Intelligent Integrated Systems of Automated Designing Laboratory ( IISAD )

Ready developments

Conceptions of maintenance of complex systems safety: "zero risk", "non-zero risk" and mixed approach.
Complex approach in problem of maintenance of safety with the decision on uniform basis of tasks of risk, diagnosing and debugging.
Support system of economic models for simulation of distribution of resources in problems of safety, diagnosing, debugging.
Support system of engineering models to obtain the new facts and making decisions in problems of safety, debugging, diagnosing, control.
Support system of graphic construction and analysis of structural models in problems of safety, debugging, diagnosing.
Alhorithmic and analytical methods and appropriate software for training of the risk models with using of statistical data.
Program for fast orthogonalization of logical function in conujnctional or disujnctional normal form.

SOFTWARE PACKAGE (SP)
"Logic and probabilistic evaluation and analysis of risk"

We have developed new approaches and methods in modeling and analysis of risk in banking and financial sphere. Our methods are based on the logical probabilistic risk evaluation models and implemented in software package (SP) "LP-evaluation of risk". The distinguishing features of our methods and SP could be summarized as follows:

1. Novelty and advantages of SP "LP-evaluation of risk"

In SP the new mathematical LP-theory of risk with groups of incompatible events is used.
LP-model of credit risk had shown twice more accuracy and seven times more robustness in classification of credits on good and bad than others known classification techniques. Decreasing of the risk allows reduce the losses of the bank and percent for credit, attract more clients and increase bank's competitiveness.
Our SP can be used for risk evaluation and analysis of several new important tasks.

2.Comparison of different methods accuracy in credits classification

Used method

Bad credits classification
error, E_b

Good credits classification
error, E_g

The average error, E_m

Linear discriminant analysis 1 (LDA)

0,26

0,279

0,273

Linear discriminant analysis 2 (LDA)

0,287

0,291

0,29

Quadratic discriminant analysis 1 (QDA)

0,183

0,283

0,253

Quadratic discriminant analysis 2 (QDA)

0,283

0,34

0,323

Cluster analysis (CART)

0,273

0,289

0,285

Neural network 1 (NN)

0,38

0,24

0,282

Neural network 2 (NN)

0,24

0,312

0.29

LP-model with GIE Var.1

0,167

0,201

0,191

LP-model without GIE Var.2

0,1433

0,190

0,176

LP-model with GIE after structural
identification - Var. 3

0,126

0,174

0,155

3. Functions of SP

Determination of credit's risk:
   1) evaluation of credit's risk;
   2) classification of credits on good and bad;
   3) estimation of the risk's price;
   4) analysis of credit's risk;
Analysis of bank credit activity:
   5) determination of contributions of signs (characteristics) and gradations of credits in the bank's average risk of crediting;
   6) determination of contributions of signs (characteristics) and gradations of credit signs in the accuracy of credits        classification;
   7) optimization of the number of signs and gradations and their distributions on intervals (the sum of credit, duration of        credit, age of client) for increasing the accuracy of credits classification;
Training (identification) and quality evaluation of the risk LP-model:
   8) preliminary statistical analysis of the risk model;
   9) preliminary combinatorial analysis of the risk model;
   10) evaluation of probabilities of events-gradations and admissible risk for LP-model;
   11) evaluation of the risk LP-model accuracy;
   12) evaluation the risk LP-model robustness;
   13) the choose of ratio of number of wrong classification of bad and good credits.

4. Users of SP "LP-evaluation of risk":

Banks of all countries in the world; with the help of SP every bank makes the training of its risk LP-models using own statistical data.

5. Developers:

We have large experience in developing and practical application of algorithms. We are the united team of qualified specialists in the area of risk management with the western experience and has possibility to make applications on Windows and Java platforms. With the low labor cost and highly productive and qualified programmers we can keep the low prices in comparison with competitors. Within several months we could develop the commercial version our SP "LP-evaluation of risk". We offer also the maintenance and further development of SP "LP-evaluation of risk" according to the client's needs and demands, the issue of ad-hoc version, reports, etc.

6. Realization:

SP "LP-evaluation of risk" consists of modules, developed on C ++. Software core of SP "LP-evaluation of risk" consists of about 3000 operators. At present time there are modules of various functionality in DELPHI-4, C++ and JAVA. Fragments of possible realization SP can be found in demo-version.

7. Our Suggestion

We suggest to make joint project and develop software tools, based on our LP-models and know-how for risk evaluation in banking and financial sphere. To estimate the possible cost of that job we need addition information from your side:

Structure of the information for training of the risk LP-model (txt-file, Excel-sheet, database, etc);
The level and means of integration with the existing software;
Preferred platform. Currently our applications are created using C++, VB, Delphi and Java under Windows platform.

8. Prospect of cooperation

To adapt the SP "LP-evaluation of risk" for evaluation and analysis risk:

VaR theory of portfolio risk;
credit risk of companies;
bank's bankruptcy;
bribes;
frauds;
life insurance;
technical analysis, etc.

The list of the main publications about offered project:

1. Solojentsev E.D., Karassev V.V., Solojentsev V.E. Logic and probabilistic risk models in banks, business and quality / Edited by E.Solojentsev. SPb.: Nauka, 1999.- 120 p. (in Russian).

2. Solojentsev E.D., Karassev V.V. (2001) Risk logic and probabilistic models in business and identification of risk models. -Informatica 25 (2001)49-55 (in English).

3. Modelling and Analysis of Safety, Risk and Quality in Complex Systems / Proceeding of the International Scientific School MA SRQ-2001, St. Petersburg, Russia, June 18-22, 2001 / by Editors I.Ryabinin and E.Solojentsev. - SPb.: Omega, 370 p. (in English)

4. Karassev V.V., Solozhentsev E.D. Identification of the Logical-and-Probabilistic Risk Models of Structurally Complex Systems with Groups of Inconsistent Events. - Automation and Remote Control, No 3, 2002, pp. 433-448.(in English)

5. Modelling and Analysis of Safety and Risk in Complex Systems // Proceeding of the International Scientific School MA SR-2002, St. Petersburg, Russia, July 2-5, 2002 / by Editors B.Ivchenko, I.Ryabinin and E.Solojentsev. - SPb.: Business-Pressa, 540p. (in English)

6. Solojentsev E.D., Karassev V.V. Logical and probabilistic risk models in business with groups of incompatible events.- Economics and mathematical methods, No 1, 2003, p.p.90-105 (in Russian).

7. E.D.Solojentsev. Scenario logic and probabilistic management of risk in business and engineering. - SPRINGER, 2005. Pages - 391 p., Figures - 70; Tables - 40; Refers - 118. (in English).

IISAD Main Page

Phone: 7(812) 321 4766
Fax: 7(812) 321 4771
E_mail: risk@sapr.ipme.ru

Used method	Bad credits classification error, E_b	Good credits classification error, E_g	The average error, E_m
Linear discriminant analysis 1 (LDA)	0,26	0,279	0,273
Linear discriminant analysis 2 (LDA)	0,287	0,291	0,29
Quadratic discriminant analysis 1 (QDA)	0,183	0,283	0,253
Quadratic discriminant analysis 2 (QDA)	0,283	0,34	0,323
Cluster analysis (CART)	0,273	0,289	0,285
Neural network 1 (NN)	0,38	0,24	0,282
Neural network 2 (NN)	0,24	0,312	0.29
LP-model with GIE Var.1	0,167	0,201	0,191
LP-model without GIE Var.2	0,1433	0,190	0,176
LP-model with GIE after structural identification - Var. 3	0,126	0,174	0,155