Вы находитесь на странице: 1из 6

A New Look at Clustering of Stock Prices

Author(s): Victor Niederhoffer


Source: The Journal of Business, Vol. 39, No. 2 (Apr., 1966), pp. 309-313
Published by: The University of Chicago Press
Stable URL: http://www.jstor.org/stable/2351480
Accessed: 19-03-2015 17:28 UTC

REFERENCES
Linked references are available on JSTOR for this article:
http://www.jstor.org/stable/2351480?seq=1&cid=pdf-reference#references_tab_contents

You may need to log in to JSTOR to access the linked references.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to The Journal of
Business.

http://www.jstor.org

This content downloaded from 148.210.115.53 on Thu, 19 Mar 2015 17:28:04 UTC
All use subject to JSTOR Terms and Conditions
A NEW LOOK AT CLUSTERING OF STOCK PRICES*
VICTOR NIEDERHOFFERt

T HE psychology of numbersis a fas- cated analyst would do well to study the


cinating subject which should re- effect of investor psychology upon price
ceive more attention from stock movementsin stable situations as well as
marketanalysts. Until recentlytherewas in times of bandwagonbuying or stam-
little systematic research in the field. pede selling.
Leads can be gathered from the work in- Such an approachis particularlyrele-
other disciplines.Motivational research- vant to an understandingof the contro-
ers have convincedmerchandisersthat a versy surroundingthe theory that stock
sales price of $2.99 is a far more attrac- market prices ending in even eighths
tive figure for the purchaserthan one of occur far more frequently than prices
$3.00. It is common knowledge among endingin odd eighths.The evidenceseems
policy bookies that thousandsof players to show that this peculiar phenomenon
will flock to place their bets on the same depends to a large degree on the predi-
number at the occurrenceof some un- lection of investors to place orders at
usual symbolic event. roundnumbers.Almost all marketorders
The literatureof the stock marketcon- are matched with limit orders. Hence,
tains similar illustrations: from Gann the record of transaction prices follows
writing in 1930,1who found that "The closely the record of limit orders. But
populartrading prices are 25, 40, 50, 60, some 84 per cent of limit orders on the
75, 100, ... ."; to Wyckoff in 1963,2 who books of specialists are typically at even
declaredthat "We think in round num- eighths.' Therefore, we might naively
bers and try to sell at round numbers." expect a certain amount of clusteringat
The violent, short-runmovements of round numbers, which are, of course,
the stock marketaveragesare frequently numbers ending in an even number of
ascribedto heavy publicbuying or selling eighths.
as the average moves through certain Osborneappearsto have been the first
round numbers. However, the sophisti- one to test this rigorously.4He sampled
* The author wishes to express his gratitude to the closingpricesof stocks in which there
Professors James Lorie and Nathan Keyfitz of the were sales of more than fifty round lots
University of Chicago and Arthur Niederhoffer of in a week. Of all closing prices, 60.8 per
Hofstra University for their helpful suggestions.
Financial support for this study was provided by the cent were at even eighths. Hence, we
Center for Research in Security Prices, University of christen this tendency for even eighths
Chicago. to appear more frequently than odd
t Graduate School of Business, University of eighths the "Osborne Effect." Further
Chicago.
confirmationof the Osborne Effect was
1 W. D. Gann, Truth of the Stock Tape (New
York: Financial Guardian Publishing Co., 1930), 3V. Niederhoffer. "Clustering of Stock Prices,"
pp. 63-64. Operations Research, XIII (1965), 258-65.
2 Peter Wyckoff, The Psychology of Stock Market 4 M. F. M. Osborne, "Periodic Structure in the

Timing (Englewood Cliffs, N.J.: Prentice-Hall, Inc., Brownian Motion of Stock Prices," Operations Re-
1963), pp. 106-7. search, X (1962), 345-79.
309

This content downloaded from 148.210.115.53 on Thu, 19 Mar 2015 17:28:04 UTC
All use subject to JSTOR Terms and Conditions
310 THE JOURNALOF BUSINESS
obtained by the present author whose eighth removed from the previous
data included the closing prices of five transaction.
Stratum 3. All transactions at a price one-quar-
consecutive samples of two hundred ter or more removed from the previ-
stocks whose prices remainedunchanged ous transactions.
for the day and were selling at a price
below 50. The proportionof all closes at For example, the prices of the first
the even eighths was .66.1 eight transactions in American Airlines
Godfrey and his co-workers6also at- on November 13, 1964,were respectively,
tempted to test the frequenciesof even 451, 45k, (455), 451, (451), 451, (455),
and odd eighths. To do this they exam- 451. The last price would be placed in
ined all ticker transactions in Bell and Stratum 1; the prices inside of paren-
Howell,and InternationalTelephoneand theses, in Stratum2; the unmarkedprices
Telegraph during April, 1963. They in Stratum 3. It turns out that the clos-
found no significantdifferences. ing price for American Airlines on No-
Although Godfrey's sample covered vember 12, 1964, was 461. Hence, the
all prices and that of Osborne and this openingprice on November 13, 453, also
author included only the closing prices, belongs in Stratum 3. Within each stra-
the conclusionsseem contradictory.This tum, it is important to observe that
paper explores the contradiction, sug- the occurrencesof consecutive fractional
gests that Godfrey'sresults are attribu- prices in a given stock on a single day
table to special characteristicsof the two are not independentof each other. Con-
stocks he studied, and offers a few sug- sider, for example, the transactionsone-
gestions for trading proceduresbased on eighth removedfrom the previous trans-
the findings. action (Stratum2) in a given security. It
is obvious here that if the first transac-
CLUSTER SAMPLING FORCLUSTERED tion one-eighth removed from the pre-
TRANSACTIONS vious price is at an odd eighth, then the
We turn now to a sampling procedure chancesare extremelyhigh that the next
for ascertaining the proportion of all transactionfalling into Stratum 2 will be
transactions in the stock market at even at an even eighth. The proportion of
eighths. The data consisted of the com- transactions falling in Stratum 2 in the
plete record of ticker transactions in cluster units varies less from its true
1964. Our source of data was the Francis value, which turns out to be 0.51, than
Emory Fitch Daily Stock Market Sales would be expected from the binomial
(Blue Sheets) on which the price and formulafor the variance of a proportion
volume of each transactionon the ticker
V(P) _(P) (1-P)
tape are arrangedalphabeticallyby com-
pany name. These transactions were
then grouped into three strata as follows: where P equals the observed proportion
of transactionsat even eighths. Converse
Stratum1. All transactionsat a priceequal to
the previoustransaction. reasoningapplied to sampling in Strata
Stratum2. All transactionsat a price one- 1 and 3 indicates that the binomial for-
5 Niederhoffer,
mula will understatethe true varianceof
op. cit. the proportionof all transactionsat even
6M. D. Godfreyet al., "The Random-WalkHy-
pothesisof Stock MarketBehavior,"Kyklos,XVII eighths.
(1964),1-30. A type of single-stageclustersampling

This content downloaded from 148.210.115.53 on Thu, 19 Mar 2015 17:28:04 UTC
All use subject to JSTOR Terms and Conditions
CLUSTERINGOF STOCKPRICES 311

of units was adopted to adjust for this transactions in 1964 falling into Strata
intercorrelation between transactions. 1, 2, and 3 were, respectively, .47, .41,
Seven days were chosen randomly for and .12. Thus our estimate of the pro-
each stratum, and a specifiednumber of portion of all transactions on the New
stocks, say (M1, n2, . .. , n7) were chosen York Stock Exchange at even eighths is
by proceedingdown the sheets in alpha- given by .47 X P1 + .41 X P2 + .12 X
betical order. For each day, we ascer- P3 where Pi, i = 1, 2, 3 equals the ob-
tained the proportionof all transactions served proportionof transactionsfalling
at even eighths within the specifiedstra- at even eighths in that stratum.
tum in each stock. Hence, for each stra- Table 1 summarizesthe results of this
tum an estimate of the proportion and procedure.For each stratum it shows (a)
variance of all transactions at even the size of the sample, (b) the proportion
eighths is provided by the usually used of transactionsat an even eighth, (c) the
TABLE 1
INFORMATIONON PROPORTIONOFTRANSACTIONSAT EVEN EIGHTHS

A B C D Percentage
Classification No. of Proportion Correct Binomial in Total
at Even Variance Variance
Observations Population
Eights of (b) of (b)

Stratum 1 .........10..... I 0 .62 .00074 .00024 .47


Stratum 2 .............. 12,800 .51 .0000029 .00019 .41
Stratum 3 .............. O11,000 .771 .000029 .000018 .12

Total sample estimate.. 24,800 .585 .00017 .000056 .........

formulas for cluster campling with n = correct variance of (b), (d) the incorrect
nl + n2 ... + n7 cluster units.7 That is, binomial variance, and (e) the fraction
n of the universe of all transactionsin the
stratum. Entries in the lowest row give
the estimates for the population values
n
of (a), (b), (c), and (d). (The samplesizes
were determinedlargely by the dictates
and of another investigation; hence they are
not optimum.)
E a - 2PE aimi+P2 m The second line (Stratum 2) of Table
VtP)- (n-1)Xnm2 1 indicates that only 51 per cent of all
transactions one-eighth removed from
wherea, equals, respectively,the number
previoustransactionoccurredat even
of transactionsin the ith cluster unit at the
for the esti-
even eighths, ml equals the total number eighths. The true variance
of transactionsin the ith clusterunit and mate is a mere 19 per cent of the corre-
m = Im /n. sponding binomial estimate. Fully 71 per
Fortunately, from previous investiga- cent of transactions one-quarter of a
tions, we knew that the proportionof all point or more removed from the previous
7W. G. Cochran,SamplingTechniques(2d ed.; transaction
took place at even eighths.
New York:John Wiley & Sons, 1961),p. 65. It is notable that 92.6 per cent of 819

This content downloaded from 148.210.115.53 on Thu, 19 Mar 2015 17:28:04 UTC
All use subject to JSTOR Terms and Conditions
312 THE JOURNALOF BUSINESS
transactionsone-halfof a point removed vey of the two stocks chosenby Godfrey
from the previous price fell at even et al., but in the month of October, 1964.
eighths. However, it must be recalled It was reassuringto find that 56.1 per
that Stratum 3 accounts for only 12 per cent of the 581 transactionsin I.T. & T.
cent of all transactions.In line 1 (Stra- occurred at even eighths. In Bell and
tum 1) we observethat 62 per cent of all Howell, 51.3 per cent of the 254 transac-
transactionsat the same price as the pre- tions fell at even eighths.
cedingtransactionare at an even eighth. Instead of limiting himself to two
The gross underestimate that the bi- stocks, Osbornetook a systematic sam-
nomial formula provides for the true ple of closing prices during the day of
variance in Stratum 1 is the result of a April9, 1964,of all stocks beginningwith
discernibletendency for transactions at A-K, in which there were sales of more
even (odd) eighths to follow each other. than fifty round lots in the week. It
Finally, a glance at the bottom row turned out that approximately 55 per
(total sample estimate) discloses the cent of these transactionswereno eighths
answer to our question. Approximately removed from the previous transaction.
58.5 per cent of all transactions in the About 26 per cent of the transactions
stock market are at an even eighth. The were one-eighth away, and 19 per cent
coefficientof variation for this estimate were two-eighths or more removed.9
is 2.2 per cent. Supposewe considerthese finalchanges
Now it becomes a simple matter to duringthe day as randomdrawingsfrom
reconcilethe results of the variousinves- the population of all changes between
tigators. Godfrey and his co-workers transactions.Then an estimate, based on
used the completemonthly data for only Table 1, of the proportionof daily closing
two New York Stock Exchangestocks to prices at even eighths in Osborne'ssam-
test the hypothesisat issue. One of these, ple is
Bell and Howell, was relatively low .55 X .62 +.26 X .51 +.19, and
priced, and low-priced stocks tend to
show less clustering than high-priced X .71 = .609.
stocks.8The other stock, I.T. & T. was
The estimate obtained agrees well with
then (April, 1963) relatively stable and
active. For example,all of the more than the value of .608 reported by Osborne.
nine hundred transactions in I.T. & T. Our conclusionsare that prices ending
during that month were executed at in even eighths do actually occur with
prices between 473 and 46.' Therefore, greater frequency than those ending in
where the total range of variation was odd eighths, and that Godfrey and his
only five-eighthsof a point, it is reason- co-workerswere misled by the special
able to assume that very few of those characteristicsof the prices they studied.
transactions occurred more than one-
DERIVED TRADING METHODS
eighth of a point away from the previous
transaction. This would tend to lower The sophisticated investor may be
the expected proportion of transactions able to utilize this knowledgeof relative
at even eighths in Godfrey'ssample. frequenciesfor his own advantage.There
The present author undertook a sur- is a further discussion of this subject in
8 Niederhoffer,op. cit. 9Osborne,op. cit.

This content downloaded from 148.210.115.53 on Thu, 19 Mar 2015 17:28:04 UTC
All use subject to JSTOR Terms and Conditions
CLUSTERINGOF STOCKPRICES 313
a forthcomingarticle.'0It is well to re- 100, just long enough to cover all the sell
memberthat the congestionresults from limits at 100 except for an amount suffi-
the tendency for the great majority of cient to match their purchaseorder.The
limit ordersto be placed at even eighths. reason for this is that so many sell-limit
Within this group there is a special pref- ordersare placed at the fascinatingfigure
erence for familiar whole numbers like of 100 that the great volume of these
10, 25, 50, 75, and 100. For example,the orders would act as a resistance to an
author in anotherstudy receivedpermis- advance beyond 100.
sion to examinethe book of the specialist A rigoroustest of the validity of these
in a well-knowncorporationthen trading suggested trading rules is not possible.
at 79. Forty per cent of the limit orders In general,the tradingrules are based on
to purchasea total of five hundredround the supposition that a reversal in direc-
lots rested at the attractive number 75, tion of price change becomes likely as
and 26.5 per cent of 337 roundlots to sell price moves to the round number. Some
were placed at 80.11 informationcontained in an above-men-
Let us considerthe case of a small in- tioned work12shows that this hypothesis
vestor who recently bought a stock at 72 is true for intervals as short as between
and would like to sell somewhereabout consecutivetransactions.The readercan
80. He would be well advised to place verify this effect for yearly highs and
his sell-limit order at 797 rather than at lows by counting, say, the number of
80. On the average,therewill be a greater yearly highs or lows from 49' to 497 and
concentration of sell orders at 80 than comparing this to the number of highs
at any other nearby prices. The sell or- or lows from 501 to 50. There are more
ders create a bearish effect, but the in- highs than lows just below 50 and more
vestor must decide whether the benefits lows than highs just above. Of course,
from a quick sale at 7987are worth the the proposed rules will change if too
possiblesacrificeof one-eighthof a point. many investors follow them.
On the other hand, if a mutual fund The introductionof psychologicalvari-
contemplatedthe purchaseof this stock ables adds one more dimension of com-
after it brokethrough80 and rose to 998, plexity to a market system which re-
it might be wiser for them to postpone sponds so sensitively to pressures from
the purchase until the issue traded at every front of society. If stock market
analysis is to attain the status of a scien-
10V. Niederhofferand M. F. M. Osborne,"Mar- tific discipline rather than remain an
ket Makingand Folkloreon the Stock Exchange,"
Journal of American Statistical Association (in
esoteric cabalistic ritual, we must open
press). our mathematically oriented models to
11V. Niederhoffer,"Non-Randomnessin Stock embracerealmsof data which have been
Prices:A New Modelof PriceMovements"(unpub- almost ignoreduntil recently.
lished thesis, Departmentof Economics,Harvard
University,1964). 12 Niederhofferand Osborne,op. cit.

This content downloaded from 148.210.115.53 on Thu, 19 Mar 2015 17:28:04 UTC
All use subject to JSTOR Terms and Conditions

Вам также может понравиться