基于模糊积分与支持向量机的房地产企业资本风险评估模型.pdf

《基于模糊积分与支持向量机的房地产企业资本风险评估模型.pdf》由会员分享，可在线阅读，更多相关《基于模糊积分与支持向量机的房地产企业资本风险评估模型.pdf（7页珍藏版）》请在淘文阁 - 分享文档赚钱的网站上搜索。

1、http:/-1-Study on Capital Risk Assessment Model of Real Estate Enterprises Based on Support Vector Machines and Fuzzy Integral Wu Chong，Wang Dong School of Management，Harbin Institute of Technology，Harbin（150001）E-mail： Abstract As the main intermediary of financial transaction,the Real Estate enter

2、prises are the weatherglass of states economic condition.Real Estate enterprises also play an important role in reducing economical risk and the unstable factor,guarantying the national economy healthily.The Real Estate enterprises itself undertakes various types risk in the operation,including cred

3、it risk,fluid risk,capital risk and policy risk and so on.Capital risk holds the special important status in each kind of risk,which is the most primary factor that causes Real Estate enterprises bankrupt.Traditional capital risk evaluating methods always only estimate the risk factors which exist i

4、n corporate itself,while seldom consider the influence which the enterprises deposit,loan structure and capital risk condition bring upon.These methods lead to the absence of the assessment subject and results in inevitable systematic error.In this essay,support vector machines(SVM)and fuzzy integra

5、l is introduced.The result of basing on voting ensemble SVM,single SVM,fuzzy nerve network and basing on fuzzy integral ensemble SVM are compared in the essay.support vector machines and fuzzy integral is better than three of others.We can conclude from figures and tables that the result of basing o

6、n fuzzy integral support vector machines is satisfaction Keyword：Real Estate enterprises，Capital Risk，SVM，Fuzzy integral，Fuzzy nerve network 1.Background 1.1 Loan growth trend Real estate loans grew very fast and the proportion is very high1-4.The potential risk implied in the real estate industry i

7、s accumulating.In recent years,the growth rate of real estate loans is too high,and far higher than the growth of RMB loans from all financial institutions5.We can see from fig.1 clearly,in the peak period,the loan growth rate of real estate developer can reach more or less 55%,although the governme

8、nt has promulgated the 195,121 and relevant documents to regulate real estate development loans and the loan growth rate of real estate developer has slightly decreased,the loan growth rate of real estate loans in the first quarter of 2005 compared to the first quarter of 2004 increased by 25.7%,inc

9、reased 2.9%compared the end of last year,and higher than the growth rate of RMB loans,which is 14.9%over the same period6-8.Fig.1 Loan growth rate 1.2 Non-performing loans situation Some real estate loans have become non-performing loans since 2004.Generally speaking,the real estate price over our c

10、ountry still continues to rise,and still have not produce major negative This work is supported sponsored by sponsored by the Heilongjiang Social Science Foundation of China,Grant No.05B0060.http:/-2-impact on the developer and the solvency of the man who buy houses9.But,real estate loans in the ban

11、king sector have started to produce bad loans.Take the four major state-owned banks as an example;we can clearly see from Tab.1 that total real estate loans reached 1.9042 trillion Yuan in 2004,while the number reached 201.7714 trillion Yuan in the first quarter of 2005.Under such circumstances,the

12、rate of non-performing real estate loans in 2004 has reached 41.6%,which implies that 875.1932 trillion in real estate loans are non-performing loans in 2004.Whats more,the number reached 41.5%in the first quarter of 2005.In other word,the non-performing loans are 907.1983 trillion in the first quar

13、ter of 200510-14.Tab.1 Real estate loans non-performing rate 2.Real estate capital risk evaluation indictors system 2.1 The principle of selecting First,the selection of indicators should not only conform to Chinas national conditions but also must be feasible.Second,because developed countries have

14、 accumulated a great deal of experience in Risk management,we must study abroad experience,the maturity of the real estate industry and international standards.At last,because the purpose of indicator system and risk assessment is to find the message of Real estate,the indicators should have Predict

15、ability,which should be able to reflect the future development trend of Real Estate.2.2 Risk Assessment Model of capital The traditional capital risk assessment indicator system contains only a few financial indicators,and cash flow analysis has not been introduced for evaluation In fact.So,the resu

16、lts of evaluation were not convincing.The paper,which based on a careful analysis to the risk of our real estate capital,takes into account the risk factors of the loan,the risk factors of the banks and the New Basel Accord.In addition,because he particularity of the real estate capital risk and the

17、 availability of real estate capital data in China,the paper put the cash flow to analyze real estate Indicator system and establish a capital risk evaluation Indicator system that formed by the twenty-seven indicators,including:liquidity rate,quick-moving rate,Super quick-moving rate,Working capita

18、l/total assets,Debt-to-asset rate,Net rate of return,Assets yield rate,Sales net interest,Sales income/total assets,Cost profit rate,Inventory turnover rate,Accounts receivable turnover rate,Total turnover rate,liquidity turnover rate,Fixed turnover rate,Cash debt rate,Cash flow debt rate,Sales cash

19、 rate,Total cash debt rate,Net cash flow 2004 2005 Bank name Total loans(billion)Real estate loans(billion)Total loans non-performing rate(%)Real estate loans non-performing Total loans(billion)Real estate loans(billion)Total loans non-performing rate(%)Real estate loans non-performing ICBC 5810.3 1

20、685 3 7.4 5998.0 1764 3 7.1 Agricultural Bank of China 4099.1 1723.4 8.1 16.6 4258.5 1826 7.9 16.2 The peoples Bank of china 3783.7 1017.7 4.8 12.8 3983.6 1022.9 4.4 12.3 CCB 5708.9 2278.0 3.7 7.3 5937.4 2402 3.5 6.9 Total 19402.0 6704.1 4.6 10.5 20177.4 7014.9 4.5 10.1 http:/-3-per share,Cash recov

21、ery of all assets,Cash investment rate,cash dividend security multiples,The total weighted risk assets/total assets,Weighted average loan duration/Total weighted average loan period,Total profit/earnings,Total profit/total assets,Overdue loans/total loans,Overdue loans/month growth.We can inspect th

22、e financial situation of the real estate from different perspective.3.Support Vector Machines(SVM)choice and expression 3.1 Integrate theory Integrated SVM means that finite sub-Support Vector Machines were integrated by some way so as to class new samples.Bagging technology should be used to integr

23、ate SVM in the paper,which based on Fuzzy Integral Method.Assumptions,the sub-Support Vector Machines are,(1,2,)eiki=L15-18,K is the number of these sub-Support Vector Machines.,1,2,C CCM=L is the set of gathered category indicators.Mis the number of these category indicators.To the Fuzzy Integrate,

24、(1,2,)iumMm=L represents the output of each sub-Support Vector Machine.So,ium is the sub-Support Vector Machine and ei is evaluation model for each category.Assumption:1,2,M L is a finite set,:0,1h is a function and()()()12hhhML According to the fuzzy integral formula:()max min(),()1nh x gh xgAXiii=

25、g Fuzzy Integral calculated as follows:()max min(),)1KFIhgAiii=In the above formula,1,2,Aii=L and gcan be calculated as follows:()()11igAgxg=()()(),111iigAggAg gAiniii=+Meanwhile,can be calculated as follows:()111Kigi+=+=,()1,0 Through the above formulas,it is possible to calculate an input model fo

26、r a specific category of fuzzy integral.According to the same method,other types of fuzzy integral can be calculated.If we use()()1,2,FIxmMm=L denote various types of fuzzy integral,then SVM system model are as follows:()()classarg max1MxFIxmm=Algorithm:Step 1for x input model,each sub-SVM must be j

27、udged,then the model should output each category of membership relative to x.Step 2for each Cmcategory,()hmkand()ghk should be calculated by each sub-SVM,Fuzzy Integral-FIm relative to Cmshould be calculated too.Step 3make a judgment on the decision-making model.http:/-4-3.2 Using sub-SVM to express

28、 the bank risk and real estate risk Assumption:()()(),11lTxyxyXYii=L is training set,1,2,1,nxXRyYMilii=LL sub-SVM discrimination algorithm can be written as follow:sgn()yx=+g,:weight vector;:deviation Then,sub-SVM should solve the optimization problem below:()1min,2,()1,()1,0,1,lTiiiCjiiij lbjTiiixi

29、f yijjjTiiixif yijjjijlj+=+=+=L i:relaxation variables;C:constant,0C;After solving the optimization problems,we can get K decision functions()(),Tllx+()(),Tkkx+The quantitative indicators of Real Estate Risk and bank were integrated by sub-SVM for the real estate business credit rating.We put sample

30、xinto the k Decision-making functions,and can obtain k numerical results.The class that corresponds to the maximum of the function is the one that contains this sample.Finally,the model take into account both qualitative and quantitative indicators to evaluate the real estate enterprise.4.Integrated

31、 SVM model based on fuzzy integral 4.1 Input and output factors based on integrated SVM This model has eight input vectorsProfitability factor,Solvency factor,Operating capacity factor,Cash flow factor,Capital adequacy rate factor,Capital quality factor,Loan time factor,Made cash factor,and one outp

32、ut vectorOverdue loans/total loans.The indictors which are contained in each factor can be shown in the tab.2 Tab.2 The relation of factors and indicators Factor name Indicators name Operating capacity factor Working capital/total assets、Debt-to-asset rate、Inventory turnover rate、Accounts receivable

33、 turnover rate、Total turnover rate、liquidity turnover rate、Fixed turnover Profitability factor Net rate of return、Assets yield rate、Sales net interest、Sales income/total assets、Cost profit rate http:/-5-Solvency factor liquidity rate、quick-moving rate、Super quick-moving rate Cash flow factor Cash de

34、bt rate、Cash flow debt、Total cash debt rate、Sales cash rate、Cash recovery of all assets、Cash Dividend security multiples、Cash investment rate、Net cash flow per share Capital adequacy rate factor The total weighted risk assets/total assets Capital quality factor Weighted average loan duration/Total w

35、eighted average loan period Loan time factor Total profit/earnings、Total profit/total assets Made cash factor Overdue loans/total loans、Overdue loans/month growth 4.2 Risk assessment model Types of classified loans were determined reversely by overdue loans/total loans.Therefore,it is possible to pr

36、edict accurately the classification results,and also have better guidance to real estate loans.The whole risk assessment model can be shown as fig.2 Fig.2 Flowchart http:/-6-4.3 Evaluation criteria In this paper,the output vector of real estate capital risk assessment model is overdue loans/total lo

37、ans calculated by SVM.Obviously,the greater overdue loans/total loans are,the higher capital risk is.The output result can be classified based on the existing five-category evaluation criteria developed by Central Bank.The five-category evaluation criteria should be shown as tab.3.Finally,we compare

38、 classification result of SVM with the actual result of classification so as to judge the validity of SVM.Tab.3 five-category evaluation criteria Output result Normal Concern Poor Suspicious Loss overdue loans/total loans 00.03 0.03-0.10 0.10-0.35 0.35-0.80 0.80-1.00 5.Conclusion Through some exampl

39、es,the paper has compared the analytical result among SVM Based on fuzzy integral,SVM based on the voting law SVM and Fuzzy Neural Network.The following table shows the results of comparison.Tab.4 Classification results of the various classifications SVM Based on fuzzy integral SVM based on the voti

40、ng law SVM Fuzzy Neural NetworkActual results Normal 32 37 33 40 37 Concern 47 40 48 40 45 Poor 32 33 30 31 29 Suspicious 29 25 22 27 29 Loss 15 20 22 17 15 Accuracy 87.09%85.16%84.52%82.58%100%We can clearly see from the above table,integrated Support Vector Machine classification results are more

41、effective than other classifications,which implies that it is feasible to use the integrated Support Vector Machine to evaluate the risk of Real Estate enterprises.References 1 Cheung SO,Lam TI,Wan YW,Lam KC,Improving objectivity in procurement selection,IEEE trans,Eng,vol.17,no.3,132139,mar.2001 2C

42、hang C C,Hus C W,Lin C J,The analysis of decomposition methods for support vector machines,IJCAI trans,Chinese,no.9,155-156,Sept.2001.3John J.Mingo,Policy Implications of the Federal Reserve Study of Credit Risk Models,IEEE trans,Eng,vol.24,15-33,2002.4Osuna E,Freund R,Girosi F,Training support vect

43、or machines,ISTP trans,Eng,no.5,25-27,May.1998.http:/-7-5Duc Thanh Luu1,S.Thomas Ng,Swee Eng Chen,Formulating Procurement Selection Criteria through Case-Based Reasoning Approach,IJCAI trans,Chinese,vol.19,no.3,155-156,269-276,Mar.2005.6Ng,S.T.,Luu,D.T.,Chen,S.E.,Lam,K.C,Fuzzy membership functions o

44、f procurement selection criteria,IEEE trans,chinese,vol.20,no.3,trans,132139,mar.2002.7Rowlinson,S,A definition of procurement systems.Procurement systems,IEEE trans,Eng,no.6,trans,276299,June.1999(a).8Kim H C,Bang S Y,Constructing Support Vector Machine Ensemble,NJ:ISTP press,2001,ch7.257-267 9Anna

45、 K J,James D M,Marek F Ronald,Computer-aided polyp Detection in CT Colonography Using an Ensemble of Support Vector Machine,IJCAI trans,Eng,vol.1256,1019-1024,1998.10Valentini G,Muselli M,Cancer Recognition with Bagged Ensemble of Support Machines,IEEE trans,Eng,vol.56 461-466,July.2001 11Platt J,Pr

46、obalistic Outputs for Support Vector Machines and Comparison to Regularized Like hood Method,In Advance in Large Margin Classifiers,NJ:MIT Press,2001,ch.6,155-164.12Wu T F,Lin C J,Wang R C,Probability Estimate for Multi-class Classification by Pairwise Coupling,IEEE trans,Chinese,vol.5,975-1005,Sept

47、.2001(5).14Chkir,I E,Cosset J,Diversification strategy and capital structure of multinational corporations,IEEE trans,Eng,vol.1,no.1,17-37,Jan.2001.15Altman E I,Caoutte J B,Narayanan P,Credit-risk Measurement Management,IEEE trans,Eng,vol.9,no.11,711,Nov.2005 16Fishelson-Holstine H,Case studies in c

48、redit risk model development,IEEE trans,Eng,no.4,69-180,Jan.2002.17Altinkemer,Chaturvedi,Kondareddy,Business progress reengineering and organization performance,IEEE trans,Eng,Vol.18,no.6,381-382,Dec,1998 18Segars A,Grover V,Kettinger W,Strategic users of information technology,IEEE trans,Eng,Vol.3,no.4,157-169,May.2005 作者简介：吴冲（1971-），男，黑龙江人，哈尔滨工业大学博士生导师，博士后，研究方向：金融系统工程、企业系统工程、人工智能在管理中的应用；王栋（1982-），男，山西人，哈尔滨工业大学硕士研究生，研究方向：金融系统工程。