Method and system for modeling future action impact in credit scoring

• US 7,970,676 B2
• Filed: 08/01/2007
• Issued: 06/28/2011
• Est. Priority Date: 08/01/2007
• Status: Active Grant

• Current Assignee: Fair Isaac Corporation
• Original Assignee: Fair Isaac Corporation
• Inventors: Feinstein, Jeffrey A.
• Primary Examiner(s): Colbert; Ella
• Assistant Examiner(s): Tran; Hai
• Application Number:
  US11/832,610
• Publication Number:
  US 20090037308A1
• Time in Patent Office: 1,427 Days
• Field of Search: 705 35- 45
• US Class Current: 705/35
• CPC Class Codes: G06Q 20/10 specially adapted for electronic funds transfer [EFT] systems; specially adapted for home banking systems
• CPC Class Codes: G06Q 20/108 Remote banking, e.g. home banking
• CPC Class Codes: G06Q 40/00 Finance; Insurance; Tax strategies; Processing of corporate or income taxes
• CPC Class Codes: G06Q 40/03 Credit; Loans; Processing thereof
• CPC Class Codes: G06Q 40/06 Asset management; Financial planning or analysis

Abstract

A method and system for predicting impact of future actions on subsequent performance involves developing a prediction model that predicts a statistical interaction of performance expectation with likely behavior. In one embodiment, sensitivity to new, post-scoring date credit behaviors in the analytic solution greatly improves snapshot score predictions. The modeling approach involves multiple snapshots: predictive and performance snapshots, plus an intermediate snapshot shortly after the predictive snapshot to quantify interim behavior. Predictive interaction variables are calculated on the predictive data using simulated profiles before and after an action.

Claims

1. A method for implementation by one or more data processors comprising:

generating, by at least one data processor, a predictive model from a sample population based on;

a predictive snapshot gathered at a past sample scoring date, the predictive snapshot comprising a credit score;

a performance snapshot gathered at a sample performance date subsequent to the scoring date, the performance snapshot representing one of a good performance and a bad performance; and

intermediate information gathered at a sample intermediate date between said scoring date and said performance date, the intermediate information representing an action of an individual out of the sample population at the sample intermediate date, the performance being in response to the action of the individual out of the sample population, wherein a time period between the past sample scoring date and the sample intermediate date is a predetermined first time interval;

inputting, by at least one data processor, a profile of a consumer to a software implementation of said generated predictive model executing on a computing device;

determining, by at least one data processor, a current credit score on current date;

simulating, by at least one data processor using the generated predictive model, an impact on the consumer'"'"'s profile of likely future actions of the consumer, the likely future actions performed on a future intermediate date, the likely future actions causing the impact on a future performance date, the simulating comprising;

computing, by at least one data processor, a capacity index that rank-orders consumers according to the impact of the likely future actions, the capacity index representing a capacity of a consumer to take additional debt without defaulting later, the capacity being represented as a second score;

adjusting, by at least one data processor, said second score to control for covariate effects; and

outputting, by at least one data processor, said adjusted score,

wherein the generated predictive model predicts a statistical interaction of the impact of the likely future actions with the likely future actions.

2. The method of claim 1, wherein said generating a predictive model comprises:

identifying, by at least one data processor, data sources for said information;

generating, by at least one data processor, a stratified sample from said information

acquiring, by at least one data processor, said snapshots and said intermediate information;

calculating, by at least one data processor, sample weights;

adjusting, by at least one data processor, sample weights by including intermediate behavior;

building, by at least one data processor, adjustor model weights, including a base score;

removing, by at least one data processor, the base score from said adjustor model;

validating, by at least one data processor, said predictive model; and

creating, by at least one data processor, reason codes, wherein the reason codes are output with the adjusted score.

3. The method of claim 2, wherein said generating sample weights comprises:

generating, by at least one data processor, predictive variables from the predictive snapshot;

generating, by at least one data processor, a base score on said predictive snapshot;

generating, by at least one data processor, performance on said performance snapshot;

generating, by at least one data processor, interaction of predictive variables with simulated predictive variables on said predictive snapshot; and

generating, by at least one data processor, intermediate behavior on the intermediate information.

4. The method of claim 1, wherein said predictive snapshot comprises a credit risk score and prediction characteristics for each member of said sample population at the beginning of a predetermined performance window;

wherein performance comprises performance behavior on reported trade lines for duration of said performance window; and

wherein said intermediate information comprises incremental balance incurred during said intermediate interval, wherein said performance is weighted by said intermediate information.

5. The method of claim 1, wherein said covariate comprises a base score, and wherein the adjusting of the second score comprises statistically partialing out the base score from the second score such that the capacity of the consumer to take additional debt is independent of the base score.

6. The method of claim 1, wherein one or more profile types likely to be underrepresented in a random sample are over sampled relative to said source population and wherein profile types are selectively weighted in a manner so that the weighted distribution of profile types mirrors the raw distribution of profile types in the source population.

7. The method of claim 1, wherein time period between the past sample scoring date and the sample performance date is a predetermined second time interval, and

wherein a time period between the current date and the future intermediate date is equal to the predetermined first time interval and time period between the current date and the future performance date is equal to the predetermined second time interval.

8. The method of claim 1, wherein the good performance comprises repayment of additional debt undertaken by the individual;

and

wherein the bad performance comprises the consumer being involved in at least one of the following;

bankruptcies, charge-offs, loan defaults, repossessions, serious delinquency, and other derogatory status indicators.

9. The method of claim 1, wherein the predictive snapshot further comprises a credit score report, the credit score report representing prediction variables at the past sample scoring date,

wherein the prediction variables include previous credit performance, current level of indebtedness, amount of time credit has been in use, pursuit of new credit, and types of credit available.

10. An apparatus for modeling future action impact in credit scoring comprising:

a computing device, said computing device comprising a processing component and a storage component; and

computer-readable instructions residing in said storage component which, when executed by said processing component instruct said processor to perform operations comprising;

generating a predictive model from a sample population based on;

a predictive snapshot gathered at a past sample scoring date, the predictive snapshot comprising a credit score;

a performance snapshot gathered at a sample performance date subsequent to the scoring date, the performance snapshot representing one of a good performance and a bad performance; and

intermediate information gathered at a sample intermediate date between said scoring date and said performance date, the intermediate information representing an action of an individual out of the sample population at the sample intermediate date, the performance being in response to the action of the individual out of the sample population, wherein a time period between the past sample scoring date and the sample intermediate date is a predetermined first time interval;

inputting a profile of a consumer to a software implementation of said generated predictive model executing on a computing device;

determining a current credit score on current date;

simulating, using the generated predictive model, an impact on the consumer'"'"'s profile of likely future actions of the consumer, the likely future actions performed on a future intermediate date, the likely future actions causing the impact on a future performance date, the simulating comprising;

computing a capacity index that rank-orders consumers according to the impact of the likely future actions, the capacity index representing a capacity of a consumer to take additional debt without defaulting later, the capacity being represented as a second score;

adjusting said second score to control for covariate effects; and

outputting said adjusted score.

11. The apparatus of claim 10, wherein generating a predictive model comprises:

identifying data sources for said information;

generating a stratified sample from said information

acquiring said snapshots and said intermediate information;

calculating sample weights;

adjusting sample weights by including intermediate behavior;

building adjustor model weights, including a base score;

removing base score from said adjustor model;

validating said predictive model; and

creating reason codes, wherein the reason codes are output with the adjusted score.

12. The apparatus of claim 10, wherein generating sample weights comprises:

generating predictive variables from the predictive snapshot;

generating a base score on said predictive snapshot;

generating performance on said performance snapshot;

generating interaction of predictive variables with simulated predictive variables on said predictive snapshot; and

generating intermediate behavior on the intermediate information.

13. The apparatus of claim 10, wherein said predictive snapshot comprises a credit risk score and prediction characteristics for each member of said sample population at the beginning of a predetermined performance window;

wherein performance comprises performance behavior on reported trade lines for duration of said performance window; and

wherein said intermediate information comprises incremental balance incurred during said intermediate interval, wherein said performance is weighted by said intermediate information.

14. The apparatus claim 10, wherein said covariate comprises a base score, and wherein the adjusting of the second score comprises statistically partialing out the base score from the second score such that the capacity of the consumer to take additional debt is independent of the base score.

15. The apparatus of claim 10, wherein one or more profile types likely to be underrepresented in a random sample are over sampled relative to said source population and wherein profile types are selectively weighted in a manner so that the weighted distribution of profile types mirrors the raw distribution of profile types in the source population.

16. A non-transitory computer-readable storage medium having computer-readable instructions for simulating impact of future events on a consumer'"'"'s profile embodied therein, said computer-readable instructions comprising instructions for:

generating a predictive model from a sample population based on;

a predictive snapshot gathered at a past sample scoring date, the predictive snapshot comprising a credit score;

a performance snapshot gathered at a sample performance date subsequent to the scoring date, the performance snapshot representing one of a good performance and a bad performance; and

intermediate information gathered at a sample intermediate date between said scoring date and said performance date, the intermediate information representing an action of an individual out of the sample population at the sample intermediate date, the performance being in response to the action of the individual out of the sample population, wherein a time period between the past sample scoring date and the sample intermediate date is a predetermined first time interval;

inputting a profile of a consumer to a software implementation of said generated predictive model executing on a computing device;

determining a current credit score on current date;

simulating, using the generated predictive model, an impact on the consumer'"'"'s profile of likely future actions of the consumer, the likely future actions performed on a future intermediate date, the likely future actions causing the impact on a future performance date, the simulating comprising;

computing a capacity index that rank-orders consumers according to the impact of the likely future actions, the capacity index representing a capacity of a consumer to take additional debt without defaulting later, the capacity being represented as a second score;

adjusting said second score to control for covariate effects; and

outputting said adjusted score.

17. The computer-readable medium of claim 16, wherein said instructions for generating a predictive model comprise instructions for:

identifying data sources for said information;

generating a stratified sample from said information

acquiring said snapshots and said intermediate information;

calculating sample weights;

adjusting sample weights by including intermediate behavior;

building adjustor model weights, including a base score;

removing base score from said adjustor model;

validating said predictive model; and

creating reason codes, wherein the reason codes are output with the adjusted score.

18. The computer-readable medium of claim 17, wherein said instructions for generating sample weights comprise instructions for:

generating predictive variables from the predictive snapshot;

generating a base score on said predictive snapshot;

generating performance on said performance snapshot;

generating interaction of predictive variables with simulated predictive variables on said predictive snapshot; and

generating intermediate behavior on the intermediate information.

19. The computer-readable medium of claim 16, wherein said predictive snapshot comprises a credit risk score and prediction characteristics for each member of said sample population at the beginning of a predetermined performance window;

wherein performance comprises performance behavior on reported trade lines for duration of said performance window; and

wherein said intermediate information comprises incremental balance incurred during said intermediate interval, wherein said performance is weighted by said intermediate information.

20. The computer-readable medium of claim 16, wherein said covariate comprises a risk score, wherein said risk score is statistically partialed out from said score indicative of the impact of the simulated action on the consumer'"'"'s profile.

21. The computer-readable medium of claim 16, wherein one or more profile types likely to be underrepresented in a random sample are over sampled relative to said source population and wherein profile types are selectively weighted in a manner so that the weighted distribution of profile types mirrors the raw distribution of profile types in the source population.