Taylor Menkhoff Technical Analysis

8/6/2019 Taylor Menkhoff Technical Analysis

1/37

Journal of Economic LiteratureVol. XLV (December 2007), pp. 936972

The Obstinate Passion ofForeign Exchange Professionals:Technical Analysis

LUKAS MENKHOFF AND MARK P. TAYLOR

Technical analysis involves the prediction of asset price movements from inductiveanalysis of past movements. We establish a number of stylized facts, including that

technical analysis is widespread in the foreign exchange market and that it may beprofitable. We then analyze four arguments that have been put forward to explainthis: that the market may not be fully rational; that technical analysis may exploit theinfluence of official interventions; that it may be an efficient form of information pro-cessing; and that it may inform on nonfundamental influences. While each may have

some validity, the latter is the most plausible.

As for the foreign exchange, it is almost as romantic as young love, and quite asresistant to formulae.

H. L. Mencken

936

1. Introduction

Technical analysis involves the predictionof future exchange rate (or other asset-price) movements from an inductive analysisof past movements, using either qualitativemethods (e.g., the recognition of certain pat-terns in the data from visual inspection of atime-series plot) or quantitative techniques

(e.g., based on an analysis of moving aver-ages), or a combination of both. For many

professional economists, the widespread,continuing use of these techniques in theforeign exchange market (Mark P. Taylorand Helen L. Allen 1992; Yin-Wong Cheungand Menzie D. Chinn 2001) is somewhatpuzzling since technical analysis eschewsscrutiny of economic fundamentals andrelies only on information on past exchangerate movements that, according to the weak-est notion of market efficiency, should alreadybe embedded in the current exchange rate,making its use either unprofitable orimplying that any positive returns that aregenerated are accompanied by an unac-ceptable risk exposure.1 On the other hand,despite an apparent emerging consensus

Menkhoff: University of Hannover. Taylor: Universityof Warwick, Barclays Global Investors, and the Centre forEconomic Policy Research. The authors are grateful toCarol Osler, Stephan Schulmeister, Roger Gordon (theeditor), and five anonymous referees for helpful com-ments on an earlier version of this paper. Menkhoff grate-fully acknowledges financial support from the GermanResearch Foundation (Deutsche Forschungsgemein-schaft DFG).

1 In other words, the ratio of expected return to risk(the volatility of returns) is unacceptably low.


2/37

that fundamentals such as relative prices orrelative monetary velocity are capable ofexplaining long-term exchange rate move-ments (Alan M. Taylor and Taylor 2004),there is still no fundamentals-based

exchange rate model available that is capa-ble of forecasting exchange rate behaviorover the shorter term (e.g., over a horizonof twelve months or less: Jeffrey A. Frankeland Andrew K. Rose 1995; Taylor 1995b).Hence, the suggestion of Burton G.Malkiel (1996, p. 154), that technicalstrategies are usually amusing, often com-forting, but of no real value is perhaps alittle too dismissive, and this has been rec-ognized by a number of researchers.2

Indeed, over the past twenty years or so,international financial economists haveincreasingly turned their attention to thestudy of technical analysis in an attempt tounderstand both the behavior of foreignexchange rates and the behavior of foreignexchange market participants; so much so,in fact, that quite an extensive literature hasdeveloped on this topic.

Although the literature on the application

of technical analysis to the foreign exchangemarket is sufficiently developed to warrant asurvey of its own, however, the foreignexchange market cannot be viewed in totalisolation from other financial markets andso we shall occasionally stray into the litera-ture on the application of technical analysisto financial markets more generally and toequity markets in particular. The foreignexchange market differs from equity mar-kets in some important aspects, however.

First, total turnover in the global foreignexchange market is very high, at some 2,000billion U.S. dollars per day (Bank forInternational Settlements 2005), which isseveral times greater than the combineddaily turnover of the largest stock exchanges

in the world.3 Second, foreign exchange mar-kets consist almost entirely of professionaltraders (Michael J. Sager and Taylor 2006), sothat the impact of individual private investorsmay be neglected without loss of generality

(in contrast to equity marketssee, e.g., JohnM. Griffin, Jeffrey H. Harris, and SelimTopaloglu 2003). Third, the share of short-term interdealer trading is much higher in theforeign exchange market than it is in stockmarkets (Richard K. Lyons 2001). Finally, onecan probably say that there is less confidenceamong traders in models of fair value in theforeign exchange market compared to equitymarkets (Frankel and Rose 1995; Taylor1995b; John Y. Campbell, Andrew W. Lo, and

A. Craig MacKinlay 1996). The greater lack ofconsensus in models of fair value in the for-eign exchange market and the greater con-centration on shorter trading horizons mightsuggest that the use of technical analysis

would be more popular in the foreignexchange market, although the high propor-tion of professional traders and deeper liquid-ity of the foreign exchange market mightsuggest the opposite. However, we do not

want to dwell too long on the differencesbetween the foreign exchange market andequity markets, but rather to emphasize thefact that foreign exchange is increasingly seenas a separate asset class (Mark Snyder 2005).

In this paper, we provide a selective andcritical overview of the literature on technical

2 As we discuss in more detail below, Malkiels dismis-sive treatment of technical analysis is at odds with the evi-dence that technical analysis is widely used by financial

market traders.

3 Comparing spot market turnover yields a ratio of aboutthree in favor of the foreign exchange market compared to

equities. We calculate this by taking daily spot trading inequities and currencies in 200405 in the seven largestasset trading centers in the world. For equities, these werethe New York Stock Exchange, Nasdaq, the London StockExchange, the Tokyo Stock Exchange, Euronext (Paris,Brussels, Amsterdam, Lisbon), the Deutsche Brse(Frankfurt) and the Bolsas y Mercados Espaoles (Spain)(see the website www.world-exchanges.org), while for for-eign exchange the major centers were London, New York,Tokyo, Singapore, Frankfurt, Hong Kong, and Sydney(Bank for International Settlements 2005). In these cen-ters, daily spot market turnover was about 485 billion U.S.dollars in foreign exchange versus 160 billion U.S. dollars inequities; both figures represent more than 75 percent of

the respective world markets.

937Menkhoff and Taylor: The Obstinate Passion of Foreign Exchange Professionals


3/37

938 Journal of Economic Literature, Vol. XLV (December 2007)

4 In this paper, we use the terms technical analysisand chartist analysis and their derivatives largely as syn-onyms. This usage is not universal, although it is notunusual among practitioners (see, e.g., Callum Henderson2002) and has some precedence in the academic literature(e.g., Charles Goodhart 1988; Frankel and Kenneth A.Froot 1990b; Taylor and Allen 1992). It should be noted,however, that some practitioners and some authors differ-entiate chartist analysis as denoting the use of largely

visual analysis of charts and therefore see it as a subset ofthe methods denoted by technical analysis (see, e.g.,Christopher J. Neely 1997). See Allen and Taylor (1992)

for an outline of the origins of technical analysis.

5 A variant would be to use exponential moving aver-ages rather than simple arithmetic moving averages. Also,analysts may smooth the data prior to any analysis byapplying very short-run (e.g., one-day) moving averages orexponential moving averages to the data, in order to

reduce the effect of noise on trading signals.

analysis in the foreign exchange market. Atthe forefront of our discussion throughout isthe question as to why technical currencyanalysis is so intensively and widely used byforeign exchange professionals. As an organ-

izing device, we develop a set of stylized factsconcerning the importance and profitabilityof technical currency analysis. We then offerfour different kinds of explanation for thepersistent use of technical analysis and ana-lyze the supporting evidence in each case.

2. The Nature of Technical Analysis

Technical analysis or, as it is also some-times called, chartist analysis, is a set of

techniques for deriving forecasts of financialprices exclusively by analyzing the history ofthe particular price series plus perhapstransactions volumes.4 This analysis can beperformed in a qualitative form, relyingmainly on the analysis of charts of past pricebehavior and loose inductive reasoning, or itcan be strictly quantitative, by constructingtrading signals or forecasts through a quanti-tative analysis of time series data. In prac-tice, technical analysts often employ a

combination of both qualitative and quanti-tative techniques.

Clearly, technical analysis assumes thatprice developments display regular, recurringpatterns, otherwise such a purely inductivetechnique would be useless. A second condi-tion for the profitability of technical analysis isthat these patterns must last long enough,first to be recognized, and second to make upfor transaction costs and false signals.

The more qualitative aspect of technicalanalysis involves recognizing patterns in thedata that are thought to herald trend rever-sals, such as flags, head and shoulderspatterns, and so on (see, e.g., Allen and

Taylor 1992).The more widely used quantitative formsof technical analysis generally involve meth-ods such as moving averages in order toexploit trends in the data. They thus attemptto distinguish trends from noise, i.e., fluctua-tions around a trend, by smoothing currencyreturns.

A simple moving average rule would sig-nal an imminent break in trend, or the emer-gence of a new trend, when the moving

average is crossed by the spot rate or by ashorter moving average. Thus, an imminentupward break in trend for the spot rate, st,might be signaled by a short moving averageof length m > 1, MAt(m), intersecting frombelow a longer moving average of length

n(n >m), MAt(n), i.e.,

where

.

Conversely, a downward break in trend wouldbe signaled by the short moving averagecrossing the long moving average fromabove.5 Indicators of this kind will tend to beprofitable in markets exhibiting definitetrends and so they are generically known as

trend-following or momentum indicators.Another widely used device is the over-bought/oversold indicator, or oscillator.Oscillators are measures designed to indi-cate that price movements in a particular

MA jj

s j m nt t ii

j( ) , ,

10

1

=

=

MA n m nt

>


4/37

direction have recently been too rapid andthat a correction in the opposite direction isimminent; they may take a number of pre-cise forms. One popular form is the relativestrength indicator (RSI) (J. W. Wilder 1978),

for example, which is defined as:,

where Ut denotes the cumulated up move-ment (i.e., the close-to-close increase on aday when the exchange rate has closed high-er than the previous days closing rate) over acertain period, and Dt denotes the cumulatedabsolute down movement (the absoluteclose-to-close decrease on a day when the

exchange rate has closed lower than the pre-vious days closing rate) over the same period(often fourteen days):6

,

where () is an indicator variable that takesthe value one when the statement in paren-

theses is true, and zero otherwise.7

The RSIthus attempts to measure the strength ofup movements relative to the strength ofdown movements, and is normalized to liebetween 0 and 100; common values at

which a particular currency is deemed tohave been overbought (signaling an immi-nent downward correction) or oversold (sig-naling an imminent upward correction) are70 and 30, respectively (see, e.g.,Henderson 2002). Indicators of this kind are

also referred to as reversal indicators,

s s s st i t i t i t i

< ( 1 10) | |Dt im

== 1

s s s st i t i t i t i

> ( )( )1 10Ut im

== 1

RSIU

U Dtt

t t

=+

100

since they are designed to anticipate areversal in trend.

A third standard quantitative technique oftechnical analysis, the filter rule, involvesbuying a currency against another currency

whenever the exchange rate has risen bymore than a given percentage above its mostrecent low and selling it when the rate dropsby more than the same percentage below itsmost recent high. An x-percent filter rulemay be expressed thus:8

Buy if:

Sell if:

.

Obviously, the variety of both qualitativeand quantitative techniques varies enor-mouslya fact which makes it quite difficult

to provide a systematic assessment of techni-cal analysis. Moreover, empirical tests ofspecific rules and their associated tradingsignals are not fully satisfactory as tests of theefficacy of technical analysis more generally,since users typically do not apply a singlerule but rather a range of technical indica-tors which they update on a nonregular


6 Some expositions define Ut and Dt in terms of averagerather than cumulated up and down movements. This isequivalent to our definition, however, since it just involvesdividing by the total number of days and this factor can-cels out when the RSI is calculated.

7 The RSI is sometimes equivalently defined as

,

where RS, or relative strength, is defined as

.RS U D

t t t /

RSIRSt

t

+

100 1001

1

8 Note that the min conditions in these filter-rule for-

mulae minimize with respect to the time period ratherthan the exchange rate, so that they find the most recentperiod when the conditions indicated are met. For exam-ple, the formula for the buy signal may be expressed thus:Starting at timet, find the most recent period in whichthe exchange rate was less than it is at timet but where ithad been falling compared to the previous period (i.e., theexchange rates most recent low) and if the exchange ratehas risen by more than x percent since then and time t,buy. The nonnegativity condition on thei subscripts is toensure that the formulae apply to lags rather than leads.(Naturally, it is understood that buy means buy the cur-rency whose price in terms of the second currency is rep-resented by the exchange rate in question and sell is to

be interpreted similarly.)

1000s s i i s s

t t i t i t = > { min[ ( )| |

0 = > t i t i ts i i s s{ min[ ( )| |

0 s st t i t

< ) & ( 11 0

0

| |{ min[ ( )= > 0s i i s st i t i t| |

0{ min[ ( )s i i s st i t i t = > | |

01

& (s st i t i > )) ]}

) & (

> >

0

0

s

st

t t iss

xt i >

>1

0) ]}%

& (> 0 st i sst i >1 0) ]}


5/37


basis. In addition, many technical analystswill also apply considerable market intuitionto complement their quantitative conclu-sions, so there remains always a major ele-ment of subjectivity with the application of

technical analysis.3. The Importance of Technical Analysis in

the Foreign Exchange Market

The widespread use of technical analysisby foreign exchange professionals was firstbrought to the attention of academicresearchers by Stephen H. Goodman (1979),Group of Thirty (1985), Frankel and Froot(1986, 1990a, 1990b), and Goodhart (1988).9

However, the existence of technical analysisand even its use did not provoke sustainedacademic interest as long as the availableevidence was not of a more systematicnature. The skepticism with which academiceconomists initially viewed (and to someextent continue to view) technical analysis

was derived largely from the intellectualstanding of the efficient markets hypothesis(EMH), which, even in its weak form(Eugene F. Fama 1970), maintains that all

relevant information should already beembodied in asset prices, making it impossi-ble to earn excess returns on forecasts basedon historical price movements, once suitablerisk-adjustment is made.10

Nevertheless, during the 1990s, beginningwith the work of Allen and Taylor (1990), anumber of academic studies appeared thatreported the results of surveys of foreignexchange market participants concerning

the use of technical analysis. The salientcharacteristics of these studiesin terms ofsurvey year, target group, response rate,location, etc.are given in table 1. The firstsurvey, discussed in Allen and Taylor (1990)

and Taylor and Allen (1992), was carried outamong chief foreign exchange dealers atfinancial institutions located in London in1988; the most recent was conducted in2001 by Thomas Gehrig and Lukas

Menkhoff (2004) among foreign exchangedealers and fund managers located inGermany. In all, the various surveys havecovered foreign exchange professionalsbased in London, Frankfurt, Hong Kong,Singapore, Tokyo, New York and Zurich. In1995the midpoint between the earliestand latest studiesthe Bank for Inter-national Settlements (2005) ranked theseven locations covered by the survey stud-ies as first to seventh in terms of daily

turnover in foreign exchange dealing, mak-ing up about 78 percent of the total globalturnover; the combined market share ofthese centers was virtually unchanged until2004. Although the response rates of thestudies differ markedly, the results areremarkably invariant.

The studies of Allen and Taylor (1990) andTaylor and Allen (1992) documented for thefirst time systematically that technical analy-

sis is, indeed, an important tool in decisionmaking in the foreign exchange market.They further established a perceived com-plementarity among market practitioners inthe use of technical and fundamental analy-sis, and showed that reliance on technicalanalysis was skewed toward shorter tradingor forecast horizons. These three basic find-ings are also features of the results reportedin the remaining six survey studies (see table2), and therefore form the first three of our

stylized factsSF1, SF2, and SF3:Stylized Fact 1 (SF1): Almost all foreign

exchange professionals use technical analysisas a tool in decision making at least to some

degree.Stylized Fact 2 (SF2): Most foreign

exchange professionals use some combination of technical analysis and fundamentalanalysis.

Stylized Fact 3 (SF3): The relative weight

given to technical analysis as opposed to

9 The early study of technical analysis in the foreignexchange market of William Poole (1967a) can be seen as

very much ahead of its time.10 The more extreme form of the EMH assumes that

agents are risk neutral so that significantly nonzero excess

returns cannot be earned.


6/37

fundamental analysis rises as the trading or

forecast horizon declines.SF1 can be established from the third col-umn in table 2. Those surveys which askedrespondents whether or not they used tech-nical analysis at some horizon found thataround 90 percent or more did so. The factthat practitioners use technical analysis doesnot by itself, however, establish that theyregard it as of major importancethey mayattach some weight to it, but only a low

weight. Although not identical in design,

most of the surveys therefore also askedrespondents to quantify the weight given totechnical analysis relative to fundamentalanalysis at various horizons (SF2); the aver-age relative weight assigned to technicalanalysis is shown in column five of table 2,and ranges from around 30 percent to a littleover 50 percent.

A further aspect of the importance oftechnical analysis concerns its use among

various groups of market participants, since

a high average score could mask its concen-tration in small subgroups in the market.Table 3 reveals, however, that technicalanalysis is perceived as important relative tofundamentals across a range of practitionergroups such as chief foreign exchange deal-ers, international portfolio managers, andothers, whatever their specific role in foreignexchange trading may be.11

Early analytical studies of the foreignexchange market that allocated a role totechnical analysts or chartists tended to view

chartists and fundamentalists as competingfactions, either in their own right as traders


11 Overall foreign exchange trading operations will beheaded by a chief dealer who is, however, due to his man-agement role, not necessarily the most active trader. Thenthere will be core traders, such as those responsible forspot trades in a given exchange rate, and finally there areother foreign exchange traders who may focus on furtherobjectives such as trading forwards and futures, or are

junior and thus have less leeway in their decision making.See Sager and Taylor (2006) for a comprehensive analysis

of the structure of the foreign exchange market.

TABLE 1INFORMATION ON QUESTIONAIRE SURVEY STUDIES

Study time financial target group number of responseof survey center responses rate

Taylor and 1988 London chief 213 60.3%Allen (1992) FX dealers

Menkhoff 1992 Germany FX dealers; 205 41.3%(1997) intl fund

managers

Lui and Mole 1995 Hong Kong FX dealers 153 18.8%(1998)

Cheung and 1995/96 Hong Kong, FX dealers 392 20.0%Wong (2000) Singapore,

Tokyo

Oberlechner 1996 Switzerland, FX dealers; 321 53.5%(2001) United Kingdom, (financial (59) (29.5%)

(Austria, Germany) journalists)

Cheung and 1996/97 United States FX dealers 142 8.1%Chinn (2001)

Cheung, Chinn 1998 United Kingdom FX dealers 110 5.8%and Marsh (2004 )

Gehrig und 2001 Germany, (Austria) FX dealers; 203 51.9%Menkhoff (2004) intl fund managers

Notes: FX stands for foreign exchange.


7/37


or as advisers to traders (Goodhart 1988;Frankel and Froot 1990a, 1990b). SF2,however, (Most foreign exchange profes-

sionals use some combination of technicalanalysis and fundamental analysis) chal-lenges this adversarial view of chartism andfundamentalism. Figure 1 demonstratesthat other studies have basically reproducedthis finding of perceived complementarity(i.e., a reliance on fundamental and techni-

cal analysis) established by Taylor and Allen

(1992). In particular, the weight given tostrong mutual exclusiveness of chartism andfundamentalism, i.e., a reliance on eitherfundamental or technical analysis (repre-sented in figure 1 by values 9 and 10), is atmost ten percent of respondents in all stud-ies.

Finally, SF3 states that technical analysistends to be perceived as less important atlonger horizons in comparison with funda-

mental analysis (see explicitly Taylor and

TABLE 2THE IMPORTANCE OF TECHNICAL ANALYSIS ACCORDING TO QUESTIONNAIRE SURVEYS

Study form of some use share of share of the relationanalysis for of technical technical plus technical analysis between thedecision making analysis fundamental to technical plus weight of technical

analysis to total fundamental analysis andforms(2) analysis(2) horizon

Taylor and fundamental 89.4% 100% 32%(4) strictly negativeAllen (1992) analysis;

technical analysis

Menkhoff fundamental; >90% 82% 45% weakly (1997) technical; hump-shaped

flow analysis

Lui and fundamental; ~100% 100% 51%(5) strictly negativeMole (1998) technical

Cheung and fundamental; n.a. 62%(3) 40%(3) stronglyWong (2000)(1) technical; hump-shaped

bandwagon;overreaction;speculative forces

Oberlechner fundamental; >98% 100% 49% strictly negative(2001) technical

Cheung and see Cheung and n.a. 56% 29% strongly Chinn (2001)(1) Wong (2000) hump-shaped

Cheung, Chinn see Cheung and n.a. 49%(6) 47%(6) stronglyand Marsh Wong (2000) 54%(7) 29%(7) hump-shaped(2004)

Gehrig and fundamental; >90% 77% 53% weakly Menkhoff (2004) technical; hump-shaped

flows analysis

Notes: (1) These studies do not directly ask for analytical tools but for factors determining exchange rate move-ments.

(2) Unweighed averages of values for different horizons.(3) Data based on Hong Kong only (values for Singapore and Tokyo are similar).(4) Share is calculated as ratio of scale values 0 to 4 / scale values 0 to 4 plus 6 to 10 (i.e., preference for

technical analysis to total preferences); weighed with share of respondents at respective horizon (seeTaylor and Allen 1992, table 3B, first column).

(5) Share is calculated as ratio of importance given to technical analysis to total.(6) Traders were asked to select the technique which best characterizes their dealing method.(7) This value is a more indirect indication and is derived from the same question as mentioned in foot-

note (1).


8/37

Allen 1992, table 3B). A graphical presenta-tion of the research regarding SF3 can beseen from the work of Taylor and Allen(1992), Yu-Hon Lui and David Mole (1998),and Thomas Oberlechner (2001) in figure 2A.These three studies relate the perceived rela-tive importance of technical and fundamentalanalysis with forecast or trading horizon. Ifone takes, however, the other studiesfea-tured in figure 2Binto account, the resultremains unchanged for the medium and

longer horizons but becomes less clear for the very short horizon. There is, however, anobvious reason for this apparent difference inperceived relative importance at the veryshort horizon, in terms of their coverage ofanalytical tools or price-determining factors.In particular, studies featured in figure 2Balso take into account other factors such asthe perceived importance attached by marketpractitioners to information on order flow

(i.e. on information relating to the value of

foreign exchange transactions signed accord-ing to the originator of the tradesee, e.g.,Takatoshi Ito, Lyons, and Michael T. Melvin1998; Lyons 2001; Lucio Sarno and Taylor2001b; Martin D. D. Evans and Lyons 2002).The inclusion of factors other than technicaland fundamental analysis in the menu ofchoices offered to survey participants thusdilutes the relative score given to technicalanalysis in the shorter-term domain (see onthese factors columns two and four in table

2).12 Considering only fundamental and


12 It is possible that order flow might better be inter-preted as fundamental rather than as technical informa-tion since, although it is clearly not on the list of standardmacroeconomic fundamentals, it may in some senseembody the net effect of fundamental influences on theforeign exchange market (Lyons 2001; Evans and Lyons2005b). Nevertheless, we rely here on studies in whichorder flow forms a third category and which may to someextent represent the current perception of order flow byforeign exchange professionals (Henderson 2002; Gehrig

and Menkhoff 2004).

TABLE 3THE IMPORTANCE OF TECHNICAL ANALYSIS IN SEVERAL SUB-GROUPS AND AT TYPICAL FORECASTING HORIZONS

Question: Please evaluate the importance of the three following information types for your typical decisionmaking, by distributing a total of 100 points. For information types which you do not use, please give0 points.

. . . Fundamentals (economic, political)

. . . Technical Analysis (charts, quantitative methods)

. . . Flows (who is doing what, which customer orders are existing)

Menkhoff (1997, 1998) Gehrig and Menkhoff (2006)

Horizon chief FX core FX other FX intl fund chief FX other FX intl funddealers dealers dealers managers dealers dealers managers

Intraday 30.5 36.6 23.2 n.a. 45.0 37.3 n.a.

Few days 37.8 38.6 44.0 45.0 45.9 45.1 52.5

Few weeks 34.3 42.5 40.6 35.9 46.9 37.3 32.8

2 to 6 months 42.6 50.0 29.3 36.1 28.3 31.7 31.7

6 to 12 months (20) n.a. (20) 30.0 (0) n.a. (15.0)>12 months n.a. n.a. (40) n.a. (100) (30) n.a.Mean 35.4 38.4 39.9 36.1 44.9 40.0 37.0

n 44 66 39 50 42 102 58

Notes: Data are from the studies Menkhoff (1997, 1998) and Gehrig and Menkhoff (2006). The first value of 30.5says that chief FX dealers who have a typical intraday forecasting horizon give technical analysis a weight of 30.5%(out of 100% for fundamental, technical and order flow analysis). Shaded cells mark the typical horizon (medianvalue) for decision making of the respective group (e.g. 49% of core FX dealers mark intraday as their typicalhorizon). Numbers in parenthesis refer to groups with 1 to 3 responses. FX stands for foreign exchange.


9/37


technical analysis for the purpose of com-parison indeed reproduces the earlier find-ing in figure 2C (also supported by Gehrigand Menkhoff 2006 with a differentmethodology).

As a final remark it may be reassuring thatthe stylized facts shown for foreign exchangedealers and fund managers from the mainfinancial centers by and large also hold for

financial journalists (Oberlechner 2001) and

dealers in an emerging market (N. R.Bhanumurthy 2004).

4. Profitability of Technical Analysis:Measures and Results

The evidence concerning the profitabilityof technical currency analysis tends to beinconclusive. From a theoretical point of

view, this is perhaps unsurprising, since if

Figure 1. On the Perceived Complementarity of Technical and Fundamental Analysis

Note: The regressions are calculated as best fit of a polynomial of second order. The data from Menkhoff (1997) andGehrig and Menkhoff (2004) are transformed in the following way: the individual weight given to fundamental analysis(f) and to technical analysis (t) is put into one measurex.x = |ft | : (f+t) 100. The percentage is then put into the scale0 to 10 according to:x < 10% 0; 10% x < 20% 1; . . .; 90% x < 100% 9; 100% 10. The data fromOberlechner (2001) are transformed as follows: < 6 (strong complementarity) 0, < 5 2, < 4 and < 7 4, < 3 and< 8 6, < 2 and < 9 8, < 1 and < 10 10 (each value multiplied by 0.6 to account for different scaling).

0 1 2 3 4 5 6 7 8 9 10

25

20

15

10

5

0

5

stronglycomplementary

mutuallyexclusive

percentageof responses

Taylor and Allen (1992; Table 4)

Menkhoff (1997)

Oberlechner (2001, Figure 2)Lui and Mole (1998, Figure 1)

Gehrig and Menkhoff (2004)


10/37


Figure 2. The Relative Importance of Technical Analysis Depending on the Horizon of Decision Making

Notes: The relative weight of technical analysis is calculated as the weight of technical analysis to the sum of technicalplus fundamental analysis. Horizons are taken from Taylor and Allen (1992) and Lui and Mole (1998); importance (scale010) in Oberlechner (2001); Figure is transformed into percentage points; Menkhoff (1998) and Gehrig and Menkhoff(2004) are transformed: few days into 1 week, few weeks into 1 month, 2 to 6 months into 3 months and 6to 12 months into 1 year, data for 6 months is interpolated; Cheung and Wong (2000), Cheung and Chinn (2001),and Cheung, Chinn and Marsh (2004) are transformed: medium run (< 6 months) into 1 month and long run (> 6months) into 1 year, data for 1 week, 3 months, and 6 months are interpolated.

A. Studies considering technical and fundamental analysis only80%

70%

60%

50%

40%30%

20%

10%

0%Intraday 1 week 1 month 3 months 6 months 1 year > 1 year

Lui and MoleOberlechnerTaylor and Allen

B. Studies considering also further influences50%

45%

40%35%

30%

25%

20%

15%

10%

5%

0%Intraday 1 week 1 month 3 months 6 months 1 year

Cheung and ChinnCheung, Chinn and MarshCheung and WongGehrig and Menkhoff

Menkhoff

relativeweight oftechnicalanalysis

relative

weight oftechnicalanalysis

C. Studies in B, here without further influences, i.e., relating only the weight oftechnical and fundamental analysis to each other

100%90%80%70%60%50%40%30%20%10%

0%Intraday 1 week 1 month 3 months 6 months 1 year

Cheung and ChinnCheung, Chinn and MarshCheung and WongGehrig and MenkhoffMenkhoff

relativeweight oftechnicalanalysis


11/37


technical analysis was never profitable, itswidespread use (see section 2) would be hardto understand; if, on the other hand, techni-cal analysis was always profitable, it wouldperhaps imply that the foreign exchange

market is inefficient to a degree that manyeconomists would not find credible.Indeed, the EMH does not imply in this

respect that returns to applying a technicaltrading strategy have to be zero. Rather, effi-cient markets rule out the possibility oftrading systems based only on [current andpast] information [having] expected profitsor returns in excess of equilibrium expectedprofits or returns (Fama 1970, p. 385). Inthis context, equilibrium expected returns

must be calculated after allowing for a rea-sonable return to risk and after allowance fortransactions costs.13

In assessing the profitability of technicalanalysis, therefore, three methodologicalaspects have to be addressed carefully. First,any examination should define appropriatealternatives, i.e., on the one hand the techni-cal analysis strategies and on the other handa strategy relying on the EMH. Important

features of this comparison must includetransaction costs and interest rate carrycosts.14

Second, the issue of statistical significancehas to be tackled. Independently of the dis-tribution of exchange rate returns, theremust always exist a technical analysis strate-gy that is able to exploit characteristics of thetime series in any particular sample. Thus, itis not profitability per se that is interestingbut the possible significance of the result

that challenges the EMH.

Third, it is probably the form of risk con-siderationan essential element of Famasequilibrium expected profitsthat dividesadvocates and opponents of technical analy-sis in the interpretation of their empirical

work.If one compares the overview of earlierstudies in table 4 with the three require-ments just mentioned, it becomes clear thatthese studies are all characterized by short-comings to a greater or lesser extent: manyof them examine only one currency, some donot consider all kinds of costs and most arehandicapped by a short period of investiga-tion which does not allow for appropriateout-of-sample calculations. Thus, only the

studies of Michael P. Dooley and Jeffrey R.Shafer (1983) and Richard James Sweeney(1986) have been consistently cited in theliterature (see, e.g., Ramazan Gencay 1999;Blake LeBaron 1999; Neely 2002; DennisOlson 2004).

It is interesting to note that most of thestylized facts that can be drawn from prof-itability examinations are already found inthis early literature (table 4) and that they

are supported by later studies (e.g.,Patchara Surajaras and Sweeney 1992;Menkhoff and Manfred Schlumberger1995; Keith Pilbeam 1995; Neely 1997;LeBaron 1999; Peter Saacke 2002). Theycan be gathered together here as StylizedFacts 4, 5 and 6:

Stylized Fact 4 (SF4): The consideration oftransaction costs and interest rate costs actu-ally faced by professionals does not necessar-

ily eliminate the profitability of technical

currency analysis.

13 It may also allow for tax payments, where tax treat-ment differs across investor groups.

14 It must also constitute a study of profitability from anex ante rather than an ex post perspective, so that there isperceived profit from gathering and utilising informationin a superior fashion. This is important because it could beargued that, in equilibrium, there should be no gain fromutilising superior information since such information

would already be embodied in market prices. If this werethe case, however, traders would presumably not bother to

gather costly information, in which case it is difficult to see

how the information would be imparted into marketprices at all, so that the proposed no-profit-from-costly-information equilibrium implies a contradiction and socannot in fact be an equilibrium. This is the so-calledGrossmanStiglitz paradox, and in the resolution of thisparadox, both Sanford J. Grossman and Joseph E.Stiglitz (1980) and W. Bradford Cornell and Richard Roll(1981) have shown that a sensible financial market equi-librium must leave some incentive for costly information

acquisition and analysis.


12/37

Stylized Fact 5 (SF5): Technical analysistends to be more profitable with volatile cur-rencies.

Stylized Fact 6 (SF6): The performance of technical trading rules is highly unstableover time.

Evidence on SF4the profitability oftechnical trading rules after allowance fortransactions costs and interest rate carryisprovided by, among others, Cornell and J.Kimball Dietrich (1978), Sweeney (1986),

Stephan Schulmeister (1987), LeBaron(1999), Saacke (2002), and Neely, Paul A.

Weller, and Joshua M. Ulrich (forthcoming).Studies supporting the hypothesis that tech-nical trading rules are more profitable forcurrencies experiencing relatively higher

volatility (SF5) include Cornell and Dietrich(1978), Dooley and Shafer (1983), Chun I.Lee and Ike Mathur (1996), and Neely and

Weller (1999). Work suggesting that techni-

cal trading rule performance is unstable over

time (SF6) include Dennis E. Logue,Sweeney, and Thomas W. Willett (1978),Dooley and Shafer (1983), Menkhoff andSchlumberger (1995), LeBaron (2000),Michael Dueker and Neely (2007), andNeely, Weller, and Ulrich (forthcoming).15

The rapidly growing empirical literatureon the profitability of technical analysis in theforeign exchange market that has appearedsince the late 1990s has, if anything, furthersubstantiated these older stylized facts (see,


15 As an anonymous referee has pointed out that theevidence for the instability of technical trading rulesshould be interpreted with care, however, since exchangerate returns are noisy relative to sample length, tests forunknown structural breaks have notoriously low powerand test for structural breaks at known breakpoints aresubject to data snooping bias. Moreover, the popularity oftechnical analysis may be sustained not by its consistentperformance but by itsperceived performance across var-ious prominent episodes and instability in performanceover time is also a characteristic of fundamentals-basedexchange rate models (Cheung, Chinn, and Antonio

Garcia Pascual 2005).

TABLE 4EARLIER STUDIES EXAMINING THE PROFITABILITY OF TECHNICAL ANALYSIS IN FOREIGN EXCHANGE MARKETS

Study period number form of consideration of excesscovered of exchange technical returns of

rates analysis transaction interest risk technical

costs rates analysis

Poole (1967) 1919 9 10 filters no no no +24/29

Poole (1967a) 195062 1 12 filters no no no +Dooley and 197375 8 Filter +Shafer (1976)

Logue and 197074 1 14 filters yes no no +Sweeney (1977)

Logue, Sweeney and 197376 7 11 filters no no no +Willett (1978)

Cornell and 197375 6 13 filters, yes no yes +Dietrich (1978) 27 mov.

Averages

Dooley and 197381 9 7 filters yes yes no +Shafer (1983)

Sweeney (1986) 197380 10 7 filters yes partially yes +Schulmeister (1987) 197386 1 9 filters, yes partially no +

9 mov. av.,5 momentum,

1 point & figure


13/37


e.g., Cheol-Ho Park and Scott H. Irwin2007). In addition, it is possible to discern anumber of developments among the morerecent literature.

First, a major methodological innovation

has been the introduction of the bootstrapapproach addressing the problem of insignif-icant evidence (Richard M. Levich and LeeR. Thomas 1993; LeBaron 1999, 2000; CarolL. Osler 2000, 2003) and, more recently, theintroduction of methods for testing forpotential data-snooping bias (Park and Irwin2005; Min Qi and Yangru Wu forthcoming).

Second, the range of technical analysistools and trading rules considered has beenincreased far beyond filter rules, moving

averages, or point-and-figure indicators, andnow includes the possible psychological bar-riers of round figures, the closely relatedissue of support and resistance levels (PaulDe Grauwe and Danny Decupere 1992;Goodhart and Riccardo Curcio 1992; Osler2000, 2003, 2005), or of momentum-basedstrategies (John Okunev and Derek White2003).16

Third, the longer span of data available

for the floating rate period since the early1970s has stimulated the question as to whether profits from technical analysis aredeclining over time. There is indeed evi-dence that the foreign exchange market hasbecome more efficient over time in thesense that the application of traditionalmoving average rules, which was shown tobe profitable for the 1970s (e.g., Logue andSweeney 1977; Cornell and Dietrich 1978;Dooley and Shafer 1983; Sweeney 1986),

became much less profitable in the 1990s(LeBaron 2000; Olson 2004), even after

allowing for a reduction in transactions costsover time (Neely, Weller, and Ulrich forth-coming). A possible reason for the declinein profitability of technical trading rules inrecent years is the growth in the hedge fund

industry to a level estimated to be of theorder of more than $1 trillion in assets undermanagement in the early to mid 2000s andstill growing (Securities and ExchangeCommission 2003).17 Foreign exchange hasbeen increasingly seen as a separate assetclass by hedge funds (Deutsche Bank 2005)and many hedge funds apply quantitativestrategies that may employ one or moretechnical trading rules (Sager and Taylor2006); insofar as technical analysis provides

a short-term informational advantage, alarge amount of trades employing similartechnical trading rules will clearly erodethat advantage much faster. Nevertheless,significant evidence of profitabilityalbeiton a reduced levelremains and may evenhave been increasing during recent years ineuro-dollar trading (Park and Irwin 2005;Schulmeister 2005). Also, there may bemore complex forms of technical analysis

that did not become less profitable overtime (e.g., Okunev and White 2003; Neely,Weller, and Ulrich forthcoming).18

Fourth, there have been attempts to avoidpotential selection bias by letting actors statetheir preferred rules prior to any profitabili-ty analysis (Allen and Taylor 1990; Goodhart

16 The technical trading rules that have been examinedin this literature include those based on head and shoul-ders patterns (P. H. Kevin Chang and Osler 1999; BerndLucke 2003), candlestick formations (Norbert M. Fiessand Ronald MacDonald 2002), neural networks (Gencay1999), genetic programming (Neely, Weller, and RobertDittmar 1997; Neely and Weller 1999, 2001), Markovswitching models (Ian W. Marsh 2000; Dueker and Neelyforthcoming) and real-time trading models (Gencay et al.

2003).

17 A hedge fund may be defined as a lightly regulatedinvestment fund that engages in active and leveraged

financial portfolio management, usually involving a rangeof securities in which they may employ complex tradingstrategies (not just buy and hold) in which both long andshort positions are typically held, and whose objective is toprovide a return much greater than the return from a pas-sive investment strategy. They are lightly regulated large-ly because, in most countries, they are only open toinstitutional investors and wealthy accredited investors,rather than the general public.

18 Similar results are reported by Po-Hsuan Hsu andChung-Ming Kuan (2005) for stock markets, providingsupport to the interpretation of Neely, Weller, and Ulrich(forthcoming) that markets may need time to becomeaware of and then to arbitrage away profit opportunities

generated by technical trading rules.


14/37

and Curcio 1992, Curcio and Goodhart1993; Osler 2000).

Fifth, studies have explored the relationbetween nonlinear exchange rate modelingand technical analysis (William C. Clyde and

Osler 1997; Fiess and MacDonald 1999;Lutz Kilian and Taylor 2003; De Grauweand Marianna Grimaldi 2006a, 2006b;Stefan Reitz and Taylor forthcoming).

Sixth, stimulated by the apparent successof longer-term exchange rate modeling via aMarkov switching approach (Charles Engeland James D. Hamilton 1990), studies havefound some links between regime switchesand technical trading rules (HansDewachter 1997, 2001; Robert Vigfusson

1997). However, profitability does not seemto be better than for simple moving averagerules (Dueker and Neely 2007) although anadvantage may be gained by the fact thatprofits remain more stable over time (Neely,

Weller, and Ulrich forthcoming).Finally, some studies (Curcio et al. 1997;

Osler 2000, 2003; Neely and Weller 2003;Roman Kozhan and Mark Salmon 2006)have examined the profitability of technical

analysis on a very high-frequency (intraday)basis, with mixed results.On balance, however, the literature on the

profitability of technical trading rules tendsto support the existence of significant profitsto be had by employing these rules in the for-eign exchange market (see also Park andIrwin 2007 for a concurring reading of theliterature). Of course, this in itself raises asample selection bias issue, since it is well

known that positive results are generallymuch easier to report and publish than neg-ative results. In particular, these studies maybe subject to data snooping (Halbert White2000). Data snooping occurs when a given

set of data is used more than once for pur-poses of inference or model selection, so thatthe possibility arises that any satisfactoryresults obtained may simply be due tochance rather than to any merit or skill inher-ent in the method yielding the good results.

White (2000) develops a bootstrap simula-tion techniquethe reality checkforexamining whether it is inherent skill or purechance that leads to the best rule being cho-sen out of any given universe of rules.

Intuitively, a reality check involves replicat-ing, by Monte Carlo methods, many artificialdata sets that in some sense match the prop-erties of the original data sets, and testing the

various trading rules for profitability on eachdata set. If there is a tendency for the samerule to be selected as the most profitable foreach data set, then this suggest that it reallyis a good rule; if there is no tendency toselect that particular rule for each of the arti-

ficial data sets, then this indicates that it wasselected as the most profitable rule in theoriginal data set purely by chance.19

The first application of Whites realitycheck to technical trading rules in the for-eign exchange market is due to Qi and Wu(forthcoming). These authors examine alarge number of technical trading rules andapply them to daily data on seven dollarexchange rates over the period 197398. The


19 An anonymous referee has pointed out a number ofissues that may be raised with respect to Whites realitycheck. In particular, while the reality check is clearly animprovement over earlier approaches that ignored data-snooping bias, the group of trading rules making up theuniverse within which the reality check is carried out muststill be chosen and that brings back the danger of datasnooping in a different guise. Indeed, there may even bea systematic bias involved as researchers may, consciouslyor unconsciously, rely on rules that have been implicitlytested on similar data in previous research. Moreover,merely adding a large number of poor rules into the real-

ity check universe will tend to raise the critical values for

a given nominal test size while the performance of thebenchmark trading rule does not change. An alternativeapproach would be to carry out an ex ante search for prof-itable trading rules using artificial intelligence such as agenetic algorithm that learns trading rules and appliesthem, as in the equity-market study of Franklin Allen andRisto Karjalainen (1999), although this approach wouldalso potentially be subject to sample-selection bias.Alternatively, one can perform true out-of-sample tests byretesting rules that have found to be profitable in earlierstudiesas for example in LeBaron (2000) or Neely,

Weller, and Ulrich (forthcoming).


15/37


technical rules are various calibrations offour classes of rules: filter rules (buy or sell acurrency if it moves more than a certain per-cent from its most recent high or low); mov-ing average rules (as discussed above);

support and resistance rules (buy or sell acurrency when it breaks above or below themaximum or minimum level, the resistancelevel, over a stipulated recent period); andchannel breakout rules (but or sell a curren-cy when it breaks out of a channel, definedas occurring when the high price of a foreigncurrency over the previousn days is withinxpercent of the low over the previousn days).Using standard tests, Qi and Wus resultsindicate significant profitability of moving

average and channel breakout rules forseven dollar exchange rates. They then apply

Whites (2000) reality check bootstrapmethodology to evaluate these rules and tocharacterize the effects of potential data-snooping biases. They find significant prof-itability at the one percent level for all sevencurrencies even, after data-snooping biases(as well as transactions costs) are properlytaken into account (Park and Irwin 2005 find

a similar result for euro and yen futures).Moreover, employing the Japanese yen orthe German mark as a vehicle currency(instead of the U.S. dollar) yields evenstronger results.

Even if the existence of significantly prof-itable technical trading rules can be estab-lished, however, there is still the possibilitythat all that is being measured is a risk premi-um, so that the risk-adjusted returns from therule would on average be non-positive. Table

4 revealed already that earlier studies usuallyignored this issue but more recent studieshave elaborated on it (see table 5). The pio-neering attempt in this respect is Cornell andDietrich (1978) who suggest a risk adjustmentaccording to the international capital assetpricing model (ICAPM). Their empiricalrealization is limited, however, by practicalconstraints: first, the world portfolio is prox-ied by a U.S. market stock index (the S&P

500) and, second, they generally calculate the

beta of foreign currencies with this indexrather than the beta of currency positions thatresult from technical trading rules (for the lat-ter, see, e.g., Stephen J. Taylor 1992 with thesame result).20 Nevertheless, their very low

beta estimates suggest that investing in for-eign currency provides a good hedge for aninvestor whose portfolio is primarily centeredon U.S. stocks (see also Neely 1997).

The first study systematically integratingrisk-adjustment into the empirical examina-tion of certain rules, however, was due toSweeney (1986). This study characterizes aquite different approach to that of Cornelland Dietrich (1978), as it compares tradingrules based on technical currency analysis

rules to buy-and-hold strategies. If, forexample, deviations from uncovered interestrate parity (the condition that the expectedexcess return, net of interest rate carry, frombuying and holding foreign currency shouldbe zero) simply represent risk premia, then apossible implication is that apparently prof-itable technical trading rules are simply pick-ing up these risk premia. Sweeney indeedcalibrated his work under the assumption of

a constant risk premium, i.e., the averagereturn on foreign exchange holdings isadjusted by the foreigndomestic interestrate differential. Then the excess return onfollowing the technical analysis rule, i.e.,gross return minus return from buy-and-hold, is adjusted by the share of days that thetrading rule is invested in foreign currencyand has thus to earn a risk premium.According to this procedure, Sweeney didnot find a risk-based explanation for excess

returns. Levich and Thomas (1993), apply-ing a similar methodology, found a similarresult.

However, these results may be ques-tioned on at least two grounds. First, it isnot clear why one should expect a positive

20 From todays perspective, the choice of a U.S. port-folio may seem less of a shortcoming, taking account ofthe well-documented preference of investors for home

assets (Karen K. Lewis 1999).


16/37

risk premium for investing in one currencyin a bilateral exchange rate as this impliesthat investment in the second currency (ora short position in the first currency) earnsa negative risk premium.21 Second, theassumption of a constant foreign exchangerisk premium is not a very realistic one(Taylor 1995b). The first study to relax thisassumption in the context of technical trad-ing rules was Taylor (1992), who allows fora time-varying risk premium in the form of

a first-order autoregressive process.Parameter values of this process, justifiedby results from other studies, are used toenter into a pricing model. For severalcombinations of parameter values, hun-dreds of time series are then simulated on

which technical analysis rules are evaluat-ed. It is found that there does not seem to

be a reasonable constellation of parametersfor the time-varying risk premium which would be needed to explain observedreturns as a compensation for risk (see alsoOkunev and White 2003). On the otherhand, all that this evidence may be reveal-ing is that the wrong parameterization ofthe risk premium was assumed.

A much more extensive approach in deriv-ing time-varying foreign exchange risk pre-mia in this context is adopted by Bong-Chan

Kho (1996). He relates possible excessreturns to a world stock portfolio (the MSCIindex) in a conditional ICAPM framework.

Within his framework, there are basicallythree factors which are assumed to deter-mine world excess returns: interest rate dif-ferentials against the U.S. dollar, theconditional variance of world excess returnsand a moving average term. The empirical

work uses econometric models in which the

conditional variance is allowed to affect the


21We thank two of the anonymous referees for encour-

aging us to make this argument.

TABLE 5SUGGESTED RISK ADJUSTMENTS IN ASSESSING TECHNICAL ANALYSIS EXCESS RETURNS

Study period number standard of risk adjustment risk-adjustedcovered of cases(1) comparison excess returns

Cornell and 197375 6 S&P 500 beta of currency with +Dietrich (1978)(2) S&P 500Sweeney (1986) 197380 70 B&H (buy constant risk premium +

and hold) equivalent to uncoveredinterest parity-notation

Taylor (1992) 198187 16 S&P 500 B&H beta with S&P 500; +time-varying risk premia +estimated on AR(1) premiaprocesses and the UIP

Menkhoff and 198191 129 B&H Sharpe ratio; riskreturn +Schlumberger ratio of monthly return (1995) differences against B&H

Kho (1996) 198091 72 MSCI (in excess covariation of currency of one week $ returns with world marketinterest rates) portfolio excess returns

Chang and 197394 24 S&P 500, Sharpe ratio with S&P 500; +Osler (1999) Nikkei, DAX beta with national index +Neely (1997) 197497 40 S&P 500 Sharpe ratio; beta +

with S&P 500 +

Notes: (1) Cases are the product of currencies times rules times models (if applicable).(2) Incomplete documentation of results; favorable outcomes refer to ex post selection of best technical

analysis rules.


17/37


conditional mean (i.e., GARCH-m models)in order to calculate expected risks. Khofinds that much of the technical analysisreturns can indeed be explained as compen-sation for the high risks involved.

The above approaches to incorporate riskinto profitability measurement implicitlyneed a benchmark model of asset pricing inequilibrium. Obviously, the ICAPM is mostpopular in this respect although empiricalfinance may tentatively prefer multifactormodels, such as the Fama and Kenneth R.French (1996) approach. From a theoreticalpoint of view consumption-based asset pric-ing seems more advantageous to the CAPM(John H. Cochrane 2005). However, neither

of these approaches has been applied to theforeign exchange market.22 Given the failurein identifying meaningful time-varying riskpremia in international finance in general(Taylor 1995b), this shortcoming may beexcusable. This lack of knowledge has, more-over, fueled other ways of addressing theriskiness inherent in the use of technicalanalysis.

Some studies circumvent the problem of

measuring the world portfolio and derivingrisk premia. Instead, they directly comparethe returnrisk profile of a speculative cur-rency portfolio to a benchmark portfolio byusing the ratio of annualized excess returns(relative to a benchmark strategy) to thestandard deviation of those returns, i.e., theSharpe or information ratio (William F.Sharpe 1966). Alternative benchmarks inthis respect are either a buy-and-hold cur-rency strategy (e.g., Menkhoff and

Schlumberger 1995) or the return from

holding a broad portfolio index such as themarket index (e.g., Neely 1997; Chang andOsler 1999; LeBaron 2000; Saacke 2002).23

The results of these studies show higher risk-adjusted returns to technical analysis rules

than to the benchmark portfolios.24

The popular Sharpe or information ratio(IR) has its own problems, however, when itis used as a criterion by which to measurethe performance of trading rules. Supposethat the mean excess return of the tradingrule over a period ofTyears is , with stan-dard deviation . Then the IR will bedefined as

IR .

Now, it can easily be shown that = T IR is approximately equal to at-ratiofor a test of the hypothesis that the excessreturn is zero.25 A common benchmark for agood trading rule in the finance industry ingeneral is an IR of 0.5 (see, e.g., Richard C.Grinold and Ronald N. Kahn 2000). But thismeans that an IR of 0.5 must be sustainedover about eleven years before the tradingrule can be said to have generated excessreturns significantly greater than zero at the 5percent significance level, since this wouldgive a value of (= 11 0.5 = 1.658,)greater than the critical value for a one-sided test at the 5 percent level (i.e., 1.645).Suppose that a trader selects a certain rulebecause it has an IR of 0.5 according to abacktest with ten or more years of data.As we discussed earlier, there is a stronglikelihood that the rule will be subject to

data-snooping, and a true out-of-sample

22 It is interesting to note in this respect that thesemore advanced approaches are also confronted with evi-dence that questions their explanatory power (JonathanLewellen, Stefan Nagel, and Jay Shanken 2006).

23 These alternative benchmarks are second-best solu-tions adopted from the equity market literature. Thus, thebuy-and-hold benchmark implicitly uses a national, one-sided perspective whereas trading rules in foreignexchange are typically symmetric with respect to the twocurrencies involved. Regarding the index benchmarks,

they implicitly assume that the trading rule would be an

alternative to another investment. Accordingly, suchbenchmarks should not be taken literally.

24 In this vein, Dewachter and Marco Lyrio (2005) findthat the application of moving average rules can provide asignificant return to investors.

25 This result is independent of the distribution ofexcess returns and follows from the Central LimitTheorem, which states that whenever a random sampleof size Tis taken from any distribution with mean and

variance 2, then the sample mean will be approximately

normally distributed with mean and variance

2

/T.


18/37


test would require that the trader keeps therule and monitors its performance over theensuing ten or eleven years or sowhich is a

very long time in the financial markets.This picture changes, however, if one

measures risk not in the traditional sense ofthe variability of returns but if one triesinstead to integrate the professionals per-ception of risk as relating to relative per-formance in comparison with the market(see, e.g., Goodhart 1988, p. 457). Here, SF6comes into play, namely that profitability isunstable over time. In summary, applyingtechnical analysis involves a high probabilityof making wrong decisions, i.e., perform-ing below the market, at least during some

periods (see, e.g., William L. Silber 1994, p.44; Neely 1997). Thus, Menkhoff andSchlumberger (1995) suggest addressing therisk inherent in using technical analysis byfocusing on the monthly difference betweenthe rules profitability performance and abuy-and-hold performance (this may beunderstood as a form of myopic loss aver-sion, see Shlomo Benartzi and Richard H.Thaler 1995). Due to the high level of insta-

bility of technical analysis returns, theexcess returnas shown by raw returns orby a Sharpe ratio criterionceases to be sig-nificant at the 5 percent level.

In a recent paper, Maxime Charlebois andStephen G. Sapp (2007), using daily data ondollarmark over the period 198898, findthat moving-average trading rules generatesignificant excess returns and that the excessreturns increase when information is includ-ed on the open interest differential on cur-

rency options (i.e., the net differencebetween the cumulative value in dollar termsof all put options that are still active on agiven day less the cumulative value of allactive call options). They interpret this aspartly reflecting risk premia and partly asreflecting extra fundamental information thatis reflected in options prices, since optionsmay be the instrument of trading of choice ofmore informed traders because of the lever-

age advantage provided (David Easley,

Maureen OHara, and P. S. Srinivas 1998).Some evidence supporting this view is pro-

vided by the fact that when the authorsexclude the fifty largest absolute dailyreturns, all of the trading rules incorporating

the open interest differential become lessprofitable (implying that the excess returns oftechnical trading rules to some extent reflectcompensation for risk), but many of themnevertheless remain strongly profitable.

In summary, looking at the last column oftable 5, where risk-adjusted profitability isdisplayed, the majority of studies concludethat the profitability of technical currencyanalysis holds in a risk-adjusted sense. Goingmore into detail, there is, first, evidence that

time-varying risk premia might explain someof the excess return of technical analysis butnot all or even most of it (Taylor 1992; Kho1996).26 Furthermore, even the correctdetermination of appropriate risk premia isquestionable with the present state of knowl-edge. A second line of argument is that it isperhaps possible to explain some of theexcess returns using a measure of risk as per-ceived by market participants. Indeed, the

available evidence indicates that technicalanalysis is quite risky in this respect.

5. Explaining the Continued Use ofTechnical Analysis in the Foreign Exchange

Market

Technical analysis is an important tool inreal-world decision making in foreignexchange markets (SF1, SF2, and SF3). Inaddition, it appears that applying certain

technical trading rules to volatile foreignexchange markets over a sustained periodmay lead to significant positive excess

26 One must admit, however, that there is not muchguidance as to whether the measures of risk premia usedin the empirical literature are fully convincing from a the-oretical point of view. For example, Taylors (1992) AR(1)risk premium model may simply be too restrictive,

whileas an anonymous referee has commentedtheKho (1996) study is based on a limited sample and has notto date been replicated for other currencies or sample

periods.


19/37


returns, although it is not clear that the per-formance of these rules is stable over time orthat the excess returns earned significantlyoutweigh the associated risk premia (SF4,SF5, and SF6).

There remains a need for further explana-tion of the continued and passionate obses-sion of foreign exchange professionals withtechnical analysis, however, as profitabilitystudies have not to date arrived at a clear

verdict. The organizing idea here is to allowexplicitly for heterogeneous agents andasymmetric information in the foreignexchange market, which makes market effi-ciency a more complex concept. It may,however, be reassuring in this context that

this complexity can indeed by rooted inFamas (1970, p. 388) seminal paper onfinancial market efficiency, as he discussesdisagreement among investors about theimplications of given information as apotential source of inefficiency. So, in what

way may disagreement (i.e., heterogeneity)help in resolving our puzzle? We group the

various explanations that have been suggest-ed into four positions, which we shall briefly

describe before we relate them to rationalbehavior of agents and efficient markets (seefigure 3 for an overview).

If one follows the traditional understand-ing of the EMH and regards foreignexchange markets as at least weakly efficientin the sense of Fama (1970)i.e., in thesense that significant profits cannot be gen-erated using forecasts based on past pricemovements alonethen one would assessthe use of technical analysis as evidence of

irrational behavior. This is the first explana-tion for the continued use of technical analy-sis in the foreign exchange market.

However, the assumption that most pro-fessionals in the market behave consistentlyirrationally does not fit the EMH either:according to the EMH, they should quicklybe driven out of the market as they makelosses at the expense of rational traders.But, if there is an important set of foreign

exchange market participants who are not

directly interested in generating profit butnevertheless have a significant influence onthe market, then these participants maygenerate profit-making opportunities fortechnical analysts over sustained periods of

time, allowing them to survive in the mar-ket. One such group that has been proposedin this context is comprised of the majorcentral banks, and the behavior of centralbanks in intervening in the foreign exchangemarket has been posited as a second expla-nation for the persistence of technicalanalysis.

A third position is that if it takes time forthe effects of economic fundamentals tofeed through fully into market exchange

rates, then technical analysis may serve as ameans of detecting these kinds of influ-ences earlier than would otherwise be thecase.

Fourthly and finally, it has often beenargued that financial prices may not onlyreflect the information from fundamentalsbut also influences from other sources, suchas the influence of noise traders or the self-fulfilling influences of technical analysis

itself.Among these four explanations, it is onlythe first that directly refers to irrationalbehavior of agents: either the users of tech-nical analysis are simply irrational and will bedriven out of the market (as suggested byMilton Friedman 1953) or they systematical-ly underestimate risk (as suggested by J.Bradford De Long et al. 1990). The otherthree explanations do not rely on techni-cians irrationality but on Famas (1970)

argument that not all market participantsneed to interpret all information at the sametime in the same way.

With regard to the foreign exchange inter-vention explanation, it would be the centralbank that distorts markets and technicaltraders profit from this inefficiency.Regarding the third explanation, technicalanalysis is seen as an instrument via which tolearn about the revelation of fundamentals

that cannot be recognized from observing


20/37


fundamentals directly.27 Here, neither tech-nical traders nor the market need be ineffi-

cient except according to a very strict form ofthe EMH requiring that market prices shouldreflect new information instantaneously; inthe real world, it takes time to learn and tech-nical analysis may be one method of learning.Finally, in the fourth strand of explanations,

there are not-fully-rational traders in the mar-ket who have price impact and whose behav-

ior can be detected and exploited by technicalanalysis. Obviously, in this case markets arenot efficient and technicians who are rationalin the sense of exploiting all available infor-mation for trading purposes (whether it isinformation about fundamentals or nonfunda-mental influences) will profit at the expenseof noise traders who are irrational in the senseofnot using all available information.

It seems noteworthy that, in the three

latter explanations discussed here, the

27We agree with an anonymous referee that order flowseems to have a fundamental component and appears tobe related to movements in fundamentals (Evans and

Lyons 2005b).

Figure 3. An Overview of Explanations for the Use of Technical Analysis on Foreign Exchange Markets

temporarily suboptimalbehavior

users underestimatethe risk involved

a marketing instrument toimpress noninformedclients

time-consuming revelationof fundamentals

using patterns from orderflows

sentiment, psychologicalinfluences on prices

technical analysis as a self-fulfilling decision process

as an indication of not-fully-rational behavior

as a means of profitingfrom foreign exchangeintervention

as a means ofprocessingfundamentalinfluences onexchange rates

as a means ofprocessing informationon nonfundamentalinfluences onexchange rates

four positionsexplaining the use

of technicalcurrency analysis


21/37


commonality is that there are price-relevant influences that cannot be addressedby conventional fundamental analysis,either because the central bank intervenesor because fundamentals cannot be

observed or else because nonfundamentalsimpact on prices.28 By using these piecesof information technical analysis does notnecessarily yield excess returns.

We review below the available empiricalevidence with respect to each of these fourpositions.

5.1 Technical Analysis as ReflectingIrrational Behavior

The charge of not-fully-rational behavior

on the part of those applying technical analy-sis is probably the most common position inexplaining its use, since in its reliance onextrapolation and/or visual pattern recogni-tion, technical analysis is inconsistent with

weak efficiency of the foreign exchange mar-ket. However, as mentioned earlier, thisposition has the paradoxical implication thatthe market is in fact not efficient since tech-nical analysis is so widely used in the market

(SF1). Thus, there must be more subtle rea-sons for using technical analysis rather thanjust an outright lack of rationality. In effect,there seem to have been three argumentsput forward in the literature.

First, that the irrational behavior may beof a largely temporary nature.

Second, that it may be the case that usersof technical analysis systematically underes-timate the risks involved in its use.

Third, that the application of technical

analysis may in fact be a form of marketingor window dressing on the part of financialinstitutions in order to impress and attractless-informed clients.

Regarding the first argument, concerningtemporarily irrational behavior, one wouldneed information about the behavior of par-ticipants in the time-series domain to testdirectly whether behavior changes over

timethis data is not available so far. Analternative is to test the cross-sectional impli-cations of this approach, for example thattraders relying on technical analysis tend tobe less experienced and will in some senselearn to use fundamental analysis as theirexperience grows over time. Learning meanshere the same as learning the right model,i.e., the lesson of avoiding technical analysisin the future. Another implication seems tobe that not-fully-rational behavior will not

lead to market success, so that chartists donot reach senior positions as often as others.Finally, nonrationality may be a consequenceof a lower level of education, since it may beargued that technical analysis does notrequire any level of economic understand-ing but isquite the contraryeasilyunderstandable on an intuitive basis.

These three implications of the assump-tion of temporary irrationality on the part of

traders using technical analysis have beentested with survey data of Gehrig andMenkhoff (2006). The details, given in table6, reveal that those market participants whoprefer the use of technical analysis are not,in fact, characterized by symptoms of a pos-sibly suboptimal behavior (see alsoMenkhoff 1997; Cheung and Clement Yuk-Pang Wong 1999; Cheung, Chinn, andMarsh 2004).

This leads to the next argument, put for-

ward by De Long et al. (1990) in the moregeneral context of noise traders, that theapplication of technical analysis may berelated to an underestimation of the riskinvolved by its users. Again, there is nodirect evidence available which could informabout risk preferences and risk perception ofchartists. Moreover, the studies examiningrisk-adjusted profitability do not come to aunanimous conclusion (see section 4). The

only study that directly compares the

28 It is thus only the fourth explanation that requires thepresence of outright nonfundamental forces in the market.Intervention itself, referring to the second explanation,may react on nonfundamental prices or create them. Thethird explanation may be related to nonfundamental prices

when market participants are affected by a preference for

certain figures.


22/37


consequences of relying on technical ratherthan fundamental analysis is that of Curcioand Goodhart (1993). In their experiment,the profit of technical traders and of funda-mentalists was similar but the volatility waslower for the users of technical analysis. This

might indicate, therefore, that technicalanalysis can in fact serve as a risk-reducinginstrument. As this is, however, only a singlestudy, the significance of this result shouldnot be overstated.

Fortunately, there is another piece of evi-dence which can be drawn from the form oftechnical analysis that is preferred. Bothstudies asking this question come to thesame conclusion that trend-following formsdominate rate of change indicators (see

Taylor and Allen 1992, table 1A; Lui andMole 1998, table 4). If one assumes thatmost people would regard going with the

wave as less risky than betting against it, thispreference of available instruments does notindicate risk-loving behavior.

Overall, therefore, the evidence presentedis unavoidably thin. Nevertheless, availableinformation does not support the notion ofchartists being a selection of people who

underestimate risk in general.

There is, finally, a third argument in favorof not-fully-rational behavior on the part oftechnical tradersthe marketing argument.The claim here is not that technical analysiscan provide any useful information in fore-casting but that it generates buy and sell sig-

nals which translate into fee and commissionincome for financial intermediaries (see,e.g., Richard Sylla 1992, p. 343). This viewmay characterize the motivation of thoseselling technical analysis, but it does notexplain why others buy such services. Iftechnical analysis were particularly popular

with small investors or other less profession-al market participants (e.g., day traders),this argument would come close to the firstargument discussed above, i.e., that of sub-

optimal behavior. Unfortunately, there is noevidence that small investors are in fact par-ticularly heavy users of technical analysis,although it is known that a large number ofprofessionals adhere to this tool. In additionto the evidence presented in section 3 it canbe said that, according to the survey ofTaylor and Allen (1992, table 1), most insti-tutions subscribe to some form of externalchartist advice. Moreover, about 25 percent

employ an in-house technical analyst in

TABLE 6THE USE OF TECHNICAL ANALYSIS AS A SIGN OF TEMPORARILY SUBOPTIMAL BEHAVIOR

Hypothesis being tested aggregated figures Pearson 2 probabilityfor chartists v. others

1. Chartists have the same younger than 35 years: 0.645 (0.419)age as other market chartists 55.56% v.participants. others 49.61%

2. Chartists reach senior senior positions reached: 1.395 (0.237)positions as often as chartists 31.94% v.other market participants. others 24.22%

3. Chartists have achieved the university level achieved: 1.254 (0.263)same level of education as chartists 24.64% v.other market participants. others 32.28%

Notes: The source is Gehrig and Menkhoff (2006). Chartists are defined as respondents who attach a greaterweight to technical analysis than to either fundamental or flow information. The number of chartists according tothis criterion was 72 and the number of other market participants was 129 (exact numbers may differ slightly dueto incomplete replies). The achieved university level compounds graduation from university as well as from uni-

versity of applied sciences. The 2-test exploits not only the aggregated figures being presented here, but allavailable information.


23/37


comparison to 39 percent who employ an in-house economist (Taylor and Allen 1992,table 2).

In summary, the evidence regarding thenot-fully-rational behavior position in explain-

ing the use of technical analysis is mostlyquite indirect. Nevertheless, it is interestingto note that available information pointsagainst rather than in favor of this position.29

5.2 Technical Analysis as Exploiting theImpact of Central Bank Interventions

As the central bank is not part of the reg-ular market process and, in particular, for-eign exchange intervention is not generally

motivated by profit considerations, theprocess of central bank intervention in theforeign exchange market may provide anexplanation as to why financial markets areactually efficient although excess returnscan be earned. This idea was formulatedlong agosee, for example, Dooley andShafer (1983, p. 65), Levich (1985), orSweeney (1986)but it had not been testedin a rigorous way until quite recently.

The seminal paper in this respect is the

study by LeBaron (1999). He applies a sim-ple moving average rule to a fourteen-yearperiod (1979 to 1992) of daily as well as

weekly time series of D-Mark/U.S.-Dollarand Yen/U.S.-Dollar exchange rates. Thisrule generates considerable returnsof theorder of more than 5 percent per year(LeBaron 1999, table 4). LeBaron calcu-lates, however, the effect of removing days

when official foreign exchange interventions

took place. The result is that the formerlyhighly profitable technical analysis rulesdiminish in attractiveness (LeBaron 1999,figure 2). This indicates that interventionhas something to do with the observed pre-dictability (LeBaron 1999, p. 137). In order

to address the issue of a possible third factorin this analysis, LeBaron undertakes severalchecks to investigate the existence of com-mon factors that might drive interventionsand profitability of technical analysis at the

same time. One finding is that periods ofhighest expected volatility (calculated usingGARCH models) are not those of highestprofitability of the moving average rules.

The thrust of this literature suggests thatofficial interventions may distort the rela-tionship between standard fundamentalsand exchange rate movements and therebydisadvantage fundamentals-oriented traders

while possibly favoring technical traders, forexample if the intervention creates trends in

the exchange rate or support and resistancelevels.

LeBarons (1999) analysis has beenextended by Saacke (2002). Saacke not onlyconsiders U.S. data but also interventions bythe Deutsche Bundesbank. Moreover, thisstudy covers two additional years and con-siders a wide range of technical analysis rulesand confirms LeBarons findings.

Other studies that make similar argu-

ments concerning the influence of centralbank intervention on technical analysis prof-itability include Silber (1994) and Andrew C.Szakmary and Mathur (1997). Silber (1994)generally links markets where technicalanalysis is profitable with the fact that theseare the markets where central banks inter-

vene. Szakmary and Mathur (1997) examinefive major foreign exchange markets but relyon the IMFs International FinancialStatistics to infer the degree of intervention

from data on foreign exchange reserves.These monthly figures cannot reveal higherfrequency interventions and are influencedby nuisance components, such as revalua-tions or interventions in third currencies.Interestingly, however, they neverthelessreach basically the same conclusion asLeBaron (1999).

This interpretationi.e., that centralbank intervention may be the source of the

profitability of technical analysishas been

29 Despite this interpretation of the available systemat-ic evidence, we do not wish to claim that there isno irra-tionality in the market (Oberlechner 2004; Oberlechner

and Osler 2006).


24/37


cautiously questioned by Neely (1998). Hestresses the point that, due to LeBaronsmethodology, most of the profits from tech-nical analysis rules occur concurrently withintervention operations (see Neely 1998, p.

7f). If official interventions tend to occurwhen markets are trending, this would alsoexplain the findings of LeBaron and others.In particular, if intervention days are those

where markets move heavily (and mightpossibly move even more strongly withoutinterventions), then interventions and tech-nical analysis profitability may be positivelycorrelated. The decisive step necessary to testthis competing interpretation is the use ofintradaily as opposed to daily data.

Neely (2002) has performed this task bycombining several sources of daily data avail-able at different times during the tradingday. He checked the timing of technicalanalysis profitability and intervention for fiveexchange rates, mostly over the period from1983 to 1998. Neely finds that interventionreacts to the same strong short-run trendsfrom which the trading rules have recentlyprofited (Neely 2002, p. 230). The result is

confirmed for a high-frequency analysis ofBundesbank interventions (Michael Frenkeland Georg Stadtmann 2004). It is also com-patible with Neely and Wellers (2001) resulton daily data, namely that their genetic pro-gramming rules are most profitable on theday before interventions take place.30

Moreover, information about central bankinformation does not increase profitability.

In a recent study using daily data on themarkdollar exchange rate and foreign

exchange intervention data from theFederal Reserve and the Bundesbank, Reitzand Taylor (forthcoming) analyze the inter-action of chartism, fundamentalism, andcentral bank intervention and provide evi-

dence that intervention is most likely tooccur and to be effective after a period ofsustained trending away from the equilibri-um level suggested by purchasing powerparity. They argue that this is evidence ofthe coordination channel of interventioneffectiveness, which has been put forwardby Taylor (1995a, 2004) and Sarno andTaylor (2001a). According to the coordina-tion channel, if technical analysts are capableof driving the exchange rate away from its

fundamental equilibrium level over a sus-tained period, then fundamentalist analysis

will not be profitable and fundamentalists will lose credibility in the market, or confi-dence in the fundamentals. Hence, funda-mentalists will reduce their trades based onfundamental analysis and the exchange rate

will tend to stick away from (and perhaps stilltrending away from) the fundamental equi-librium. (This is an example of the limits of

arbitrage effect, as suggested in a more gen-eral setting by Andrei Shleifer and Robert W.Vishny 1997.) When this occurs, the centralbank may at some point intervene publicly inthe hope that the intervention will act as acoordinating signal to fundamentalists toenter the market at the same time and soreturn the exchange rate to its fundamentalequilibrium level. To the extent that funda-mentalists rally to the central banks clarioncall, the intervention will then be effective.

Using a nonlinear microstructural model ofexchange rate behavior, Reitz and Taylor(forthcoming) find evidence supportive ofthe existence of a coordination channel ofintervention effectiveness.

The coordination channel therefore pro-vides a rationale as to why intervention, theuse (or profitability) of technical analysis,and trending exchange rates may all coin-cide. Note, however, that the coordination

channel implies that intervention may be

30 One reason that technical analysis may be profitablebefore interventions was revealed by Bettina Peiers(1997), indicating that one bank had superior forecastingperformance with respect to later interventions. It seemsplausible that this bank had an informa

Taylor Menkhoff Technical Analysis

Documents