I2, PAGES 3995-4002, DECEMBER 1993

The Hurst Effect: The Scale of Fluctuation Approach
Programasde Postgradoen Aprovecharniento de RecursosHidrauticos, UniversidadNacional de Colombia, Medellin

After more than 40 years the so-called Hurst effect remains an open problem in stochastic hydrology.Historically, its existence has been explained either by preasymptotic behaviorof the

resca!ed adjusted rangeR'n, certain classes of nonstationarity in time series,infinitememory,or
erroneous estimation of the Hurst exponent. Various statistical tests to determine whether an observedtime seriesexhibitsthe Hurst effectare presented. The testsare basedon the fact that for

thefamily of processes in theBrownian domain of'attraction, R*n/((On))•/2 converges in distribution
to a nondegeneraterandom variable with known distribution(functional central limit theorem). The

scaleof fluctuation0, definedas the sumof the correlation function,plays a key role. Applicationof the teststo severalgeophysical time seriesseems to indicatethat they do not exhibit the Hurst effect, althoughthoseserieshave beenusedas examples of its existence,and furthermore the traditionalpox diagrammethodto estimatethe Hurst exponentgivesvalueslargerthan 0.5. It turned out that the

coefficient in therelation ofR*n versus n, which is directly proportional to thescale of fluctuation, was
more important than the exponent.The Hurst effect motivated the popularizationof 1/f noisesand related ideas of fractals and scaling.This work illustrateshow delicate the proceduresto deal with infinity must be.



lems such as turbulence (see, for instance, Mandelbrot

The Hurst effect is one of the most important unsolved

problems in stochastic hydrology. There is ampleevidence
tosuppo_rt this statement.Ever sinceHurst' s [1951]original work,there has been a proliferation of papersabout it. Some ofthe most important awards of the hydrologiccommunity havegone to contribution toward its solution. Hydrologists havebeen divided into two schoolsin the attempt to interpret the alleged existence of this anomaly in geophysical records. Some of the most famouspersonalities in probability theory, like Feller [195!], have devoted time to this problem.Mandelbrot [1982] has declared that his original investigation into the Hurst effect was one of the sourcesof
inspirationfor his, now in vogue, fractal theory, whose importance in chaos theory is Without doubt.



Let X•, X2, "', Xn be a sequence of randomvariables representing, for instance,the inflows into an infinite reservoir. Denote the partial sum series(cumulativeinflows)by

So=O St= • Xm

t= 1,2,'", n. (1)

The sample mean represents the ideal release from the reservoir and therefore the adjusted partial sum sequence

S•, defined by

S• = St - (t/n)Sn t = O, 1, 2, ..., n, (2) Despitethe activity cited above, the problemin hydrology is stagnant.There are no clear winners between the short- representsthe fluctuations in the content of this ideal reserandlong-memoryschools. Hurst's original motivation was voir. In his studies for the Aswan Dam on the Nile River, thepracticalproblem of reservoir design,but it was later Hurst [1951] considered the "adjusted range," defined by discovered that only in someparticular casesdid the soRn = max S'•- min S• for 0 -< t <_n. (3) calledHurst effect have significantpractical implications [Klemds et al., 1981]. Nevertheless,economic implications He showed that the adjusted range is a measure of the of the use of differenthydrologic modelsare not insignifi- reservoir capacity required under idealized conditions and cant,andthe way the persistence is modeledis very impor- therefore the study of its properties becomesvery pertinent. tantfrom an economicpoint of view [Zapata, 1987;Mejfa He was particularly interestedin the dependence of R n on andMilldn, 1982;Pereira et al., 1984]. the samplesize n. Obviously, R,• increases with n, but how We presentvarious tests for the existenceof the Hurst fast? The fact that the length of the existing streamflow effect. Applications of theseteststo Hurst's originaldata records is rarely of the order of 100 years motivated Hurst showsthat either there is no Hurst effect or the series is not [1951] to look at various series of different geophysical long enough to providedefinite answers. Evenin the latter phenomena. For that purpose, he defined the "resca!ed case, the Hurst effect is not a natural interpretation. In adjustedrange"
addition to the importance of the studyof Hurst'sgeophysR*n= R•/Dn (4) other applications whereinfinitememorymodels (nonsumwhichis a dimensionless quantity;D n is the samplestandard mable correlograms) havebeenproposed for physical prob-

icalseries, theproposed testmayprove to beimportant in

Copyright 1993 by the American Geophysical Union.
Paper number93WR01686. 0043-1397/93/93WR-01686505.00

deviationof the X i. Hurst [ 1951]used 690 different time seriesof 75 geophysical variablessuchas temperature,rainfall, solar spot numbers, mud varves, tree rings,etc. His empiricalfindingswere




that, in general, R*ngrows like n to a powerof the orderof
0.72 with some variations for the different records, but in all

standard deviationof the Xi, say, or.In that case,R*• is

scaled bycr because it isthelimit ofR*•which isscaled by

the cases the exponentwaslargerthan0.5. This resultwasin Dn. A consequence of the above definition and of the invaricontradictionwith his own theoretical analysiswhich indicatedthat the exponentshouldbe asymptotically 0.5. Feller ance principle is that for a sequenceto exhibit the Hurst that at leastone of the conditions ofthe [ 1951 ] was quick to providefirm theoreticalcomputations for effect,it is necessary of the the caseof independentidenticallydistributed(IID) random functionalcentral limit theorem be violated. Because variableswith finite secondmoments.He computedthe limit generality of these hypotheses something quite dramatic mean and variance of the rescaled adjusted range, and must be happeningfrom the physical point of view anda suggestedthat Hurst's [1951] empirical findings might be searchfor the cause of the violation is very much in order. or a explained by some kind of Markovian dependence.The The discoveryof anomaliessuchas a nonstationarity discrepancybetween the empirical observationshowingthe strong dependence in that large a class of geophysical increaseof the rescaledadjustedrangelike a power of 0.7 or recordswill have profound physical implications. As may be expected,from a retrospectiveviewpoint,the so, and the theoretical expectation that for a wide class of processesthe exponent is asymptotically 0.5 has becometo explanationsproposed by different authors were relatedto be known as the Hurst effect. The exponent H in the the violationof the hypotheses of the invariance principle.
empirical relation There have been some theories essentially related to the

a*n"- nH


dependence structureof the process,to its stationarity, to
the existence of an infinite second moment, and to the

is called the Hurst exponent. To appreciate the different argumentsthat have been put forward as possibleexplanationsof the Hurst effect, it is necessary to use a more precisedefinition.Suchprecisionis important also for the proper understandingof the estimation problem of the next section. To that end, we will use the definition of Bhattacharya et al. [1983, p. 651]: "A sequence
of random variables

preasymptotic behavior of the limiting process.
The strongdependenceexplanation was initiated by Hurst [1951] himself, and by Feller's [195!] suggestionaboutthe Markovian character of the series being responsiblefor the
Hurst effect. However, soon Barnard [1956] showed the existence of a 0.5 convergence for that case. Nevertheless,

Matalas and Huzzen [ 1967]presentedreports of estimates of
H between 0.58 and 0.87 for simulation of first-order autore-

is said to exhibit

the Hurst

effect with

cients. At that time the Markovian explanation seemed n goesto infinity, to a non-zero random variable." This is in adequate, and the autoregressive models were becoming contrast with a very general result, known as the invariance fashionable.However, Mandelbrot and Van Ness [ 1968] put principle or the functional central limit theorem [Ibragimov, things back in proper perspective by recalling that for 1962;BilIingsley, 1968], which implies that under conditions of stationarity andweakdependence R*•/n ø'5converges in processesin the domain of attraction of the Brownian motion, the rescaled adjusted range grows like n to the 0.5 distribution to a random variableR*•/n0.5with mean power (notice that Barnard's result goes back to 1956,and that the invarianceprincipledatesback to Ibragimov [1962]). (6) E(n-ø'SR*•) = (0•r/2)1/2 Mandelbrot and Wallis [!968, 1969] and Mandelbrot and
and variance

exponent H > 0 5 ff (1/nU)R * converges in distribution, as gressiveMarkovian models with different correlation coeffi-

Var (n-ø'SR*•) = 0 (z-2/6- z-/2)


Van Ness [1968]proposed an explanation of the Hursteffect by the strong dependenceof the geophysical seriesand
introduced the so-called fractional Brownian motion and the

where 0 is a positive constant, the so-calledscale of fluctuation, or correlation length scale, a parameter first introduced by Taylor [1921]. It can be shown that 0 is the sum of
the correlation function

fractional Brownian noise (1/f noise), which are processes

0= Z p(m)


with infinitememory, as modelsthat may be usedto simulate the Hurst effect. These important theoretical contributions were not accompanied by either a physical explanation nor an investigationof the structure of dependenceof the different geophysical series.The lackof physical justification of thistheory was pointedout by authorssuchas Scheideger [1970]and Klem•s [1974]. There has not been appreciation for thiskind of questionas the followingquote shows: "I am

where p(m) denotesthe correlationcoefficientbetweenXn and Xn+ m. Equations (6) and (7) correspondto Feller's [195!] previous result, for in that case 0 is one. Besides the mean and the

prepared to argue thata lackof serious motivation ina model that fitsand workswell is muchpreferable to lack of fit in a
model that seems well motivated" [Mandelbrot, !982, p.

variance, the whole asymptotic distribution of the rescaled adjustedrange are ready availablefrom known resultsin the BrownJan motion case [Bhattacharya et al., 1983]. The condition of weak dependenceimplies a correlation function decreasing fast enoughto ensureconvergenceof the seriesin (8). This explains the use of terms such as strong dependence, infinite memory (infinite 0), and short memory(finite 0). Normally, existence of finite second moments for the sequence of Xi is assumed. Then the ergodic theorem

On the estimation side, the contribution of Mandelbrot

and Wallis[1968]wasalsoimportant. They pointed outthat
the way Hurst [195!] estimated the exponent H was not adequate. Hurst [1951] used the equation

H l = log(R*n)/log (n/2)


whichpresupposes that the relation(R*) versusn in loga-

rithmic paper passes through thepointn = 2 R* = I This wasmotivated by weakempirical arguments, because it is

applied toX•2 implies thatDn converges inprobability tothe easy to see that using the biased estimator for O'n, one



obtains R*n= 1 when n = 2. Mandelbrot andWallis [1968] els were used in hydrology to model Hurst's exponents introduced the so-called "pox" diagram. This and other largerthan 0.5. The broken line model [Rodrlguez-Iturbeet ways ofestimating H will bediscussed in thenextsection. al., 1972; Mejia et al., 1972] and the ARMA(1, 1) model
Attempts to explainthe Hurst effectas a consequence of [O'Connell, 1974]are the standardpractice. Indeed, Hipel that ARMA(p, q) models an infinite second momentwereproposed by Moran [ 1964] and McLeod [1978]demonstrated preservethe rescaledadjusted rangeor equivaand Boesand Salas [1973]. However, Mandelbrotand statistically

Taqqu [1979] demonstrated thatanasymptotic relation with lently the Hurst exponentestimatedas (9). One can interpret
of thesemodelsas an attemptto increasethe //= 0.5 holds for a sequence of IID random variables with the philosophy stable distribution and characteristic exponentstrictly less correlation length scale 0 so as to obtain preasymptotic than 2 (seeFeller [1971, p. 169] for a discussion of stable estimates of the Hurst coefficient similar to the observed in
dist•butions). Hurst [1951] recognized the nonstationarities in his origi-

nal geophysical series. He evendesigned anexperiment with "probability cards" which produced sudden changes in the
mean of the process,and obtainedempiricalestimations of the exponent H near 0.71. Klemds[1974]andPotter [1975]

developed simulations with nonstationary models that produced empirical series exhibiting the Hurst effect. Something similar wasobtained by Boesand Salas[1978]with the shifting levels model. It is worth noting that all of these
nonstationary models belong to the Brownian domain of

Bhattacharya et al. [1983] provided clear mathematical demonstration of the existence of the Hurst effect for weakly

empirical records. The conclusionsof the chapter on the Hurst effect in the textbook by Bras and Rodriguez-Iturbe [1984,p. 265]are very illustrativeof the standard practicein engineeringhydrology. This is most noteworthy if one considersthat the physical problem is unsolved. In fact, there is no physical explanation yet for the occurrence of infinitememoryin geophysical series.There is no systematic (physicalor empirical)studyof the correlation structureof theseprocesses. Even thoughthe nonstationarity of some geophysicalprocessesmay be argued on some physicalgrounds[Leopold et al., 1964, p. 61], it is only in very generalterms and it remains to be explained why the trendsproducethe sameHurst exponentin variousgeophysical series. Moreover, the issues of estimation raised by the

explanationsare very relevant and have no dependent processes perturbedby smallmonotonic trends. preasymptotic
Asan example, let Y• be a sequenceof weakly dependent definiteanswers.The importance of the estimationproblem random variables, say IID normal variables with zero mean is reinforced if one considers the claims about the robustness andunit variance, then of the range analysis in current literature on fractals [Mandelbrot, 1982, p. 382; Feder, 1988, p. 194].

Xn -- Yn+ c(m+ n)•


willexhibit the Hurst effect with exponent H dependent on



•e valueof the parameter/3as follows:for - 1/2 </3 < 0, H Notice that the whole puzzle rests on the empirical eviisequalto 1 + /3, for/3 > 0, the exponentH is 1, and for /3< -1/2, and for /3 = 0, the Hurst exponentis I/2 [see dence of the exponent H being larger than 0.5. Also, recall thequestion isnotrelated tominor things: no.7 isalmost Bhattacharya et al., 1983]; notice the discontinuityat/3 = 0. that
with n. With all the history behind but the perspective of First, the class of processeswith infinite memory is no huge unsolved questions present, the least that can be done longerthe only theoretically proved class of processes is to look further into the empirical evidence. exhibiting the Hurst effect. Second, the estimationproblem 3.1. Estimators of H becomes very important, for it is possiblethat very small As was pointed out before, Hurst [1951] originally estitrends may be responsiblefor the appearanceof the Hurst
mated H by means of (9), but this practice was shortly In additionto the infinite memory and the nonstationarity abandonedand substitutedby the least squaresslope in the explanations, there have been theories that present the linear relationof log (n) versuslog R* alongwith some Hurst effectonly as a preasymptotic behavior.This means other variations. How good are those estimators?Very little theconvergence to the theoretical 0.5 exponent is slow,and theoretical work has been done along these lines. As an therefore the empiricalobservations for finitesamplesizes alternative for the complex theoretical issues involved, maygive Hurst exponentslarger than 0.5 [Lloyd, 1967]. extensivecomputer experimentswere performed by Poveda SaMsand Boes [1974] consideredthe equation for the [1987] using Bhattacharya et al.'s [1983] nonstationary expected value of R n for finite n and the case of IID model(equation(10)), given its capacityto producevaluesof variables showingthe preasymptotic behaviorof the ad- H, at will, by fixing fl. This analysis consideredmost of the justed range, andproposed a different way of estimation of estimators reported in the literature [see also Poveda and the exponent H. Gomide[1975,1978]considered the caseof Mesa, 1993]; Table 1 presents Poveda's [1987] results for MarkovJan processes and showed how the preasymptoticone caseand someof the estimators.The performanceof all region is expanded with valuesof p(1) near one, and pro- estimatorsin all caseswas poor. Slightly better resultswere posed a new way of estimation of the exponent H. Salas et obtainedwith a new estimator proposed by Poveda [1987], of the slope Sn of the regression of sample al. [1979a, hi, using autoregressive movingaverage which consists

In all the cases,m and c are arbitrary parameters(c > 0 and m -> 0). The significanceof this demonstrationis twofold.

4 times n0.sfor n = 1000, andthe factorkeeps increasing

of R*m versus m taking onlyvalues of m larger thann, (ARMA)(1, 1)models, showed thatthepreasymptotic region values isexpanded because of either asymmetric marginal distribu- but some degree of arbitrariness remains regarding the tion, large butfinitememory (large 0) andnonstationarity.choice of n. It is worth notingthat parallelto the preasymptotic A conclusionof these computerexperimentsis that all the
explanation of the Hurst effect,variousshort-memory rood- estimatorsof the Hurst's [1951] exponent H performed


3.3. Statistical Tests

TABLE 1. EstimatedValues of the Hurst ExponentH for a SimulatedSequenceof the Bhatthacharya et aI.'s [1983] Nonstationary Model, Equation (10), With /3 = -0.3, c = 1, and m -- 1,000 and
20,000 Record Length
Wallis and Matalas

The visualtest in the GEOS diagram may be improved substantially if an independent estimate of the limit of the

sequence {R*n/n 0.5} isknown. Under thehypotheses ofthe



[ 1951 ]
5 10 25 100 250 500 0.7172 0.6901 0.6571 0.6195 0.6049 0.5939

[ 1970]
0.6544 0.6216 0.5889 0.5760 0.5661

[ 1975]
0.1343 0.3773 0.4137 0.5487 0.4788 0.4660

[ 1987]
0.5408 0.5370 0.5357 0.5399 0.5508 0.5760

functional centrallimit theorem this limit is a random vail. able with known distribution; therefore samplevalues should be around the mean,with deviations of theorder of

thestandard deviation. In fact,forshort memory stationary
processes themean andthevariance aregiven by (6)and (7). As a consequence, standard statistical techniques maybe
employedto test the hypothesisof absenceof the Hurst effect. The only extra parameter needed is the scaleof fluctuationor correlationlength scale 0. Clearly, because of the emphasison the exponent previous studies have over looked the proportionalityconstantin the asymptotic ex.

1,000 2,500 5,000 10,000

0.5753 0.5563 0.5474 0.5637

0.5517 0.5346 0.5223 0.5248

0.5020 0.4966 0.4710 0.5037

0.6257 0.7086 0.7998 0.7921


0.5855 0.5812

0.5372 0.5421

0.5198 0.5177


3.3.1. Test1. Given a sequence X1, X2, ßß' of random

variables,with known scale of fluctuation0, a sample
sequence x l, x2, "',
poorly. This contradicts the alleged robustnessof the range analysis [Mandelbrot, 1982, p. 386].
3.2. Visual Tests

and a level of confidence a, the

sequence does not exhibit the Hurst effect if the sample

values ofR*n/n 0.5 remains intheinterval (q,, - q, +) forlarge
enough n (where q• andq•+ arethe 1 - od2 andaY2 quan-

tiles oftheasymptotic distribution ofR*•/n 0.5, respectively).
In practicalapplications q,- and q,+ canbe approximated by the mean asymptotic value (equation (6)) ---2.3timesthe standard deviation(squareroot of (7)). The value2.3 surely exceedsthe values correspondingto the confidencelevelof 0.95 in the asymptotic distribution. More precise valuesof + maybe computed if desired q,- and qa Test 1 is an immediate consequenceof the definitionof the Hurst effect. Obviously, a test with a not known valueof 0is needed. Various alternative ways for estimating 0 from stationary random time series have been presentedin the literature [Vanmarcke, 1988, p. 327]. A short summary is presented next. The first procedure is by means of the samplecorrelation function using (8). However, this estimator is inconsistent, since its variance does not vanish when the record length becomesvery large; indeed, it exhibits a high coefficient of variation [Vanmarcke, 1988, p. 325]. On the other hand, consistent estimators of 0 can be obtained by usingthe
variance function F( ) and the known fact that under a

Thebehavior of R*n/n H (H = 0.5 andH > 0.5) is the
most natural thing to examine to test for the Hurst effect in

a geophysicaltime seriesinstead of the logarithmicregression of R*n on n. A useful set of diagramswas designedfor that purpose, the so-called "GEOS" (geophysicalrecord) diagrams and R* , with n on the abscissa n/n0.5 on the

ordinates. Recall thatif there exists theHurst effect R*•/n 0.5
will eventually diverge to infinity, whereas if there is no

Hurst effect R*n/n o.5 willconverge toafinite limit, with small
random variation around it. Therefore samplepoints for a
time series which exhibits the Hurst effect will increase

indefinitely in the GEOS diagram. Instead, a time serieswill not possessthe Hurst effect when its G EOS diagram converges to a finite limit.

In a similar way, it is possibleto check the convergence of

R*n/n •, H > 0.5, byscaling thevertical axis bynH. In this
case a geophysical time series which exhibits the Hurst effect, with exponent H, will converge to a nonzero limit, whereasa time serieswithout the Hurst effect will converge
to zero.

condition of weak dependence(finite first moment of [Vanmarcke, 1988, p. 188]) the scale of fluctuationis als0 given by

GEOS diagrams are visual tests for the existence of the Hurst effect in any time series with a long enoughrecord. These diagrams become more powerful tools than the socalled pox diagrams.This superiorityis due to the fact that the diagramis scaleddownproperly, not only with respectto


lim TF(T).
T---> oc


Recallthat the variancefunctionF(T) is simplythe variance of the T averageof the original process. the meanof R* but alsowith respect to the variance and Nevertheless, the ordinaryvariancefunctionestimator is other moments as well. Therefore deviations from the exbiasedand a correction is requiredfor the estimation of 0. pected behavior have the proper significance throughthe Following Vanmarcke [1988,p. 336],the expected value of whole range of n values. Besides, no slope estimationis the estimator of the variance function for an n long zero involvedand, as was pointed out, factors of the order of 4 or mean unit variance sample is more are involved providing a magnifyingview that should
help discriminate the existence of the Hurst effect. In applications, the main limitation of these visual tests is that


- r(n)

if the lengthof the recordis not longenoughit may not be
easy to draw definite conclusions. For instance a conver-

1 - F(n)

r_< n.

gencefrom below may be wrongly interpretedas a continuous increase.

This motivates a corrected estimate F*c as follows ß

r'AT) = r(n) +

- r(n)].





2 o ................................. ;............... .......... ........ ............................................. []

' ' "" '-' "i0











Fig. I•. GEOS dia•m for a time se•s of tree •ngs of a •u•ss Hr(Snake •ver). HoHzont• dashed linescodespond to the •ymptotic me• •d •2 stand•d deviations.

Fig. 1c.

GEOS diagram for St. Lawrence river discharges.

the factorsmultiplyingq,• and q, are affectedby the new
SinceF(n) is not known, finding an estimate 0* of the scale of fluctuation will require some iteration. Using an situationof 0* being an estimate. In a processexhibiting the


Hurst effect, theincrease of R*•/n ø'5withn will eventually
dominate. The hypothesis of the test is also weak dependence and stationarity.

approximate modelfor the variance function(F(T) • O/T,
forlargeT) provides the following expression[Vanmarcke, 1988, p. 337]:
O* =

Notice that the estimation of 0 is by itself a test of the existence of the Hurst effect for stationary processes. In
fact, because of the functional central limit theorem, if 0 is


- r*(r)]n-




finite the exponent h is 0.5 and there is no Hurst effect.
Otherwise, if 0 is infinite then there is Hurst effect and it is not necessaryto perform the tests. In fact, for long memory time series the estimation of the scale of fluctuation in (14) does not converge to a finite limit and therefore test 2 is not suitable. For the nonstationary process of Bhattacharya et al. [1983] the estimation of 0 by any of the means presented above also show divergence to infinity, in concordance with
the theoretical result about the existence of the Hurst effect.

Another possibleway of estimating 0 is by means of the one-side unit area spectral density function at zero. Also, if a shortmemory theoretical model is adjusted to the data, 0 could be estimated from the model, according to the theoreticalexpressionsfor 0 in terms of the parameters of the

For those processestest 2 is not applicable either. In view of the above, the recommendation to deal with an observed time series is to proceed first to the estimation of 0 variance of R*n/(O*n) ø'5 andperform a testin classical the variancefunctionapproachof (14) (see terms. Even though some technical details need to be usingfor instance worked out, the test is in the same spirit of test 1. For this Figure 3b). Stabilization of the estimator with n indicates reason, formal substitution of 0* for the scale of fluctuation finite memory and test 2 may be performed. If there is no intest1 is proposed. The idea is that this will not affectthe stabilizationthere may be three possible causes:the series power of the test significantly.Only small modifications in comesfrom an infinite memory process, or it comes from a nonstationaryprocess or the length of the record is insufficient to estimate 0. As an easy cheek, if the value of n (the length of the record) divided by the estimated 0 is less than, say, 15, the record is short. In these cases, if possible, the length of the record should be increased, and the estimation of 0 repeated. On the other side, there are various ways of

3.3.2. Test2 (Outline). Supposethe scaleof fluctuation 0 is not given, estimate it by means of 0* using any of the methodsdiscussed above, estimate also the size of the

testing and removing nonstationarities.If the problem remains, no conclusion can be inferred from the data alone. In

fact, all extrapolations of the behavior of either the range or the estimator of 0 are equally arbitrary from a statistical point of view, and any decisionshouldbe basedon physical
reasoning. In a related problem, Burg [1967] observed that the

problemwith conventionalFourier spectralanalysisof finite time seriesis that only a finite numberof correlationlagsare estimableand that the truncation in lag spaceresults in a of the true spectralfunctionin frequencyspace. 0 5• 1•)0 "1• 2(•)0 '2•0 300 smoothing n Burg [1967]arguedthat the criterion for extrapolationshould Fig. lb. GEOS diagram fortemperature incentral England series. be to obtainthe spectraldensityestimatethat corresponds to



TABLE 2. Estimationof H According to Siddiqui[1976]
Mstouis Neumunas Danubio Rhin

96 132 120 150

1.6 1.4 1.0 1.0

Poveda [1987]
0.451 0.499 0.495 0.4984

Hipel and McLeod [1978]
0.591 0.591 0.495 0.484

andMcLeod[1978] on ARMA(p, q) models fittedto some ofthose series allowed estimation of 0 using (15).In Figures l a, lb, and lc threeof thoseGEOS diagrams areshown. Forthemajority ofthecases, sample values ofR*n/n o..• seem to indicate convergence to the asymptotic theoretical distil. butionof the fittedARMA(p, q) model.Moreover, estima.

tionof the Hurstexponent H in the way suggested by
Siddiqui [ 1976] alsoshows the lackof existence of theHurst






Espafiola Temp Precip
Minimum Snake Exshaw Naramata Dell Lakeview Ninemile

350 255 100
848 669 506 515 655 544 771

44.8 1.6 1.0
24.6 3.9 3.9 2.1 3.8 5.9 7.4

0.455 0.521 0.473
0.462 0.475 0.420 0.435 0.475 0.499 0.466

0.927 0.640 0.473
0.746 0.663 0.580 0.543 0.667 0.706 0.642

effect inthesetofgeophysical timeseries analyzed byHipel
andMcLeod[1978](seeTable 2). According to theestimated

scaleof fluctuation all the series but OdgenandEspafiola haveratiosof n over 0 largerthan30, indicating adequate
length of the records.

Nevertheless, time series corresponding to mud varyes exhibit GEOS diagramsthat always increasewith the value

of n (seefor instance Figure 2), although the asymptotic
value given by (6) for the fitted short memory modelis unknown in these cases. Two facts could explain this situa.

Eaglecol Navajo

858 700

9.3 2.9

0.485 0.468

0.701 0.584

Tioga Bigcone

661 509

3.5 3.6

0.498 0.404

0.691 0.691

tion:thetimeseries is notlongenough to reach theasymptoticmean of R*•/n ø'5or those timeseries actually exhibit the Hurst effecteitherbecause of nonstationarity or long

For further illustration of the ideas presented,a 18,000 long series of vertical wind velocity sampledevery 0.I s collected with a very precise instrument was analyzed. H the most random or unpredictable stochastic process whose Figure 3a shows the pox diagram with the exponent correlationfunction is consistentwith the given information. estimatedusing traditional estimationindicatingthe pres. Using the maximum entropy method he derived a widely enceof the Hurst effect (least squaresslopeof 0.773). Figure used procedure for the estimation of the spectral density. 3b shows the estimation of the scale of fluctuation; 0 canbe to be of the order of 55 notwithstanding sampling The techniqueis basedon obtaininga data modelthat is least estimated informative with respectto data that are not available. This fluctuationsand after observing stabilization. Figure 3c maximum entropy method for estimation of the spectral showsthe GEOS diagram and the proposed test indicates density is equivalent to the linear prediction method that that there is no Hurst effect that can be inferred from this assumethat the underlyingdata model is an autoregressive record(asymptoticmean around 9.3). A straightline of slope 0.5 and intercept equalto ((0•-/2)) 1/2was alsodrawn in (finite memory) model [Roy et al., 1991].
Figure 3a for further clarification of the test. The least

squaresline in the pox diagram has both the slopeand intercept free and fits the preasymptotic domain. The 0.5 to predict At this point an obvious question arises, Do the classical slopeline hasno fittingparameterand is supposed time series which have been used to illustrate the existence the asymptotic behavior of the rescaled adjusted range. of the Hurst effect really possessit? To elucidate the Nevertheless, some may extrapolate the 0.773 powerlaw and othersmay considerthat the lengthof the series questionGEOS diagramswere plottedfor severalgeophys- increase ical time series. Basically, the formal version of test 2 was doesnot permitdefiniteconclusions yet. In any case,thetest performedusing GEOS diagramswith 0 estimators basedon may be repeated if a longer series is available. the parameters of short memory models fitted to the observations.Indeed, for ARMA(p, q) modelsin the senseof Box 7 and Jenkins [1970], the scale of fluctuation is given by [Siddiqui, 1976]

1--Ea j


0* = -1i=1

qb i



where •'0 is the ratio of the variance of the processto the variance of the noise, and qband a are the ARMA(p, q) parameters. This procedure to estimate 0 was used because the originalserieswere not available.Values of R*n versus n for variousgeophysicalserieswere taken from the extensive work by Hurst et al. [1965]. Additionally, resultsby Hipel





Fig. 2. G EOS diagramfor Lake Saki mud varves.






m c•



•oo .....

i'6'oo ....

' t

i 6•





i •36ooo

Fig. 3c.

3_ •


1'0 1'2' 1'•' 1'6 1'8' 20


TIME (0.1 SECONDS) (Thousands)

Fig. 3a. Pox diagramfor a vertical wind velocity time series. The dotted line is a least squares fit with slopeof 0.773,and the solid

GEOS diagram for a vertical wind velocity time series.

line isa 0.5slope with((•r0/2))1/2 intercept.
gence to a finite limit, the tests presented here are not conclusive.In these casesany extrapolation of the asympTo determine if a finite time series exhibits the Hurst effect totic behaviorof the rescaledadjustedrangeis not supported isa delicatematter. Pox diagramsalone are not sufficientand by observations and should be based on physical evidence. further testsare proposedin that respect. Various statistical Application of the tests to data used by Hurst do not show tests were developedto determine the existenceof the Hurst existenceof the Hurst effect. Only in the case of mud varyes effect. They are formulated in precise mathematicalterms. is there spacefor some speculation. However, even in that Their basis is the so-called GEOSdiagrams of R*n/n ø'5 case, long memory does not appear to be the natural versus n that for a short memory stationary processwill explanation of the evidence which might be related to convergeto a known distribution, whereas for a series nonstationarity of the records. exhibiting the Hurst effect will divergeto infinity. The scale of fluctuation is, with the mean and the variance, Theestimation of the scaleof fluctuationis itself a way of one of the most important parametersin hydrological stotesting for shortmemory stationarity and for computing the chasticmodeling, since they completely describestatistical asymptotic value for the GEOS diagram.Additionally, the propertiesof a time series such as central tendency, fluctuscale of fluctuation providesa way of determining the length ations around the mean, the whole correlation structure, the of a time series. "Hurst characteristics" of the time series, the low- and

The testsintroducedin this work do not have adjustable high-frequency componentsof the signal power spectrum, the so-called Noah and Joseph effects [Mandelbrot and byindependent means,but evenwithoutit GEOS diagrams Wallis, 1968], etc. Additional advantagesof a better knowlmayserveto indicate the presenceof the Hurst effect. If the edge of the scale of fluctuation of a stochasticprocess, estimationof the scale of fluctuation does not show conver- besidesits role in the definition of the Hurst effect, lies on its
parameters.The scale of fluctuation needs to be estimated

importance to study some characteristics of a time series suchas runs,run lengths,level crossings, passage times and, in general, probability distribution of extreme values.
The so-called Hurst effect and other related anomalities in

geophysical time series are probably the result of a mixture

of scalesmore than infinite memory. Knowledgeof those scalesis a more fundamentalissue from a physical viewpoint. Modelingthat mixture of scalesis more simpleand down to earth than infinite memory modeling.
Acknowledgment. Wind data used in this work were kindly provided by Roger Shaw (The University of California at Davis) from an experiment funded by Environment Canada under the direction of Hartogand Neumannof the Atmospheric Environment
Service of Canada.









552-553, 1956.



Barnard,G. A., Discussion of Hurst, Proc. Inst. Civ. Eng., 5(5),

Bhattacharya, R. N., V. K. Gupta, and E. C. Waymire,The Hurst velocity timeseries. The solid curve is TF(T), andthedotted curve effectundertrends, J. Appl. Probab., 20(3), 649-662, 1983.
0. Is.

Fig. 3b. Scale of fluctuation estimation for a vertical wind

isthe estimator given by(14).Units ofthescale offluctuation arein

Billingsley, P., Convergence of Probability Measures, JohnWiley,
New York, 1968.



Boes, D.C., and J. D. Salas, On the expectedrange of partial sums of exchangeablerandom variables,J. Appl. Probab., 10, 671-677,

sidad Nacional, Facultad de Minas, Medellfn, 1982.

con t•nfasis enel Problema dela Informaci6n Escasa, Univer.

Boes, D.C., and J. D. Salas, Nonstationarity of the mean and the Hurst phenomenon, Water Resour. Res., 14(1), 135-143, 1978. Box, G. E. P., and G. Jenkins, Time Series Analysis, Forecasting and Control, Holden-Day, Oakland, Calif., 1970. Bras, R., and I. Rodriguez-Iturbe, Random Functions and Hydrology, Addison-Wesley, Reading, Mass., 1984.

Mejfa, J. M., I. Rodrfguez-Iturbe, and D. R. Dawdy, Streamflow simulation, 2, The brokenline process as a potential model for hydrologic simulation, WaterResour.Res., 8(4), 931-941, 1972. Moran,P. A. P., On therange of cumulative sums, Ann.Inst.Stat.
Math., 16, 109-112, 1964.

O'Connell, P. E., Stochastic modelling of longtermpersistence in streamflow sequences, Ph.D. thesis, Dep. of Civ. Eng.,Imperial
Coil., London, 1974.

Burg, J.P., Maximum entropyspectralanalysis,in Proceedings of
the 37th Annual International SEG Meeting, Society of Exploration Geophysicists,Oklahoma City, Okla., 1967. Feder, J., Fractals, Plenum, New York, 1988. Feller, W., The asymptotic distribution of the range of sums of independent random variables, Ann. Math. Stat., 22, 427-432,

Pereira,M. V. F., G. C. Oliveira,C. G. Costa,and J. Kelman.
Stochastic streamflowmodelingfor hydroelectricsystems, Water
Resour. Res., 20(3), 379-390, 1984.

Potter,K. W., Comment on "The Hurst phenomenon: A puzzle:"
by V. Klem•,s, Water Resour. Res., ••(2), 373-374, 1975. Poveda,G., El Fen6menode Hurst, M.S. thesis,Univ. Nacional de
Colombia, Medellfn, 1987.

Feller, W., An Introduction to Probability Theory and Its ApplicaPoveda,G., and O. J. Mesa, Estimation of the Hurst exponent and tions, vol. 2, John Wiley, New York, 1971. GEOS diagramsfor a non-stationarystochasticprocess,in ProGomide, F. L. S., Range and deficit analysisusingMarkov chains, ceedings of Stochastic and StatisticalMethodsin Hydrology and Hydr. Pap. 79, Colo. State Univ., Fort Collins, 1975. Environmental Engineering: An International Conferencein Gomide, F. L. S., Markovian inputs and the Hurst phenomenon, J. Honor of Professor T. E. Unny, University of Waterloo, WaterHydrol., 37, 23-45, 1978. loo, Ont., Canada, 1993. Hipel, K. W., and A. I. McLeod, Preservation of rescaledadjusted Rodrfguez-Iturbe, I., J. M. Mejfa, and D. R. Dawdy, Streamflow range, 2, Simulation studies using Box-Jenkins models, Water simulation, 1, A new look at Markovian models, fractional Gaus. Resour. Res., 14(3), 509-516, 1978. siannoiseand crossing theory, Water Resour. Res., 8(4), 921-930, Hurst, H. E., Long term storagecapacity of reservoirs, Trans. Am. 1972. Soc. Civ. Eng., 116,770-779, 1951. Roy, R., B. G. Sumpter, G. A. Pfeifer, S. K. Gray, and D. W. Noid, Hurst, H. E., R. P. Black, and V. M. Simaika, Long-Term Storage: Novel methodsfor spectral analysis, Phys. Rep., 205(3), 109-152, An Experimental Study, !45 pp., Constable,London, 1965. 1991. Ibragimov, I. A., Some limit theorems for stationary processes, Salas,J. D., and D.C. Boes, Expected range and adjusted range of Theor. Probl. Appl., 7, 349-382, 1962. hydrologicsequences,Water Resour. Res., I0(3), 457--463, 1974. Klem•.s, V., The Hurst phenomenon--a puzzle?, Water Resour. Salas,J. D., D.C. Boes, V. Yevjevich, and G. G. S. Pegram,Onthe Res., 10(4), 675-688, 1974. Hurst phenomenon, in Modeling HydrologicProcesses, edited by Klem•s, V., R. Srikanthan, and T. A. McMahon, Long-memory H. J. Morel-Seytoux, Water ResourcesPublications,Fort Collins, Colo., 1979a. flow models in reservoir analysis:What is their practicalvalue?, Water Resour. Res., 17(3), 737-751, 1981. Salas, J. D., D.C. Boes, V. Yevjevich, and G. G. S. Pegram,Hurst Leopold, L. B., M. G. Wolman, and J.P. Miller, Fluvial Processes phenomenonas a pre-asymptoticbehavior, J. Hydrol., 44, 1-15, 1979b. in Geomorphology, W. H. Freeman, New York, 1964. Lloyd, E. H., Stochastic reservoir theory, in Advances in HydroScheideger,A. E., Stochasticmodelsin hydrology, WaterResour. Res., 6(3), 750-755, 1970. science, vol. 4, edited by V. T. Chow, pp. 281-339, Academic, Siddiqui, M. M., The asymptotic distributionof the rangeandother San Diego, Calif., !967. Mandelbrot, B. B., Intermittent turbulence in self-similar cascades: functionsof partial sumsof stationary processes,Water Resour. Res., •2(6), 1271-1276, 1976. divergence of high moments and dimensionof carrier, J. Fluid Mech., 62, 331-358, 1974. Taylor, G.I., Diffusion by continuousmovements, Proc. London Math. Soc., 2(20), !96-211, 1921. Mandelbrot, B. B., The Fractal Geometry of Nature, W. H. Freeman, New York, 1982. Vanmarcke, E., Random Fields: Analysis and Synthesis,MIT Press, Cambridge, Mass., 1988. Mandelbrot, B. B., and M. Taqqu, Robust R/S analysisof long run of hand serialcorrelation, paperpresentedat 42nd international Statistical Wallis, J. R., and N. C. Matalas, Small sampleproperties estimators of the Hurst coefficient h, Water Resour. Res., 6(6), Institute, Manila, 1979. 1583-1594, 1970. Mandelbrot, B. B., and J. W. Van Ness, Fractional Brownian motions, Fractional Gaussian noises and applications, SIAM Zapata, A.D., El Problema de la Informaci6nEscasaen la Rev., 10(4), 422--437, 1968. Operaci6nde un SistemaHidrotdrmico, M.S. thesis,Univ. Nacional de Colombia, Medellfn, 1987. Mandelbrot, B. B., and J. Wallis, Noah, Joseph, and operational hydrology, Water Resour. Res., 4(5), 909-918, 1968. Mandelbrot, B. B., and J. Wallis, Roboustness of rescaledrangeR/S O. J. Mesa and G. Poveda, Programas de Postgradoen in measurementof noncyclic long run staticaldependence,Water Aprovechamiento de Recursos Hidraulicos, Universidad Nacional Resour. Res., 5(5), 967-988, 1969. de Colombia,ApartadoA•reo 1027, Medellin, Colombia. Matalas, N. C., and C. S. Huzzen, A property of the range of partial sums, Proc. Int. Hydrol. Syrup., 1,252-257, 1967. (Received October 6, 1992; Mejfa, J. J., and J. MilMn, Efecto de la memoria larga de los revised June 1, 1993; procesoshidro16gicossobre la operaci6n de los sistemashidroaccepted June 21, !993.) t6rmicosde generaci6ne16ctrica,in Seminario Sobre Hidrologia