Average Total of Support and Resistance Levels per Reporting Day

German Japanese BritishFirm Overall Mark Yen Pound

1 10.0 10.0 10.0 10.02 6.0 6.0 6.0 6.03 17.8 18.1 18.0 17.54 9.0 9.2 8.8 —5 4.2 4.8 3.8 4.06 2.5 2.7 2.3 —

Source: Author’s calculations

TABLE 2

Average Distance between Support and Resistance Levels


1 30 36 29 262 61 75 52 553 54 58 49 544 57 64 49 —5 162 156 184 1446 42 47 36 —


Note: Distances are measured in points, or 0.0001 marks/dollar, 0.01yen/dollar, and 0.0001 dollars/pound.

On any given day, technical indicators were likely to be received fromabout five of the six firms. Firms failed to report for reasons such asvacations, sickness, and equipment problems. For individual firms, theaverage number of support and resistance levels (combined) thatwere listed per reporting day per currency ranges from two toeighteen (Table 1).

The support and resistance levels were quite close together for somefirms and quite far apart for others. As shown in Table 2, the averagedistance between levels varied from 30 to 162 points on average(where a point is the smallest unit used in quoting an exchange rate).

Firms also varied dramatically in the range over which they chose topresent support and resistance levels applicable to a given day (Table3). For Firm 6, the outermost support and resistance levels weretypically only about 100 points away from current spot rates, whilefor Firm 3 the outermost support and resistance levels were typicallymore than 700 points away. The final two rows of the table show theaverage and maximum daily exchange rate moves away from theiropening rates over the period. For all firms, the outermost supportand resistance levels were substantially farther away from theopening rates than this average move. The correlation between dailyexchange rate ranges and the gap between opening rates andoutermost support and resistance levels was not statisticallysignificant for any firm-currency pair.


TABLE 3

Average Gap between Current Spot Rates and OutermostSupport and Resistance Levels

German Japanese BritishFirm Mark Yen Pound

1 3.45 2.56 2.382 3.71 2.38 2.633 8.32 6.73 7.154 5.52 3.79 —5 4.96 5.13 4.526 1.24 0.74 —Daily ranges

Average 0.45 0.33 0.37Maximum 3.23 3.88 3.28


Note: Distances are measured in units of 100 points, or 0.01 marks/dollar, 1.0yen/dollar, and 0.01 dollars/pound.

Use of Round Numbers

More than 70 percent of the support and resistance levels in thesample end in 0, and a full 96 percent end in either 0 or 5 (Table 4).These proportions greatly exceed the proportions we would observeif levels were chosen randomly, which would be 10 or 20 percent,respectively. Levels ending in 00 or 50 were also disproportionatelyrepresented. This may be a manifestation of the psychologicalinterpretation of support and resistance levels mentioned earlier. It isinteresting to note that Goodhart and Figliuoli (1991) observed thatround numbers were also disproportionately represented in bid-askspreads for major currencies.

TABLE 4

Support and Resistance Levels Ending in Round NumbersPercent

Support and Resistance Levels Ending in

00 00 or 50 0 0 or 5

Naturalfrequency 1.0 2.0 10.0 20.0

Firm 1 12.1 17.9 65.8 96.3Firm 2 13.2 22.4 58.1 84.4Firm 3 12.5 16.8 82.4 96.6Firm 4 7.8 15.7 52.9 92.2Firm 5 49.4 66.4 97.1 99.4Firm 6 22.9 42.9 74.3 100.0

All firms 13.5 19.6 70.1 95.5


Continuity

To analyze the extent to which support and resistance levelspublished by a given firm vary from day to day, I counted thenumber of support and resistance levels shared across days for agiven firm, and compared it with the maximum number of levels thatcould have been shared. That maximum depends on the number ofsupport and resistance levels provided on the two days: if the firmprovided three support levels on the first day and four on the secondday, the number of shared support levels, or matches, could notpossibly exceed three. The maximum number also depends on thesize of the exchange rate move from the first to the second day: ifthe exchange rate falls substantially between days one and two, thensome of the support levels provided on day one might be irrelevanton day two. A shared level, or “match,” was defined as a pair ofsupport and resistance levels on contiguous days that differed by lessthan 5 points.

On average, about three-quarters of the still-applicable support andresistance levels from one day would be used again the next day(Table 5). This average masks a clear division of the firms into twogroups. Firms 3, 4, and 5 showed the strongest continuity: more than

three-quarters of their still-applicable support and resistance levelswere used again the next day. For the remaining three firms, thecorresponding proportions were lower, ranging from about one-halfto two-thirds. These results do not change qualitatively if a match isdefined as two levels within 2 points of each other.

TABLE 5

Continuity in Support and Resistance Levels


1 64.4 62.4 62.9 67.92 54.0 51.0 56.2 54.53 91.4 89.8 91.7 92.64 81.9 81.2 82.7 —5 80.8 77.5 85.0 81.36 56.5 56.0 57.4 —

All firms 77.8 76.5 78.2 79.1

Source: Author’s calculations.

Notes: The table shows the percentage of support and resistance levelsshared across adjacent days. A pair of levels on adjacent days is defined asshared if the levels differ by at most 5 points. Numbers for firms showingparticularly high continuity are italicized.

Agreement across Firms

Firms do not agree extensively on the relevant support and resistancelevels for a given day. To examine the extent of agreement acrossfirms, I first counted the number of matches across each of thefifteen pairs of firms. For each day, the number of actual matcheswas then compared with the number of possible matches for thatday. For a given day, the number of possible matches among support(resistance) levels was taken to be the minimum number of support(resistance) levels provided across the two firms.

On average, roughly 30 percent of all possible matches were realizedas actual matches under the basic definition of a match (a maximumdifference of 5 points), as shown in Table 6. Across firm pairs, thefrequency of agreement varied from 13 to 38 percent (Table 7). Firm5 stands out as the least likely to agree with its peers. Firms 1 and 2stand out as agreeing particularly frequently with each other. Amongthe other firms, no strong patterns are distinguishable.

TABLE 6

Overall Agreement on Support and Resistance Levelsacross Firms

German Japanese BritishOverall Mark Yen Pound

Average possible matchesper day 112.0 47.9 43.0 21.1

Average actual matchesper day 33.5 14.3 12.3 6.9

Average actual matchesper day as a percentageof possible matches 29.9 29.9 28.6 32.7


Notes: The table shows the number of times all firms’ support and resistancelevels actually match as a percentage of the total number of possiblematches. A match is defined as a pair of support and resistance levels thatdiffer by at most 5 points.

If a match is defined as two levels within 2 points of each other, thenroughly 18 percent of possible matches are actually realized. Thesame Firm 5 still stands out as the least likely to agree with its peers;the only strong agreement appears to be between Firms 3 and 6. 2


TABLE 7

Pairwise Agreement across Firms on Support andResistance Levels

All Currencies

Firm

Firm 2 3 4 5

1 38 33 27 172 35 31 133 32 254 235


Notes: The table shows the number of times a pair of firms’ support andresistance levels actually match as a percentage of the number of possiblematches. A match is defined as a pair of support and resistance levels thatdiffer by at most 5 points. Numbers representing firm pairs for whichagreement falls at or below 23 percent (mean overall agreement of 30 percentminus one standard deviation) are in bold. The italicized number representsfirm pairs for which agreement falls at or above 37 percent (mean overallagreement of 30 percent plus one standard deviation).

EXCHANGE RATE DATA AND METHODOLOGY

This section presents the exchange rate data, some importantdefinitions, and the statistical methodology used to test the ability ofsupport and resistance levels to predict intraday trend interruptions.

Exchange Rate Data

The exchange rate data comprise indicative bid-ask rates posted onReuters, captured at one-minute intervals from 9 a.m. to 4 p.m. NewYork time. The prices for a given minute were taken to be the lastquote made prior to that minute.

The analysis in Goodhart, Ito, and Payne (1996) suggests that theseindicative quotes are likely to correspond closely to actual transactionprices. The major divergences between quotes and actual prices seemmost likely to occur at times of large, rapid price movements. Duringthe sample period, these divergences often occurred at times ofmacroeconomic data announcements from the United States, whichtended to happen at 8:30 a.m., before the exchange rate data usedhere begin. Recent research by Danielsson and Payne (1999) finds thatquotes may differ from actual transaction prices in other, potentiallyimportant ways. This point is discussed in greater detail below.

The construction of the exchange rate data set was driven primarilyby the need to capture as closely as possible the price sequence thatwould be observed by traders operating in the market. This explainswhy an interval of one minute was selected, rather than the morecommon interval of five minutes (for example, see Andersen andBollerslev [1998]). It also explains why prices were not taken as anaverage of the immediately preceding and following quotes, anothercommon technique in the literature (for example, see Andersen andBollerslev [1998]): traders operating in real time could not know theimmediately following quote.

The starting time for the data was chosen as 9 a.m. New York timefor two reasons. First, by 9 a.m., the support and resistance levels inthe data set have been transmitted to customers, including thosefrom New York firms. Second, by 9 a.m., the reaction tomacroeconomic data announcements from all the countries involved– including the United States, where, as noted, majorannouncements generally occurred at 8:30 a.m. New York time –would largely be over (see Andersen and Bollerslev [1998]).

The data end at 4 p.m. because very little trading takes placebetween then and the beginning of the next trading day in Asia. The4 p.m. cutoff was also chosen because, in the underlying tick-by-tickexchange rate data set, quotes are not captured after 4 p.m. onFridays.

Some Definitions

The exchange rate was defined as hitting a support (resistance) levelif the bid (ask) price fell (rose) to within 0.01 percent of that level

(see the chart for an illustration). Because the 0.01 percent figure issomewhat arbitrary, more than one definition was tried: the gap wasalso set at 0.00 percent and 0.02 percent in alternative tests. A trendinterruption was defined as follows: once the exchange rate hit asupport (resistance) level, the trend was interrupted if the bid (ask)price exceeded (fell short of) the support (resistance) level fifteenminutes later. Since the cutoff at fifteen minutes is also somewhatarbitrary, an alternative of thirty minutes was examined as well.

For brevity, a trend interruption will be referred to frequently as a“bounce,” and the ratio of times the exchange rate bounces to thenumber of actual hits will be referred to as the “bounce frequency.”In formal terms, the goal of the test described below is to ascertainstatistically whether the bounce frequencies for the published supportand resistance levels are high, as claimed by technical analysts.

Statistical Methodology

The statistical test evaluates whether or not published support andresistance levels are able to identify points of likely trendinterruptions, as claimed by technical analysts. The central or “null”hypothesis is that the published levels have no special ability toidentify such points. I begin with a summary of the methodology andthen present details.

Summary

The statistical methodology used to test this null hypothesis is aspecific application of the bootstrap technique (Efron 1979, 1982). Toapply this technique, I first calculate bounce frequencies for each firmfor each month in the sample. I then build a statistical representationof what bounce frequencies for the published support and resistancelevels would look like if the null hypothesis were true. In the presentcontext, this representation is constructed by first creating 10,000sets of artificial support and resistance levels for each day. For each ofthese artificial sets of support and resistance levels, I then calculatebounce frequencies for each month, using the criteria for hits andbounces listed above.

At this point, I have twenty-eight bounce frequencies for each firm,one for each month of the sample, and twenty-eight average bouncefrequencies for the artificial support and resistance levels. In the finalstep of the test, I determine the number of months in which thebounce frequency for a given firm exceeds the average bouncefrequency for artificial levels. If this number of months is quite high, Iconclude that the published support and resistance levels have someability to predict intraday trend interruptions. Additional details onthis methodology are presented below.

Calculating Artificial Support and ResistanceLevels

For each day, the artificial support and resistance levels are chosen atrandom from exchange rates within a certain range of the day’sopening rate. The range for each month is based on the exchangerate’s actual behaviour, as follows: for a given month, I calculate thegap between the opening rate and intraday highs and lows for eachday. The absolute value of the largest of these gaps is used as the

Exchange rate

ypo e ca c a ge a e a s

Support

A “hit” if rate crosses support +

0.01 percent

A “bounce” if rateis greater than support after

fifteen minutes

t t + 15Time

Support + 0.01 percent

HYPOTHETICAL EXCHANGE RATE PATHS



range for calculating artificial support and resistance levels for thatmonth.

For each day, twenty artificial support and twenty artificial resistancelevels are calculated using the following algorithm:

Rti = Ot + bti range ,Sti = Ot – ati range .

Here, t represents time, Sti(Rti) is the ith artificial support (resistance)level, Ot is the day’s opening rate, ati and bti are random numbersgenerated from a uniform distribution over [0,1], and range is therange for that month. These levels are then rounded off so that theyhave the same number of significant digits to the right of the decimalpoint as actual quoted exchange rates.3

The Statistical Test

The statistical test is based on comparing the bounce frequencies forthe published support and resistance levels (BP) with the averagebounce frequencies for the artificial levels (BA), month by month. Tounderstand the test intuitively, suppose published support andresistance levels provide no more information than arbitrary levels. Inthis case, BP should not consistently be higher or lower than BA.However, if published support and resistance levels can predict pointsof likely trend interruptions, as claimed by technical analysts, then BP

should usually be higher than BA.

This idea can be formalized into a rigorous statistical test. Thecomparison for each month can be viewed as a “Bernoulli trial,” inwhich a random event occurs with a certain probability. The randomevent here would be BP>BA. Under the null hypothesis that publishedlevels have no special predictive power, the likelihood of that event is50 percent. Over the entire twenty-eight month sample, if it is truethat published levels are not informative, the chance that BP>BA forany given number of months will conform to the binomialdistribution. This distribution is symmetrical around a single peak atfourteen months, where the probability is about 15 percent.

To understand how to use this distribution, suppose we find thatBP>BA in twenty of the twenty-eight months of the sample. Wemight naturally ask: Would it be unusual to get such an extremeoutcome if the published levels are truly not informative? Moreconcretely, what is the likelihood, if the published levels are notinformative, of finding BP>BA in twenty or more of the twenty-eightmonths of the sample? This likelihood is the area under the tail of thedistribution to the right of the number 20. This is a very smallnumber: in fact, it is 1.8 percent.

The likelihood of finding BP>BA in twenty or more of the twenty-eight months, under the assumption that published levels are notinformative, is called the “marginal significance level” associated withthe number 20.4 If the marginal significance level of some result issmaller than 5 percent, it is consistent with standard practice in theliterature to conclude that the published numbers are better thanarbitrary numbers at predicting trend interruptions. Such a result issaid to be “statistically significant.”

To summarize our example: it would be extremely unusual to findthat BP>BA in twenty or more of the twenty-eight months if thepublished support and resistance levels were truly not informative. Infact, we would realize such an outcome only 1.8 percent of the time.Since 1.8 percent falls below the common critical value of 5 percent,we would conclude that the predictive power of the published levelsexceeds that of the arbitrary levels to a statistically significant degree.

RESULTS

Consistent with the market’s conventional wisdom, exchange ratesbounced quite a bit more frequently after hitting published supportand resistance levels than they would have by chance. Exchangerates bounced off arbitrary support and resistance levels 56.2 percentof the time on average.5 By contrast, they bounced off the publishedlevels 60.8 percent of the time on average (Table 8). Looking moreclosely, we find that in all sixteen firm-currency pairs, average bouncefrequencies for published levels (across the entire sample period)exceeded average bounce frequencies for artificial levels.

The month-by-month breakdown shows that for most firm-currencypairs, bounce frequencies for the published levels exceeded average

bounce frequencies for artificial levels in twenty or more months. Asnoted above, these outcomes would be extremely unlikely if thesupport and resistance levels were truly not informative. Morerigorously, the marginal significance levels indicate that the results arestatistically significant at the 5 percent level for all but three firm-currency pairs.

The firms’ ability to predict turning points in intraday trends seems tohave been stronger for the yen and weaker for the mark and thepound. On average, bounce frequencies for published support andresistance levels exceeded those for arbitrary levels by 4.2 percentagepoints for the mark, 5.6 percentage points for the yen, and 4.0percentage points for the pound. This relative ranking wasmaintained fairly consistently for individual firms.

Although all six firms seem to have the ability to predict exchangerate bounces, their performance varied considerably. The bouncefrequencies of the best and worst firms differ by 4.0 percentagepoints on average. At one extreme, Firm 1’s support and resistancelevels for the yen had a bounce frequency 9.2 percentage pointshigher than that of the arbitrary levels.

Differences across firms are evaluated statistically in Table 9. Firm 1 isclearly the best overall: it had the highest bounce frequency for twoof the three currencies and the second-highest bounce frequency forthe third currency. Furthermore, the differences between Firm 1 andthe other firms are statistically significant at the 5 percent level inseven of the thirteen possible firm-to-firm comparisons and arestatistically significant at the 10 percent level in another comparison.Firm 5 did quite well for the mark, but did not do noticeably well forthe other two currencies. No firm was consistently worst.

TABLE 8

Ability of Support and Resistance Levels to PredictInterruptions of Intraday Exchange Rate Trends

Artificial Levels Published by Firm

Levels 1 2 3 4 5 6

Bounce Frequency(Number of Hits)

German 54.9 60.1 56.6 58.0 58.5 62.0 59.3mark (6,291) (4,102) (8,111) (3,570) (1,262) (2,296)

Japanese 57.3 66.5 63.6 62.3 60.7 61.6 62.6yen (4,558) (3,874) (6,271) (2,679) (859) (1,396)

British 56.3 63.0 58.8 59.6 60.0pound (5,409) (3,920) (6,056) (1,039)

Months BP>BA/Total Months(Marginal Significance)

German 24/28 17/28 21/28 20/27 20/26 19/28mark (0.000) (0.172) (0.006) (0.010) (0.005) (0.044)

Japanese 26/28 24/27 22/28 23/27 15/23 20/27yen (0.000) (0.000) (0.002) (0.000) (0.105) (0.010)

British 27/28 19/28 24/28 16/26pound (0.000) (0.044) (0.000) (0.163)


Notes: The table compares the ability of published support and resistancelevels to predict intraday trend interruptions with the distribution ofpredictive ability for 10,000 sets of arbitrary support and resistance levels.The measure of predictive ability is based on the “bounce frequency,” or thenumber of times the exchange rate stopped trending after reaching support orresistance levels compared with the total number of times the rate actuallyreached such levels.

The table shows the bounce frequency for published and artificial support andresistance levels, the number of hits, the number of months in which thebounce frequency for published levels (BP) exceeds the bounce frequency forartificial levels (BA), and the marginal significance of this number of monthsunder the null hypothesis that published support and resistance levels are notinformative. “Total months” varies across firms and currencies becauseoccasionally a firm contributed too few support and resistance levels to haveany hits at all.


TABLE 9

Differences in Firms’ Ability to Predict Exchange RateBounces

Firm A

Firm B 1 2 3 4 5

German mark2 3.5**3 2.1** -1.44 1.5 -2.0 -0.6***5 -2.0** -5.5** -4.1*** -3.56 0.8 -2.7 -1.3** -0.7 2.8

Japanese yen2 2.8*3 4.1*** 1.34 5.7*** 2.9 1.65 4.9 2.0 0.7 -0.96 3.8 1.0 -0.3 -1.9 -1.0

British pound2 4.2***3 3.5** -0.8*5 3.1* -1.2 -0.4


Notes: The table compares different firms’ ability to predict intraday trendinterruptions in exchange rates. The measure of predictive ability is based onthe “bounce frequency,” or the number of times the exchange rate stoppedtrending after reaching support or resistance levels compared with the totalnumber of times the rate actually reached such levels. The table presents thedifference between bounce frequencies (measured as Firm A minus Firm B).

* Statistically significant at the 10 percent level. ** Statistically significant at the 5 percent level. *** Statistically significant at the 1 percent

Robustness

These results are robust to changes in the test methodology.6 Theyare not changed qualitatively if a hit is defined more broadly or morenarrowly (as described earlier) or if one looks thirty minutes ratherthan fifteen minutes beyond a hit. The results are also unchanged ifone splits the sample into morning and afternoon sessions (where themorning session is defined to include positions entered before noon).

Interestingly, the results change somewhat if the sample is split inhalf chronologically. During the first half of the sample period, whenvolatility was fairly low by historical standards, bounce frequencieswere statistically significant for published levels in all but one case. Inthe second half, when volatility returned to more normal levels, firms’bounce frequencies still exceeded those for the artificial levels, butthe differences were no longer statistically significant in half thecases.7 This outcome is consistent with the market’s conventionalwisdom that rates tend to “range trade” in periods of low volatility,thus making this type of trend reversal more common.

Quotes versus Transaction Prices

At this point, it is possible to discuss more fully the potentialimplications of the differences noted by Danielsson and Payne (1999)between exchange rate quotes and actual transaction prices. The firstimportant difference they note is that quotes tend to be more volatilethan actual transaction prices. This should not be critical here,because these results concern the direction of price changes, not theirmagnitude.

Second, Danielsson and Payne (1999) find that quotes tend to benegatively autocorrelated while transaction prices are not. In theory,this could affect the absolute frequency of bounces presented inTable 8. Fortunately, the important qualitative conclusions of thepaper are based on the difference between bounce frequencies forpublished and simulated levels, rather than the absolute size of thosebounce frequencies. Furthermore, the negative autocorrelation inquote data may largely have dissipated by the end of the fifteen-minute horizon of interest. The reason is that if the exchange rate isrequired to reach the actual level rather than some nearby level to

achieve a hit, bounce frequencies in the simulated support andresistance levels fall slightly short of 50 percent. If negativeautocorrelation at the fifteen-minute horizon were an issue, thatproportion would presumably exceed 50 percent.

Duration of Predictive Power

Could an analyst using support and resistance levels published todayhave any success predicting intraday trend reversals one week fromtoday? The answer seems to be yes. Five days after their publication,bounce frequencies for our six firms still exceeded those fromarbitrary levels for all firms and currencies, and the differences werestatistically significant in nine of the sixteen cases (Table 10). Notsurprisingly, the published levels were not quite as useful atpredicting intraday trend reversals five days after publication as theywere on their actual publication day. On average, five days afterpublication, rates bounced at published levels 1.7 percentage pointsless frequently than they did on the actual publication day.8

TABLE 10

Ability of Support and Resistance Levels to PredictInterruptions of Intraday Exchange Rate Trends after FiveTrading Days

Artificial Levels Published by Firm

Levels 1 2 3 4 5 6

Bounce Frequency(Number of Hits)

German 53.8 58.0 59.0 57.5 56.9 58.3 55.5mark (4,078) (2,464) (5,902) (2,930) (1,004) (1,148)

Japanese 55.4 62.5 60.5 61.2 59.5 59.7 64.3yen (3,531) (2,112) (4,392) (2,373) (703) (635)

British 54.0 57.5 59.0 56.5 57.4pound (3,661) (2,400) (4,641) (622)

Months BP>BA/Total Months(Marginal Significance)

German 22/27 22/27 21/27 16/26 16/23 16/25mark (0.001) (0.001) (0.003) (0.163) (0.047) (0.115)

Japanese 21/27 20/27 22/27 19/26 12/22 16/24yen (0.003) (0.010) (0.001) (0.014) (0.416) (0.076)

British 19/27 17/27 17/27 13/19pound (0.026) (0.124) (0.124) (0.084)


Notes: The table shows the results of using published support and resistancelevels to predict intraday trend interruptions five business days after thelevels’ publication date. The measure of predictive ability is based on the“bounce frequency,” or the number of times the exchange rate stoppedtrending after reaching support or resistance levels compared with the totalnumber of times the rate actually reached such levels.

The table shows the bounce frequency for published and artificial support andresistance levels, the number of hits, the number of months in which thebounce frequency for published levels (BP) exceeds the bounce frequency forartificial levels (BA), and the marginal significance of this number of monthsunder the null hypothesis that published support and resistance levels are notinformative. “Total months” varies across firms and currencies becauseoccasionally a firm contributed too few support and resistance levels to haveany hits at all.

The Power of Agreement

If many analysts agree that a particular level is likely to be important,does this imply that the level is more likely than others to beimportant? I addressed this question by comparing the predictivepower of support and resistance levels provided by more than onefirm (“agreed levels”) on a given day with the predictive power ofsupport and resistance levels provided by only one firm.

As shown in Table 11, the bounce frequencies associated with agreedlevels are quite close to the bounce frequencies associated with levels


provided by just one firm. Although the agreed levels tend to havehigher bounce frequencies, the differences are generally notstatistically significant. The one difference found to be statisticallysignificant implies that agreed levels have less predictive power thanother levels. Overall, these results suggest that, if agreed levels doprovide additional predictive power, the benefit is too small to be ofmuch practical importance.9

TABLE 11

Is There Power in Agreement?

Narrow Agreement Broad Agreement

German Japanese British German Japanese BritishMark Yen Pound Mark Yen Pound

Agreed levels 58.0 65.9 63.1 59.1 65.2 61.7

Other levels 59.0 63.7 60.6 58.6 63.5 60.3

Monthsagreed>other/totalmonths 10/28 15/27 17/28 11/28 14/27 14/28

MarginalSignificance 0.96 0.35 0.17 0.91 0.50 0.57


Notes: The table compares the predictive ability of support and resistancelevels on which two or more firms agree (“agreed levels”) with the pre-dictive ability of support and resistance levels provided by only one firm.The measure of predictive ability is based on the “bounce frequency,” or thenumber of times the exchange rate stopped trending after reaching supportor resistance levels compared with the total number of times the rateactually reached such levels. If agreed levels were better able to predictintraday trend interruptions, the numbers would be positive and statisticallysignificant. Two levels were in “narrow agreement” if they were within 2points of each other; they were in “broad agreement” if they were within 5points of each other.

Reliability of Estimated “Strengths”

Three of the firms regularly provided estimates of the “strength” oftheir published support and resistance levels. For example, levelscould be categorized as having strength numbers “1,” “2,” or “3,”with 3 being the strongest. The strength of a particular level can beinterpreted as a crude measure of the likelihood that an exchangerate that arrives at the level will actually bounce off it.

TABLE 12

The Meaning of Reported Strength Ratings

Comparison ofStrengths 1 and 2 German Mark Japanese Yen British Pound

Firm 1 -2.3** -4.1*** -6.510/27 8/26 12/28(0.04) (0.01) (0.17)

Firm 2 -2.7** -5.5 -3.9*10/27 10/25 10/26(0.04) (0.11) (0.08)

Firm 3 2.4 0.2 -0.114/25 11/24 11/23(0.35) (0.27) (0.34)


Notes: The table evaluates whether support and resistance levels consideredsomewhat strong by their publishers actually predict intraday trendinterruptions better than those considered least strong. The measure ofpredictive ability is based on the “bounce frequency,” or the number of timesthe exchange rate stopped trending after reaching support or resistance levelscompared with the total number of times the rate actually reached suchlevels. Strength 1 corresponds to the support and resistance levels at whichtrend interruptions are least likely; strength 2 corresponds to support andresistance levels at which trend interruptions are more – but not most – likely.

For each firm listed on the left side of the table, the first row of numbersrepresents the difference between the predictive ability of support and

resistance levels of the two different strengths. If the reported strength levelswere reliable, then the numbers would be positive and significant. The firstnumber in each second row represents the months in which the bouncefrequency for strength 2 actually exceeded the bounce frequency for strength1; the second number in each row (following the slash) represents thenumber of months in which the comparison was valid; the third row ofnumbers gives the marginal significance of the second row under the nullhypothesis that there is no difference between the two sets of numbers.

* Statistically significant at the 10 percent level.** Statistically significant at the 5 percent level. *** Statistically significant at the 1 percent level.

Were the estimated strengths of support and resistance levelsmeaningful? To answer this question, I examined the relativefrequency of bounces off support and resistance levels in threestrength categories: (1) least strong, (2) somewhat strong, and (3)strongest. Unfortunately, in many months there were fewobservations in strength category 3, so the only reliable comparisonwas between categories 1 and 2.

Results for this comparison are shown in Table 12, where thereported differences would be positive and statistically significant ifthe strength categories were meaningful. In fact, the reportedstrength levels seem to have no consistent correspondence with theactual frequency with which exchange rates bounced off support andresistance levels. All but two of the differences are negative, and thethree that are statistically significant are negative. In short, publishedestimates of the strength of the levels do not seem to be useful.

CONCLUSION

This article has examined the predictive power of support andresistance levels for intraday exchange rates, using technical signalspublished by six active market participants from January 1996through March 1998. The statistical tests, which use the bootstraptechnique (Efron 1979, 1982), cover support and resistance levels forthree currency pairs: dollar-mark, dollar-yen, and dollar-pound.

The results indicate that intraday exchange rate trends wereinterrupted at published support and resistance levels substantiallymore often than would have occurred had the levels been arbitrarilychosen. This finding is consistent across all three exchange rates andacross all six firms studied. The predictive power of published supportand resistance levels varies considerably across firms and acrossexchange rates. It lasts at least one week. The strength estimatespublished with the levels are not meaningful. These results are highlystatistically significant and are robust to alternative parameterizations.

The predictive power of support and resistance levels has manypossible sources, some of which are discussed in Osler (2000).Central bank intervention has been cited as a possible source of thepredictive power of other technical trading strategies (Szakmary andMathur 1997; LeBaron 1999). However, central bank interventionseems unlikely to be an important source of the predictive power ofsupport and resistance levels since there was no reported interventionfor the mark and the pound during the sample period. Other possibleexplanations include clustered order flow, which receives support inOsler (2000), and self-fulfilling prophecies.

The ability of support and resistance levels to predict trend reversalssuggests that the intraday currency markets may not be fullyefficient. To investigate this possibility, it would be natural to examinewhether traders could profit from these predictable bounces on afairly consistent basis. If it were indeed profitable to trade on thesereadily available technical signals, there would seem to be someincentive for rational traders to trade the profits away. This would bean appropriate subject for future research. It might also beappropriate to examine the claim of technical analysts that trendstypically are sustained once support and resistance levels are“decisively” crossed.

ENDNOTES

1. Support and resistance levels are related to but not identical totrading ranges. A trading range has just one support level and oneresistance level. The firms examined here usually provided multiplesupport levels and multiple resistance levels each day.

2. These results are available from the author upon request.


3. That is, all artificial support and resistance levels for the mark andthe pound had the form x.xxxx00, while all artificial support andresistance levels for the yen had the form xxx.xx00000.

4. For some firms, there were few support and resistance levels insome months, and thus few hits and bounces. These months wereexcluded from the sample for those firms.

5. If intraday exchange rates followed a random walk, the tendencyto bounce would, in the abstract, be about 50 percent. Thetendency to bounce in the actual data exceeds this benchmark fortwo reasons. First, changes in the actual and the simulated datahave a fairly strong negative first-order autocorrelation, as notedby Goodhart and Figliuoli (1991). Second, to “bounce,” theexchange rate must first reach a level a little above (below) theactual support (resistance) level, and then remain above (below)the actual support (resistance) level for a certain interval. Thus, theexchange rate can continue trending slightly after officially hittingthe level yet still be considered as having “bounced.”

6. Results from these sensitivity tests are available from the authorupon request.

7. The standard deviation of daily exchange rate changes rose byone-third on average between the first and second halves of thesample period. In the first half, these standard deviations were0.199, 0.216, and 0.260 for the mark, yen, and pound,respectively. In the second half, the corresponding standarddeviations were 0.252, 0.362, and 0.277 (all figures E+3).

8. The reader may also be interested to know whether the tendencyof support and resistance levels to be selected as round numbersor as local highs/lows has any influence on the levels’ predictivepower. In Osler (2000), I examine whether round numbers or localminima/ maxima (both of which are known to be sources ofpublished support and resistance levels) have predictive power forexchange rate bounces. I find that they do, from which I concludethat at least some of the predictive power in the published levelscomes from the firms’ tendency to choose these types of numbers.I also show that the size of the typical move following a hit differssubstantially between the published levels of some firms and theartificial levels. I conclude from this that round numbers and localminima/maxima do not incorporate as much information aboutintraday trend reversals as do some published support andresistance levels.

9. It would be desirable here to weight the advising firms by theirorder flow. However, order information is very closely guarded bythe firms in question. Furthermore, some of the firms do notactually take orders.

The views expressed in this paper are those of the author and do notnecessarily reflect the position of the Federal Reserve Bank of NewYork or the Federal Reserve System. The Federal Reserve Bank ofNew York provides no warranty, express or implied, as to theaccuracy, timeliness, completeness, merchantability, or fitness for anyparticular purpose of any information contained in documentsproduced and provided by the Federal Reserve Bank of New York inany form or manner whatsoever.

Carol Osler is a senior economist at theFederal Reserve Bank of New York.

This article originally appeared in the Federal Research Bank of NewYork’s Economic Policy Review. The author thanks Franklin Allen,Alain Chaboud, and Charles Jordan for thoughtful comments, GijoonHong for excellent research assistance, and two anonymous referees.

The views expressed are those of the author and do not necessarilyreflect the position of the Federal Reserve Bank of New York or

the Federal Reserve System.

REFERENCES

Allen, Helen, and Mark P. Taylor. 1992. “The Use of TechnicalAnalysis in the Foreign Exchange Market.” Journal of InternationalMoney and Finance 11, no. 3 (June): 304-14.

Andersen, Torben, and Tim Bollerslev. 1998. “Deutsche Mark–DollarVolatility: Intraday Activity Patterns, Macroeconomic Announce-

ments, and Longer Run Dependencies.” Journal of Finance 53(February): 219-65.

Arnold, Curtis M. 1993. Timing the Market: How to Profit in Bull andBear Markets with Technical Analysis. Chicago: Probus PublishingCompany.

Brock, W., et al. 1992. “Simple Technical Trading Rules and theStochastic Properties of Stock Returns.” Journal of Finance 48(December): 1731-64.

Chang, P. H. Kevin, and C. L. Osler. 1998. “Methodical Madness:Technical Analysis and the Irrationality of Exchange RateForecasts.” Economic Journal 109, no. 458 (October): 636-61.

Cheung, Yin-Wong, and Menzie Chinn. 1999. “Traders, MarketMicrostructure, and Exchange Rate Dynamics.” Unpublishedpaper, University of California at Santa Cruz, January.

Cheung, Yin-Wong, and Clement Yuk-Pang Wong. 1999. “ForeignExchange Traders in Hong Kong, Tokyo, and Singapore: A SurveyStudy.” Advances in Pacific Basin Financial Markets 5: 111-34.

Curcio, Richard, et al. 1997. “Do Technical Trading Rules GenerateProfits? Conclusions from the Intraday Foreign Exchange Market.”Unpublished paper, London School of Economics.

Danielsson, Jón, and Richard Payne. 1999. “Real Trading Patternsand Prices in Spot Foreign Exchange Markets.” Unpublished paper,London School of Economics, March.

Dooley, Michael P., and Jeffrey Shafer. 1984. “Analysis of Short-RunExchange Rate Behaviour: March 1973 to November 1981.” InDavid Bigman and Teizo Taya, eds., Floating Exchange Rates andthe State of World Trade and Payments, 43-70. Cambridge, Mass.:Ballinger Publishing Company.

Edwards, Robert, and John Magee. 1997. Technical Analysis of StockTrends. 5th ed. Boston: John Magee.

Efron, B. 1979. “Bootstrap Methods: Another Look at the Jackknife.”Annals of Statistics 7: 1-26.

———. 1982. The Jackknife, the Bootstrap, and Other ResamplingPlans. Philadelphia: Society for Industrial and Applied Mathematics.

Goodhart, C., and L. Figliuoli. 1991. “Every Minute Counts inFinancial Markets.” Journal of International Money and Finance10, no. 1 (March): 23-52.

Goodhart, C., T. Ito, and R. Payne. 1996. “One Day in June 1993: AStudy of the Working of the Reuters 2000-2 Electronic ForeignExchange Trading System.” In Jeffrey Frankel, Gianpaolo Galli, andAlberto Giovannini, eds., The Microstructure of Foreign ExchangeMarkets, 107-79. National Bureau of Economic ResearchConference Report Series. Chicago: University of Chicago Press.

Hardy, C. Colburn. 1978. The Investor’s Guide to Technical Analysis.New York: McGraw-Hill.

Kaufman, P. 1978. Commodity Trading Systems and Methods. NewYork: Ronald Press.

LeBaron, Blake. 1999. “Technical Trading Rule Profitability of ForeignExchange Intervention.” Journal of International Economics 49(October): 125-43.

Levich, R., and L. Thomas. 1993. “The Significance of TechnicalTrading Rule Profits in the Foreign Exchange Market: A BootstrapApproach.” Journal of International Money and Finance 12(October): 451-74.

Lo, Andrew W., et al. 2000. “Foundations of Technical Analysis:Computational Algorithms, Statistical Inference, and EmpiricalImplementation.” NBER Working Paper no. 7613, March.

Lui, Yu-Hon, and David Mole. 1998. “The Use of Fundamental andTechnical Analyses by Foreign Exchange Dealers: Hong KongEvidence.” Journal of International Money and Finance 17, no. 3(June): 535-45.

Menkhoff, Lukas, and Manfred Schlumberger. 1995. “PersistentProfit-ability of Technical Analysis on Foreign Exchange Markets?”Banca Nazionale del Lavoro Quarterly Review 193 (June): 189-

One of the characteristics of a bear market is that support levels donot work but resistance levels do and this is what we are seeing inthe current sell-off. Ominously the Dow Jones has traced out atdiamond top which gives a target of 8,000.

On the S&P 500 the next big support level is 980 and below that 930.

The most useful guide to the sentiment in the US markets is,however, currently being given by the NYSE Composite index. If the575 level gives way, the next support level is 475.

Our main message is to steer well clear of TMT stocks at themoment. But it is not all bad news. The food retailers are still lookinggood and some of the drug companies are still in uptrends on boththe price and relative charts.

Pharmaceutical sector relative to the UK market

Nick Glydon and Nicola Merrell, technical analysts, JP Morgan


DOW JONES INDEX

S&P COMPOSITE INDEX

NYSE COMPOSITE INDEX

PHARMER/TOTMKER

The TMT Massacreby Nick Glydon & Nicola Merrell

continued from page 17

REFERENCES

215. Murphy, John J. 1986. Technical Analysis of the Futures Market:A Comprehensive Guide to Trading Methods and Applications.New York: Prentice Hall.

Osler, C. L. 2000. “Are Currency Markets Efficient? Predictable TrendReversals in Intraday Exchange Rates.” Unpublished paper, FederalReserve Bank of New York, February.

Pring, M. 1991. Technical Analysis Explained: The SuccessfulInvestor’s Guide to Spotting Investment Trends and Turning Points.3rd ed. New York: McGraw-Hill.

Shabacker, R. W. 1930. Stock Market Theory and Practice. New York:B. C. Forbes Publishing Company.

Sklarew, Arthur. 1980. Techniques of a Professional CommodityChart Analyst. New York: Commodity Research Bureau.

Sweeney, R. J. 1986. “Beating the Foreign Exchange Market.”Journal of Finance 41 (March): 163-82.

Szakmary, Andrew, and Ike Mathur. 1997. “Central Bank Interventionand Trading Rule Profits in Foreign Exchange Markets.” Journal ofInternational Money and Finance 16, no. 4 (August): 513-35.


Positioning of traders is often used as a guide for a potential trend inmarket prices. Speculative traders, whose views are cast with an eyeon the outright direction of the market, are of particular interest. Astretched long or short position among speculative traders couldforetell an abrupt reversal in the direction of the market in order torebalance demand and supply.

We find that changes in market positioning are highly correlated withfluctuations in foreign-exchange markets, but that positioningprovides virtually no leading information on the direction of themarket. We also illustrate that simple trading rules based on thepositioning of speculative traders are not very profitable on average.It stands to reason that the positioning of speculative traders onlymatters when taken in conjunction with other technical indicatorsand the state of fundamentals.

Market Positioning – Measurement

We measure the positioning of speculative traders through theCommitment to Traders’ report published by the Commodity FuturesTrading Commission (CFTC). Once a week the CFTCdecomposes the open interest on futures contractsinto long and short positions by the type of trader.Reportable positions are divided into two categories:commercial traders, who are involved for hedgingpurposes, and non-commercial traders, who arespeculating on the outright direction of the market.

We focus on the difference between long and shortpositions of speculative traders – the ‘net long’position – and its properties with market directionover a short one-month horizon. We examinepositioning of futures in the Euro, Yen, Swiss Franc,Pound Sterling, C$and A$all versus the US$. For thepurposes of historical analysis we normalise the netlong position of speculative traders by the openinterest on the futures contract. We also create a USDollar sentiment index, based on the average positionacross these currency futures weighted by openinterest.

Correlation Structure – Positioning Matters

We first consider the correlation between changes in the positioning

of speculative traders and changes in the spot exchange rate. Table 1(on the next page) illustrates this relationship over three differentperiods: the coincident change in the two variables, the laggedchange in market positioning, and the lead in positioning changes.The lag-lead relationship is presented over a four-week horizon,which roughly coincides with the half-life of positioning in currencyfutures. A positive correlation implies that a rise in net long positionsis associated with an appreciation in the currency.

There is a strong coincident correlation between changes in marketpositioning and changes in the exchange rate. The averagecorrelation across the six currencies is 48%, suggesting that a rise inlong positions accompanies a coincident appreciation in the exchangerate. However, there is virtually no correlation between laggedchanges in market positioning and changes in the currency. Theaverage correlation across the six currencies is merely 2%, rangingfrom a high of 8%for the Yen to a low of –8% for the A$. Thenegative correlation between lead changes in positioning and thecurrency is somewhat stronger (–13%) and is consistent with tradersfollowing profitable trading rules. Long positions are reduced after anappreciation in the currency and vice versa.

Positioning as a Signalling Device

We next examine the historical performance ofmarket positioning as a signalling device for thefuture direction of the foreign exchange market. Toavoid noisy signals we focus on ‘large’ marketpositions, which we define as more than onestandard deviation away from the historical average.The attached table illustrates the percentage of timesthat a large long position led to an appreciation in thecurrency after one week and one month (vice versa).A value of 50% means that a large long position wasequally likely to be followed by a rise in the currencyas a fall. Likewise, values below 50% imply that alarge long position accompanied a fall in the currency– a reverse signal on the future direction of themarket.

The signals across the six currency pairs are relativelyweak. The average signal across all of the positionsconsidered was a modest 51%, which suggests thatmarket positioning is not much better than a coin flip

IMM Positions and Currency TrendsGoldman Sachs Economic Research


on average. The irregularity of the signals is also noteworthy. Of the24 signals examined, 12 were greater than 50% while 12 were at orbelow 50%. Speculative traders were generally more successful withshort than long positions. However, there is no good reason tobelieve that accuracy should be systematically better with shortsversus longs.

Profitability of Trading Rules

Thus far we can say that there is no correlation between thepositioning of speculative traders and the future direction of themarket. However, we have yet to consider the size of the marketmovements. Even if signals are right only 50% of the time, thisprovides no information on the average profitability of positions andtrading rules based on this information. To address this issue weexamine the returns of a trading rule that takes the opposite view ofspeculative traders – long when speculators are short and vice versa.We focus once again on large positions as the unwinding of crowdedtrades could require an abrupt reversal in prices to rebalance demandand supply. Hence the trading rule is triggered when the position ofspeculative traders is more than one standard deviation from thehistorical average.

Table 3 below summarises the one-month returns for this reversetrading rule. The trading rule was profitable 75% of the time in theaggregate (9 out of 12). However, the profits were very small andnot significantly different from zero most of the time. Only 3 out of12 times was the trading rule right and profits significant. The tradingrule was most successful for the C$, with profits significantly differentfrom zero for both short and long positions, while least successful forthe Yen, resulting in losses for both positions. The fact that rigidtrading rules are not very profitable should not be surprising. Afterall, the timing of changes in market positioning is very difficult todetermine a priori.

Positioning Does Matter

The fact that there is not a simple leading relationship betweenpositioning and currency markets does not suggest that positioning ofspeculative traders should be ignored. Combined with other indicators,like risk reversals for instance, positioning can be judged against macrofundamentals as a guide for trading views. When deviations betweensentiment and fundamentals are wide, one would expect a turn inpositioning and a coincident move in currency markets.

These discrepancies are indeed quite substantial in the currentenvironment. Sentiment remains bullish for the US$, with speculativetraders holding net long positions against all of the currenciesexamined other than the Euro. Taken together, the currency sentimentindex is at its most bullish posture for the US Dollar since the end oflast November. Coupled with bearish valuation, the mix of positioningand fundamentals support a strategic short position in the US Dollar.

Acknowledgement:Source: ‘The Weekly Analyst’, 6 March 2001, pp 8-10. GoldmanSachs Economic Research, London.

This article was submitted by Michael Feeny

Market Positioning*

+4 Weeks Current -4 Weeks

Euro -21% +57% +4%

Yen -10% +56% +8%

CHF -11% +59% +6%

GBP -17% +54% -1%

C$ -14% +57% +1%

A$ -5% +7% -8%

Avg -13% +48% +2%

Table 1: Correlation of Positioning and Currencies

*Correlation coefficient between the four-week change in positioning andfour-week % change in the currency period.

Short Long

1 Week 4 Weeks 1 Week 4 Weeks

Euro 51 56 44 45

Yen 63 50 54 64

CHF 59 53 50 42

GBP 45 38 50 39

C$ 56 64 55 42

A$ 54 66 42 45

Avg 55 55 49 46

Avg for position 55 48

Avg for all 51

Table 2: Signalling Performance*

*Percentage of times that a ‘large’ long position followed an appreciation inthe currency (vice versa).

Euro Yen CHF GBP C$ A$

Short Long Short Long Short Long Short Long Short Long Short Long

Rule ✔ ✔ ✘ ✘ ✔ ✔ ✘ ✔ ✔ ✔ ✔ ✔

+0.1% +0.2% -0.2% -1.1% +0.2% +0.4% -0.5% +0.4% +0.4% +0.2% +0.7% +0.1%

Profit (0.3) (0.6) (0.5) (3.3) (0.5) (1.2) (1.6) (1.4) (2.7) (1.7) (2.4) (0.4)

Significant ✘ ✘ ✘ ✔ ✘ ✘ ✘ ✘ ✔ ✔ ✔ ✘

Table 3: Profitability of Trading Rules

*Trading rule takes the opposite view of speculative traders. The rule is triggered when positions are greater than one standard deviation from the average. Profit isthe one-month return, t-stat is in parentheses.

Documents

Market Technician No40