- PDF Free Download

Size: px

Start display at page:

Download ""

つねときさどひら
5 years ago
Views:

1 Vol. 9 yasuhisa.toyosawa@mizuho-cb.co.jp 3 3

3 Altman968 Z Kaplan and Urwitz 979 Merton974 Support Vector Machine SVM

4 SVM i s i x b si t = b x i i r i R * R r (R,R, L,R ), R < R < L < R i 2 K 2 K s i a k Pr(r i = Rk ) = + exp( (s i + a )) + exp( (s k i + a k )) N(*) Pr( r i = Rk) = N(a k s i) N(a k s ) i i rˆ i = arg maxpr(r i = x) x Multi model 20Multi model 3

5 One model AIC SBCSchwarz One model EBITDA/ CF CF Debt/EBITDA EBITDA / / / 4

6 X: Log( ) Debt EBITDA EBITDA/ EBITDA = ( ) EBITDA EBITDA/ X 2 : X 3 : EBITDA EBITDA 2 X 4 : EBITDA X 5 : / X 6 : 3 5 NTT JR JT 2 Microsoft Excel Stdev 5

7 I * X7 : X8 : X9 : I I I {NTTJRJT} { } { } I * X0 : I { } AAAAA+BBB. 2. Accuracy RatioAR AR AR 6

8 Cumulative accuracy profilecap CAP X 30% Y 30% %, X 30% Y 00% 2 CAP X 30%Y 30% CAP X 30%Y 30% CAP Cumulative accuracy profiles %% AR A B 0 AR = B A + B Area Under CurveAUCAUC AR Engelmann and Tasche 2003 AR AUC 7

9 AR = 2AUC AR R k R k+ AR AR(R k,r k+ ) AR AAR K AAR = ( AR(R k= k,r k+ )) K AAR b AAR bˆ = arg max AAR( b) b 2009AR AUC Simulated annealing SA b f(r k,r k + c ) c i q i HR HR N = I N { f(rˆ, q ) } i i i = HR a HR aˆ = arg maxhr( a) a 3 Moody s Investors Service 8

10 SA a R&I 395 AAA AA+ AA AA- A+ A A- BBB+ BBB BBB- BB+ BB

11 AAA AAA BB+ BB 9 4,5,6 Log() / 2.0x 8 0.5x 4 0.0x AA+ AA A+ A A- BBB+ BBB BB* AA+ AA A+ A A- BBB+ BBB AA- BBB- AA- BBB- BB* 75 %th 25%th 2 BB*BB+ Pacific Data EBITDA/ EBITDA / 40% 0.30x 20% 0.5x 0% 0.00x AA+ AA A+ A A- BBB+ BBB BB* AA+ AA A+ A A- BBB+ BBB AA- BBB- AA- BBB- BB* 75 %th 25%th 2 BB*BB+ Pacific Data 0

12 / / 0.8x 2x 0x 0.4x 2x 0.0x 4x AA+ AA A+ A A- BBB+ BBB BB* AA+ AA A+ A A- BBB+ BBB AA- BBB- AA- BBB- BB* 75 %th 25%th 2 BB*BB+ Pacific Data AAR SA AAR 7 95% % 85% 0,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000 9,000 0,000 AAR % AR % AA AA- A+ A A- BBB+ BBB BBB AAR 93%

13 8 AR 87% X Log() X X2 / X7 NTT,JR,JT 4.8 X3 EBITDA/ X4 EBITDA/ 27.4 X5 / X8, 2 3. X X s = 7X + 4X 4X X + 2X X 9 27X 4 0X 3X X Multi model 88.9% 93.9% 00% 90% 93.9% 80% 70% 0,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000 9,000 0,000 2

14 90% % 4 > % > % AA+ AA AA- A+ A A- BBB+ BBB BBB- BB* > % BB*BB+ 2 AA+ BBB- 90% One model 4 20,000 5,000 5,000 76% 3

15 3,000,000,000,000,000 z XX4 0 = 7ΔX Δ 4 ΔX2 + 2 ΔX3 27 X4 X3EBITDA/ X4EBITDA / 0 = 7(Log(,000 +,000z) Log(,000)),000 +,000( z) 4( 3,000,000 ) 2, ΔX 27Δ 3 X Log( + z) + z = 2ΔX + 27Δ X 4 3 X4=+5% 5 5 EBITDA EBITDA / 4

16 X3= 5% 42%,000,420 % X4: EBITDA/ X EBITDA/ % 2%,000,20,000,790 3 One model 5

17 94% Altman, Edward L. 968, Financial Ratios, Discriminant Analysis and Prediction of Corporate Bankruptcy, The Journal of Finance, Vol.23, pp Engelmann, Bernd and D. Tasche 2003, Testing rating accuracy, CREDIT RISK, RISK JANUARY 2003, pp Gentle, James E. 2005, Random Number Generation and Monte Carlo Methods, Springer Hardle, Wolfgang, R. Moro and D. Schafer 2007, Estimating Probabilities of Default With Support Vector Machines, SFB 649 ECONOMIC RISK Kaplan, R. S. and G. Urwitz 979, Statistical model of bond ratings: A methodological inquiry, The Journal of Business, Vol.52, pp ,,, 20,,, Liu, Jon S. 2008, Monte Carlo Strategies in Scientific Computing, Springer Merton, Merton C. 974, On the Pricing of Corporate Debt: The Risk Structure of Interest Rates, The Journal of Business, Vol.29, pp

18 2009, AUC AR, FSA 5, pp ,, 50 2, 7

19 9 202 No Tel. (03)

大規模データベースを用いた信用リスク計測の問題点と対策(変数選択とデータ量の関係)

大規模データベースを用いた信用リスク計測の問題点と対策(変数選択とデータ量の関係) ( ) CRD 450,000 86 2, CRD,, CRD 1 1 4 2 5 2.1........................................ 5 2.2............................... 6 3 2 7 3.1 2.................................... 7 3.2.............................