Loading [MathJax]/jax/output/SVG/jax.js
Research article

Resolutions to flip-over credit risk and beyond-least squares estimates and maximum likelihood estimates with monotonic constraints

  • Given a risk outcome y over a rating system {Ri}ki=1 for a portfolio, we show in this paper that the maximum likelihood estimates with monotonic constraints, when y is binary (the Bernoulli likelihood) or takes values in the interval 0≤y≤1 (the quasi-Bernoulli likelihood), are each given by the average of the observed outcomes for some consecutive rating indexes. These estimates are in average equal to the sample average risk over the portfolio and coincide with the estimates by least squares with the same monotonic constraints. These results are the exact solution of the corresponding constrained optimization. A non-parametric algorithm for the exact solution is proposed. For the least squares estimates, this algorithm is compared with “pool adjacent violators” algorithm for isotonic regression. The proposed approaches provide a resolution to flip-over credit risk and a tool to determine the fair risk scales over a rating system.

    Citation: Bill Huajian Yang. Resolutions to flip-over credit risk and beyond-least squares estimates and maximum likelihood estimates with monotonic constraints[J]. Big Data and Information Analytics, 2018, 3(2): 54-67. doi: 10.3934/bdia.2018007

    Related Papers:

    [1] Wangming Lu, Zhiyong Yu, Zhanheng Chen, Haijun Jiang . Prescribed-time cluster practical consensus for nonlinear multi-agent systems based on event-triggered mechanism. Mathematical Biosciences and Engineering, 2024, 21(3): 4440-4462. doi: 10.3934/mbe.2024196
    [2] Yuhan Su, Shaoping Shen . Adaptive predefined-time prescribed performance control for spacecraft systems. Mathematical Biosciences and Engineering, 2023, 20(3): 5921-5948. doi: 10.3934/mbe.2023256
    [3] Jiabao Gu, Hui Wang, Wuquan Li . Output-Feedback stabilization for stochastic nonlinear systems with Markovian switching and time-varying powers. Mathematical Biosciences and Engineering, 2022, 19(11): 11071-11085. doi: 10.3934/mbe.2022516
    [4] Leonid Shaikhet . Stability of a positive equilibrium state for a stochastically perturbed mathematical model ofglassy-winged sharpshooter population. Mathematical Biosciences and Engineering, 2014, 11(5): 1167-1174. doi: 10.3934/mbe.2014.11.1167
    [5] Linni Li, Jin-E Zhang . Input-to-state stability of stochastic nonlinear system with delayed impulses. Mathematical Biosciences and Engineering, 2024, 21(2): 2233-2253. doi: 10.3934/mbe.2024098
    [6] Dongxiang Gao, Yujun Zhang, Libing Wu, Sihan Liu . Fixed-time command filtered output feedback control for twin-roll inclined casting system with prescribed performance. Mathematical Biosciences and Engineering, 2024, 21(2): 2282-2301. doi: 10.3934/mbe.2024100
    [7] Stephan Gerster, Michael Herty, Elisa Iacomini . Stability analysis of a hyperbolic stochastic Galerkin formulation for the Aw-Rascle-Zhang model with relaxation. Mathematical Biosciences and Engineering, 2021, 18(4): 4372-4389. doi: 10.3934/mbe.2021220
    [8] Lin Zhang, Yongbin Ge, Zhi Wang . Positivity-preserving high-order compact difference method for the Keller-Segel chemotaxis model. Mathematical Biosciences and Engineering, 2022, 19(7): 6764-6794. doi: 10.3934/mbe.2022319
    [9] Yue Song, Yi Zhang, Song Yang, Na Li . Investigation on stability and controller design for singular bio-economic systems with stochastic fluctuations. Mathematical Biosciences and Engineering, 2021, 18(3): 2991-3005. doi: 10.3934/mbe.2021150
    [10] Rinaldo M. Colombo, Mauro Garavello . Optimizing vaccination strategies in an age structured SIR model. Mathematical Biosciences and Engineering, 2020, 17(2): 1074-1089. doi: 10.3934/mbe.2020057
  • Given a risk outcome y over a rating system {Ri}ki=1 for a portfolio, we show in this paper that the maximum likelihood estimates with monotonic constraints, when y is binary (the Bernoulli likelihood) or takes values in the interval 0≤y≤1 (the quasi-Bernoulli likelihood), are each given by the average of the observed outcomes for some consecutive rating indexes. These estimates are in average equal to the sample average risk over the portfolio and coincide with the estimates by least squares with the same monotonic constraints. These results are the exact solution of the corresponding constrained optimization. A non-parametric algorithm for the exact solution is proposed. For the least squares estimates, this algorithm is compared with “pool adjacent violators” algorithm for isotonic regression. The proposed approaches provide a resolution to flip-over credit risk and a tool to determine the fair risk scales over a rating system.


    Stochastic nonlinear control is a hot topic because of its wide application in economic and engineering fields. The pioneer work is [1,2,3,4,5,6,7,8]. Specifically, [1,2,3] propose designs with quadratic Lyapunov functions coupled with weighting functions and [4,5,6,7,8] develop designs with quartic Lyapunov function, which are further developed by [9,10,11,12]. Recently, a class of stochastic systems (SSs) whose Jacobian linearizations may have unstable modes, has received much attention. Such systems are also called stochastic high-order nonlinear systems (SHONSs), which include a class of stochastic benchmark mechanical systems [13] as a special case. In this direction, [14] studies the state-feedback control with stochastic inverse dynamics; [15] develops a stochastic homogeneous domination method, which completely relaxes the order restriction required in [13], and investigates the output-feedback stabilization for SHONSs with unmeasurable states; [16] investigates the output-feedback tracking problem; [17] studies the adaptive state-feedback design for state-constrained systems. It should be emphasized that [13,14,15,16,17], only achieve stabilization in asymptotic sense (as time goes to infinity). However, many real applications appeals for prescribed-time stabilization, which permits the worker to prescribe the convergence times in advance.

    For the prescribed-time control, [18,19] design time-varying feedback to solve the regulation problems of normal-form nonlinear systems; [20] considers networked multi-agent systems; [21,22] investigates the prescribed-time design for linear systems in controllable canonical form; [23] designs the output-feedback controller for uncertain nonlinear strict-feedback-like systems. It should be noted that the results in [18,19,20,21,22,23] don't consider stochastic noise. For SSs, [24] is the first paper to address the stochastic nonlinear inverse optimality and prescribed-time stabilization problems; The control effort is further reduced in [25]; Recently, [26] studies the prescribed-time output-feedback for SSs without/with sensor uncertainty. It should be noted that [24,25,26] don't consider SHONSs. From practical applications, it is important to solves the prescribed-time control problem of SHONSs since it permits the worker to set the convergence times in advance.

    Motivated by the above discussions, this paper studies the prescribed-time design for SSs with high-order structure. The contributions include:

    1) This paper proposes new prescribed-time design for SHONSs. Since Jacobian linearizations of such a system possibly have unstable modes, all the prescribed-time designs in [24,25,26] are invalid. New design and analysis tools should be developed.

    2) This paper develops a more practical design than those in [13,14,15,16,17]. Different from the designs in [13,14,15,16,17] where only asymptotic stabilization can be achieved, the design in this paper can guarantee that the closed-loop system is prescribed-time mean-square stable, which is superior to those in [13,14,15,16,17] since it permits the worker to prescribe the convergence times in advance without considering the initial conditions.

    The remainder of this paper is organized as follows. In Section 2, the problem is formulated. In Section 3, the controller is designed and the stability analysis is given. Section 4 uses an example to explain the validity of the prescribed-time design. The conclusions are collected in Section 6.

    Consider the following class of SHONSs

    dx1=xp2dt+φT1(x)dω, (2.1)
    dx2=udt+φT2(x)dω, (2.2)

    where x=(x1,x2)TR2 and uR are the system state and control input. p1 is an odd integral number. The functions φi:R2Rm are smooth in x with φi(0)=0, i=1,2. ω is an m-dimensional independent standard Wiener process.

    The assumptions we need are as follows.

    Assumption 1. There is a constant c>0 such that

    |φ1(x)|c|x1|p+12, (2.3)
    |φ2(x)|c(|x1|p+12+|x2|p+12). (2.4)

    We introduce the function:

    μ(t)=(Tt0+Tt)m,t[t0,t0+T), (2.5)

    where m2 is an integral number. Obviously, the function μ(t) is monotonically increasing on [t0,t0+T) with μ(t0)=1 and limtt0+Tμ(t)=+.

    Our control goal is to design a prescribed-time state-feedback controller, which guarantees that the closed-loop system has an almost surely unique strong solution and is prescribed-time mean-square stable.

    In the following, we design a time-varying controller for system (2.1)–(2.2) step by step.

    Step 1. In this step, our goal is to design the virtual controller x2.

    Define V1=14ξ41, ξ1=x1, from (2.1), (2.3) and (2.5) we have

    LV1(ξ1)=ξ31xp2+32ξ21|φ1|2ξ31xp2+32c2ξp+31ξ31(xp2xp2)+ξ31xp2+32c2μpξp+31. (3.1)

    Choosing

    x2=μ(2+32c2)1/pξ1μα1ξ1, (3.2)

    which substitutes into (3.1) yields

    LV1(ξ1)2μpξp+31+ξ31(xp2xp2), (3.3)

    where α1=(2+32c2)1/p.

    Step 2. In this step, our goal is to design the actual controller u.

    Define ξ2=xkx2, from (3.2) we get

    ξ2=x2+μα1ξ1. (3.4)

    By using (2.1)–(2.2) and (3.4) we get

    dξ2=(u+mTμ1+1/mα1ξ1+μα1xp2)dt+(φT2+μα1φT1)dω. (3.5)

    We choose the following Lyapunov function

    V2(ξ1,ξ2)=V1(ξ1)+14ξ42. (3.6)

    It follows from (3.3), (3.5)–(3.6) that

    LV22μpξp+31+ξ31(xp2xp2)+ξ32(u+mTμ1+1/mα1ξ1+μα1xp2)+32ξ22|φ2+μα1φ1|2. (3.7)

    By using (3.4) we have

    ξ31(xp2xp2)|ξ1|3|ξ2|(xp12+xp12)|ξ1|3|ξ2|((β1+1)μp1αp11ξp11+β1ξp12)=(β1+1)μp1αp11|ξ1|p+2|ξ2|+β1|ξ1|3|ξ2|p, (3.8)

    where

    β1=max{1,2p2}. (3.9)

    By using Young's inequality we get

    (β1+1)μp1αp11|ξ1|p+2|ξ2|13μpξp+31+13+p(3+p3(2+p))(2+p)((β1+1)αp11)p+3μ3ξp+32 (3.10)

    and

    β1|ξ1|3|ξ2|p13μpξ3+p1+pp+3(3+p9)3/pβ(3+p)/p1μ3ξp+32. (3.11)

    Substituting (3.10)–(3.11) into (3.8) yields

    ξ31(xp2xp2)23μpξp+31+(pp+3(p+39)3/pβ(p+3)/p1+1p+3(p+33(p+2))(p+2)((β1+1)αp11)p+3)μ3ξp+32. (3.12)

    From (2.3), (2.4), (3.2) and (3.4) we have

    |φ2+μα1φ1|22|φ2|2+2μ2α21|φ1|24c2(|x1|1+p+|x2|1+p)+2c2α21μ2|x1|1+pβ2μ1+p|ξ1|p+1+4c22p|ξ2|1+p, (3.13)

    where

    β2=4c2(1+2pαp+11)+2c2α21. (3.14)

    By using (3.13) we get

    32ξ22|φ2+μα1φ1|213μpξ3+p1+(23+p(p+33(1+p))(1+p)/2(32β2)(3+p)/2+6c22p)μ3(1+p)/2ξp+32. (3.15)

    It can be inferred from (3.7), (3.12) and (3.15) that

    LV2μpξp+31+ξ32(u+mTμ1+1/mα1ξ1+μα1xp2)+β3μ3(p+1)/2ξp+32, (3.16)

    where

    β3=p3+p(3+p9)3/pβ(3+p)/p1+13+p(p+33(p+2))(p+2)((β1+1)αp11)3+p+23+p(p+33(1+p))(1+p)/2(32β2)(3+p)/2+6c22p. (3.17)

    We choose the controller

    u=mTμ1+1/mα1ξ1μα1xp2(1+β3)μ3(p+1)/2ξp2, (3.18)

    then we get

    LV2μpξp+31μ3(p+1)/2ξp+32. (3.19)

    Now, we describe the main stability analysis results for system (2.1)–(2.2).

    Theorem 1. For the plant (2.1)–(2.2), if Assumption 1 holds, with the controller (3.18), the following conclusions are held.

    1) The plant has an almost surely unique strong solution on [t0,t0+T);

    2) The equilibrium at the origin of the plant is prescribed-time mean-square stable with limtt0+TE|x|2=0.

    Proof. By using Young's inequality we get

    14μξ41μpξp+31+β4μ3, (3.20)
    14μξ42μ3(p+1)/2ξp+32+β4μ3, (3.21)

    where

    β4=p1p+3(3+p4)4/(p1)(14)(3+p)/(p1). (3.22)

    It can be inferred from (3.19)–(3.21) that

    LV214μξ4114μξ42+2β4μ3=μV2+2β4μ3. (3.23)

    From (2.1), (2.2) and (3.18), the local Lipschitz condition is satisfied by the plant. By (3.23) and using Lemma 1 in [24], the plant has an almost surely unique strong solution on [t0,t0+T), which shows that conclusion 1) is true.

    Next, we verify conclusion 2).

    For each positive integer k, the first exit time is defined as

    ρk=inf{t:tt0,|x(t)|k}. (3.24)

    Choosing

    V=ett0μ(s)dsV2. (3.25)

    From (3.23) and (3.25) we have

    LV=ett0μ(s)ds(LV2+μV2)2β4μ3ett0μ(s)ds. (3.26)

    By (3.26) and using Dynkin's formula we get

    EV(ρkt,x(ρkt))=Vn(t0,x0)+E{ρktt0LV(x(τ),τ)dτ}Vn(t0,x0)+2β4tt0μ3eτt0μ(s)dsdτ. (3.27)

    Using Fatou Lemma, from (3.27) we have

    EV(t,x)Vn(t0,x0)+2β4tt0μ3eτt0μ(s)dsdτ,t[t0,t0+T). (3.28)

    By using (3.25) and (3.28) we get

    EV2ett0μ(s)ds(Vn(t0,x0)+2β4tt0μ3eτt0μ(s)dsdτ),t[t0,t0+T). (3.29)

    By using (3.6) and (3.29) we obtain

    limtt0+TE|x|2=0. (3.30)

    Consider the prescribed-time stabilization for the following system

    dx1=x32dt+0.1x21dω, (4.1)
    dx2=udt+0.2x1x2dω, (4.2)

    where p1=3, p2=1. Noting 0.2x1x20.1(x21+x22), Assumption 1 is satisfied.

    Choosing

    μ(t)=(11t)2,t[0,1), (4.3)

    According to the design method in Section 3, we have

    u=3μ1.5x11.5μx3257μ6(x2+1.5μx1)3 (4.4)

    In the practical simulation, we select the initial conditions as x1(0)=6, x2(0)=5. Figure 1 shows the responses of (4.1)–(4.4), from which we can obtain that limtt0+TE|x1|2=limtt0+TE|x2|2=0, which means that the controller we designed is effective.

    Figure 1.  The responses of closed-loop system (4.1)–(4.4).

    This paper proposes a new design method of prescribed-time state-feedback for SHONSs. the controller we designed can guarantee that the closed-loop system has an almost surely unique strong solution and the equilibrium at the origin of the closed-loop system is prescribed-time mean-square stable. The results in this paper are more practical than those in [13,14,15,16,17] since the design in this paper permits the worker to prescribe the convergence times in advance without considering the initial conditions.

    There are some related problems to investigate, e.g., how to extend the results to multi-agent systems [27], impulsive systems [28,29,30] or more general high-order systems [31,32,33,34].

    This work is funded by Shandong Provincial Natural Science Foundation for Distinguished Young Scholars, China (No. ZR2019JQ22), and Shandong Province Higher Educational Excellent Youth Innovation team, China (No. 2019KJN017).

    The authors declare there is no conflict of interest.

    [1] Anderson R, (2007) The Credit Scoring Toolkit: Theory and Practice for Retail Credit Risk Management and Decision Automation, Oxford: Oxford University Press.
    [2] Barlow RE, Bartholomew DJ, Bremner JM, et al. (1972) Statistical inference under order restrictions: The theory and application of isotonic regression, New York: Wiley.
    [3] Best MJ, Chakravarti N, (1990) Active set algorithms for isotonic regression; A unifying framework. Math Program 47: 425–439. doi: 10.1007/BF01580873
    [4] Eichenberg T, (2018) Supervised weight of evidence binning of numeric variables and factors, R-Package Woebinning.
    [5] Friedman J, Tibshirani R, (1984) The monotone smoothing of scatterplots. Technometrics 26: 243–250. doi: 10.1080/00401706.1984.10487961
    [6] Kotlowski W, Slowinski R, (2009) Rule learning with monotonicity constraints. Proceedings of the 26th Annual International Conference on Machine Learning, 537–544.
    [7] KotsiantisS, KanellopoulosD, (2006) Discretization techniques: A recent survey. GESTSInt Trans Comput Sci Engin 32: 47–58.
    [8] Leeuw JD, Hornik K, Mai P, (2009) Isotone optimization in R: Pool-adjacent-violators algorithm (PAVA) and active set methods. J stat software 32.
    [9] Nocedal J, Wright SJ, (2006) Numerical Optimization, 2 Eds.,New York: Springer.
    [10] Papke LE, Wooldrige JM, (1996) Econometric methods for fraction response variables with application to 401 (k) plan participation rates. J Appl Econometrics 11: 619–632. doi: 10.1002/(SICI)1099-1255(199611)11:6<619::AID-JAE418>3.0.CO;2-1
    [11] Potharst R, Feelders AJ, (2002) Classification trees for problems with monotonicity constraints. SIGKDD Explor 14: 1–10.
    [12] Ramsay JO, Wickham H, Graves S, et al. (2018) Package 'fda'-CRAN.R-Project, 265–267
    [13] Robertson T, Wright FT, Dykstra RL, (1998) Order Restricted Stat Inference, New Jersey: John Wiley and Sons.
    [14] SAS Institute Inc (2014) SAS/STAT(R) 13.2 User's Guide.
    [15] Siddiqi N, (2006) Credit risk scorecards: Developing and implementing intelligent credi scoring. Hoboken, New Jersey: John Wiley and Sons.
    [16] Tasche D, (2013) The art of PD curve calibration. J Credit Risk 9: 63–103. doi: 10.21314/JCR.2013.169
    [17] Van der Burgt M, (2008) Calibrating low-default portfolios, using the cumulative accuracy profile. J Risk Model validation 1: 17–33. doi: 10.21314/JRMV.2008.016
    [18] Yang BH, (2018) Smoothing algorithms by constrained maximum likelihood. J Risk Model Validation 12: 89–102.
    [19] You S, Ding D, Canini K, et al. (2017) Deep lattice networks and partial monotonic functions. 31st Conf Neural Inf Process Syst (NIPS).
  • This article has been cited by:

    1. Ran Liu, Hui Wang, Wuquan Li, Prescribed-time stabilization and inverse optimal control of stochastic high-order nonlinear systems, 2024, 67, 1674-733X, 10.1007/s11432-022-3842-2
    2. Liuliu Zhang, Xianglin Liu, Changchun Hua, Prescribed-Time Control for Stochastic High-Order Nonlinear Systems With Parameter Uncertainty, 2023, 70, 1549-7747, 4083, 10.1109/TCSII.2023.3274680
    3. Yang Xu, Xisheng Dai, 2024, Prescribed-Time Stabilizaition of Linear Stochastic Parabolic Systems, 979-8-3503-8778-0, 4059, 10.1109/CCDC62350.2024.10588138
    4. Jianyang Du, Guici Chen, Song Zhu, Junhao Hu, Prescribed-time stabilisation of stochastic high-order nonlinear systems with time-varying powers, 2025, 0020-7179, 1, 10.1080/00207179.2025.2451712
  • Reader Comments
  • © 2018 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(6217) PDF downloads(2155) Cited by(0)

Figures and Tables

Tables(1)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog