Methods and systems for generating transition probability matrices through an optimization framework

• US 8,249,981 B2
• Filed: 12/16/2008
• Issued: 08/21/2012
• Est. Priority Date: 12/16/2008
• Status: Expired due to Fees

• Current Assignee: General Electric Company
• Original Assignee: GE Corporate Financial Services, Inc. (General Electric Company)
• Inventors: Keenan, Sean Coleman, Avasarala, Vishwanath, Black, Jason Wayne, Chalermkraivuth, Kete, Ellis, John Andrew, Neagu, Radu, Subbu, Rajesh Venkat, Zhang, Jingjiao, Li, David Chienju
• Primary Examiner(s): VEZERIS, JAMES A
• Application Number:
  US12/336,360
• Publication Number:
  US 20100153299A1
• Time in Patent Office: 1,344 Days
• Field of Search: 705/35, 705/36.R, 705/40
• US Class Current: 705/38
• CPC Class Codes: G06N 3/006 based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
• CPC Class Codes: G06N 3/126 Evolutionary algorithms, e.g. genetic algorithms or genetic programming
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks
• CPC Class Codes: G06Q 40/06 Asset management; Financial planning or analysis

Abstract

A method for generating an optimized transition probability matrix (OTPM) is provided. The method is performed using a computer system coupled to a database. The method includes storing in the database financial data including obligor credit ratings, generating multi-period empirical transition probability matrices (ETPMs) for a selected time horizon using the financial data stored within the database, generating a mathematical expression to minimize a difference between target ETPM values and candidate OTPM values, and calculating the OTPM from the generated mathematical expression and the financial data stored within the database, wherein the calculated OTPM includes a first set of optimized transition probability values for predicting a likelihood that a credit rating of an obligor will migrate from one credit state to another credit state during a first time interval in the future.

Claims

1. A method for generating an optimized transition probability matrix (OTPM), said method performed using a computer coupled to a database, said method comprising:

storing in the database financial data including obligor credit ratings;

generating multi-period empirical transition probability matrices (ETPMs) for a selected time horizon using the financial data stored within the database, wherein each generated ETPM includes transition probability values indicating a likelihood that a credit rating of an obligor will migrate from one credit state to another credit state during a time period included within the selected time horizon, the transition probability values included within the generated ETPMs are defined as target ETPM values;

generating a mathematical expression to minimize a difference between target ETPM values and candidate OTPM values, wherein the mathematical expression is generated using an objective function including at least one of a probit transform function and an exponential function, and by applying predetermined constraints to the objective function;

calculating the OTPM from the generated mathematical expression and the financial data stored within the database, wherein the calculated OTPM includes a first set of optimized transition probability values for predicting a likelihood that a credit rating of an obligor will migrate from one credit state to another credit state during a first time interval in the future, wherein the calculating is performed by the computer; and

predicting the likelihood that the credit rating of the obligor will migrate from one credit state to another credit state during the first time interval based at least in part on the calculated OTPM.

2. A method in accordance with claim 1 wherein generating a mathematical expression further comprises generating a constrained optimization problem by applying the predetermined constraints to the objective function.

3. A method in accordance with claim 2 further comprising creating a representation schema for the constrained optimization problem that satisfies the predetermined constraints.

4. A method in accordance with claim 3 further comprising:

optimizing the constrained optimization problem based on the representation schema using at least one of a nonlinear programming method, an evolutionary algorithm method, and a particle swarm optimization method.

5. A method in accordance with claim 1 further comprising:

calculating additional sets of optimized transition probability values by raising the calculated OTPM to successive powers wherein each successive power represents a subsequent future time interval.

6. A method in accordance with claim 5 wherein calculating additional sets of optimized transition probability values comprises calculating additional sets of optimized transition probability values by raising the calculated OTPM to successive powers wherein each successive power represents a subsequent future time interval and each additional set of optimized transition probability values predicts a likelihood that a credit rating of an obligor will migrate from one credit state to another credit state during the corresponding future time interval.

7. A method in accordance with claim 1 further comprising:

selecting a business or geographical sector for creating the ETPMs, wherein the generated ETPMs and the calculated OTPM relate to the selected business sector or selected geographical sector.

8. A method in accordance with claim 1 further comprising:

selecting a time horizon for creating the ETPMs, wherein the time horizon includes a plurality of historical years, and each multi-period ETPM is for a one-year historical time period during the time horizon.

9. A method in accordance with claim 1 further comprising:

selecting a business or geographical sector for creating the ETPMs;

selecting a time horizon for creating the ETPMs; and

retrieving financial data from the database for creating the ETPMs, wherein the retrieved financial data includes obligor credit ratings for obligors included within the selected sector during the selected time horizon.

10. A method in accordance with claim 9 wherein generating multi-period ETPMs comprises:

selecting a sampling technique for creating the ETPMs, wherein the sampling technique includes either overlapping or non-overlapping sampling.

11. A method in accordance with claim 1 wherein generating multi-period ETPMs comprises:

retrieving financial data from a database for creating the ETPMs, wherein the retrieved financial data includes obligor credit ratings for obligors included within a selected sector during the selected time horizon; and

generating the ETPMs for the selected sector based on the retrieved financial data and a selected sampling technique, wherein the generated ETPMs are based purely on known historical financial data for the selected sector and the selected time horizon.

12. A method in accordance with claim 1 wherein calculating the OTPM comprises calculating an optimized transition probability value for each of a plurality of credit states, wherein each value indicates a likelihood that a credit rating of an obligor will migrate from one credit state included within the plurality of credit states to another credit state included within the plurality of credit states.

13. A method in accordance with claim 1 wherein selecting an objective function comprises:

selecting an objective function including a probit transform function, wherein the probit transform function is defined by

f=|Φ

<sup>−

1</sup>[(y+ε

)/2−

Φ

<sup>−

1</sup>(x+ε

)/2]|

wherein the variable “

x”

is defined as a target transition probability value in the ETPMs, the variable “

y”

is defined as a candidate transition probability value in the OTPM, and ε

is an adder used to comply with the robustness property when the value of x is zero.

14. A method in accordance with claim 1 wherein selecting an objective function comprises:

selecting an objective function including an exponential function, wherein the exponential function is defined by

f=(e<sup>|y−

x|</sup>−

1)/e^x

wherein the variable “

x”

is defined as a target transition probability value in the ETPMs, and the variable “

y”

is defined as a candidate transition probability value in the OTPM.

15. A method in accordance with claim 1 wherein generating a mathematical expression comprises:

generating a mathematical expression to minimize a difference between target ETPM values and candidate OTPM values by applying predetermined constraints to the selected objective function, wherein the predetermined constraints include at least one of strict equality constraints, monotonic decreasing property constraints, and monotonic increasing rating categories constraints.

16. A method in accordance with claim 1 wherein generating a mathematical expression comprises generating the following mathematical expression for calculating the OTPM:

17. A method in accordance with claim 1 wherein generating a mathematical expression comprises generating the following mathematical expression for calculating the OTPM:

18. A system for generating an optimized transition probability matrix (OTPM), said system comprising:

a computer comprising a processor for executing computer-readable instructions, said computer configured to;

store financial data including obligor credit ratings within a database;

generate multi-period empirical transition probability matrices (ETPMs) for a selected time horizon using the financial data stored within the database, wherein each generated ETPM includes transition probability values indicating a likelihood that a credit rating of an obligor will migrate from one credit state to another credit state during a time period included within the selected time horizon, the transition probability values included within the generated ETPMs are defined as target ETPM values;

generate a mathematical expression to minimize a difference between target ETPM values and candidate OTPM values, wherein the mathematical expression is generated using an objective function including at least one of a probit transform function and an exponential function, and by applying predetermined constraints to the objective function; and

calculate the OTPM from the generated mathematical expression and the financial data stored within the database, wherein the calculated OTPM includes a first set of optimized transition probability values for predicting a likelihood that a credit rating of an obligor will migrate from one credit state to another credit state during a first time interval in the future.

19. A system in accordance with claim 18 wherein said computer is further configured to generate a constrained optimization problem by applying the predetermined constraints to the objective function.

20. A system in accordance with claim 19 wherein said computer is further configured to create a representation schema for the constrained optimization problem that satisfies the predetermined constraints.

21. A system in accordance with claim 20 wherein said computer is further configured to optimize the constrained optimization problem based on the representation schema using at least one of a nonlinear programming method, an evolutionary algorithm method, and a particle swarm optimization method.

22. A system in accordance with claim 18 wherein said computer is further configured to:

calculate additional sets of optimized transition probability values by raising the calculated OTPM to successive powers wherein each successive power represents a subsequent future time interval.

23. A system in accordance with claim 22 wherein each additional set of optimized transition probability values predicts a likelihood that a credit rating of an obligor will migrate from one credit state to another credit state during the corresponding future time interval.

24. A system in accordance with claim 18 wherein said computer is further configured to:

receive a selected business or geographical sector for creating the ETPMs, wherein the generated ETPMs and the calculated OTPM relate to the selected business sector or selected geographical sector.

25. A system in accordance with claim 18 wherein said computer is further configured to:

receive a selected time horizon for creating the ETPMs, wherein the time horizon includes a plurality of historical years, and each multi-period ETPM is for a one-year historical time period during the time horizon.

26. A system in accordance with claim 18 wherein said computer is further configured to:

receive a selected business or geographical sector for creating the ETPMs;

receive a selected time horizon for creating the ETPMs; and

retrieve financial data from the database for creating the ETPMs, wherein the retrieved financial data includes obligor credit ratings for obligors included within the selected sector during the selected time horizon.

27. A system in accordance with claim 18 wherein said computer is further configured to generate multi-period ETPMs by automatically selecting a sampling technique for creating the ETPMs, wherein the sampling technique includes either overlapping or non-overlapping sampling.

28. A system in accordance with claim 18 wherein said computer is further configured to generate multi-period ETPMs by:

retrieving financial data from said database for creating the ETPMs, wherein the retrieved financial data includes obligor credit ratings for obligors included within a selected sector during the selected time horizon; and

generating the ETPMs for the selected sector based on the retrieved financial data and a selected sampling technique, wherein the generated ETPMs are based purely on known historical financial data for the selected sector and the selected time horizon.

29. A system in accordance with claim 18 wherein said computer is further configured to calculate an optimized transition probability value for each of a plurality of credit states, wherein each value indicates a probability that a credit rating of an obligor will migrate from one credit state included within the plurality of credit states to another credit state included within the plurality of credit states.

30. A system in accordance with claim 18 wherein said computer is further configured to select an objective function including a probit transform function, wherein the probit transform function is defined by

f=|Φ

<sup>−

1</sup>[(y+ε

)/2−

Φ

<sup>−

1</sup>(x+ε

)/2]|

wherein the variable “

x”

is defined as a target transition probability value in the ETPMs, the variable “

y”

is defined as a candidate transition probability value in the OTPM, and ε

is an adder used to comply with the robustness property when the value of x is zero.

31. A system in accordance with claim 18 wherein said computer is further configured to select an objective function including an exponential function, wherein the exponential function is defined by

f=(e^|y−

x|−

1)/e^x

wherein the variable “

x”

is defined as a target transition probability value in the ETPMs, and the variable “

y”

is defined as a candidate transition probability value in the OTPM.

32. A system in accordance with claim 18 wherein said computer is further configured to generate a mathematical expression to minimize a difference between target ETPM values and candidate OTPM values by applying predetermined constraints to the selected objective function, wherein the predetermined constraints include at least one of strict equality constraints, monotonic decreasing property constraints, and monotonic increasing rating categories constraints.

33. A system in accordance with claim 18 wherein said computer is further configured to calculate the OTPM using the following mathematical expression:

34. A system in accordance with claim 18 wherein said computer is further configured to calculate the OTPM using the following mathematical expression:

35. A non-transitory computer readable medium comprising instructions thereon, said instructions configured to instruct a computer to generate an optimized transition probability matrix (OTPM), said instructions comprising at least one code segment that:

stores in a database financial data including obligor credit ratings;

generates multi-period empirical transition probability matrices (ETPMs) for a selected time horizon using the financial data stored within the database, wherein each generated ETPM includes transition probability values indicating a likelihood that a credit rating of an obligor will migrate from one credit state to another credit state during a time period included within the selected time horizon, the transition probability values included within the generated ETPMs are defined as target ETPM values;

generates a mathematical expression to minimize a difference between target ETPM values and candidate OTPM values, wherein the mathematical expression is generated using an objective function including at least one of a probit transform function and an exponential function, and by applying predetermined constraints to the objective function; and

calculates the OTPM from the generated mathematical expression and the financial data stored within the database, wherein the calculated OTPM includes a first set of optimized transition probability values for predicting a likelihood that a credit rating of an obligor will migrate from one credit state to another credit state during a first time interval in the future.

36. A non-transitory computer readable medium in accordance with claim 35 further comprising instructions comprising at least one code segment that generates a constrained optimization problem by applying the predetermined constraints to the objective function.

37. A non-transitory computer readable medium in accordance with claim 36 further comprising instructions comprising at least one code segment that creates a representation schema for the constrained optimization problem that satisfies the predetermined constraints.

38. A non-transitory computer readable medium in accordance with claim 37 further comprising instructions comprising at least one code segment that optimizes the constrained optimization problem based on the representation schema using at least one of a nonlinear programming method, an evolutionary algorithm method, and a particle swarm optimization method.

39. A non-transitory computer readable medium in accordance with claim 35 further comprising instructions comprising at least one code segment that calculates additional sets of optimized transition probability values by raising the calculated OTPM to successive powers wherein each successive power represents a subsequent future time interval.

40. A non-transitory computer readable medium in accordance with claim 39 wherein each additional set of optimized transition probability values predicts a likelihood that a credit rating of an obligor will migrate from one credit state to another credit state during the corresponding future time interval.

41. A non-transitory computer readable medium in accordance with claim 35 further comprising instructions comprising at least one code segment that receives a selected business or geographical sector for creating the ETPMs, wherein the generated ETPMs and the calculated OTPM relate to the selected business sector or selected geographical sector.

42. A non-transitory computer readable medium in accordance with claim 35 further comprising instructions comprising at least one code segment that receives a selected time horizon for creating the ETPMs, wherein the time horizon includes a plurality of historical years, and each multi-period ETPM is for a one-year historical time period during the time horizon.

43. A non-transitory computer readable medium in accordance with claim 35 further comprising instructions comprising at least one code segment that:

receives a selected business or geographical sector for creating the ETPMs;

receives a selected time horizon for creating the ETPMs; and

retrieves financial data from the database for creating the ETPMs, wherein the retrieved financial data includes obligor credit ratings for obligors included within the selected sector during the selected time horizon.

44. A non-transitory computer readable medium in accordance with claim 35 further comprising instructions comprising at least one code segment that selects a sampling technique for creating the ETPMs, wherein the sampling technique includes either overlapping or non-overlapping sampling.

45. A non-transitory computer readable medium in accordance with claim 35 further comprising instructions comprising at least one code segment that:

retrieves financial data from the database for creating the ETPMs, wherein the retrieved financial data includes obligor credit ratings for obligors included within a selected sector during the selected time horizon; and

generates the ETPMs for the selected sector based on the retrieved financial data and a selected sampling technique, wherein the generated ETPMs are based purely on known historical financial data for the selected sector and the selected time horizon.

46. A non-transitory computer readable medium in accordance with claim 35 further comprising instructions comprising at least one code segment that calculates an optimized transition probability value for each of a plurality of credit states, wherein each value indicates a likelihood that a credit rating of an obligor will migrate from one credit state included within the plurality of credit states to another credit state included within the plurality of credit states.

47. A non-transitory computer readable medium in accordance with claim 35 further comprising instructions comprising at least one code segment that selects an objective function including a probit transform function, wherein the probit transform function is defined by

f=|Φ

<sup>−

1</sup>[(y+ε

)/2−

Φ

<sup>−

1</sup>(x+ε

)/2]|

wherein the variable “

x”

is defined as a target transition probability value in the ETPMs, the variable “

y”

is defined as a candidate transition probability value in the OTPM, and ε

is an adder used to comply with the robustness property when the value of x is zero.

48. A non-transitory computer readable medium in accordance with claim 35 further comprising instructions comprising at least one code segment that selects an objective function including an exponential function, wherein the exponential function is defined by

f=(e^|y−

x|−

1)/e^x

wherein the variable “

x”

is defined as a target transition probability value in the ETPMs, and the variable “

y”

is defined as a candidate transition probability value in the OTPM.

49. A non-transitory computer readable medium in accordance with claim 35 further comprising instructions comprising at least one code segment that:

generates a mathematical expression to minimize a difference between target ETPM values and candidate OTPM values by applying predetermined constraints to the selected objective function, wherein the predetermined constraints include at least one of strict equality constraints, monotonic decreasing property constraints, and monotonic increasing rating categories constraints.

50. A non-transitory computer readable medium in accordance with claim 35 further comprising instructions comprising at least one code segment that calculates the OTPM using the following mathematical expression:

51. A non-transitory computer readable medium in accordance with claim 35 further comprising instructions comprising at least one code segment that calculates the OTPM using the following mathematical expression: