17-03-2014, 09:12 PM
Abstract—Modern credit risk management aims at assess the
default probability (DP) of a debtor according to his historical
and current financial data. Due to its prominent importance in
credit loan decisions, the DP assessment becomes a research
focus in the filed of financial data mining. To tackle this
problem, we propose a nonlinear multi-parameter model
(NMM) based on domain knowledge. Additionally, the
parameters of the model are estimated using one of recent
evolutionary algorithm - differential evolution (DE). In the
experiment, real-world financial data is utilized to test the
performance of NMM. Experimental results reveal that NMM
performs much better than back-propagation neural network
(BPNN).
Keywords-credit risk; nonlinear multi-parameter model;
evolutionary algorithm; differential evolution
I. INTRODUCTION
Credit risk is the risk of causing economic losses to
creditors, due to the breach of contracts or inability to fulfill
contracts by debtors [1]. In a narrow sense, it can be called
default risk, as credit risk is concerned essentially with the
economic losses caused by debtors breaking contracts. As
one of the most important financial risks in modern economy,
credit risk is a major challenge confronted by economic
entities, especially financial organizations, investors and
consumers, which has tremendous influence on economic
entities’ operation and management, financial system’s
stabilization, and what’s more, the national economy’s
healthy development.
Traditional credit risk management aims at evaluating a
credit rating for a bank debtor. Credit rating is assessed
according to a debtor’s financial history, current assets and
liabilities. Normally, it is complex and time-consuming for
manual assessment by bank workers, as the evaluation
process deals with huge financial data. From a data mining
perspective, credit rating evaluation is a typical classification
problem. During the last several decades, many researchers
have applied various data mining techniques to tackle this
problem. Rong-zhou et al. [2] established a neural network
credit-risk evaluation model by using back-propagation
algorithm. Kyung-Shik Shin et al. [3] applied the support
vector machine (SVM) to evaluate the credit risk and they
summarized the superiorities of SVM compared with backpropagation
neural network (BPNN). Li-wei Wei et al. [4]
discussed the applications of the support vector machine with
mixture of kernel to design a credit evaluation system. Lean
Yu et al. [5] proposed multistage neural network ensemble
learning model to assess credit risk at the measurement level.
Jian-guo Zhou and Tao Bai [6] also introduced a classifier,
hybridizing rough set approach and improved support vector
machine (SVM) using genetic algorithm (GA), to the study
of credit risk assessment.
As the credit loans decisions become increasingly
complicated, modern credit risk management is more
advanced than traditional methods. Its main proposes are to
assess portfolio risks of a commercial bank, rationalize the
distribution of resources and meet the needs of supervision.
Therefore, limited economic information reflected by the
credit ratings cannot meet the needs of risk management. At
least we must know the default probability (DP) of a debtor.
In terms of DP calculation, KMV model [7], developed
by KMV Company, is famous in credit risk management.
KMV model is based on pricing theory of option. According
to the changes in the stock market and analytical results of
stock price information, the DP of a listing company can be
derived from KMV model. The underlying principle is that
in an effective stock market, stock prices contain investors’
expectations of various factors, which influence stock prices
from different aspects. So stock prices are forward-looking,
and have advanced predictability.
Inspired by pricing theory of option and KMV model, we
propose a nonlinear multi-parameter model (NMM) to
address the credit risk evaluation problem. The proposed
model involves four variables: volatility of stockholders’
equity, value of stockholders’ equity, book value of current
liabilities and long-term liabilities. Moreover, the parameters
of the model are estimated by considering the estimation as a
numerical optimization problem. In the field of optimization,
evolutionary algorithms (EAs) have been increasingly
regarded as a class of effective and efficient techniques and
yielded a wide range of applications. In this paper, we apply
differential evolution (DE) [8] to estimate the model
parameters, as DE is a highly efficient and robust EA.
In the experiment, NMM is tested using the financial data
from a bank in China. The results show that NMM performs
much better than BPNN. Furthermore, the parameters of
NMM are discussed in a business perspective, resulting in
several useful points for credit risk management.
default probability (DP) of a debtor according to his historical
and current financial data. Due to its prominent importance in
credit loan decisions, the DP assessment becomes a research
focus in the filed of financial data mining. To tackle this
problem, we propose a nonlinear multi-parameter model
(NMM) based on domain knowledge. Additionally, the
parameters of the model are estimated using one of recent
evolutionary algorithm - differential evolution (DE). In the
experiment, real-world financial data is utilized to test the
performance of NMM. Experimental results reveal that NMM
performs much better than back-propagation neural network
(BPNN).
Keywords-credit risk; nonlinear multi-parameter model;
evolutionary algorithm; differential evolution
I. INTRODUCTION
Credit risk is the risk of causing economic losses to
creditors, due to the breach of contracts or inability to fulfill
contracts by debtors [1]. In a narrow sense, it can be called
default risk, as credit risk is concerned essentially with the
economic losses caused by debtors breaking contracts. As
one of the most important financial risks in modern economy,
credit risk is a major challenge confronted by economic
entities, especially financial organizations, investors and
consumers, which has tremendous influence on economic
entities’ operation and management, financial system’s
stabilization, and what’s more, the national economy’s
healthy development.
Traditional credit risk management aims at evaluating a
credit rating for a bank debtor. Credit rating is assessed
according to a debtor’s financial history, current assets and
liabilities. Normally, it is complex and time-consuming for
manual assessment by bank workers, as the evaluation
process deals with huge financial data. From a data mining
perspective, credit rating evaluation is a typical classification
problem. During the last several decades, many researchers
have applied various data mining techniques to tackle this
problem. Rong-zhou et al. [2] established a neural network
credit-risk evaluation model by using back-propagation
algorithm. Kyung-Shik Shin et al. [3] applied the support
vector machine (SVM) to evaluate the credit risk and they
summarized the superiorities of SVM compared with backpropagation
neural network (BPNN). Li-wei Wei et al. [4]
discussed the applications of the support vector machine with
mixture of kernel to design a credit evaluation system. Lean
Yu et al. [5] proposed multistage neural network ensemble
learning model to assess credit risk at the measurement level.
Jian-guo Zhou and Tao Bai [6] also introduced a classifier,
hybridizing rough set approach and improved support vector
machine (SVM) using genetic algorithm (GA), to the study
of credit risk assessment.
As the credit loans decisions become increasingly
complicated, modern credit risk management is more
advanced than traditional methods. Its main proposes are to
assess portfolio risks of a commercial bank, rationalize the
distribution of resources and meet the needs of supervision.
Therefore, limited economic information reflected by the
credit ratings cannot meet the needs of risk management. At
least we must know the default probability (DP) of a debtor.
In terms of DP calculation, KMV model [7], developed
by KMV Company, is famous in credit risk management.
KMV model is based on pricing theory of option. According
to the changes in the stock market and analytical results of
stock price information, the DP of a listing company can be
derived from KMV model. The underlying principle is that
in an effective stock market, stock prices contain investors’
expectations of various factors, which influence stock prices
from different aspects. So stock prices are forward-looking,
and have advanced predictability.
Inspired by pricing theory of option and KMV model, we
propose a nonlinear multi-parameter model (NMM) to
address the credit risk evaluation problem. The proposed
model involves four variables: volatility of stockholders’
equity, value of stockholders’ equity, book value of current
liabilities and long-term liabilities. Moreover, the parameters
of the model are estimated by considering the estimation as a
numerical optimization problem. In the field of optimization,
evolutionary algorithms (EAs) have been increasingly
regarded as a class of effective and efficient techniques and
yielded a wide range of applications. In this paper, we apply
differential evolution (DE) [8] to estimate the model
parameters, as DE is a highly efficient and robust EA.
In the experiment, NMM is tested using the financial data
from a bank in China. The results show that NMM performs
much better than BPNN. Furthermore, the parameters of
NMM are discussed in a business perspective, resulting in
several useful points for credit risk management.