United States: (12) Patent Application Publication (10) Pub. No.: US 2013/0046710 A1
United States: (12) Patent Application Publication (10) Pub. No.: US 2013/0046710 A1
1 0O 4‘
FINANCIAL msmumsm
swam suscroa
user; q =
.1215? HQ 155
Li m’remcnvs RESULTS
135' \
Rmum
ADDITIONAL _ _ .
RECEIVE
3Q m TRADING 0am
Puauc DATA or FINANCIAL msmumems Lg
Patent Application Publication Feb. 21, 2013 Sheet 3 0f 7 US 2013/0046710 A1
324 / ans
' aaa/
Fig. 3
Patent Application Publication Feb. 21, 2013 Sheet 4 0f 7 US 2013/0046710 A1
400
,v-"aaz
CHQGSE m mmm.
rams was
GENERATE AYAELE 0F
CGMPAQAELE HNAHQAL
WSTRUMENTli DANE F051 THE
"5W5 SUCE
J 1163
4
12 CHQDSE AH
AWL? A DEQHQN TREE V V V V
ALQQRUHM “ME SUCKS UM’RC‘CESSED
ii??? TiME SLMCE
AL
,414
FM‘ER. OUT MOESY RESULTS 7 1r , 7
END
Fig. 4
Patent Application Publication Feb. 21, 2013 Sheet 5 0f 7 US 2013/0046710 A1
71%.“
am
JEUZd m3.
m ./“
Sm
,3»
4.52? mhzwang 13m
wmm
m.QE
12m
com
$2
SJNEQH 33.33V.2a
Patent Application Publication Feb. 21, 2013 Sheet 6 0f 7 US 2013/0046710 A1
1/
$15,92.1?4 /!
/{5./1,
E8E
Patent Application Publication Feb. 21, 2013 Sheet 7 0f 7 US 2013/0046710 A1
US 2013/0046710 A1 Feb. 21,2013
METHODS AND SYSTEM FOR FINANCIAL cial instruments based on the price performance of ?nancial
INSTRUMENT CLASSIFICATION instruments, i.e., the daily prices of ?nancial instruments and
the change in the daily prices.
FIELD OF INVENTION [0007] In another embodiment, the prices of the ?nancial
[0001] The present invention relates to ?nancial instrument instruments considered for classi?cation are adjusted and
classi?cation and more particularly, the present invention take into account bene?ts, such as the impact of dividends for
relates to ?nancial instrument classi?cation that is able to stocks and interest rates for bonds.
classify different ?nancial instruments based on similarities [0008] In another embodiment, classi?cation is based on
in behavior patterns. the returns of the ?nancial instruments. The return is de?ned
as the gain or loss of a ?nancial instrument in a particular
1. BACKGROUND AND PRIOR ART period and consists of the income and the capital gains of an
investment. The return is quoted as a percentage.
[0002] Classi?cation methods for ?nancial instruments
[0009] In another embodiment, the methods provide simi
such as mutual funds, exchange-traded funds, stocks, and
larities betWeen time series representing other information,
bonds, are commonly used to identify investments that meet
not necessarily limited to prices or returns of ?nancial instru
one’s personal criteria. Such methods aim to save time by
ments.
narrowing one’s search from hundreds of thousands of the
Worldly available investment choices doWn to a manageable [0010] In another embodiment, the methods provide simi
number of speci?c investments for further research and larities betWeen ?nancial instruments as trading occurs, i.e.,
examination. These classi?cation methods (e.g., ?nancial the user speci?es a ?nancial instrument, and he or she
instrument screeners) facilitate a user to create a list of spe receives a list of ?nancial instruments that behave similarly to
ci?c ?nancial instruments he or she desires to further com the speci?ed ?nancial instrument during a pre-de?ned time
pare and analyZe. This is achieved by letting the user specify period (e. g., one minute). The updated prices and additional
comparison criteria applied to the list of ?nancial instruments characteristics such as description, sector and stock exchange
he or she is considering. Criteria include parameters such as of the speci?ed ?nancial instrument and those found to be
performance history, investment style and category, and fees, similar to the ?nancial instrument are plotted in a user inter
to name a feW. face display.
[0003] One disadvantage of current ?nancial instrument [0011] According to the teachings of the present invention,
classi?cation systems is the lack of ability to classify different there is provided a classi?cation method for selecting ?nan
?nancial instruments based on similarities in behavior pat cial instruments, performed by a computer processor. The
terns.An example of a behavior pattern Would be a time series classi?cation method includes the steps of: specifying a par
of a ?nancial instrument considered in a speci?c time period, ticular ?nancial instrument, specifying one or more screening
Wherein the time series is a sequence of data points that criteria, querying a database, coupled to operate With the
represent the daily change in the ?nancial instrument price. computer processor, With the particular ?nancial instrument
The level of similarity betWeen tWo ?nancial instruments is and the screening criteria, and retrieving ?nancial instru
determined by calculating a Similarity Rank value and ments from the database that behave similarly to the particu
described in more detail in the Detailed Description section. lar ?nancial instrument and the screening criteria, to thereby
Another disadvantage of current ?nancial instrument classi obtain acquired ?nancial instruments.
?cation systems is that they require the user to be ?nancially [0012] Optionally, one of the screening criteria is a time
knowledgeable enough to create a list of ?nancial instruments range determined by a starting time and an ending time.
of interest and to have the ability to pick the appropriate [0013] Optionally, the similarity in behavior of the particu
criteria. Another disadvantage is the inability to classify lar ?nancial instrument is determined by calculating a rank
?nancial instruments from different classes, for example, to ing measure, Wherein the higher is the ranking measure,
?nd behavioral similarities betWeen a certain stock and a betWeen the particular ?nancial instrument and one of the
certain mutual fund or betWeen a certain exchange-traded acquired ?nancial instruments, the more similarly behaving
fund and a certain bond. Another disadvantage is the inability the tWo ?nancial instruments are.
to classify ?nancial instruments from different stock [0014] Optionally, the acquired ?nancial instruments are
exchanges and/or from different countries, for example, to presented in a descending order, according to similarity rank
?nd behavioral similarities betWeen a certain Israeli mutual results, While the most similar is presented ?rst.
fund and a certain American exchange-traded fund. [0015] Optionally, the particular ?nancial instrument and
the acquired ?nancial instruments include sets of time-depen
2. SUMMARY
dent numbers that represent prices for the ?nancial instru
[0004] One embodiment of the ?nancial instrument classi ments, Wherein the prices for a ?nancial instrument, selected
?cation methods and system described herein facilitates a from the group consisting of the particular ?nancial instru
user to specify a ?nancial instrument and one or more screen ment and the acquired ?nancial instruments, are adjusted to
ing criteria such as a time range, and receive ?nancial instru represent the effect of bene?ts provided by the ?nancial
ments that behave similarly to it. instruments.
[0005] In another embodiment, the historical and current [0016] Optionally, the particular ?nancial instrument
prices of the ?nancial instruments considered are plotted as a includes a set of time-dependent numbers that represents
graph in a user interface display, for example, as price vs. prices for a market index, Wherein the market index is an
time. This facilitates further comparison and analysis of the aggregated value obtained from a Weighted sum of the
behavior of the ?nancial instruments. acquired ?nancial instruments and expressing the total values
[0006] In another embodiment, the methods employ of the acquired ?nancial instruments against a base value
machine learning algorithms to classify the behavior of ?nan from a speci?c date.
US 2013/0046710 A1 Feb. 21,2013
[0017] Optionally, each of the acquired ?nancial instru [0030] The user computer facilitates a user to send a request
ments is coupled With one or more indicators associated With to the classifying server and Wherein the request includes a
the particular ?nancial instrument and the screening criteria. speci?c ?nancial instrument and one or more screening cri
[0018] Optionally, the indicator is selected from a group of teria. The classifying server is facilitated to identify in the
expressions including an expression that represents the dif public ?nancial instruments database ?nancial instruments
ference in fees betWeen the speci?ed ?nancial instrument and that behave similarly to the speci?c ?nancial instrument
the acquired similarly behaving ?nancial instrument, an according to the screening criteria; to calculate a similarity
expression that represents the difference in return betWeen the ranking measure betWeen every tWo ?nancial instruments to
speci?ed ?nancial instrument and the acquired similarly thereby create classi?cation results; to store the classi?cation
behaving ?nancial instrument, and an expression that repre results in the classifying database; and to send the classi?ca
sents the difference in risk betWeen the speci?ed ?nancial tion results to the user computer.
instrument and the acquired similarly behaving ?nancial [0031] An aspect of the present invention is to provide a
instrument. method for grouping time series over a pre-de?ned time
[0019] Optionally, the classi?cation method further range, Wherein a time series is a sequence of values. The
includes the step of displaying the acquired ?nancial instru method includes the steps of: splitting the time range into a
ments on a display unit coupled to operate With the computer collection of time slices, Wherein for each time series in each
processor. of the time slices, the method performs the folloWing steps:
[0020] Optionally, the acquired ?nancial instruments are generating a modi?ed time series including value differences
acquired from a remote database, over a data netWork. betWeen every tWo subsequent values of the time series, and
[0021] Optionally, each of the ?nancial instruments calculating a numerical value representing the time series
includes a set of derived time-dependent numbers that repre denoted as a label, Wherein the numerical value is a summa
sent returns for each of the respective ?nancial instrument. tion of the values of the modi?ed time series at the time slice
[0022] Optionally, the particular ?nancial instrument and/ considered.
or the acquired ?nancial instruments are abbreviations used to [0032] The grouping method further includes the steps of:
uniquely identify publicly traded ?nancial instruments, or applying a classi?cation algorithm on the time slice data
abbreviations used to uniquely identify custom generated points Where the inputs for the algorithm are the modi?ed
time series representing hypothetical trading. time series and the respective calculated labels, thereby cre
[0023] An aspect of the present invention is to provide a ating different groups of time series, Wherein each group
computer softWare product for interactively selecting ?nan contains similarly behaving time series, and storing the
cial instruments, the computer softWare product embodied in groups of time series.
a non-transitory computer-readable medium in Which pro [0033] Optionally, the grouping method further including
gram instructions are stored, Wherein the program instruc the steps of: ?nding similarities for a particular time series
tions, When read by a computer processor, perform a classi during a partial period of the pre-de?ned time range, applying
?cation methodthat includes the steps of: selecting a ?nancial a decision tree classi?cation algorithm on each time slice,
instrument, specifying one or more screening criteria, query Wherein each time slice is represented as a decision tree data
ing a database, coupled to operate With the computer proces structure, and for each decision tree data structure associated
sor, With the selected ?nancial instrument and the screening With a time slice at the partial time range, the grouping
criteria, and retrieving matched ?nancial instruments that method performs the folloWing steps: ?nding the nodes that
behave similarly to the selected ?nancial instrument and the contain the particular time series, for each node that contains
screening criteria, from the database. the particular time series, ?nding other time series and
[0024] Optionally, the computer softWare product further increasing by one a counter value associated With each time
includes the step of storing the matched ?nancial instruments. series found, and sorting the time series in a descending order
[0025] Optionally, in the computer softWare product, said according to the total counter value, Wherein the higher each
screening criteria comprise a time range. of the counters is, the more similarly behaving the respective
[0026] Optionally, the computer softWare product further time series is, to the particular time series.
includes the step of storing in the database additional behav [0034] Optionally, in the grouping method, the time series
ioral descriptors for the speci?ed ?nancial instrument and for includes a set of time-dependent numbers that represent
the matched ?nancial instruments. prices for ?nancial instruments.
[0027] Optionally, the computer softWare product further [0035] Optionally, in the grouping method, each of the time
includes the step of sending a ?nancial instrument and addi series includes a set of derived time-dependent numbers that
tional criteria over a netWorkbetWeen the computer processor represent returns for ?nancial instruments.
and the database.
[0028] Optionally, the computer softWare product further 3. DESCRIPTION OF DRAWINGS
includes a user interface that facilitates a user to specify a
?nancial instrument and additional criteria, as Well as to vieW [0036] The speci?c features, aspects, and advantages of the
similarly behaving ?nancial instruments and additional disclosure Will become better understood With regard to the
behavioral descriptors. folloWing description, appended claims, and accompanying
[0029] According to further teachings of the present inven draWings Where:
tion, there is provided a system for classifying ?nancial [0037] FIG. 1 is exemplary system architecture for employ
instruments. The system includes a classifying server having ing one exemplary embodiment of the ?nancial instrument
a computer processor and a classifying database, at least one classi?cation methods and system described herein.
user computer terminal, including a display, and a public [0038] FIG. 2 depicts an exemplary ?oW diagram for
?nancial instruments database operatively connected to the employing one embodiment of the ?nancial instrument clas
classifying server. si?cation methods and system described herein.
US 2013/0046710 A1 Feb. 21,2013
[0039] FIG. 3 depicts a user interface employed by one stocks, and bonds. Some background information on major
exemplary embodiment of the ?nancial instrument classi? ?nancial instrument categories is provided in the paragraphs
cation methods and system described herein. beloW.
[0040] FIG. 4 depicts an exemplary ?oW diagram for [0050] A mutual fund is a type of investment that pools
employing one embodiment of the ?nancial instrument clas money from many investors in stocks, bonds, money-market
si?cation methods and system described herein. instruments, other securities, or cash. Partial criteria for
[0041] FIG. 5 is a partial representation of an exemplary mutual funds include categories such as: 1) Fund Obj ectivei
decision tree for providing classi?cation results in one each fund has a predetermined investment objective that tai
embodiment of the ?nancial instrument classi?cation meth lors the fund’s assets, regions of investments, and investment
ods and system described herein. strategies. The fund’s objectives are de?ned by factors, such
[0042] FIG. 6 is an example for price time series represent as hoW steady its cash How is, hoW risky it is, and hoW
ing several doZens of ?nancial instruments. diversi?ed its assets are; 2) Momingstar Ratingia rating
system created by Momingstar, Inc., ranking mutual funds
[0043] FIG. 7 is an example for grouping of price time based on the risk-adjusted performance over various periods,
series representing several groups of ?nancial instruments.
ranging from one as the Worst to ?ve as the best; 3) Year-to
Date, 1-Year, 3-Year, 5-Year, and 10-Year Performance; 4)
4. DETAILED DESCRIPTION Expenses and Expense Ratiosiassociated fees such as man
agement fees, non-management expenses, investor fees and
4.1 Preface expenses, brokerage commissions, etc.; and 5) Assets. Addi
[0044] In the folloWing detailed description, reference is tional data may be provided With research tools for the speci
made to the accompanying draWings that shoW, by Way of ?ed ?nancial instruments, for example, performance history,
illustration, speci?c embodiments in Which the invention may loads, redemption fees, etc.
be practiced. It is to be understood that other embodiments [0051] The stock or capital stock of a business entity rep
may be utiliZed and structural changes may be made Without resents the original capital paid into or invested in the busi
departing from the scope of the claimed subject matter. ness by its founders. Partial criteria for stocks include catego
ries such as: 1) Price Informationiincludes parameters such
[0045] In machine learning, classi?cation refers to an algo
as market value and current last sale (CLS); 2) Trade Infor
rithmic procedure for assigning a given piece of input data to
mationiincludes parameters such as volume, 50 average
one of a given number of categories. One example is assign
daily volume, and beta, de?ned as a measure of the volatility
ing a candidate for a university program to “accepted” or
“denied” admission classes or assigning a “diabetic” or “non
of a stock relative to the overall market; 3) Earnings; 4)
diabetic” medical diagnosis to a patient based on values of
Dividends, and; 5) Analyst Informationiincludes criteria
such as forecast earnings groWth, industry forecast earnings
certain characteristics such as gender, age, vital signs, lab
groWth, and groWth rate relative to industry.
observations, etc.
[0052] A bond is a debt security in Which the authoriZed
[0046] An algorithm that implements classi?cation is issuer oWes the holders a debt and, depending on the terms of
knoWn as a “classi?er.” The term classi?er refers to the math
the bond, is obliged to pay interest (the coupon) and/or repay
ematical function implemented by a classi?cation algorithm the principal at a later date, Which later date is termed matu
that maps input data to a category. The piece of input data is rity. A bond is a formal contract to repay borroWed money
formally termed an “instance,” and the categories are termed With interest at ?xed intervals. Partial criteria for bonds
“classes.” The instance is formally described by a vector of
include categories such as: 1) Nominal, Principal, or Face
features, Which together constitute a description of all knoWn Amountithe amount on Which the issuer pays interest, and
characteristics of the instance.
Which interest, most commonly, has to be repaid at the end of
[0047] Classi?cation normally refers to a supervised pro the term; 2) Issue Priceithe price at Which investors buy the
cedure, i.e., a procedure that classi?es neW instances based on bonds When they are ?rst issued, Which price Will typically be
learning from a data set of instances that have been properly approximately equal to the nominal amount. The net proceeds
labeled With the correct classes. The corresponding unsuper that the issuer receives are the issue price, minus issuance
vised procedure is knoWn as clustering, Which clustering fees; 3) Maturity Dateithe date on Which the issuer has to
involves grouping data into classes based on a measure of repay the nominal amount. As long as all payments have been
similarity, such as the distance betWeen instances. made, the issuer has no more obligations to the bondholders
[0048] The folloWing sections provide a background of after the maturity date. The period of time until the maturity
?nancial instrument comparison in general, an overvieW of date is often referred to as the term, or maturity of a bond. The
the proposed ?nancial instrument classi?cation methods and maturity can be any length of time, although debt securities
system, as Well as an exemplary architecture. A layout for a With a term of less than one year are generally designated
user interface for one exemplary embodiment of the system is money-market instruments rather than bonds. Most bonds
also provided. Lastly, a detailed description of the compo have a term of up to 30 years. Some bonds have been issued
nents and the features of the methods and system, as Well as With maturities of up to 100 years, and some never mature;
alternate embodiments, are provided. and 4) Couponithe interest rate that the issuer pays to the
[0049] Numerous investment institutions (e.g., Fidelity bondholders.
Investments, Vanguard, etc.), softWare companies (e.g., [0053] An exchange-traded fund (ETF) is an investment
Google, Yahoo!, etc.), banks (e.g., Bank of America), and fund traded on stock exchanges, much like stocks. An ETF
Websites (e.g., Bloomberg.com, NASDAQcom, etc.) offer holds assets such as stocks, commodities, or bonds, and trades
Internet-based interactive research tools to facilitate users to at approximately the same price as the net asset value of its
evaluate and compare, i.e., to classify, a variety of ?nancial underlying assets over the course of the trading day. Most
instruments, such as mutual funds, exchange-traded funds, ETFs track an index, such as the S&P 500.
US 2013/0046710 A1 Feb. 21,2013
[0054] An embodiment is an example or implementation of munication method in Which method electronic messages are
the inventions. The various appearances of “one embodi sent betWeen people, and received at some later time, not
ment,” “an embodiment” or “some embodiments” do not necessarily in real-time. Occasionally the length of time
necessarily all refer to the same embodiments. Although vari betWeen sending and receiving a particular email is in the
ous features of the invention may be described in the context range of several seconds to several hours. Another scenario
of a single embodiment, the features may also be provided for non-real-time is When a user receives classi?cation results
separately or in any suitable combination. Conversely, periodically, for example, several hours after a pre-de?ned
although the invention may be described herein in the context trading period Was ended, daily-based, Weekly-based,
of separate embodiments for clarity, the invention may also be monthly-based, etc. In such cases there is no signi?cant
implemented in a single embodiment. importance to the immediacy of receiving the classi?cation
[0055] Reference in the speci?cation to “one embodi results.
ment”, “an embodiment,” “some embodiments” or “other
embodiments” means that a particular feature, structure, or [0060] FIG. 1 provides an exemplary system architecture
characteristic described in connection With the embodiments 100 for employing one embodiment of the ?nancial instru
is included in at least one embodiment, but not necessarily all ment classi?cation methods and system. As shoWn in FIG. 1,
embodiments, of the inventions. It is understood that the the system architecture 100 employs a client computer 102
phraseology and terminology employed herein are not to be and a classifying server 104. The client computer 102 facili
tates a user 106 to specify a ?nancial instrument 108 and a
construed as limiting and are for descriptive purpose only.
time range 110 via a user interface 112 presented, With no
[0056] Methods of the present invention may be imple
mented by performing or completing manually, automati limitation, on a display 114, coupled to operate With client
cally, or a combination thereof, selected steps or tasks. The computer 102. The ?nancial instrument 108 and the time
order of performing some methods’ steps may vary. The range 110 speci?ed by the user 106 are sent to a classifying
descriptions, examples, methods and materials presented in database 122, operatively coupled With classifying server
the claims and the speci?cation are not to be construed as 104, preferably in a textual format. The classifying database
limiting but rather as illustrative only. 122 contains several tables including one or more data struc
tures such as tables of classi?cation results 140, and one or
[0057] Meanings of technical and scienti?c terms used
more data structures such as tables of comparable ?nancial
herein are to be commonly understood as to Which the inven
tion belongs, unless otherWise de?ned. The present invention instrument data 142 and 148 (the structure and functionality
of the tables are described With greater detail in Section 4.3).
can be implemented in the testing or practice With methods
and materials equivalent or similar to those described herein. Once a user-request 116 is received by the classifying server
104, the classifying server 104 processes the user-request 116
4.2 System Description and sends processed classi?cation results 118 back to the
client computer 102.
[0058] The folloWing paragraphs provide an exemplary
description for employing the ?nancial instrument classi?ca [0061] In response to receiving the processed classi?cation
tion methods and system. It shouldbe understood that in some results 118, including a list of ?nancial instruments, the client
cases, the order of actions can be interchanged, and in other
computer 102 provides on the display 114 interactive results
ones, some of the actions may even be omitted.
120 that include a representation of the ?nancial instruments,
[0059] In one embodiment of the ?nancial instrument clas
preferably hyperlinked textual representation, Wherein the
?nancial instruments behave most similarly to the ?nancial
si?cation methods and system, the user (e.g., an investor)
instrument 108 during the time range 110 speci?ed. The user
speci?es a ?nancial instrument and screening criteria such as
106 can act on these results and sort them. In addition, the
a time range at a client computer Which ?nancial instrument
client computer 102 may request 124 additional ?nancial
and screening criteria are sent to a local or remote server
details 126 associated With the speci?ed ?nancial instrument
computer. Additional screening criteria may include an
108 and the similarly behaving ?nancial instruments, i.e., the
objective such as “Municipal Bonds,” “Blend,” or “Diversi processed classi?cation results 118. Such additional ?nancial
?ed Emerging Markets.” Screening criteria may also include
details 126 are available at one or more public databases 128
a speci?c stock exchange and/or a speci?c country in Which
and are provided by variety of resources such as NASDAQ or
stock exchange and/or country the ?nancial instruments are
NYSE stock exchanges. The additional ?nancial details 126
traded at. Screening criteria may also include a speci?c type
and trading data (such as prices and volumes) 130 associated
such as “Stocks,” “Mutual Funds,” or “Exchange-Traded With the ?nancial instruments are received 132 by the client
Funds.” The server computer provides the user in real-time a
computer 102 and presented on the display 114.
list of ?nancial instruments that behave similarly to the speci
?ed ?nancial instrument during the speci?ed time range. It [0062] The table of classi?cation results 140 is formed by
should be noted that the list of ?nancial instruments and applying classi?cation procedures. A classi?cation module
additional details associated With them can be acquired either 134 includes the classi?cation procedures and is a component
in real-time or not in real-time, Wherein real-time, as used of the classifying server 104. The classi?cation module 134
herein, is as quickly as the ?nancial instrument and time range requests (136) and receives (138) trading data of publicly
are typed, and non-real-time is a delayed display, and Wherein traded ?nancial instruments and uses the data to generate the
delayed is referred to some later time. Additionally, it should content of the table of classi?cation results 140. To generate
be noted that the time range and/ or the ?nancial instrument table of classi?cation results 140, additional tables are formed
could be default values determined in advanceiin this case including tables of comparable ?nancial instrument data 142
the user is not required to specify the time range and/ or the and 148, and a table of raW patterns 144. The classi?cation
?nancial instrument. An example for non-real-time interac module 134 and the classi?cation procedures Will be
tion is having the user receiving delayed classi?cation results described in greater detail further in the text referring to FIG.
attached to an email message. Email is considered as a com 4. It should be noted that if desired, the classi?cation module
US 2013/0046710 A1 Feb. 21,2013
[0069] The time range selection ?xture 316, including its parison With the speci?ed ?nancial instrument 328. An indi
sub-componentsithe left button 318, the right button 320, cator is an expression that represents a bene?t betWeen the
and the time range selection pad 322imay be aligned on any speci?ed ?nancial instrument and each of the ?nancial instru
location on the time range selection canvas 304. One Way to ments found. For example, one of the results, “VBIRX,” has
align the time range selection ?xture 316 is to click With a loWer expense ratio and a higher 5-year average return in
mouse 115 on any invisible vertical line on the time range comparison With “FFXSX.” The indicators for “VBIRX” Will
selection canvas 304.Another Way to align the range selection be then “LoWer Expense Ratio” and “Higher 5YAvg Return.”
?xture 316 is to use buttons 324 and 326. Pressing on button Another example for an indicator, “LoWer Beta,” represents
324 moves time range selection ?xture 316, including its the difference in the ?nancial risk, or beta, betWeen tWo
sub-components, one time period back. Pressing on button ?nancial instruments. Financial risk is de?ned as the risk
326 moves time range selection ?xture 316, including its resulting from the existence of debt in the ?nancing structure
sub-components, one time period ahead. In one embodiment, of the ?nancial instrument. Financial instruments With high
a single time movement is one Week. market risk Will have required returns above the market rate,
[0070] Once the user 106 speci?es a time range using the While those With loW market risk Will have loWer rates of
various controls included in time range selection panel 302, return. The indicators mentioned are examples and are not
he or she may type a ?nancial instrument in an input text box intended to suggest any limitation of the scope of use or
328. In one embodiment, pressing on button 330 sends the functionality of the ?nancial instrument classi?cation meth
speci?ed ?nancial instrument 328 and the speci?ed time ods and system.
range selected 312 to the classifying database 122. In an
additional embodiment, button 330 is not necessary, and [0073] In another embodiment, additional data associated
With the speci?ed ?nancial 328 and the ?nancial instruments
sending the speci?ed ?nancial instrument 328 and time range
selected 312 is achieved by pressing a pre-de?ned key such as found at panel 332 may be presented in a chart shoWing, for
“ENTER” at a keyboard 117 coupled to operate With client
example, price/performance information such as nominal
computer 102. In another embodiment the user 106 may not
price, price change betWeen tWo time steps, earnings, divi
have to type the entire string for the ?nancial instrument in dends, descriptive information such objective, analyst infor
328, instead, an autocomplete feature may be provided to pull mation, etc. Charts may ease understanding of the large quan
tities of data and the relationships/ similarities betWeen the
?nancial instruments from the classifying database 122 upon
partial string typing of a ?nancial instrument. In another ?nancial instrument patterns. Line charts, bar charts and his
to grams are only a feW examples that may be presented on the
embodiment the user 106 may not have to type a ?nancial
instrument and/or time range, instead, a microphone Would user interface 300 (see also 112).
acquire the user’s voice to specify the ?nancial instrument
and/ or the time range. In another embodiment a camera 4.3 Classi?cation Method
coupled With a gesture recognition module Would alloW the
user to specify the ?nancial instrument and/or the time range [0074] To classify ?nancial instruments, the classi?cation
via hand gestures and/or other human gestures. It should be module 134 is used as shoWn in FIG. 4. The classi?cation
noted that specifying time range in time range selection panel module 134 is facilitated to perform a method that generates
302 and specifying the ?nancial instrument 328 can be of any classi?cation results stored in one or more tables in the clas
order, meaningithe user 106 may specify a ?nancial instru sifying database 122. The classi?cation method is applied on
ment ?rst and then a time range, or vice versaihe or she may all of the price patterns of all ?nancial instruments available.
specify a time range and then a ?nancial instrument. In one embodiment the available patterns are of all ?nancial
[0071] Panel 332 includes informative representations of instruments traded in NASDAQ, NYSE, AMEX, and of
the results as returned from the classifying database 122. approximately 20,000 American mutual funds traded over
Panel 332 contains a list of the ?nancial instruments that are approximately one decade (2000-2010). In addition to table
found behaving similarly to the speci?ed ?nancial instrument of classi?cation results 140, classifying database 122
328 in the time range selected 312. Additional characteristics includes tables of comparable ?nancial instrument data 142
and the characteristics’ corresponding values associated With and 148, and raW price patterns 144 for the ?nancial instru
the ?nancial instruments found, such as historical perfor ments considered. The original patterns, i.e., trading patterns
mance, fees and ranking are available at public database 128 of ?nancial instruments such as prices/volumes are requested
and also presented in panel 332 next to each result, for 136 and received 138 by the classifying server 104. Once
example as in 334. Examples include Description, Type, Total received, the patterns are stored and modi?ed in classifying
Assets, Category, Expense Ratio, Beta, and Morningstar Risk database 122 using a data preparation procedure as described
Rating to name a feW. Additional characteristics are also through expressions 4.1-4.6. Real-time and daily ?nancial
presented for the speci?ed ?nancial instrument 328 at 336. instrument prices, fundamental company data, historical
The additional characteristics and values 334 and 336 asso chart data, daily updates, fund summary, fund performance
ciated With the ?nancial instruments are pulled from the pub and dividend data stored in classifying database 122 are pro
lic database 128 and/or from the classifying server 104. One vided for example by companies such as Capital IQ, Com
of the characteristics 334 in panel 332 is a “Read More” modity Systems, Inc. (CSI) and Morningstar, Inc. Addition
interactive textual link. Clicking With a mouse 115 on a “Read ally data can be acquired by using ?nancial Websites such as
More” link facilitates the user 106 to receive additional infor of the NASDAQ/NY SE stock exchanges, for example.
mation for a ?nancial instrument. The additional information [0075] Assume S1, S2, S1, . . . Sm are m ?nancial instruments
can be pulled from the public database 128 or other external considered for classi?cation during a trading time range that
?nancial information systems/Websites. In one embodiment includes n time-steps (e.g., a one time-step equals to a one
the additional information is acquired from a Website and day). Each ?nancial instrument S,- is associated With a vector
presented using a standard Web-broWser. of prices in Which vector of prices each value represents an
[0072] Next to each similarly behaving ?nancial instru adjusted
(jIl to n).closing
For a price
?nancial
for ainstrument
business day
Sl- the
ended
vector
in time-step
of prices,
ment presented at panel 332 shoWn one or more indicators
specifying the ?nancial instrument’s superiority 338 in com i.e., a signal/time series, is as folloWs:
US 2013/0046710 A1 Feb. 21, 2013
51 [C1, C2, C3, C4, C511. sllcs. C7, C8, C9, C1012, ---S1 [C144, CW3. CW2. CW1. Cnlk 4-4
silcr. C2, C3, C4, C511. silce. C7, C8. C9, C1012, ---Si[Cn*4, CW3. CW2. CW1. Cnlk
[0076] For all ?nancial instruments, generate vectors rep- Where the siZe of the total time range of n time-steps, also
resenting the change in price for every tWo subsequent trading equals to k time slices each of length of h:5. The following
days; representation, for example, is considered for the ?rst time
slice (kIl):
S [P12 1 P13 1 P14 1 Pin 1] 4.2 Sl[cl’ C2’ C3’ C4’ C511 4'5
1P11 ,P12 aP's iii-P5141
S2[C1,C2,C3.C4.C5l1
slpl
P11
1, g—1.
P12
?-l.
P13
Pt” ‘1]
P'rwl
P“ P’2 P“ P Sm[C1.C2.C3.C4.C5]l
P1 P1 P1 P In
[0080] In the classi?cation problem considered here no
. . . . .
Sm[—2 — l. —3 — l. —4 — l. — l] labels are ava1lable for the signals and there is no information
P11 P12 P13 P5141 on hoW to refer to a set of values associated With a certain time
slice. As such, a numerical value representing each signal is
generated and assigned as the label of the signal. The numeri
[0077] To slmphfy the representanon of 42 It 15 presented cal value label denoted as LS1. is calculated for each signal:
as:
h 4.6
ml = 2 s1 (C1)
S1[C1, C2, C3, C,,)] 4.3 1:1
h
S2[C1, C2, C3, C,,)] is2 _
—
S2(C1)
1s.- = Z SAC!)
sm[cl, c2, c3, C,,)] H
[0078] The representation for the ?nancial instruments S1, Lsm : i SAC!)
S2, 8,, . . . Sm as in expression 4.3 facilitates comparing H
betWeen them because this representation is price-scale and
value-scale independent. Since n could be large (e. g., if clas- [0081] The representation of self-labeling as shoWn in 4.6
si?cation for one decade is desired), time slices of a constant expressions facilitates the application of supervised learning
siZe h are de?ned. One reason to use time slices is to reduce methods on unlabeled data sets. This is achieved by providing
US 2013/0046710 A1 Feb. 21,2013
a supervised learning classi?cation algorithm With pairs of contents of each cell representing the adjusted close price of
adjusted representations of original signals (as shoWn as an the ?nancial instrument for the trading day (see 4.1 expres
example for kIl in 4.5 expressions) and the adjusted repre sions). In one embodiment, a column name for a trading day
sentations’ corresponding self-generated label (4.6 expres is in the format of “Day_Month_Year,” for example, “20*
sions). 8i2008.” An example is shoWn in Table C.
TABLE C
Daily prices for all ?nancial instruments
[0082] The procedure described through expressions 4.1- [0084] Table 146 generates table 148 (see example in Table
4.6 is applied in one embodiment by acting several tables D) With a distinct column of ?nancial instruments and addi
stored in classifying database 122. Prices and additional data tional columns, each column representing a price difference
are acquired for all the ?nancial instruments considered. The in percent betWeen the adjusted close price of tWo subsequent
data is stored in a ?rst table 144ifor each ?nancial instru days (see 4.2 and 4.3 expressions). In one embodiment, the
ment, the folloWing historical data is stored: 1) Symbol; 2) column names for the difference are in the format of “Month
Date; 3) Opening Price; 4) Closing Price; 5) Volume, and; 6) Title_TradingDay_PreviousTradingDay_Year,” for example,
Adjusted Closing Price, as seen for example in Table B. “Octoberi13i12i2010.” An example is shoWn in Table D.
TABLE D
Daily price change (%) for all ?nancial instruments
TABLE B [0085] The data stored in table 148 may serve as a data set
for a machine learning algorithm. In one embodiment, the
Daily data for all ?nancial instruments data in table 148 may serve as an input set for a supervised
learning algorithm using the stored comparable numerical
Adjusted values as input. In another embodiment as described herein,
Opening Closing Closing
Symbol Date Price Price Price Volume
the supervised learning algorithm is a decision tree algorithm.
In yet another embodiment, the data in table 148 may serve as
EBAY Jan. 3, 2000 130.13 141.25 17.66 48902400 input for an unsupervised learning algorithm or a reinforce
EBAY Jan. 4,2000 135.5 128 16 33803200 ment learning algorithm.
EBAY Jan. 5,2000 121.25 136.56 17.07 44146400 [0086] The classi?cation method 400 shoWn in FIG. 4 con
EBAY Jan. 6,2000 133.94 134.88 16.86 44147200
EBAY Jan. 7,2000 134 134.75 16.84 21574400
siders a large number of time slices. For example, if the
EBAY Jan. 10,2000 141.06 142.25 17.78 25056000 desired classi?cation time range is a certain quarter, then the
EBAY Jan. 11,2000 142 139.19 17.4 22664000 number of time slices considered is approximately tWelve
EBAY Jan. 12,2000 137.63 130.38 16.3 21400800 (assuming that the length of a time slice is one Week). Clas
EBAY Jan. 13,2000 133.5 137.81 17.23 19286400 si?cation considers all patterns of ?nancial instruments
EBAY Jan. 14,2000 140.13 133.81 16.73 23342400
stored in table 148. Each time slice has a starting date and an
AMZN Jan. 3,2000 81.5 89.38 89.38 16117600 ending date. In one embodiment, the time slice is ?ve busi
AMZN Jan. 4,2000 85.37 81.94 81.94 17487400 ness days con?gured in advance (one WeekiMonday
AMZN Jan. 5,2000 70.5 69.75 69.75 38457400 through Friday). It should be noted that in one embodiment
AMZN Jan. 6,2000 71.31 65.56 65.56 18752000 the classi?cation method 400 can be applied on ?nancial
AMZN Jan. 7,2000 67 69.56 69.56 10505400 instruments as trading occurs, and in Which ?nancial instru
AMZN Jan. 10,2000 72.56 69.19 69.19 14757900
AMZN Jan. 11,2000 66.88 66.75 66.75 10532700
ments the duration of a time slice is shorter, e.g., one milli
AMZN Jan. 12,2000 67.88 63.56 63.56 10804500 second, or longer, e.g., one month. The classi?cation method
AMZN Jan. 13, 2000 64.94 65.94 65.94 10448100 400 starts With classifying data of an initial time slice 402. If
AMZN Jan. 14, 2000 66.75 64.25 64.25 6853600 classi?cation results already exist for the time slice as evalu
ated in 404, the procedure evaluates Whether classi?cation
has not been applied yet for additional time slices considered,
as shoWn in 406. If all time slices have been processed, the
[0083] Table 144 generates a second table 146 With a dis procedure ends. If there are time slices that have not been
tinct column of ?nancial instruments and additional columns, processed yet, the next time slice is considered, as shoWn in
each representing a title for a single trading day and the block 408.
US 2013/0046710 A1 Feb. 21,2013
[0087] Exemplary time series 602 representing prices of ment records in a node (the minimum is tWo), the less this
several dozens of ?nancial instruments are presented in FIG. node varies, i.e., a node With feWer records is more likely to
6. The time range 604 shown in FIG. 6 includes approxi represent a better classi?cation betWeen the ?nancial instru
mately 52 Weeks of 201 1. Each Week, i.e., ?ve business days, ments that the node contains.
considered as a time slice. An exemplary time slice 606 is [0090] The number of nodes in a generated tree depends on
marked for one Week during November 201 1 . An exemplary the length of the time slice and the number of ?nancial instru
grouping of time series is presented in FIG. 7. The time series ments considered. The classi?cation accuracy of the algo
represent six ?nancial instruments traded over a period 714 of rithm depends on its input parameters. In one embodiment,
approximately three months in 2012. Three groups of ?nan parameters for a decision tree algorithm include complexity
cial instruments are shoWn: 1) OIL 702 and USO 704, 2) penalty, to control the groWth of the decision tree, and mini
PIREX 706 and GRERX 708, and 3) GZIIX 710 and EWZ mum support, to determine the minimal number of leaf cases
712. required to generate a split. Setting the desired values for the
[0088] For any time slice, for Which time slice the classi? decision tree algorithm parameters depends on the tradeoff
cation results are not yet available, table 142 is generated, as betWeen classi?cation accuracy and computational speed.
shoWn in block 410. Table 142 consists of a portion of table Classifying With perfect or close to perfect accuracy thou
148. The structure of table 142 depends on the occurrence and sands or hundreds of thousands of ?nancial instruments, may
duration of the time slice considered. For example, for the require many days or even many Weeks to apply a decision
time slice Apr. 20-Apr. 24, 2009 (a total of ?ve trading days), tree algorithm using the classi?cation method herein. To
table 142 consists of a ?nancial instrument symbol column reduce the calculation time, the groWth of the decision tree is
and numerical value columns denoted as “Features” (4.2 controlled by increasing the complexity penalty level (this
expressions): 1) “Aprili20i17i2009,” 2) “Aprili21i20i decreases the number of splits) and by increasing the level of
2009,” 3) “Aprili22i21i2009,” 4) “Aprili23i22i2009,” minimum support. On one hand, controlling the groWth of the
and 5) “Aprili24i23i2009.”Values in these columns are as tree improves computation performance. On the other hand,
in table 148. An additional column in table 142 is titled controlling the groWth of the tree may affect classi?cation
“Predictor,” or “Label.” “Predictor” values are a function of accuracy. A ?ltering procedure 414 is applied to each decision
the other numerical values for a ?nancial instrument. In one tree generated to partially overcome this and to avoid recog
embodiment, values in “Predictor” are a summation (4.6 niZing groups of ?nancial instruments that behave differently
expressions). In the previous example of the time periodApr. from each other but are still classi?ed as similar. In one
20-Apr. 24, 2009, numerical values in “Predictor” for a ?nan embodiment, for each tree, the predictor value of each ?nan
cial instrument are equal to summing the values of “Aprili cial instrument in a node is compared With the other predic
20i17i2009,” “Aprili21i20i2009,” “Aprili22i21i tors of the ?nancial instruments present in the node. If the
2009,” “Aprili23i22i2009,” and “Aprili24i23i2009.” variability of predictors found in a node is above a pre-de?ned
In another embodiment, values in “Predictor” are an average threshold, then the node is considered a noisy/ inaccurate clas
of the numerical values of the features. In yet another embodi si?cation, i.e., the node is pruned.
ment, time periods may exclude one or more trading days,
such as When a holiday occurs. It should be noted that the [0091] In one embodiment 28,601 ?nancial instruments are
number of records in table 142 equals the number of ?nancial considered for classi?cation including several thousands of
instruments considered. An example for table 142 for Apr. NASDAQ, NYSE, and AMEX ?nancial instruments, several
20-Apr. 24, 2009 is shoWn in Table E. market indexes, and approximately 20,000 American mutual
TABLE E
A comparable table example (values are in %) for Apr. 20-Apr. 24 2009
[0089] The data of table 142 serves as an input for a stan- funds. The total time range for classi?cation is 574 Weeks
dard supervised learning algorithm. In one embodiment, the (approximately one decade) spanning from Monday Jan. 3,
supervised learning algorithm is a decision tree algorithm 2000 to Friday Dec. 31, 2010. For most of the ?nancial
412. For each time slice, a decision tree is generated. An instruments considered trading information Was available for
example for a partial representation of a decision tree is the entire time range, hoWever, for certain stocks and mutual
shoWn in FIG. 5. A decision tree is a data structure that
funds data Was available only When they ?rst became avail
consists of branches and leaves. Leaves (also denoted as
able for trading (e.g., Google Inc. Went public in August
“nodes”) represent classi?cations, and branches represent
conjunctions of features that lead to those classi?cations. In 2004). For each of the 574 Weeks, a decision tree based
one embodiment each node has a unique title to distinguish classi?cation is performed using the data of table 142. Each
the node from other nodes that the tree is composed. A node such classi?cation results a decision tree data structure. For
contains tWo or more records. Each record represents a ?nan the amount of data considered here, a typical siZe for one
cial instrument, its feature values (4.2 expressions) and its decision tree is in the range of 5,000 to 10,000 nodes. An
predictor value (4.6 expressions). The feWer ?nancial instru exemplary partial representation for a decision tree 500 plot
US 2013/0046710 A1 Feb. 21,2013
ting only several nodes 502-522 is shown in FIG. 5. The [0098] The decision tree classi?cation results for the time
decision tree 500 includes a main node 502 that contains all slice considered, excluding noisy data, are stored 416 in table
?nancial instruments. The decision tree algorithm generates of classi?cation results 140 of classifying database 122 of the
rules as shoWn in 524-542. The rules are based on values for classifying server 104. Table G is an exemplary partial rep
the ?nancial instruments (price change given in percent) for resentation of the table of classi?cation results 140 for one
business Week. For the amount of data considered here, the
every tWo subsequent trading days; see as described through
number of records representing the nodes of one decision tree
4.1-4.6 expressions. Some nodes in the tree split to tWo sub classi?cation results is in the range of 10,000 to 70,000
nodes, i.e., children, and other nodes do not. A split, if occurs, records.
is based on the generated rules and separates a group of
?nancial instruments to tWo smaller groups. For example, for TABLE G
the main node 502 that consists of 28,601 ?nancial instru
An example for a tabular representation of
ments, tWo rules Were generatedirule “Decemberi28i one decision tree classi?cation results
27i2010>:—6.987 and <0.744” 524 and rule “Decemberi
28i27i2010<—6.987 or >:0.744” 526. Rule 524 generates a Node Name Symbol
sub-node that contains 27,327 ?nancial instruments 504 and MMEBX
rule 526 generates a sub-node that contains 1,274 ?nancial MMEKX
IU O >
DSPIX
instruments 506. Similarly, other generated rules split nodes NMIAX
across the tree as in 528-542. For a ?nancial instrument to be SHRAX
considered classi?ed to a certain node, the series of rules that TWSIX
TWClX
lead to that node are consideredifor example, the tWo ?nan FAElX
cial instruments of node 520 are classi?ed using a series of AELIX
FEIlX
?ve rules starting from the main node 502 as shoWn beloW.
GTMUX
[0092] “Decemberi28i27i2010>:—6.987 and <0.744” SSFFX
(as shoWn in 524). STCSX
[0093] “Decemberi28i27i2010>:—0.8022 and XGAMX
<—0.0291” (as shoWn in 528).
[0094] “Decemberi31i30i2010<—4.482 or >:2.915”
(as shoWn in 534). [0099] The procedure repeats itself With the next time slice
[0095] “Decemberi31i30i2010<—4.482 or >:17.709” 408 until all time slices are processed and decision trees are
(as shoWn in 538). created for them and added in a tabular format to the table of
[0096] “Decemberi28i27i2010>:—0.33834 and classi?cation results 140 as shoWn for example in Table H.
<—0.26103” (as shoWn in 540). Table 140 includes the folloWing records of data: 1) Period
[0097] Table F shoWs the content of node 520. The content lDian integer specifying the time period title considered; 2)
includes the symbols of the tWo ?nancial instruments in the Period Titleia string specifying the time period title consid
node, “DRQAX,” and “DRQLX,” change in price vectors, ered; 3) Node Nameia unique name for the node, and; 4)
and the ?nancial instruments’ corresponding Predictor. It Symbolithe ?nancial instrument symbol. For the amount of
should be noted that the decision tree algorithm applies a data considered here, the number of records in table 140 is
feature selection procedure to identify the attributes and val approximately 18 million.
ues that provide the most information. As such, it is typical for
a set of rules generated not to include all of the available TABLE H
features. For example, in generating the ?ve rules mentioned An example for a tabular representation of
in the above example, only tWo out of the ?ve possible fea all decision tree classi?cation results
tures are consideredi“Decemberi28i27i2010,”
_ and Period ID Period Title Node Name Symbol
“Decemberi31i30i2010.” It also should be ment1oned that
occasionally rules that determine a classi?cation for a certain 1 Jan' 03h“- A MMEBX
_ _ 07, 2000 A MMEKX
node may overlap. For example, for the tWo ?nanc1al 1nstru- B DSPD;
ments of node 520 only rules “Decemberi28i27i2010>:— B NMIAX
0.33834 and <—0.26103” (as shoWn in 540) and “Decemberi g gals}?
31i30i2010<—4.482 or >:17.709” (as shoWn in 538) are c TWCIX
necessary, While the other three are redundant.
TABLE F
TABLE H-continued the ?nancial instrument is to the ?nancial instrument and the
time range speci?ed, i.e., the ?nancial instrument is ranked
An example for a tabular representation of higher. As seen in Table I, ?nancial instruments “GLDIX”
all decision tree classi?cation results
and “GLDAX” are the most similarly behaving to “GOLDX”
Period ID Period Title Node Name Symbol during Jul. 20-24, 2009. “TOLCX” and “TOLLX” are also
considered as similarly behaving to “GOLDX” but less simi
FAEIX
C S IEX
lar in comparison With “GLDIX” and “GLDAX.” “TOLIX”
KNIEX and “GLDBX” are also considered as similarly behaving to
MAS RX “GOLDX” but are considered less similar in comparison With
IUO>
SWANX the rest of the ?nancial instruments speci?ed in Table I.
2 Jan. 10-Jan.
14, 2000 TABLE I
Classi?cation example for “GOLDX” for one time slice 1 Week
Symbol Counter
GOLDX
GLDIX
GLDAX
TOLCX
TOLLX
574 Dec. 27-Dec. TOLIX
:ow>'
31,2010 GLDBX