U.S. patent application number 12/338871 was filed with the patent office on 2009-04-16 for systems and methods for analyzing data.
This patent application is currently assigned to EXPERIAN-SCOREX, LLC. Invention is credited to Chuck Robida, Chien-Wei Wang.
Application Number | 20090099960 12/338871 |
Document ID | / |
Family ID | 38480111 |
Filed Date | 2009-04-16 |
United States Patent
Application |
20090099960 |
Kind Code |
A1 |
Robida; Chuck ; et
al. |
April 16, 2009 |
SYSTEMS AND METHODS FOR ANALYZING DATA
Abstract
Information regarding individuals that fit a bad performance
definition, such as individuals that have previously defaulted on a
financial instrument or have declared bankruptcy, is used to
develop a model that is usable to determine whether an individual
that does not fit the bad performance definition is more likely to
subsequently default on a financial instrument or to declare
bankruptcy. The model may be used to generate a score for each
individual, and the score may be used to segment the individual
into a segment of a segmentation structure that includes
individuals with related scores, where segments may include
different models for generating a final risk score for the
individuals assigned to the particular segments. Thus, the segment
to which an individual is assigned, which may be determined based
at least partly on the score assigned to the individual, may affect
the final risk score that is assigned to the individual.
Inventors: |
Robida; Chuck; (Roswell,
GA) ; Wang; Chien-Wei; (Irvine, CA) |
Correspondence
Address: |
KNOBBE MARTENS OLSON & BEAR LLP
2040 MAIN STREET, FOURTEENTH FLOOR
IRVINE
CA
92614
US
|
Assignee: |
EXPERIAN-SCOREX, LLC
Costa Mesa
CA
|
Family ID: |
38480111 |
Appl. No.: |
12/338871 |
Filed: |
December 18, 2008 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
11535907 |
Sep 27, 2006 |
|
|
|
12338871 |
|
|
|
|
60781391 |
Mar 10, 2006 |
|
|
|
Current U.S.
Class: |
705/38 ;
706/52 |
Current CPC
Class: |
G06Q 40/08 20130101;
G06Q 40/06 20130101; G06Q 40/025 20130101; G06Q 40/04 20130101;
G06Q 20/10 20130101; G06Q 30/0204 20130101; G06Q 40/00 20130101;
G06Q 40/02 20130101 |
Class at
Publication: |
705/38 ;
706/52 |
International
Class: |
G06N 5/02 20060101
G06N005/02; G06Q 40/00 20060101 G06Q040/00 |
Claims
1. A computer readable medium having stored thereon a computer
program that embodies a method of generating a model for
determining an individual's propensity to enter either a first
failure mode or a second failure mode, wherein the computer program
is configured for storage on a computing system in order to
transform the computing system into a special purpose computing
system configured to perform the method comprising: receiving
information defining a bad performance definition, wherein the bad
performance definition is defined to include individuals that have
characteristics of one or more of the first failure mode and the
second failure mode; receiving observation data regarding a
plurality of individuals fitting the bad performance definition,
the observation data indicating characteristics of the individuals
at an observation time; receiving outcome data regarding the
plurality of individuals fitting the bad performance definition,
the outcome data indicating characteristics of the individuals
fitting the bad performance definition during an outcome period,
the outcome period beginning after the observation time; and
storing at least some of the observation data and the outcome data
in a storage device; transforming the observation data and the
outcome data into a model configured to determine a likelihood that
an individual not fitting the bad performance definition will enter
the first failure mode or if the individual will enter the second
failure mode.
2. The computer readable medium of claim 1, wherein the observation
time is about 24 months prior to generation of the model.
3. The computer readable medium of claim 1, wherein the outcome
period is a period of about 24 months prior to generation of the
model, but exclusive of the observation time.
4. A computerized method of generating a model for determining an
individual's propensity to enter either a first failure mode or a
second failure mode, the method comprising: receiving information
that defines characteristics of a bad performance definition so
that the bad performance definition includes individuals that have
characteristics of one or more of a first failure mode and a second
failure mode; receiving observation data regarding a plurality of
individuals fitting the bad performance definition, the observation
data including a snapshot of financial and demographic information
associated with respective individuals at a previous point in time
T-X, where T is the current month and X is a number of months;
recording outcome data regarding the financial behavior of the
individuals during the time from T-X+1 to T; and comparing the
observation data associated with the individuals at time T-X and
the outcome data recorded during the period T-X to T in order to
generate a model usable to determine a likelihood that an
individual not fitting the bad performance definition will enter a
first failure mode or if the individual will enter the second
failure mode, wherein the computerized method is embodied in
computer source code that is loaded into one or more memories of a
computing systems in order to modify the content of the one or more
memories to cause the computing system to perform the computerized
method.
5. The method of claim 4, wherein the first failure mode comprises
filing for bankruptcy and the second failure mode comprises
defaulting on a financial instrument.
6. The method of claim 4, wherein the first failure mode comprises
defaulting on an installment loan and the second failure mode
comprises defaulting on a revolving loan.
7. The method of claim 4, wherein the first failure mode comprises
defaulting on a bank loan and the second failure mode comprises
defaulting on an automobile loan.
8. The method of claim 4, wherein the observation time is about 24
months prior to generation of the model.
9. The method of claim 8, wherein the outcome period is a period of
about 24 months prior to generation of the model, but exclusive of
the observation time.
10. A computer system configured to generate a behavioral model
that is indicative of an individual's propensity to enter either a
first failure mode or a second failure mode, the computer system
comprising: one or more input devices for receiving information
defining a bad performance definition so that the bad performance
definition includes individuals that have characteristics of one or
more of the first and second failure modes; one or more interfaces
for receiving observation data regarding a plurality of individuals
fitting the bad performance definition, the observation data
indicating characteristics of the individuals at an observation
time, and for receiving outcome data regarding the plurality of
individuals fitting the bad performance definition, the outcome
data indicating characteristics of the individuals fitting the bad
performance definition during an outcome period, the outcome period
beginning after the observation time; and a profile module
configured to compare the observation data and the outcome data in
order to generate a model usable to determine a likelihood that an
individual not fitting the bad performance definition will enter a
first failure mode or if the individual will enter the second
failure mode,
11. The computer system of claim 10, wherein the observation data
comprises one or more of demographic data and financial data
regarding individuals
12. The computer system of claim 10, wherein the profile module is
further configured to apply the generated model to information
regarding a modeled individual that does not fit the bad
performance definition and to determine whether the modeled
individual is more likely to later enter the first failure mode or
the second failure mode.
13. A computerized method of generating a model for determining an
individual's propensity to enter either a first failure mode or a
second failure mode, the method comprising: defining a bad
performance definition to include individuals that have
characteristics of one or more of the first and second failure
modes; receiving observation data regarding a plurality of
individuals fitting the bad performance definition, the observation
data indicating characteristics of the individuals at an
observation time; receiving outcome data regarding the plurality of
individuals fitting the bad performance definition, the outcome
data indicating characteristics of the individuals fitting the bad
performance definition during an outcome period, the outcome period
beginning after the observation time; and comparing the observation
data and the outcome data in order to generate a model usable to
determine a likelihood that an individual not fitting the bad
performance definition will enter a first failure mode or if the
individual will enter the second failure mode, wherein the method
is performed by one or more computing systems.
Description
RELATED APPLICATIONS
[0001] This application is a division of U.S. patent application
Ser. No. 11/535,907, filed on Sep. 27, 2006 and entitled "SYSTEMS
AND METHODS FOR ANALYZING DATA," which claims priority under 35
U.S.C. .sctn. 119(e) to U.S. Provisional Application Ser. No.
60/781,391, filed on Mar. 10, 2006, each of which is hereby
expressly incorporated by reference in their entirety.
BACKGROUND OF THE INVENTION
[0002] 1. Field of the Invention
[0003] This invention is related to analysis of data related to a
plurality of individuals in order to categorize the individuals.
More particularly, the invention is related to analysis of
financial and demographic information of individuals in order to
categorize the individuals, assign risks for future delinquencies
to the individuals, and return reasons for assignment of a
particular risk to an individual.
[0004] 2. Description of the Related Art
[0005] Lending institutions provide credit accounts such as
mortgages, automobile loans, credit card accounts, and the like, to
consumers. Prior to providing an account to an application, or
applicants, however, many of these institutions review credit
related data and demographic data associated with the applicant in
order to determine a risk of the applicant defaulting on the
account or filing for bankruptcy, for example. Such credit and
demographic data may be used to categorized, or segment,
individuals into one of a plurality of segments where each segment
is associated with other individuals that each have certain similar
attributes. Scoring models that may be particular to the assigned
segment may then be applied to the individual in order to determine
a risk score that is used by the lending institution to assess a
risk level associated with the applicant.
SUMMARY
[0006] In one embodiment, information regarding individuals that
fit a bad performance definition, such as individuals that have
previously defaulted on a financial instrument or have declared
bankruptcy, is used to develop a model that is usable to determine
whether an individual that does not fit the bad performance
definition is more likely to subsequently default on a financial
instrument or to declare bankruptcy. The model may be used to
generate a score for each individual, and the score may be used to
segment the individual into a segment of a segmentation structure
that includes individuals with related characteristics, where
segments may include different models for generating a final risk
score for the individuals assigned to the particular segments.
Thus, the segment to which an individual is assigned, which may be
determined based at least partly on the score assigned to the
individual, may affect the final risk score that is assigned to the
individual.
[0007] In another embodiment, a method of generating a
default/bankruptcy model for assigning an individual to particular
segments of a segmentation structure, wherein the
default/bankruptcy model is indicative of an individual's
propensity to either default on one or more financial instruments
or file for bankruptcy comprises, receiving observation data
comprising financial and demographic information regarding a
plurality of individuals, the observation data indicating
characteristics of the individuals at an observation time,
receiving outcome data comprising financial and demographic
information regarding the plurality of individuals fitting a bad
performance definition, the outcome data indicating characteristics
of the individuals fitting the bad performance definition during an
outcome period, the outcome period beginning after the observation
time, and comparing the observation data and the outcome data in
order to generate the bankruptcy/default model usable to determine
which of a plurality of segments in the segmentation structure a
particular individual should be assigned.
[0008] In another embodiment, a method of assessing a risk
associated with an individual comprises generating a model based on
data regarding a first subgroup of a population, the subgroup
comprising a first portion fitting a first failure definition and a
second portion fitting a second failure definition, and applying
the generated model to the individual, wherein the individual is
not a member of the first subgroup.
[0009] In another embodiment, a computing system for segmenting
each of a plurality of individuals into one of a plurality of
segments of a segmentation structure comprises a profile module
configured to generate a default/bankruptcy model for assigning
each individual to one or more segments of the segmentation
structure, wherein the default/bankruptcy model is indicative of an
individual's propensity to either default on one or more financial
instruments or to file for bankruptcy, and a segmentation module
configured to segment each of the individuals using the
default/bankruptcy model, wherein the individuals include
individuals satisfying a bad performance definition and individuals
satisfying a good performance definition.
[0010] In another embodiment, a method for selecting one or more
adverse action codes to associate with a final risk score assigned
to an individual, each of the adverse action codes indicating a
reason that the final risk score was assigned to the individual,
wherein the individual is assigned to a segmentation hierarchy
comprising a plurality of segments, including a final segment, in a
segmentation structure comprises determining a first penalty
associated with assignment of the individual to a final segment,
determining a first ratio of the first penalty to a difference
between a highest possible final risk score and the final risk
score for the individual, if the determined first ratio is above a
first determined threshold, allotting an adverse action code
related to assignment of the individual to the final segment.
[0011] In another embodiment, a method of generating a model for
determining an individual's propensity to enter either a first
failure mode or a second failure mode comprises defining a bad
performance definition to include individuals that have
characteristics of one or more of the first and second failure
modes, receiving observation data regarding a plurality of
individuals fitting the bad performance definitions, the
observation data indicating characteristics of the individuals at
an observation time, receiving outcome data regarding the plurality
of individuals fitting the bad performance definition, the outcome
data indicating characteristics of the individuals fitting the bad
performance definition during an outcome period, the outcome period
beginning after the observation time, and comparing the observation
data and the outcome data in order to generate a model usable to
determine a likelihood that an individual not fitting the bad
performance definition will enter a first failure mode or if the
individual will enter the second failure mode.
BRIEF DESCRIPTION OF THE DRAWINGS
[0012] FIG. 1 is one embodiment of a block diagram of a computing
system that is in communication with a network and various devices
that are also in communication with the network.
[0013] FIG. 2 is one embodiment of a flowchart illustrating an
exemplary method of analyzing data to create a model.
[0014] FIG. 2A is another embodiment of a flowchart illustrating an
exemplary method of analyzing data from multiple points in time in
order to create a model.
[0015] FIG. 3 illustrates one embodiment of a segmentation
structure having a single segment.
[0016] FIG. 4 illustrates one embodiment of a segmentation
structure having two levels of segments.
[0017] FIG. 5 illustrates one embodiment of a segmentation
structure having three levels of segments.
[0018] FIG. 6 illustrates one embodiment of a segmentation
structure having four levels of segments.
[0019] FIG. 7 illustrates one embodiment of a segmentation
structure having five levels of segments.
[0020] FIG. 8 illustrates one embodiment of the segmentation
structure of FIG. 7 replacing the segment captions with criteria
for assigning individuals to each segment.
[0021] FIG. 8A illustrates another embodiment of the segmentation
structure of FIG. 7 replacing the segment captions with criteria
for assigning individuals to each segment.
[0022] FIG. 9 is one embodiment of a flowchart illustrating an
exemplary process for development of a model using financial and/or
demographic information related to a subset of individuals, and
application of the developed model to any individual.
[0023] FIG. 10 is one embodiment of a Venn diagram showing an
exemplary division of an entire population into previous bankruptcy
and no previous bankruptcy segments, as well as a high risk segment
that overlaps portions of both the previous bankruptcy and no
previous bankruptcy segments.
[0024] FIG. 11 is one embodiment of a flowchart showing a process
of generating a model that tracks which of two or more results is
most likely.
[0025] FIG. 12 is one embodiment of a flowchart showing a process
of applying the model generated by the method of FIG. 11 in order
to assign particular individuals to segments, where each segment
may have a unique scoring model that is applied to accounts
assigned to the segment.
[0026] FIG. 13 is one embodiment of a flowchart showing a process
of developing a default/bankruptcy profile model using only data
related to individuals with accounts that are classified as default
and individuals that have previously declared bankruptcy.
[0027] FIG. 14 is one embodiment of a flowchart showing a process
of applying the default/bankruptcy profile model generated by the
method of FIG. 13 in order to segment individuals.
[0028] FIG. 15 is one embodiment of a flowchart illustrating an
exemplary method of allocating adverse action codes to various
levels of a segment hierarchy associated with an individual.
[0029] FIG. 16 is one embodiment of a flowchart illustrating an
exemplary process of determining how many adverse action codes
should be allotted to each level of the segment hierarchy of an
individual.
[0030] FIG. 17 is one embodiment of a flowchart illustrating an
exemplary process of allocating adverse action codes to various
segments in a segment hierarchy.
DETAILED DESCRIPTION OF CERTAIN EMBODIMENTS
[0031] Embodiments of the invention will now be described with
reference to the accompanying figures, wherein like numerals refer
to like elements throughout. The terminology used in the
description presented herein is not intended to be interpreted in
any limited or restrictive manner, simply because it is being
utilized in conjunction with a detailed description of certain
specific embodiments of the invention. Furthermore, embodiments of
the invention may include several novel features, no single one of
which is solely responsible for its desirable attributes or which
is essential to practicing the inventions described herein.
[0032] FIG. 1 is one embodiment of a block diagram of a computing
system 100 that is in communication with a network 160 and various
devices that are also in communication with the network 160. The
computing system 100 may be used to implement certain systems and
methods described herein. For example, in one embodiment the
computing system 100 may be configured to receive financial and
demographic information regarding individuals and generate risk
scores for the individuals. The functionality provided for in the
components and modules of computing system 100 may be combined into
fewer components and modules or further separated into additional
components and modules.
[0033] In general, the word module, as used herein, refers to logic
embodied in hardware or firmware, or to a collection of software
instructions, possibly having entry and exit points, written in a
programming language, such as, for example, C or C++. A software
module may be compiled and linked into an executable program,
installed in a dynamic link library, or may be written in an
interpreted programming language such as, for example, BASIC, Perl,
or Python. It will be appreciated that software modules may be
callable from other modules or from themselves, and/or may be
invoked in response to detected events or interrupts. Software
instructions may be embedded in firmware, such as an EPROM. It will
be further appreciated that hardware modules may be comprised of
connected logic units, such as gates and flip-flops, and/or may be
comprised of programmable units, such as programmable gate arrays
or processors. The modules described herein are preferably
implemented as software modules, but may be represented in hardware
or firmware. Generally, the modules described herein refer to
logical modules that may be combined with other modules or divided
into sub-modules despite their physical organization or
storage.
[0034] The computing system 100 includes, for example, a personal
computer that is IBM, Macintosh, or Linux/Unix compatible. In one
embodiment, the exemplary computing system 100 includes a central
processing unit ("CPU") 105, which may include a conventional
microprocessor. The computing system 100 further includes a memory
130, such as random access memory ("RAM") for temporary storage of
information and a read only memory ("ROM") for permanent storage of
information, and a mass storage device 120, such as a hard drive,
diskette, or optical media storage device. Typically, the modules
of the computing system 100 are connected to the computer using a
standards based bus system. In different embodiments, the standards
based bus system could be Peripheral Component Interconnect (PCI),
Microchannel, SCSI, Industrial Standard Architecture (ISA) and
Extended ISA (EISA) architectures, for example.
[0035] The computing system 100 is generally controlled and
coordinated by operating system software, such as the Windows 95,
98, NT, 2000, XP, Linux, SunOS, Solaris, or other compatible
operating systems. In Macintosh systems, the operating system may
be any available operating system, such as MAC OS X. In other
embodiments, the computing system 100 may be controlled by a
proprietary operating system. Conventional operating systems
control and schedule computer processes for execution, perform
memory management, provide file system, networking, and I/O
services, and provide a user interface, such as a graphical user
interface ("GUI"), among other things.
[0036] The exemplary computing system 100 includes one or more
commonly available input/output (I/O) devices and interfaces 110,
such as a keyboard, mouse, touchpad, and printer. In one
embodiment, the I/O devices and interfaces 110 include one or more
display device, such as a monitor, that allows the visual
presentation of data to a user. More particularly, a display device
provides for the presentation of GUIs, application software data,
and multimedia presentations, for example. The computing system 100
may also include one or more multimedia devices 140, such as
speakers, video cards, graphics accelerators, and microphones, for
example.
[0037] In the embodiment of FIG. 1, the I/O devices and interfaces
110 provide a communication interface to various external devices.
In the embodiment of FIG. 1, the computing system 100 is coupled to
a network 160, such as a LAN, WAN, or the Internet, for example,
via a wired, wireless, or combination of wired and wireless,
communication link 115. The network 160 communicates with various
computing devices and/or other electronic devices via wired or
wireless communication links. In the exemplary embodiment of FIG.
1, the network 160 is coupled to a financial data source 162, such
as a bank or other financial institution, a demographic data source
166, such as a government public information database, and a
customer 164, such as a financial institution that is interested in
the financial risks associated with particular individual. The
information supplied by the various data sources may include credit
data, demographic data, application information, product terms,
accounts receivable data, and financial statements, for example. In
addition to the devices that are illustrated in FIG. 1, the network
160 may communicate with other data sources or other computing
devices.
[0038] In the embodiment of FIG. 1, the computing system 100 also
includes two application modules that may be executed by the CPU
105. In the embodiment of FIG. 1, the application modules include
the profile module 150 and the adverse action module 160, which are
discussed in further detail below. Each of these application
modules may include, by way of example, components, such as
software components, object-oriented software components, class
components and task components, processes, functions, attributes,
procedures, subroutines, segments of program code, drivers,
firmware, microcode, circuitry, data, databases, data structures,
tables, arrays, and variables.
[0039] In the embodiments described herein, the computing system
100 is configured to execute the profile module 150 and/or the
adverse action module 160, among others, in order to provide risk
information regarding certain individuals or entities. For example,
in one embodiment the computing system 100 generates risk scores
for individuals, where the risk scores indicate a financial risk
associated with the individual. In one embodiment, the customer 164
is a financial institution interested in the risk of default or
late payments on a loan or credit card account that has been
applied for by an individual. Thus, the computing system 100 may be
configured to analyze data related to the individual from various
data sources in order to generate a risk score and provide the risk
score to the customer 164. In one embodiment, multiple financial
accounts, such as bank accounts, credit card accounts, and loan
accounts, are associated with each individual. Thus, the computing
system 100 analyzes data regarding multiple accounts of individuals
and determines scores for the individuals that are usable by one or
more customers. Various other types of scores, related to other
types of risks, may also be generated by the computing system 100.
Although the description provided herein refers to individuals, the
term individual should be interpreted to include groups of
individuals, such as, for example, married couples or domestic
partners, and business entities.
[0040] In one embodiment, the computing system 100 executes the
profile module 150, which is configured to analyze data received
from one or more data sources and generate a profile model that is
usable to assign individuals to groups. The groups to which
individuals may be assigned may also be referred to as segments and
the process of assigning accounts to particular segments may be
referred to as segmentation. A segmentation structure may include
multiple segments arranged in a tree configuration, wherein certain
segments are parents, or children, of other segments. A segment
hierarchy includes the segment to which an individual is assigned
and each of the parent segments to the assigned segment. FIG. 7,
described in detail below, illustrates a segmentation structure
having multiple levels of segments to which individuals may be
assigned. In one embodiment, the segments are each configured to be
associated with individuals that each have certain similar
attributes.
[0041] After assigning a score to an individual, the computing
system 100 may also select and provide reasons related to why the
individual was assigned a particular score. For example, many
customers request information regarding the factors that had the
most impact on an individual's risk score. Thus, in one embodiment
the computing system 100 selects one or more adverse action codes
that are indicative of reasons that a particular score was assigned
to an individual. In certain embodiments, the assignment of an
individual to a particular segment may be a factor that was
relevant in arriving at the risk score for the individual. Thus, in
one embodiment, one or more adverse action codes provided to a
customer may be related to the assignment of the individual to a
particular segment, or to particular segments in the segment
hierarchy. In one embodiment, the adverse action module 160 is
configured to determine how many, if any, of a determined number of
total adverse action codes should be allotted to various segments
of the individuals segment hierarchy. The adverse action module 160
may also determine which adverse action codes are returned. The
operation of the profile module 150 and the adverse action module
160 are explained further below with respect to the drawings.
I. Segmentation
[0042] FIG. 2 is one embodiment of a flowchart illustrating an
exemplary method of analyzing data to create a model. The exemplary
method of analyzing data may be stored as a process accessible by
the profile module 120 and/or other components of the computing
system 100. As described above, models may be created based on
existing data in an attempt to predict characteristics of other
related data. Depending on the embodiment, certain of the blocks
described below may be removed, others may be added, and the
sequence of the blocks may be altered.
[0043] Beginning in a block 210, financial and demographic
information is received by a computing device, such as the
computing device 100 of FIG. 1. The financial and demographic data
may be received from various data sources, including those
discussed above with reference to FIG. 1. In the embodiment of FIG.
2, financial and demographic information related to a plurality of
individuals, and a plurality of financial accounts associated with
the individuals, is obtained. Thus, for any given individual, data
regarding characteristics of multiple financial accounts may be
received. In addition, the received data may be a subset of the
available data, such as, for example males older than 40, or a
random 10% sample of the population. In an advantageous embodiment,
the received data is in a format that is easily understood and
usable by the computing system 100. It is recognized that in other
embodiments, the data could be retrieved in block 210, such as, for
example, by reading data stored on one or more data source via the
network 160
[0044] Moving to a block 220, one or more models are developed
based on a comparison of the received data. In the embodiment of
FIG. 2, a model is generated by comparing characteristics of
individuals that are classified as fitting either a good or a bad
definition. In one embodiment, for example, a bad performance
definition is associated with individuals having at least one
account that has had a 90+ days past due status within the previous
two years, for example, while the good performance definition is
associated with individuals that have not had a 90+ days past due
status on any accounts within the previous two years. It is
recognized that in other scenarios, individuals with at least one
account that is 90+ days past due may be classified as a good
performance definition. As those of skill in the art will
recognize, the specific criteria for being categorized in either
the good or bad performance definitions may vary greatly and may
consider any available data, such as data indicating previous
bankruptcy, demographic data, and default accounts associated with
an individual, for example.
[0045] Continuing to a block 230, the developed model is applied to
an individual in order to determine risks associated with the
individual. For example, the model may be used to determine if an
individual is more closely related to the individuals associated
with the good performance definition, or with individuals
associated with the bad performance definition. Thus, application
of the model on an individual may predict whether the individual
will have past due account statuses in the future, for example.
Accordingly, the generated model may be used by customers in order
to determine what types of financial services should be offered to
a particular individual, if any, and rates, such as interest rates,
for the individual may be proportional to the risk score developed
by application of the model to the individual.
[0046] FIG. 2A is another embodiment of a flowchart illustrating an
exemplary method of analyzing data from multiple points in time in
order to create a model. In this embodiment, the model may be
created based on analyzing data from a previous point in time (an
observation point) in an attempt to predict known behavior as
measured subsequent to the observation point (during an outcome
period). More particularly, the model is generated by analysis of
the data from the observation point, referred to as observation
data, in context of the data from the outcome period, referred to
as outcome data. Once generated, the model may be applied to
individuals, based on the current data related to the individual at
the time of applying the model. Depending on the embodiment,
certain of the blocks described below may be removed, others may be
added, and the sequence of the blocks may be altered.
[0047] Beginning in a block 250, a snapshot of financial and
demographic information regarding a plurality of individuals at a
particular point in time is received. In the embodiment of FIG. 2A,
the observation point is some time previous to the current time and
may be expressed generally as T-X, where T is the current time and
X is a number of months. In one embodiment, T=the date the profile
model is being generated. In this embodiment, if X=25, the
observation point is 25 months previous to the date the profile
model is being generated. In other embodiments, X may be set to any
other time period, such as 6, 12, 18, 36, or 48, for example.
[0048] Continuing to a block 260, data related to individuals
during a period subsequent and mutually exclusive to the
observation point is obtained. In one embodiment, this outcome
period may be defined generally as the period from T-X+1 to T, is
obtained. Thus, in an exemplary embodiment where X=25, data from
the individuals from 24 months previous until the date of model
generation, is obtained. Behaviors measured for individuals during
the outcome period may include, for example, repayment performance,
bankruptcy filing, and response to a marketing offer. These
behaviors may be referred to as the performance definition of the
analysis.
[0049] Moving to a block 270, the observation data and the outcome
data relative to the categories of the performance definition are
analyzed in order to develop a model. Thus, data regarding the
individuals at the snapshot date is compared to data regarding the
individuals during the outcome period.
[0050] In a block 280, the model developed in block 270 may be
applied to current data of an individual in order to predict future
behavior or attributes of the individual over a time period. In one
embodiment, the model is applied to a snapshot of the financial and
demographic data related to the individual at the time of model
application. Thus, the data used in applying the model may be
predictive during any time after T, such as T+1, T+6, T+12, or
T+24, for example. With respect to the example above, application
of a model generated using X=25 may result in information that
predicts an individual's behavior for a subsequent 24 month
period.
[0051] As described in further detail below, generation of a model
using data related to a certain subpopulation of all individuals
received may advantageously be used to predict certain
characteristics of even individuals outside the subpopulation used
in development of the model. In particular, described below are
exemplary systems and methods for generating a model for segmenting
individuals based on whether the individual is more likely to
default on one or more financial instruments, or whether the
individual is more likely to file for bankruptcy. Thus, the model
is generated by comparing individuals that are associated with
default accounts and/or bankruptcy during the outcome period, which
are each individuals classified in the bad performance definition.
However, although the model is generated using only individuals
that fit the bad performance definition, the generated model is
used to segment individuals that do not fit the bad performance
definition. For example, the model may be applied to individuals
that are not associated with default accounts or bankruptcy
observed during the outcome period. By applying a model generated
from a first subgroup of a population (for example, bad performance
definition individuals) to a second subgroup of the population (for
example, any individuals, include good and bad performance
definition individuals), certain attributes of the first subgroup
are usable to predict risk characteristics of the second subgroup
that may not be detectable using a traditional model.
[0052] FIGS. 3-7 are segmentation structures illustrating
embodiments of levels of segments that may be included in a
segmentation structures. The exemplary segmentation structure of
FIG. 3 illustrates an embodiment of a first level of a segmentation
structure, while the segmentation structures of FIGS. 3-7 each add
an additional segmentation level to the segmentation structure. In
one embodiment, the segmentation structures of FIGS. 3-7 may be
based on observation data. The description of FIGS. 3-7 also
describes exemplary steps of applying a model in order to segment
an individual to a particular segment, and then to apply a model to
the individual in order to determine an individual risk score. The
segmentation structure discussed in these drawings provides one
exemplary segmentation structure that may be use to categorize
individuals. Thus, the segmentation structures described herein are
not intended to limit the scope of segmentation structures that may
be used in conjunction with the profile model generation and
application systems and methods described herein.
[0053] FIG. 3 illustrates one embodiment of a segmentation
structure having a single segment 310. In the embodiment of FIG. 3,
all individuals are assigned to the segment 310. In one embodiment,
segment 310 comprises a scoring model that may be applied to
individuals within the segment in order to determine a preliminary
risk score for the individuals. In one embodiment, because segment
310 includes all individuals, segment 310 may be considered a
starting segment in which any individual is placed, rather than a
segment 310 to which individuals may be assigned using one or more
scoring criteria or attributes of the individuals.
[0054] FIG. 4 illustrates one embodiment of a segmentation
structure having first and second levels of segments. More
particularly, the segmentation structure 400 includes the first
level segment 310 and two second level segments 410, 420 that are
each connected as children nodes of the first level segment 310. In
the embodiment of FIG. 4, segment 410 is associated with
individuals that have one or more previous bankruptcies, while
segment 420 is associated with individuals that have no previous
bankruptcies. Thus, each individual in the entire population
segment 310 may be assigned to one, and only one, of the second
level segments 410, 420. More particularly, each individual either
has a previous bankruptcy, or does not have a previous bankruptcy,
and may therefore be assigned to exactly one of the second level
segments 410 or 420. In other embodiments, some of the individuals
may remain in the first level segment 310, while others are
assigned to second level segments, such as segments 410, 420.
[0055] FIG. 5 illustrates one embodiment of a segmentation
structure having first, second, and third level segments. In the
embodiment of FIG. 5, third level segments 510, 520 have been
associated as children nodes of second level segment 410, and third
level segments 530, 540, and 550 have been associated as children
nodes of second-level segment 420. Thus, as illustrated in FIG. 5,
individuals that are segmented to the previous bankruptcy segment
410 may be further segmented to either a higher risk segment 510 or
a lower risk segment 520. Likewise, individuals that are segmented
to the no previous bankruptcy segment 420 may be further segmented
in either a highest risk segment 530, higher risk segment 540, or
lower risk segment 550. Accordingly, the third level segments
further divide and classify the individuals that are assigned to
the second level segments. In one embodiment, assignment of
individuals to one of the third level segments is determined
according to a preliminary risk score for each particular count.
The preliminary risk score may be determined based on a model that
is developed for application to any individual in the entire
population segment 310, based on certain attributes of each
individual. In the embodiment of FIG. 5, the preliminary risk score
is used in segmenting accounts into one of the third level
segments, rather than directly as a factor in the model for
determining a final risk score.
[0056] FIG. 6 illustrates one embodiment of a segmentation
structure having first, second, third, and fourth level segments.
In the embodiment of FIG. 6, the third level higher risk segment
510 is further segmented into fourth level segments including a
higher bankruptcy risk segment 610 and a lower bankruptcy risk
segment 620. Similarly, the highest risk segment 530 is further
segmented into a default segment 630, default/bankruptcy segment
640, and a bankruptcy segment 650. The higher risk segment 540 is
further segmented into a default segment 660 and a bankruptcy
segment 670. In an advantageous embodiment, a default/bankruptcy
profile model is developed by analyzing individuals that are
associated with a default account and/or bankruptcy during the
outcome period. This default/bankruptcy profile model may then be
applied to individuals within the higher risk segment 510, highest
risk segment 530, or higher risk segment 540, in order to determine
how each of the individuals should be further segmented into one of
the fourth level segments. Thus, although the default/bankruptcy
profile model is developed using only individuals that are
associated with a previous default account and/or bankruptcy, the
model may be useful in segmenting individuals that are not
associated with default accounts or bankruptcy.
[0057] FIG. 7 illustrates one embodiment of the segmentation
structure of FIG. 6 having first through fifth level segments. In
the embodiment of FIG. 7, the bankruptcy segment 650 is further
subdivided into higher risk segment 710 and lower risk segment 720.
In one embodiment, assignment of individuals to either the higher
risk segment 710 or the lower risk segment 720 is determined
according to preliminary risk scores for respective individuals. In
other embodiments, other criteria may be used to segment
individuals into the higher risk segment 710 or the lower risk
segment 720.
[0058] FIG. 8 illustrates one embodiment of the segmentation
structure of FIG. 7 replacing the segment captions with criteria
for assigning individuals to each segment. Accordingly, the
segmentation structure 700 may be used to assign an individual to a
particular segment in the segmentation structure, based on various
attributes of accounts held by the individual at the time of
observation or application of the model. The criteria include in
FIG. 8 are exemplary and are not intended to limit the types or
ranges of criteria that may be used in segmenting individuals. In
the embodiment of FIG. 8, the preliminary risk scores assigned to
individuals range in values from 0 to 10, with 10 representing the
least amount of risk; the default/bankruptcy scores range in values
from 0 to 10, with 10 representing the greatest risk of default and
0 representing the greatest risk of bankruptcy; and the preliminary
bankruptcy scores range in values from 0 to 10, with 10
representing the greatest risk of bankruptcy and 0 representing the
least risk of bankruptcy. However, these ranges of values are
exemplary and are not intended to limit the scope of the systems
and methods described herein. Other scores, such as letter scores
from A-F may be used as preliminary risk scores, default/bankruptcy
scores, and/or preliminary bankruptcy scores. In other embodiments,
higher values may represent different attributes of an individual
than are described above. For example, in one embodiment, the
preliminary bankruptcy scores may range in values from 0 to 10,
with 0, rather than 10, representing the greatest risk of
bankruptcy and 10, rather than 0, representing the least risk of
bankruptcy.
[0059] In one embodiment, the final segment to which an individual
is assigned is associated with a scoring model that is applied to
the individual in order to develop a final risk score for the
individual. Thus, the criteria included in each of the segments
illustrated in FIG. 7 define which individuals should be associated
with each particular segment, rather than indicating a particular
final risk score associated with an individual. As described
further below, certain scoring models associated with segments may
adjust a final risk score for an individual due to assignment of
the individual to a particular segment and/or assignment to a
particular segment hierarchy. For example, in one embodiment a risk
score model for higher bankruptcy risk segment 610 may inherently
or explicitly adjust final risk scores of individuals in that
segment based on the fact that the individuals are assigned to
segment 610. In addition, the risk score model for segment 610 may
also inherently or explicitly adjust the final risk scores of
individuals in that segment based on the fact that the segment
hierarchy includes higher risk segment 510 and previous bankruptcy
segment 410. Other risk score models, however, may not adjust the
final risk score based on assignment to particular segments or
segment hierarchies, or may adjust for some, but not all,
segments.
[0060] In the exemplary embodiment of FIG. 8, at the beginning of
the segmentation process, all individuals are placed in the entire
population segment 310. The individuals are then segmented into two
groups, specifically, previous bankruptcy segment 410 and no
previous bankruptcy segment 420. Thus, the determination of a
second level segment is based only on whether the individual has
previously filed for bankruptcy. As those of skill in the art will
recognize, bankruptcy data may be obtained from various sources,
such as public records or financial account information that may be
available from one or more data sources.
[0061] Once an individual is segmented to either the previous
bankruptcy segment 410 or the no previous bankruptcy segment 420,
further segmentation according to preliminary risk scores is
performed. As noted above, in one embodiment a preliminary risk
score is determined for each of the individuals in the entire
population segment 310. In the embodiment of FIG. 8, for those
individuals assigned to the previous bankruptcy segment 410, if the
preliminary risk score is less than or equal to seven, the account
is assigned to the higher risk segment 510. If, however, an
individual from the previous bankruptcy segment 410 has an
associated preliminary risk score of greater than seven, the
individual is assigned to the lower risk segment 520. Because the
segmentation structure 800 does not include any further segments
below the lower risk segment 520, a final risk model associated
with the lower risk segment 520 may be applied to individuals
assigned to segment 520 in order to generate respective final risk
scores. However, segmentation structure 700 includes additional
segments that are configured as child nodes of the higher risk
segment 510 and, accordingly, the final risk score is not
determined by a model associated with the higher risk segment 510,
but rather by models associated with the child segments.
[0062] In the embodiment of FIG. 8, individuals in the higher risk
segment 510 are further segmented based on a bankruptcy risk score.
In one embodiment, a bankruptcy risk score is calculated for
certain, or all, of the individuals in the previous bankruptcy
segment 410. In the segmentation structure 700, individuals in the
higher risk segment 510 with a bankruptcy risk score that is
greater than or equal to nine are assigned to the higher bankruptcy
risk segment 610, while individuals in the higher risk segment 510
with a bankruptcy score that is less than nine are assigned to the
lower bankruptcy risk segment 620. In one embodiment, each of the
higher bankruptcy risk segment 610 and lower bankruptcy risk
segment 620 have respective final risk score models that are
applied to the individuals assigned to the respective segments in
order to determine a final risk score for each individual.
[0063] As shown in FIGS. 7 and 8, the previous bankruptcy segment
420 is linked to multiple child segments to which individuals may
be segmented. In particular, individuals with a preliminary risk
score of less than or equal to seven are assigned to the highest
risk segment 530, individuals with a preliminary risk score of less
than nine are assigned to the higher risk segment 540, and
individuals with a preliminary risk score of greater than or equal
to nine are assigned to the lower risk segment 550. Because the
segmentation structure 800 does not include any further segments
below the lower risk segment 550, a final risk model associated
with the lower risk segment 550 is applied to individuals assigned
to segment 550 in order to generate respective final risk scores.
However, segmentation structure 800 includes additional segments
that are configured as child nodes of the highest risk segment 530
and the higher risk segment 540 and, accordingly, the final risk
score is not determined by a model associated with the highest risk
segment 530 or the higher risk segment 510, but rather by models
associated with the child segments.
[0064] In the embodiment of FIG. 8, the highest risk segment 530
includes multiple child nodes, specifically, default segment 630,
default/bankruptcy segment 640, and bankruptcy segment 650. In one
embodiment, individuals in the highest risk segment 530 are
segmented into one of the segments 630, 640, or 650 based on a
default/bankruptcy profile score. As described in further detail
below with reference to FIGS. 9-14, a default/bankruptcy model may
be developed based on account information related to individuals
within either bankruptcy or default accounts within the outcome
period. In one embodiment, individuals associated with default
accounts includes those individuals that have had at least one 90
days past due account status in the outcome period. For example, in
one embodiment an individual is categorized as default if within
the two year outcome period, one or more accounts associated with
the individual have reported a 90 days past due status. In one
embodiment, the default category individuals and the bankruptcy
category are mutually exclusive, so that if an individual satisfies
the criteria for being categorized in both the bankruptcy and
default categories, only the bankruptcy categorization will be
applied to the individual. In other embodiments, other criteria may
be used to categorize individuals as default or bankrupt. For
example, information regarding 30 days past due, 60 days past due,
and 120 days past due accounts of an individual may be used in
categorizing individuals as default. Likewise, various time periods
may be reviewed in order to locate possible past due accounts and
bankruptcy information. For example, the outcome period may be six
months, one year, two years, three years, or five years.
[0065] As will be described in further detail below, although the
default/bankruptcy profile model is developed based on only account
data associated with individuals categorized as either default or
bankrupt, the default/bankruptcy profile model may advantageously
be applied to individuals that are not categorized as either
bankrupt or default in order to segment these individuals. For
example, as illustrated in FIG. 8, those individuals in the highest
risk segment 530 having a default/bankruptcy profile score of
greater than 8 are assigned to the default segment 630, those
individuals having a default/bankruptcy profile score of greater
than seven, but less than or equal to eight, are assigned to the
default/bankruptcy segment 640, and those individuals having a
default/bankruptcy profile score of less than or equal to seven are
assigned to the bankruptcy segment 650. In one embodiment, the
assignment of individuals to one of the segments 630, 640, or 650,
is indicative of a prediction as to whether the individual is more
likely to either default or file for bankruptcy in the future.
Thus, those individuals in the default segment 630 are more likely
to default on an account in the future then they are to go bankrupt
and those individuals in the bankruptcy segment 650 are more likely
to declare bankruptcy in the future than to default on an account.
In the embodiment of FIG. 8, those individuals in the
default/bankruptcy segment 640 are substantially equally likely to
either default on an account or to file for bankruptcy.
[0066] For those individuals in the higher risk segment 540, the
default/bankruptcy profile model is applied and the individuals are
further segmented to either the default segment 660 or the
bankruptcy segment 670 according to the score returned from
application of the default/bankruptcy profile model. More
particularly, those individuals with a default/bankruptcy profile
score of less than seven are assigned to the default segment 660,
while those individuals with a default/bankruptcy profile score of
greater than or equal to seven are assigned to the bankruptcy
segment 670. As noted above, assignment to the default segment 660
may indicate that an individual is more likely to default on an
account than to file for bankruptcy, while assignment to the
bankruptcy segment 670 may indicate that an individual is more
likely to file for bankruptcy then you default on an account.
[0067] In the embodiment of FIGS. 7 and 8, individuals assigned to
the bankruptcy segment 650 may further be segmented into the higher
risk segment 710 or the lower risk segment 720. In the embodiment
of FIG. 8, segmentation to one of segments 710 or 720 is based upon
the preliminary risk score for each individual. In the particular
example of FIG. 8, those individuals having a preliminary risk
score of less than seven are assigned to the higher risk segment
710, while those individuals having a preliminary risk score
greater than or equal to seven are assigned to the lower risk
segment 720. In one embodiment, each of the higher risk segment 710
and lower risk segment 720 are associated with a final risk score
model that is applied to individuals within the respective segments
in order to determine final risk scores for those individuals. FIG.
8A illustrates an additional embodiment of the segmentation
structure of FIG. 7.
[0068] FIG. 9 is one embodiment of a flowchart illustrating an
exemplary process for development of a model using account
information related to a subset of individuals (for example,
individuals fitting a bad performance definition) and application
of the developed model to any individual (for example, any
individuals). This exemplary method of developing and applying a
model may be stored as a process accessible by the profile module
120 and/or other components of the computing system 100. This
process of generating and applying a model may be used in
conjunction with various types of information. In one embodiment,
models may be developed using the methodology described with
reference to exemplary FIG. 9 based on data associated with two
failure groups within a group of individuals fitting a bad
performance definition. This model may then be applied to
individuals that do not fit the bad performance definition, as well
as to individuals that do fit the bad performance definition. For
example, a first failure group may include individuals that have
defaulted on installment loans and a second failure group may
include individuals that have defaulted on revolving loans, where
both failure groups fit a bad performance definition. In another
embodiment, models may be developed with the methodology of FIG. 9
using information regarding the bank loans of individuals and
information regarding auto loans of individuals. Depending on the
embodiment, certain of the blocks described below may be removed,
others may be added, and the sequence of the blocks may be
altered.
[0069] In a block 910, financial and demographic information from a
previous point in time, referred to as an observation point,
regarding a plurality of individuals is received by a computing
device, such as the computing system 100. This information may be
obtained from various sources and received in various manners. In
one embodiment, information may be received by the computing system
100 on a network connection with one or more financial data sources
162 and/or demographic data sources 166. In another embodiment, the
financial and demographic information is retrieved by the computing
system 100, such as, for example, by reading data stored on a data
source connected to the network 160. In other embodiments,
information may be received on a printed medium, such as through
the mail, or verbally. In an advantageous embodiment, any
information that is not received in an electronic format is
converted to electronic format and made accessible to the computing
system 100.
[0070] Next, in a block 920, behaviors of a subpopulation of
individuals are observed over a set time period subsequent and
mutually exclusive to the observation point. Individuals in two
subcategories of a bad performance definition, such as first and
second failure groups, are then selected for analysis in developing
a model. For example, individuals having accounts that satisfy
either default or bankruptcy criteria may be selected for use in
developing a default/bankruptcy model. In another example, a first
failure group may include individuals that have defaulted on an
installment loan and a second failure group may include individuals
that have defaulted on a revolving loan. The model generated using
these failure groups may be used to determine whether an individual
to which the generated model is applied is more likely to default
on an installment loan or a revolving load. Additionally, models
may be generated based on contrasting of data regarding individuals
in other groups that are not necessarily part of a bad performance
definition. Thus, the term failure group should not be construed as
limited to only groups of individuals that have negative credit
attributes. For example, a model may be created using information
related to individuals in each of two success groups that are each
part of a good performance definition. This model may then be used
to determine the likelihood that an individual not fitting the good
performance definition will enter the first success group or the
second success group.
[0071] In a block 930, a model is developed based on only account
information of the individuals in the selected one or more
categories. Thus, the model is developed using account information
related to only a subset of individuals, such as individuals in
first and second failure groups within a bad performance
definition. For example, a default/bankruptcy model may be
developed using data associated with only those individuals having
accounts that are classified as either bankrupt or default,
although the entire population includes many other individuals that
do not meet these criteria.
[0072] In a block 940, the developed model is applied to
individuals using current data in order to segment individuals into
groups, where each group includes individuals having one or more
related attributes. In one embodiment, the developed model is
applied to individuals that do not meet the criteria for the
selected categories that were used in developing the model, such as
individuals that fit a good performance definition. Thus, a
default/bankruptcy model may be applied to individuals that are
classified as neither default nor having a previous bankruptcy.
[0073] FIG. 10 is one embodiment of a Venn diagram showing an
exemplary division of an entire population into previous bankruptcy
and no previous bankruptcy segments, as well as a high risk
segment. As shown in FIG. 10, the entire population includes
individuals with no previous bankruptcy in segment 1010, and those
with a previous bankruptcy in segment 1020. Additionally, a high
risk segment 1030 includes some individuals from both the previous
bankruptcy segment 1020 and the no previous bankruptcy segment
1010. Thus, because there are high risk individuals in both the
previous bankruptcy and no previous bankruptcy segments, a model
developed using the high risk individuals and previous bankruptcy
individuals may provide some predictive value to those individuals
in the no previous bankruptcy segment 1010.
[0074] FIG. 11 is one embodiment of a flowchart showing a generic
process of generating a profile model that tracks which of two or
more results is more likely. The method of FIG. 11 may be applied
to various types of data sets in order to predict which of two or
more possible results is most likely. For example, the methodology
of FIG. 11 may be used in order to generate a model that predicts
whether an individual is more likely to default on a revolving loan
or if the individual is more likely to default on an installation
loan. This exemplary method of generating a profile model may be
stored as a process accessible by the profile module 120 and/or
other components of the computing system 100. Depending on the
embodiment, certain of the blocks described below may be removed,
others may be added, and the sequence of the blocks may be
altered.
[0075] Beginning in a block 1110, data related to accounts that are
associated with one or more of the results is received. For
example, if the model is intended to determine if an individual is
more likely to default on installment loans or revolving loans, the
data received by a computing device 100 may include financial and
demographic information regarding individuals that have previously
defaulted on either installment or revolving loans. Likewise, if
the model is intended to determine if an individual is more likely
to default on a bank loan or if the individual is more likely to
default on an automobile loan, the data received by the computing
device 100 may include financial and demographic information
regarding individuals that have previously defaulted on either
automobile or bank loans.
[0076] Continuing to a block 1120, a model that predicts whether a
first result is more likely that a second result is developed based
on at least a portion of the received data. In one embodiment, the
data related to the multiple results is analyzed in order to detect
similarities and differences in the data. Application of one or
more statistical models may be used in order to analyze the data
and generate a model that projects which of the multiple results is
more likely based upon attributes of an individual that are later
evaluated using the developed model.
[0077] FIG. 12 is one embodiment of a flowchart illustrating an
exemplary process of applying the model generated by the method of
FIG. 11 in order to assign particular individuals to segments,
where each segment may have a unique scoring model that is applied
to individuals assigned to that segment. This exemplary method of
applying a model may be stored as a process accessible by the
profile module 120 and/or other components of the computing system
100. As noted above with reference to FIGS. 3-8, segmentation of
individuals into two or more segments may be useful to group
individuals having one or more similar attributes, where a scoring
model developed specifically for individuals having the similar
attributes may be applied to individuals assigned to respective
segments. Depending on the embodiment, certain of the blocks
described below may be removed, others may be added, and the
sequence of the blocks may be altered.
[0078] Beginning in a block 1210, data related to individuals to be
scored is received. In one embodiment, the data received in block
1210 comprises financial and demographic information regarding one
or more accounts related to each individual to be segmented. In
other embodiments, the data regarding the individuals may comprise
any other types of data that may be useful in categorizing the
individuals into groups.
[0079] Continuing to a block 1220, individuals are divided into
groups based on a model developed using a process similar to the
process described above with reference to FIG. 11. For example, if
the developed model predicts whether in individual is more likely
to default on a revolving loan or a installment loan, the model may
be applied to each of the individuals for which data is received in
block 1210 in order to categories each of the individuals into a
revolving loan group or an installment loan group. In one
embodiment, the individuals that are classified using the model are
not necessarily individuals that meet the criteria used for
selected individuals for generation of the model. For example, a
revolving/installment default model may be applied to individuals
that have never defaulted on either a revolving loan or an
installment loan in order to categorize the individual as either
more likely to default on a revolving loan or more likely to
default on and installment loan. In the embodiment of FIG. 8, for
example, the default/bankruptcy model is applied to individuals in
order to segment the individuals into multiple groups. In the
embodiment of FIG. 8, the individuals that are categorized by the
default/bankruptcy model have not previously declared bankruptcy
and may not be in the default category either. Thus, the
individuals on which the model is applied are not necessarily
individuals that satisfy the criteria for use in model
generation.
[0080] Moving to a block 1230, a score is created for each
individual. In one embodiment, the scores for each individual are
created based on a model that is specific to a particular segment
in which the individual has been assigned. For example, if an
individual is assigned to a first segment, such as through the use
of a revolving/installment model score for the individual, a first
scoring model may be applied to the individual in order to generate
a final risk score for the individual. Likewise, if another
individual is assigned to a second segment, such as through the use
of the revolving/installment model score for the individual, a
second scoring model may be applied to the individual in order to
generate a final risk score.
[0081] FIG. 13 is one embodiment of a flowchart showing a process
of developing a profile model using only data regarding individuals
with accounts that are classified as default and individuals that
have previously declared bankruptcy. This exemplary method of
developing a profile model may be stored as a process accessible by
the profile module 120 and/or other components of the computing
system 100. In an exemplary embodiment, the profile model uses data
regarding individuals that fit a bad performance definition as
measured in the outcome period in order to generate a
default/bankruptcy profile model. Depending on the embodiment,
certain of the blocks described below may be removed, others may be
added, and the sequence of the blocks may be altered.
[0082] Beginning in a block 1310, financial and demographic data
regarding individuals with default accounts and individuals that
have previously filed for bankruptcy during the outcome period are
received by a computing device, such as the computing system 100.
As noted above, individuals may fit a bad performance definition
based on various criteria, such as a number of past due accounts
and a past due period for those accounts. In the embodiment
described herein, individuals fit a bad performance definition if
an account associated with an individual has had a 90+ day past-due
status or if the individual has filed for bankruptcy within the two
year outcome period.
[0083] Moving to a block 1320, a default/bankruptcy profile model
as to whether an individual is more likely to default or go
bankrupt is developed. The model developed by the computing system
100 in block 1320 may be applied to individuals in order to predict
whether an individual is more likely to file for bankruptcy or to
have a default account. In one embodiment, the model may also
predict that there is a similar likelihood that the individual
either declares bankruptcy or as a default account.
[0084] FIG. 14 is one embodiment of a flowchart showing a process
of applying the default/bankruptcy profile model to individuals. As
noted above, the default/bankruptcy profile model may be applied to
any individuals, regardless of whether the individuals have
associated default accounts or have filed for bankruptcy. This
exemplary method of applying a default/bankruptcy profile model may
be stored as a process accessible by the profile module 120 and/or
other components of the computing system 100. Depending on the
embodiment, certain of the blocks described below may be removed,
others may be added, and the sequence of the blocks may be
altered.
[0085] In a block 1410, data regarding individuals to be segmented
is received by the computing system 100. The received data may be
received from one or more data sources, such as the financial data
source 162 and the demographic data source 166.
[0086] Moving to a block 1420, the default/bankruptcy profile model
is applied to individuals for which current data has been received
in order to segment the individuals into two or more segments. For
example, with reference to FIGS. 7 and 8, a default/bankruptcy
profile model is applied to individuals in the highest risk segment
530 in order to further segment the individuals into default
segment 630, default/bankruptcy segments exported, or bankruptcy
segments 650. Likewise, the default/bankruptcy profile model is
applied to individuals assigned to the higher risk segment 540 in
order to further segment those individuals into either the default
segment 660 or the bankruptcy segment 670. In this embodiment, the
default/bankruptcy profile model is used only for segmenting the
individuals and not specifically in the determination of a final
risk score for the individuals. In other embodiments, the results
of application of the default/bankruptcy profile model may be used
in the development of risk scores for individuals.
[0087] Continuing to a block 1430, final risk scores are generated
for the segmented individuals according to a risk score model that
is particular to the segment in which each individual is assigned.
For example, if an individual is assigned to the default segment
630, a risk score model that has been developed specifically for
scoring those individuals that have a higher risk of defaulting,
rather than going bankrupt, is applied to the individual. If an
individual is assigned to the bankruptcy segment 670, a risk score
model that has been developed specifically for scoring those
individuals that have a higher risk of filing for bankruptcy,
rather than defaulting, is applied to the individual. Thus, for
each bottom segment of the segmentation structure 700, a separate
risk score model may be developed and applied to individuals that
are assigned to the respective segments. For example, in the
embodiment of FIG. 7, the bottom segments include the higher
bankruptcy risk segment 610, the lower bankruptcy risk segment 620,
the lower risk segment 520, the default segment 630, the
default/bankruptcy segment 640, the higher risk segment 710, the
lower risk segment 720, the default segment 660, and the bankruptcy
segment 670. Thus, each of these segments may include a unique risk
scoring model that is applied to individuals within each respective
segment. In other embodiments, a risk scoring model may be used by
multiple segments in determining final risk scores.
II. Adverse Action Codes
[0088] FIG. 15 is one embodiment of a flowchart illustrating an
exemplary method of allocating adverse action codes to various
levels of a segment hierarchy associated with an individual. In
certain embodiments, after determining a segment hierarchy for an
individual, a final risk score is returned and may be provided to a
customer, such as the customer 164. In certain embodiments, the
customer may request and/or be provided with information regarding
attributes of or other information about the individual that
contributed to any decreases in the final risk score. For example,
if a total risk score range that may be assigned to individuals is
from 0 to 100, with 100 representing the lowest risk and 0
representing the highest risk, various factors may contribute to
the actual final risk score assigned to each individual. For
example, the segment to which an individual is assigned may be
considered in determining the final risk score. In addition, the
segment hierarchy, or the parent segments to the assigned segment,
may also be considered and may affect the final risk score for the
individual. Thus, the risk scoring model used by the assigned
segment may take into account the assigned segment and the segment
hierarchy in determining a final risk score.
[0089] In one embodiment, indicators of adverse action codes are
provided to the customer, where the adverse action codes indicate a
specific reason as to why a final risk score for an individual is
less than the maximum. In certain embodiments, adverse action code
may indicate that a final risk score is less than the maximum
partly because of the segment, or segment hierarchy, to which the
individual was assigned. However, for different individuals, the
actual affect of being assigned in a particular segment or in a
segment hierarchy on the final risk score may be significantly
different. For example, for a first individual, assignment to lower
bankruptcy risk segment 620 (FIG. 7) may have had a larger
percentage impact on the individuals final risk score than for a
second individual that was also assigned to the lower bankruptcy
risk segment 620. Thus, providing an adverse action code related to
segmentation of the first individual may be appropriate, while
providing an adverse action code related to segmentation of the
second individual may not provide the most relevant information to
the customer regarding reasons for the final risk score for the
individual. Accordingly, described herein with respect to FIGS.
15-17 are exemplary methods of allotting adverse action codes
related to segmentation of an individual based on the relevance of
the segmentation decision on the final risk score assigned to the
individual. Depending on the embodiment, certain of the blocks
described below may be removed, others may be added, and the
sequence of the blocks may be altered.
[0090] Beginning in a block 1510, a number of adverse action codes
to be provided to the customer 164, for example, is determined. In
one embodiment, a predetermined number of adverse action codes,
such as 2, 4, 5, 6, 8, or 10 adverse action codes, are returned for
each individual for which a final risk score is developed. In one
embodiment, the number of adverse action codes is determined or
calculated based on attributes of the particular individual being
scored and/or the final risk score, and/or other characteristics
related to scoring of the individual.
[0091] Continuing to a block 1520, the number of adverse action
codes that should be allotted to each level of a segmentation
structure in which the individual is assigned is determined. For
example, one or more adverse action codes may be returned for the
segment in which an individual is assigned, as well as for each of
the parent segments in the segment hierarchy. The allotment of
adverse action codes for various levels of a segmentation hierarchy
may be determined based on several factors, such as the relative
impact of assignments to each level of the segment hierarchy had on
the final risk score for the individual.
[0092] Moving to a block 1530, the adverse action codes for each
allotted segment are determined. In one embodiment, the adverse
action code for being assigned to a particular segment comprises an
indication that the individual was assigned to the particular
segment. For example, an adverse action code for an individual
assigned to the higher bankruptcy risk segment 610 (FIG. 7) may
indicate that the individual was assigned to the higher bankruptcy
risk segment. Additionally, the individual assigned to the higher
bankruptcy risk segment 610 may also receive an adverse action code
indicating that the individual was assigned to a higher risk
segment, for example, the higher risk segment 510. However, based
on the allotment of adverse action codes, neither of these
segmentation adverse action codes may be reported to the customer,
and all of the adverse action codes may be related to the various
outputs of the scoring model associated with generation of the
final risk score.
[0093] FIG. 16 is one embodiment of a flowchart illustrating an
exemplary process of determining how many adverse action codes
should be allotted to each level of the segment hierarchy to which
an individual has been assigned.
[0094] Beginning in a block 1610, the total number of adverse
action codes to provide to the customer is determined. As noted
above, the number of adverse action codes returned may be a static
number used for all individuals or, alternatively, may be a dynamic
number that is determined based on attributes of the individual or
results of one or more scoring models applied to the
individual.
[0095] Continuing to a block 1620, the final segment to which the
individual was assigned is selected for allotment analysis. More
particularly, the segment in which the individual was assigned is
selected in order to determine whether one or more of the available
adverse action codes should indicate assignment to the segment.
[0096] Moving to a block 1630, a percentage drop of the final risk
score for the individual due to a penalty for assignment to the
selected segment is determined. In certain embodiments, assignment
to a particular segment decreases a total possible final risk score
that an individual may receive. For example, if a total possible
final risk score for the entire population 310 (FIG. 700) is 1000,
the total possible final risk score for individuals in the previous
bankruptcy segment 410 may be decreased, for example by 100 points,
so that the total possible final risk score for individuals
segmented in the previous bankruptcy segment 410 is 900. Similarly,
if an individual is then further segmented into the higher risk
segment 510, the total possible final risk score for the individual
may be further decreased by another penalty, for example 50 points,
reducing the total possible final risk score for individuals
segmented in the higher risk segment 510 to 850.
[0097] Continuing to a block 1640, the selected segment is allotted
one or more adverse action codes if the percentage drop of the
final risk score due to a penalty for assignment to the selected
segment is within a predetermined range. For example, in one
embodiment a single adverse action code may be allotted to the
selected segment if the percentage drop of the final risk score due
to the penalty for assignment to the selected segment is greater
than 25%. In other embodiments, the percentage drop required for
allocating an adverse action code to a particular segment may be
lower or higher than 25%, such as 10%, 12.5%, 20%, 30%, or 50%, for
example.
[0098] Moving to a decision block 1650, the computing system 100
determines if there are additional parent groups in the
segmentation hierarchy to which the individual has been assigned.
For example, the segmentation hierarchy for an individual assigned
to the higher bankruptcy risk segment 610 includes the higher risk
segment 510, the previous bankruptcy segment 410, and the entire
population segment 310. Accordingly, after allotment of adverse
action codes to the higher bankruptcy risk segment 610, the
computing device 100 determines at block 1650 that additional
parent groups in the segment hierarchy are present and additional
adverse action code allotment should be considered. If additional
parent groups are present, the process continues to a block 1660
where the parent group of the previously selected segment is
selected for allotment analysis. For example, after allotment
analysis on the higher bankruptcy risk group 610, the higher risk
segment 510 is selected at block 1660 for allotment analysis.
Likewise, after allotment analysis on higher risk segment 510, the
previous bankruptcy segment 410 is selected for allotment analysis.
After selecting the parent group for allotment analysis in block
1660, the method continues to block 1630, 1640, and 1650. Thus, the
process of determining a percentage drop of the final risk score
due to a penalty for assignment to a particular segment and
allotment of adverse action codes based on the determined
percentage may be performed for each segment in the segmentation
hierarchy for the individual. After each of the segments in the
segmentation hierarchy are considered for allotment analysis, the
method continues from block 1650 to a block 1670, where the adverse
action codes allotted to various segments are generated and
provided to the customer.
[0099] Although the embodiment of FIG. 16 begins the process of
allocating adverse action codes at the final segment to which the
individual is assigned and moves upward through the segmentation
hierarchy, it is understood that the process of allocating adverse
action codes to segments may be performed in the opposite
direction, or in any other order. In one embodiment, adverse action
code allotment begins at the first segmentation level, with the
entire population segment 310 (FIG. 7), for example, and then moves
to the children nodes, such as to the previous bankruptcy segment
410, then to the higher risk segment 510, and then to the higher
bankruptcy risk segment 610. In other embodiments, not all of the
segments in a segmentation structure are considered for allotment
of adverse action codes. For example, the entire population segment
310 and the no previous bankruptcy segment 420, among other
segments, may be excluded from adverse action code allotment
analysis, such as by using the process described above with
reference to FIG. 16.
[0100] FIG. 17 is one embodiment of a flowchart illustrating an
exemplary process of allocating adverse action codes to various
segments in a segment hierarchy. FIG. 17 also includes an example
of application of the general formulas described in the flowchart
using exemplary data related to an exemplary individual. In the
example illustrated in FIG. 17, it is assumed that the highest
final risk score possible for an individual is 100, the penalty for
being assigned to the previous bankruptcy segment 410 (FIG. 7) is
20, and the penalty for assignment to the higher bankruptcy risk
segment 610 is 15. Thus, in the example discussed with reference to
FIG. 17, for an individual assigned to the higher bankruptcy risk
segment 610, the total possible final risk score is 65. For
purposes of example, an individual assigned to the higher
bankruptcy risk segment 610 and having a final score of 50, for
example, having 15 points deducted for reasons other than being
assigned to the higher bankruptcy risk segment 610, is discussed
with reference to the adverse action code allotment method.
[0101] In a block 1710, a total number of adverse action codes to
provide to the customer is determined. In the example of FIG. 17, 4
adverse action codes are returned to the customer.
[0102] Continuing to a block 1720, an adverse action code related
to being assigned to the previous bankruptcy segment is allotted if
the ratio of the penalty for assignment to the previous bankruptcy
segment to the difference between the highest available final risk
score and the actual final risk score is larger than a
predetermined ratio. In the example of FIG. 70, the penalty for
assignment to the previous bankruptcy segment is 20 and the
difference between the highest final risk score and the actual
final risk score is 50 (for example, 100-50=50). Thus, the
determined ratio is 40%. In this example, one adverse action code
is allotted to indicate segmentation to the previous bankruptcy
segment if the ratio is greater than 12.5%. Because the determined
ratio of 40% is greater than 12.5%, an adverse action code is
assigned to indicate segmentation to the previous bankruptcy
segment. In one embodiment, this allotted adverse action code
returned to the customer indicates that the individual was assigned
to a previous bankruptcy group and assignment to that segment had a
nontrivial effect on the actual final risk score of the
individual.
[0103] Moving to a block 1730, an adverse action code related to
being assigned to a subgroup, or segment configured as a child of
the previous bankruptcy segment, is allotted if the ratio of the
penalty for assignment to the particular subgroup to the difference
in the highest available final risk score and the actual final risk
score is larger than a predetermined ratio. In the example of FIG.
17, the penalty for assignment to the higher bankruptcy risk
segment 610 is 15 and a difference between the highest final risk
score and the actual final risk score is 50 (for example,
100-50=50). Accordingly, the determined ratio is 30%. In this
example, if the ratio is between 12.5% and 37.5%, one adverse
action code is allotted to indicate segmentation to the subgroup;
and if the ratio is greater than 37.5%, two adverse action codes
are allotted to indicate segmentation to the subgroup. Using the
exemplary figures provided herein, the ratio is 30% and, thus, one
adverse action code is allotted for indicating segmentation to the
higher bankruptcy risk segment 610.
[0104] Next, in a block 1740, the allotted adverse action codes are
determined and returned to the customer. Using the exemplary
figures introduced with respect to FIG. 17, one adverse action code
has been allotted for indication of assignment to the previous
bankruptcy segment and one adverse action code has been allotted to
indicate segmentation to a subgroup, such as the higher bankruptcy
risk segment 610. In one embodiment, the reported adverse action
codes are derived from the characteristic that had the most
negative impact on segmentation to the selected segment.
Accordingly, because the total number of adverse action codes to
return to the customer is four in this example, two adverse action
codes may be allotted to indicate relevant information determined
from the segment scoring model applied to the individual. In other
examples, a different range of risk score may be used. For example,
the highest final risk score may be 990 with the minimum score at
501; the penalty for a previous bankruptcy may be 190 such that the
highest score for an individual with a previous bankruptcy is 800;
the penalty for being in the highest bankruptcy risk is 160 such
that the highest score for an individual with the highest
bankruptcy risk is 550.
[0105] The foregoing description details certain embodiments of the
invention. It will be appreciated, however, that no matter how
detailed the foregoing appears in text, the invention can be
practiced in many ways. As is also stated above, it should be noted
that the use of particular terminology when describing certain
features or aspects of the invention should not be taken to imply
that the terminology is being re-defined herein to be restricted to
including any specific characteristics of the features or aspects
of the invention with which that terminology is associated. The
scope of the invention should therefore be construed in accordance
with the appended claims and any equivalents thereof.
* * * * *