An Evaluation Method Based on Cloud Model for the Credit of High-tech SMEs
Guo-qiang Zhou1, Xue-qing Wang1, Rui Liu2, Li-guo Sun3
2
1
College of Management and Economics, Tianjin University, Tianjin, P. R. China
Nuclear and Radiation Safety Center, Ministry of Environmental Protection, Beijing, P. R. China
3
China Bohai Bank, Tianjin, P. R. China
(steven661@tju.edu.cn)
Abstract – In the process of resolving financing
difficulties of high-tech small and medium enterprises
(SMEs) in China, the evaluation of credit risk of high-tech
SMEs becomes a very challenging problem for the bank.
This paper proposes a novel evaluation method based on
cloud model to measure the credit risk of Chinese listed
high-tech SMEs. Finally, an example is provided for
illustrative purpose, and the indexes system of credit
evaluation is established of 25 key factors, embedded within
five broad categories- credit quality, organizational level,
operation level, R&D level and network position. This
research shows that it is a better way to use this method to
realize transforming qualitative terms described in a natural
language to distribution patterns of quantitative values,
especially for high-tech SMEs.
Keywords – Cloud model, credit evaluation, credit risk,
credit scoring, high-tech SMEs
I. INTRODUCTION
In recent years, the credit of high-tech small and
medium enterprises (SMEs) has been gaining much more
importance according to their high growth in financial
world [1]. However, the credit guarantee risk is very high
in SMEs due to their particular characteristics, which lead
to a low credit scoring in general [2], [3]. The evaluation
of credit risk of high-tech SMEs becomes a very
challenging problem for the bank. Therefore, it is
essential to develop an accurate credit scoring model for
high-tech SMEs for the efficient management of bank.
Most well-known evaluation models use probability
or fuzzy set theory to hold randomness or fuzziness
respectively, such as the decision trees [4], artificial
neural networks [5], genetic algorithm [6], etc. Among all
of these methods, only cloud model based models
consider both aspects of uncertainty. Cloud model is the
innovation and development of membership function in
fuzzy theory [7], which transforms qualitative terms
described in a natural language to distribution patterns of
quantitative values [8]. It is successfully used in spatial
analysis [9], target recognition [10], intelligent algorithm
improvement [11] and so on.
Therefore, in this paper, we propose an evaluation
method based on cloud model for the credit of high-tech
SMEs. As credit evaluation is a typical multi-attribute
evaluation problem, it is more significant in applying this
novel approach to credit evaluation so as to demonstrate
its usefulness.
II. METHODOLOGY
A. Basic concepts of cloud model
Suppose that r is the language value of domain U ,
and mapping Cr ( x) : U  [0,1] , x  X ( X  U ) ,
x  Cr ( x) , then the distribution of Cr ( x) in U is
called the membership cloud of r , or cloud in short, and
each projection is called a cloud drop in the distribution.
If the distribution of Cr ( x) is normal, it is named normal
cloud model.
B. Numerical characteristics of the cloud model
The normal cloud model is determined by the
following three parameters: expectation Ex , entropy En ,
and hyper entropy He .
Expectation Ex represents the values that mostly
stand for this qualitative concept, generally it is the value
x that corresponds to the gravity center of the cloud, it
should belongs to this qualitative concept hundred percent.
Entropy En is the measurement of the qualitative
concept fuzzy degree, it determines the range of the cloud
and about 99.74% cloud drops fall within the range
between [ Ex  3 En, Ex  3 En ] , and it reflects the numerical
value range acceptable by concept and represents the
margin with double-sided property.
Hyper entropy He is the entropy of entropy, it
reflects the dispersion degree of the entropy of concept
[12].
A cloud model can be denoted with vector
C ( Ex, En, He) . The numerical characteristics of the cloud
Model are shown in Fig. 1.
Fig. 1. Numerical characteristics of the cloud model
C. Normal cloud model
Let U be a quantitative universal set and r be the
x  U , which is a
random realization of the concept r , and x satisfies
x N ( Ex, En ' 2 ) , where En ' N ( En, He 2 ) , and the
certainty degree of x on r is
qualitative concept related to U. If
 e

( x  Ex )2
2( En' )2
(1)
Then the distribution of x on U is a normal cloud
[13], and every x is defined as a cloud drop. Given the
three parameters Ex , En , He , the normal cloud model
can be generated [14].
Input: Ex , En , He , and the number of the cloud
drops n .
Output: n of cloud drops x and their degree  .
Step 1. Generate a normally distributed random
2
'
number Eni with expectation En and variance He , i.e.,
'
Eni  NORM ( En, He ) .
2
Step 2. Generate a normally distributed random
number
xi with expectation Ex and variance Eni' 2 , i.e.,
xi  NORM ( Ex, Eni' 2 ) .
Step 3. Calculate
i  e
Step 4.

( xi  Ex )2
En 
(2)
xi with certainty degree of i becomes one
n cloud drops are
D. Cloud model-based credit evaluation algorithm
In the evaluation method based on cloud model,
gravity center of cloud can be denoted as:
T  ab
(3)
In the type, a means the position of gravity center of
cloud, depicting with the expectation value Ex , then if
the expectation value Ex changes, the position of gravity
center of cloud also corresponds of change; b means the
height of gravity center of cloud, depicting with the heavy
value of power, which takes often value (0.371).
Therefore, the variety that passes gravity center of
cloud can reflect the variety of system information status,
the concrete step of the evaluation method based on cloud
model is as follows:
Step1: Confirming index system and index power
weight.
Step2: Denoting the cloud model of each index.
Denotation of accurate-number type and languagedescription type are different in cloud model. Withdraw a
set of sample n to constitute to make policy matrix, so
, Exn )  min( Ex1 , Ex2 ,
max( Ex1 , Ex2 ,
, Exn )
(5)
6
And the index of language-description type can be
denoted as follow:
Ex En  Ex2 En2   Exn Enn
Ex  1 1
(6)
En1  En2   Enn
En  En1  En2 
 Enn
(7)
Step3: Denoting status of the system. n indexes can
be depicted with n cloud models, therefore the
evaluation system containing n indexes can be denoted
with
dimension
comprehensive
cloud,
n
T  (T1 , T2 ,
, Tn ) , Ti  ai  bi (i  1, 2,
n) , and when
the status of evaluation system occurrences variety, the
gravity center changes to T  (T1 , T2 ,
'
'
'
'
, Tn ) .
Step4: Measuring the variety of cloud gravity center
based on power-added deviation degree. Suppose that
each index of the ideal status of a system is given, then
the vector of cloud gravity center can be depicted as
T  a  b  (T1 , T2 ,
, a  ( Ex10 , Ex2 0 , , Exn 0 ) ,
b  (b1 , b2 , , bn ) , bi  wi  0.371 , and the normal status
of vector of n dimension comprehensive cloud gravity
0
2 Eni'2
cloud drop in the domain.
Step 5. Repeat Steps 1 to 4 until
generated.
the index of accurate-number type can be denoted as
follow:
Ex  Ex2   Exn
Ex  1
(4)
n
T
0
0
0
Tn )
center is denoted as T  (T1 , T2 ,
, Tn ) .
Generally power-added deviation degree can be used
to measure the variety of cloud gravity center between
ideal status and normal status. The vector of
n
dimension comprehensive cloud gravity under normal
status is normalized to T  (T1 , T2 ,
G
G
G
G
, Tn )
, among
them,
(Ti  Ti 0 ) / Ti 0 , Ti  Ti 0
Ti  
(i  1, 2, , n) (8)
0
0
(Ti  Ti ) / Ti , Ti  Ti
G
Therefore,
power-added
deviation
degree
 (0    1) is denoted as:
n
   ( wT
)
i i
G
(9)
i 1
Under the ideal status,   0 .
Step5: Confirming the evaluation set based on cloud
model. Generally, the more numbers of evaluation scales,
the more accurate of the evaluation results. According to
the feature of high-tech SMEs, five-scale evaluation set is
adopted (see Fig. 2):
V  (v1 , v2 , v3 , v4 , v5 )  (bad (C ), general ( B), good ( A), verygood ( AA), best ( AAA))
Select 10 high-tech SMEs in the second-board Market
in China as test sample set S  Si i  1, 2,
,10 ,
including electronics information, medical apparatus,
biological pharmacy, etc. Finance data comes from
database in CSMAR Solution [15], and other qualitative
indexes are descripted by the expert evaluation languish.
Take company S1 (NO. 300002) in the test sample set as
example, the status of each credit evaluation index are
shown in Table Ⅰ, Table Ⅱ, Table Ⅲ, Table Ⅳ and
Table Ⅴ.
TABLE Ⅰ
Fig. 2. Five-scale evaluation set based on cloud model
Status of credit quality U 1
Status
U 11
1
37920
33470815
general
good
2
37920
33470815
general
good
3
37920
33470815
good
good
4
37920
252764848
good
very good
Ideal
100000
300000000
best
best
III. Application example
A. Confirming index system and sample data
The index system of credit evaluation of high-tech
SMEs includes total 25 indexes, 5 major type as follows:
U12
U13
U14
credit quality U1 -register capital U 11 , history credit
condition U12 , equipment level U13 and guarantee U14 ;
Organizational
level
U 2 -business
strategy
U 21 ,
TABLE Ⅱ
organization system U 22 , stability of management team
U 23 , stability of R&D team U 24 , business proposal U 25 ;
Operation level U 3 - turnover ratio of accounts receivable
U 31 , turnover ratio of total assets U 32 , return on total
assets ratio U 33 , operating profit ratio U 34 , income
growth ratio U 35 , profit growth ratio U 36 , liquidity ratio
U 37 , debt asset ratio U 38 , after-sales service U 39 ; R&D
level U 4 - R&D input U 41 , intellectual property rights
U 42 , R&D character U 43 ; Network position U 5 -market
share U 51 , public relations U 52 , industry trend U 53 ,
geography position U 54 .
Status of organizational level U 2
Status
U 21
U 22
U 23
U 24
U 25
1
1.701
good
very good
good
good
2
1.789
good
good
good
good
3
1.660
general
good
good
good
4
1.912
good
very good
general
very good
Ideal
2.000
best
best
best
best
TABLE Ⅲ
Status of operation level U 3
Status
U 31
U 32
U 33
U 34
U 35
U 36
U 37
U 38
U 39
1
2.450
0.307
0.121
0.420
0.545
0.266
12.400
0.069
good
2
0.436
0.073
0.028
0.378
-0.267
-0.245
11.178
0.068
general
3
0.776
0.188
0.075
0.407
0.613
0.816
11.305
0.067
good
4
1.563
0.255
0.091
0.358
-0.277
-0.577
6.884
0.095
good
Ideal
1.000
0.100
0.030
0.500
0.200
0.100
10.000
0.050
best
TABLE Ⅳ
Status of R&D level U 4
U 41
U 42
U 43
1
19435331
66500
general
2
29018320
76110
bad
3
38857798
1076920
bad
Status
4
12724180
49268370
bad
Ideal
50000000
50000000
best
A2
 2.450
 0.436
 
 0.776
 1.563
A
3
TABLE Ⅴ
Status of network position U 5
U 51
U 52
U 53
U 54
1
general
very good
very good
best
2
general
good
best
best
3
bad
very good
very good
best
4
general
good
very good
best
best
best
best
best
Status
Ideal
1.701
1.789
 
1.660
1.912
A4
0.50
0.75
0.50
0.50
0.50
0.50
0.25
0.50
0.50
0.50
0.75
0.25

0.50 

0.50 

0.75 
0.50



0.50 

0.50 
0.307
0.121
0.420
0.545
0.266
12.400
0.069
0.50
0.073
0.028
0.378
0.267
0.245
11.178
0.068
0.25
0.188
0.075
0.407
0.613
0.816
11.305
0.067
0.255
0.091
0.358
0.277
0.577
6.884
0.095





 0.25
 0.25
 
 0
 0.25
 19435331
 29018320
 
 38857798
12724180
66500
0.25
76110
0
1076920
0
49268370
0
A5
0.75
0.75
0.50
1
0.75
0.75
0.50
0.75



1

1
1
1
Expectation Ex and entropy En of each index cloud
model are calculated by the above policy matrixes (see
Table Ⅵ, Table Ⅶ, Table Ⅷ, Table Ⅸ and Table Ⅹ).
TABLE Ⅵ
Expectation Ex and entropy En of credit quality U 1
B. Denoting the cloud model of each index
Parameter
U 11
Ex
37920
En
0
Normalize the evaluation languish set (bad, general,
good, very good, best) to (0, 0.25, 0.50, 0.75, 1), and thus
the policy matrix A1 - A5 is constituted as follow:

 37920
 
 37920
 37920
37920
A1
0.25
0.50
33470815
0.25
0.50
33470815
0.50
252764848
0.50
U13
U14
88294323.3
0.375
0.563
36549005.5
0.042
0.042
TABLE Ⅶ



0.50 

0.75 
33470815
U12
Expectation Ex and entropy En of organizational level U 2
Parameter
U 21
U 22
U 23
U 24
U 25
Ex
1.766
0.438
0.625
0.438
0.563
En
0.042
0.042
0.042
0.042
0.042
TABLE Ⅷ
Expectation Ex and entropy En of operation level U 3
Parameter
U 31
U 32
U 33
U 34
U 35
U 36
U 37
U 38
U 39
Ex
1.306
0.206
0.079
0.391
0.154
0.065
10.442
0.075
0.438
En
0.336
0.039
0.016
0.010
0.148
0.232
0.919
0.005
0.042
TABLE Ⅸ
TABLE Ⅹ
Expectation Ex and entropy En of R&D level U 4
Expectation Ex and entropy En network position U 5
U 41
U 42
U 43
Parameter
U 51
U 52
U 53
U 54
Ex
25008907.25
12621975
0.063
Ex
0.188
0.625
0.813
1
En
4355603
8200311.667
0.042
En
0.042
0.042
0.042
0
Parameter
C. Confirming the power weight of indexes
ACKNOWLEDGMENT
The power weights of indexes are shown in Table Ⅺ.
TABLE Ⅺ
The power weights of indexes
wi
wi 1
wi 2
wi 3
wi 4
wi 5
wi 6
wi 7
wi 8
wi 9
0.36
0.58
0.14
0.07
0.21
-
-
-
-
-
0.16
0.11
0.19
0.19
0.19
0.33
-
-
-
-
0.23
0.17
0.17
0.07
0.07
0.09
0.09
0.12
0.12
0.11
0.13
0.25
0.59
0.17
-
-
-
-
-
-
0.12
0.56
0.11
0.26
0.07
-
-
-
-
-
This work was partially supported by the National
Natural Science Foundation of China (Grants No.
71172148) and Soft Science Research Projects of the
Ministry of Housing and Urban-Rural Construction
(Grants No. 2011-R3-18). The authors are also grateful to
the referees for their helpful comments and valuable
suggestions for improving the earlier version of the paper.
REFERENCES
D. Result and analysis
Through cloud model computation, the credit
evaluation value of high-tech SMEs:
5
PS   ( wi PU )  0.487
1
i 1
i
Then the credit evaluation value is input into the fivescale evaluation set based on cloud model (see Fig. 3). It
will activate two cloud objects: A and B, but the
activation degree of A is far larger than B, so the credit
evaluation of company S1 (NO. 300002) obtains A.
Fig. 3. Result of credit evaluation of company S 1
IV. Conclusion
In this paper, credit evaluation applications in hightech SMEs are discussed, and an evaluation method based
on cloud model is formulated. The application of this
model is also illustrated. After this research, we find that
it is a better way to use this method to realize
transforming qualitative terms described in a natural
language to distribution patterns of quantitative values,
especially for high-tech SMEs.
[1] Gülnur Derelioglu, Fikret Gürgen, “Knowledge discovery
using neural approach for SME’s credit risk analysis
problem in Turkey,” Expert Systems with Applications, vol.
38, no. 8,pp. 9313-9318, 2011.
[2] Xiaohong Chen, Xiaoding Wang, Desheng Dash Wu,
“Credit risk measurement and early warning of SMEs: An
empirical study of listed SMEs in China,” Decision
Support Systems, vol. 49, pp. 301-310, 2010.
[3] X. Chen, W. Han, J. She, “The credit risk, company quality
and growth illusion,” Economic Theory and Business
Management, no. 12, pp. 62-67, 2006.
[4] H. Frydman, E.I. Altman and D.L. Kao, “Introducing
recursive partitioning for financial classification: The case
of financial distress,” The Journal of Finance, vol. 40, no.
1, pp. 269-291, 1985.
[5] H.L. Jensen, “Using neural networks for credit scoring,”
Managerial Finance, vol. 18, no. 6, pp. 15-26, 1992.
[6] R.F. Walker, E. Haasdijk and M.C. Gerrets, “Credit
evaluation, using a genetic algorithm,” Intelligent Systems
for Finance and Business, pp. 39-59, 1995.
[7] Di K, Deyi L, Deren L, “Cloud theory and its applications
in spatial data mining knowledge discovery,” Journal of
Image Graph, vol. 4A, no. 11, pp. 930-935, 1999.
[8] Deyi L, Haijun M., Xuemei S, “Membership clouds and
membership cloud generators,” Journal of Computer
Research and Development, vol. 32, no. 6, pp. 15-20, 1995.
[9] Haijun W., Yu D., “Spatial clustering method based on
cloud model,” Proceedings of the Fourth International
Conference on Fuzzy Systems and Knowledge Discovery,
no. 7, pp. 272-276, 2007.
[10] Fang W., Yanpeng L., Xiang L., “A new performance
evaluation method for automatic target recognition based
on forward cloud,” Proceedings of the Asia Simulation
Conference 2007, pp. 337-345, 2007.
[11] Yunfang Z., Chaohua D., Weirong C., “Adaptive
probabilities of crossover and mutation in genetic
algorithms based on cloud generators,” Journal of
Computational Information Systems, vol. 1, no. 4, pp. 671678, 2005.
[12] Pin Lv, Lin Yuan, Jinfang Zhang, “Cloud theory-based
simulated annealing algorithm and application,”
Engineering Applications of Artificial Intelligence, no. 22,
pp. 742-749, 2009.
[13] D. Y. Li, C. Y. Liu, W. Y. Gan, “A new cognitive model:
cloud model,” International Journal of Intelligent Systems,
pp. 357-375, 2009.
[14] Deyi L, Yi D, “Artificial intelligence with uncertainty,”
Chapman & Hall, 2005.
[15] CSMAR
Solution,
“CSMAR
Database,”
http://
www.gtarsc.com, 2012.