MPE Mathematical Problems in Engineering 1563-5147 1024-123X Hindawi Publishing Corporation 10.1155/2016/2371826 2371826 Research Article Sparsity-Homotopy Perturbation Inversion Method with Wavelets and Applications to Black-Scholes Model and Todaro Model http://orcid.org/0000-0001-9603-8051 Dou Yixin 1 http://orcid.org/0000-0002-1074-7971 Wang Zhihao 2 Schuster Thomas 1 School of Finance Harbin University of Commerce Harbin 150028 China hrbcu.edu.cn 2 School of Management Harbin University of Science and Technology Harbin 150080 China hrbust.edu.cn 2016 2172016 2016 17 04 2016 28 06 2016 2016 Copyright © 2016 Yixin Dou and Zhihao Wang. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Sparsity regularization method plays an important role in reconstructing parameters. Compared with traditional regularization methods, sparsity regularization method has the ability to obtain better performance for reconstructing sparse parameters. However, sparsity regularization method does not have the ability to reconstruct smooth parameters. For overcoming this difficulty, we combine a sparsity regularization method with a wavelet method in order to transform smooth parameters into sparse parameters. We use a sparsity-homotopy perturbation inversion method to improve the accuracy and stability and apply the proposed method to reconstruct parameters for a Black-Scholes option pricing model and a Todaro model. Numerical experiments show that the proposed method is convergent and stable.

National Natural Science Foundation of China 11301119 71541023 Scientific Research Fund of Heilongjiang Provincial Education Department 12541191
1. Introduction

The reconstruction of parameters plays an important role in financial mathematics, such as the reconstructions of a volatility and a policy parameter [1, 2] and, in other fields, imaging enhancement techniques [3, 4], seismic signals [5, 6], and electrocardiogram signals (ECG) [7, 8]. With the development of economy and financial mathematics, the reconstructions of a volatility and a policy parameter have been widely used in many real applications. In general, the reconstruction of parameters is ill-posed. In other words, the small noisy level of measurement data can lead to the large error of reconstruction . In order to overcome the ill-posedness, some regularization methods are developed. The most popular method is Tikhonov regularization method, which is composed of a fitting term and a penalty term in L2 norm. The aims of those two terms are to match measurement data and to suppress noises, respectively.

The numerical methods for Tikhonov regularization method have been conducted, such as a well-known Landweber method , a Gauss-Newton method , a regularizing Newton-Kaczmarz method , and a multiscale smoothing method . These methods have the ability to reconstruct smooth parameters in the case of sufficient measurement data. When measurement data are limited, smooth parameters are very difficult to be reconstructed. In the fields of economy and finance, reconstructed parameters are smooth and measurement data are limited. Hence, we need to use wavelet and curvelet transformations from smooth parameters to sparse parameters (i.e., the number of nonzero elements of parameter is very limited). Sparsity regularization methods are used to reconstruct sparse parameters. Compared with Tikhonov regularization method, sparsity regularization methods are not differentiable, and hence some specific techniques were developed to overcome this difficulty, such as Bregman iterations . For reducing the computational time, a homotopy perturbation inversion method has been widely used in real applications, such as nonlinear systems, optimal control, and heat transfer equation .

In this paper, we combine a homotopy perturbation inversion method with a sparsity regularization method, in order to improve the accuracy and stability. After introducing the proposed method, we reconstruct two economic parameters based on a Black-Scholes option pricing model and a Todaro model. From the numerical experiments, a sparsity-homotopy perturbation inversion method with wavelets is convergent and stable.

2. Sparsity Regularization Method

The reconstruction of parameters is ill-posed, and hence we should employ the regularization method to reconstruct parameters. The Tikhonov regularization method can reconstruct smooth parameters while measurement data are sufficient. However, measurement data are limited in real applications. We combine a sparsity regularization method with a wavelet method, in order to improve the performance of reconstructing smooth parameters for limited measurement data.

2.1. Tikhonov Regularization Method

Many inverse problems can be formulated as a nonlinear operator equation (1)Fx=y,where F,x,andy denote a nonlinear operator, reconstructed parameter, and measurement data, respectively. We assume measurement data contaminated by noise (2)yδ-y2δ,where yδ,δ stand for the real measurement data and noisy level in the L2 norm.

Ill-posedness means that the small noisy level included in measurement data may lead to the large error of reconstruction. We apply the regularization method in order to overcome ill-posedness. It is very important to balance a fitting term and a penalty term. Next, we introduce a Tikhonov regularization functional as follows: (3)Jx=Fx-yδ22+αx22,where F(x)-yδ2 is a fitting term, x2 is a penalty term, and α is a regularized parameter balancing the fitting term and the penalty term. For reconstruction, we should minimize the functional (3): (4)x=argminJx.

The minimizer satisfies the Euler equation: (5)FxFx-yδ+αx=0,where F(x) is the Fréchet derivative. The Landweber method is widely used for solving (5). The Landweber method can be written as follows: (6)xn+1=xn-FxnFxn-yδ+αxn,where n denotes the iteration number. Equation (6) is a well-known Landweber method that is stable; however, the convergent rate is slow and the accuracy is low.

2.2. Sparsity Regularization Method with Wavelets

The Tikhonov regularization method can obtain better reconstruction for smoothing parameters; however, its performance for reconstructing sparse parameters is worse. Hence, the classical Tikhonov regularization method is modified as the following sparsity regularization method: (7)Jx=Fx-yδ22+αx0,where the norm of ·0 means the number of nonzeros in x. The minimization problem (7) has the ability to obtain the best reconstruction for sparse parameters. However, the penalty term of (7) is nondifferentiable and minimization problem (7) is a NP problem. For speeding up minimization problem (7), we rewrite (7) as (8)Jx=Fx-yδ22+αx1,where the norm of ·1 means xdσ. In functional (8), the L1 norm replaces the L0 norm in order to overcome a NP problem. This improvement can reduce the cost time of (7) dramatically.

Because the penalty term of (8) is nondifferentiable, we modify problem (8) as follows: (9)Jx=Fx-yδ22+αx2,ε,where x2,ε=x2+ε2 replaces x1, and the auxiliary parameter ε is a positive real number. As the Tikhonov regularization method, we can apply the Landweber method to minimize functional (9): (10)xn+1=xn-FxnFxn-yδ+αxnxn2+ε21/2.

For a smooth parameter, we combine a sparsity regularization method with a wavelet method. A wavelet transformation converts a smooth parameter into a sparse parameter. A wavelet transformation is written as (11)x=Wxw,where W,xw stand for an inverse transformation matrix and the sparse representation of x, respectively. Taking (11) into (9) and (10), we arrive at (12)Jxw=FWxw-yδ22+αxw2,ε,(13)xw,n+1=xw,n-FWxw,nFWxw,n-yδ+αxw,nxw,n2+ε21/2.

3. Sparsity-Homotopy Perturbation Inversion Method

Because the Landweber method (13) is very slow, it is not realistic to reconstruct large scale real applications. We use a homotopy perturbation inversion method to modify the classical Landweber method. For convenience, we omit W and replace xw with x.

Setting a homotopy mapping (14)H:F×0,1YHx~,p=pFx~Fx~-yδ+αx~x~2+ε21/2+1-px~-x0=0,p0,1,where p is an embedding parameter and x0 is an initial guess value. Hence, (15)Hx~,0=x~-x0=0,Hx~,1=Fx~Fx~-yδ+αx~x~2+ε21/2=0.

We write x~ as the power series of p, (16)x~=x0+p1x1+p2x2+,and obtain the approximation of functional (12): (17)x=limp1x~=x0+x1+x2+.

We expand F(x) in (14) as a Taylor series near x0: (18)Hx~,p=pFx0Fx0+Fx0x~-x0+ox~-x0-yδ+αx0x02+ε21/2+ε2x02+ε23/2x~-x0+ox~-x0+1-px~-x0=0,pFx0Fx0+Fx0p1x1+p2x2+-yδ+αx0x02+ε21/2+ε2x02+ε23/2p1x1+p2x2++1-pp1x1+p2x2+=0.Following the power of p, we can get (19)p1:x1=-Fx0Fx0-yδ-αx0x02+ε21/2,p2:x2=I-Fx0Fx0-αε2x02+ε23/2-Fx0Fx0-yδ-αx0x02+ε21/2,(20)x=x0+-Fx0Fx0-yδ-αx0x02+ε21/2+I-Fx0Fx0-αε2x02+ε23/2-Fx0Fx0-yδ-αx0x02+ε21/2+.Following formula (20), the parameter for the noisy measurement data is reconstructed by the first two terms: (21)xn+1=xn+2I-FxnFxn-αε2xn2+ε23/2-FxnFxn-yδ-αxnxn2+ε21/2,where n is the iteration number.

When the parameter is reconstructed by the first term, (22)xn+1=xn-FxnFxn-yδ+αxnx2+ε21/2.Equation (22) is a well-known Landweber method. Equation (21) is a modified version of (22), which is called a homotopy perturbation inversion method. The convergent rate is faster and the accuracy is higher than a Landweber method. By using a wavelet transformation, we can obtain (23)xw,n+1=xw,n+2I-FWxw,nFWxw,n-αε2xw,n2+ε23/2-FWxw,nFWxw,n-yδ-αxw,nxw,n2+ε21/2.

4. Numerical Experiments 4.1. Reconstructing Sparse Volatility

The reconstruction of a volatility is investigated based on the Black-Scholes (B-S) option pricing model. We use a finite difference method to solve the forward problem. In order to improve the accuracy and convergent rate of a Landweber method, we design a homotopy perturbation inversion method to reconstruct a volatility. The B-S formula is widely used in the field of a derivative pricing, when the price changes of a derivative satisfy the standard geometric Brown motion. The boundary conditions of the different values vary with the different types of derivatives. When boundary conditions are given, a derivative pricing Vt can be obtained by solving the B-S formula of the derivative pricing model. When the derivative is taken as option, we define the forward problem as the determination of the option pricing. For simplicity, we take the European call option as example.

The relationship between the European call and put options shows that the reconstructed volatility should be the same by using the call options market quotes or the put options market quotes. For the European call option in the time interval [0,T], let Vt=V(St,t;σ,K,T) be the European call option pricing, and V satisfies (24)Vtt+r-qStVtSt+12σ2St,tSt22VtSt2-rVt=0,VT=ST-K+,where St is stock prices with time, K is strike price, r is interest rate, q is dividend, T is availability period, t is time, and σ is volatility.

In the B-S formula, except that the volatility is a free variable, the other parameters and variables are decided by the prevailing market conditions or contracts. Hence, the volatility σ is a very important parameter. For an option, we can infer the volatility from the B-S formula.

The inverse problem is defined as follows: reconstructing the local volatility σ from the measurement data V(S,t;σ,Ki,Ti)=Vi(i=1,2,,N) and S=S.

We define a nonlinear vector-valued function F:σV, namely, F(σ)=V. For testing the performance of sparse regularization and Tikhonov regularization methods, we set T=1 and set stock prices S=50, interest rate r=0.05, and strike price K=35,40,45. We add 0%, 0.5%, and 1% Gaussian random noises to the measurement data in order to test the stability. The exact volatility is provided as follows: (25)σ=0.15sinπTt+0.15cosπTt.

We transform the smooth volatility into the sparse volatility (26)σ=Wσw,where σw is a sparse representation. Functional (12) is rewritten as (27)Jσw=FWσw-Vδ22+ασw2,ε.

The results of the proposed method (I) and Tikhonov method (II) are listed in Table 1.

Relative errors and computational time.

Noisy level Relative error (I) Relative error (II) Time (I) Time (II)
0.00% 0.21% 1.36% 77 s 112 s
0.50% 1.28% 3.06% 196 s 332 s
1.00% 3.25% 6.98% 528 s 1052 s

From the above reconstructions, we can see that the proposed method has better performance than the Tikhonov regularization method.

4.2. Reconstructing Policy Parameter for Todaro Model

Todaro model is the famous model to describe the number of rural migrants (namely, workers in urban areas from rural areas) in the labor economics. The Todaro model is also used in development economics and welfare economics to explain some of the issues concerning rural-urban migration. The main assumption of the model is that the migration decision is based on expected income differentials between rural and urban areas . A Todaro model shows the relationship between the number of rural migrants with the income difference of urban and rural areas. A Todaro model can be written in the following form: (28)M=fd,where M,d denote the number of rural migrants and the income difference of urban and rural areas, respectively. Function f is an increasing function, that is, f>0. A modified Todaro model considers policy parameter χ: (29)M=fχ,d,where χ describes the efficiency of government policy including household registration policy, social security policy, old-supporting policy, and others.

We consider I urban areas and J rural areas. dij(1iI,1jJ) stands for the income difference between the ith urban area and the jth rural area, and gii~(1iI,1i~I) stands for the income difference between the ith and the i~th urban areas. The number of migrants into the ith urban area is denoted as Mi. For the ith urban area, policy parameter χi is split into two parts χi=χir+χiu, where χir,χiu denote the efficiency of government policy to rural and urban areas, respectively. We assume that the number of workers from urban areas into rural areas is zero. In real applications, Function f has many different representations. In this section, we focus on the performance of the proposed method, and hence we take f as a linear function. The Todaro model is modified as (30)χ1rd11+χ1rd12++χ1rd1J+χ1ug11+χ1ug12++χ1ug1I=M1χirdi1+χirdi2++χirdiJ+χiugi1+χiugi2++χiugiI=MiχIrdI1+χIrdI2++χIrdIJ+χIugI1+χIugI2++χIugII=MI,where the income differences dij and gii~ and the number Mi of migrants into the ith urban area are known. The policy parameter χ is unknown.

Let (31)Dk=j=1Jdkj,Gk=i=1Igki,1kI,A=D1G10000DiGi0000DIGIM=M1,,Mi,,MIT,X=χ1r,χ1u,,χir,χiu,,χIr,χIuT.

Equation (30) is recast as (32)AX=M.Note that (32) is underdetermined, and hence we apply the regularization method to solve this equation. We transform a smooth policy parameter into a sparse parameter X=WXw, where Xw means a sparse policy parameter. The cost functional is as follows: (33)JXw=AWXw-yδ22+αXw2,ε.In numerical tests, we take I=2n-1,J=2n. We add a 1% Gaussian random noise to the measurement data in order to test the stability. We reconstruct three policy parameters corresponding to n=4,5,6. The relative errors of the proposed method (I) and Tikhonov method (II) are listed in Table 2. Due to linearity and small scale, the difference of the computational times of the proposed method and Tikhonov method is small. Table 2 shows that the proposed method is feasible to reconstruct policy parameters.

Relative errors.

n Relative error (I) Relative error (II)
4 3.61% 5.82%
5 3.76% 6.75%
6 4.97% 8.99%
5. Conclusions

We design a sparsity regularization method to reconstruct a volatility and a policy parameter. Because parameters are often smooth in the fields of economy and finance, we apply a wavelet transformation from a smooth parameter into a sparse parameter. The homotopy perturbation inversion method is used to minimize the cost functional, and the accuracy and convergent rate of reconstruction are improved. The numerical experiments show that the proposed method can be applied to parameter identification and initial value problem in heat transfer equations [22, 23].

Competing Interests

The authors declare that there are no competing interests regarding the publication of this paper.

Acknowledgments

This work is supported by National Natural Science Foundation of China (11301119 and 71541023) and Scientific Research Fund of Heilongjiang Provincial Education Department (12541191).

Egger H. Engl H. W. Tikhonov regularization applied to the inverse problem of option pricing: convergence analysis and rates Inverse Problems 2005 21 3 1027 1045 10.1088/0266-5611/21/3/014 MR2146819 2-s2.0-18744381309 Trong D. D. Thanh D. N. Lan N. N. Uyen P. H. Calibration of the purely T-dependent Black-Scholes implied volatility Applicable Analysis 2014 93 4 859 874 10.1080/00036811.2013.800974 MR3180022 2-s2.0-84896491007 Ding L. Zhao X. R. Shearlet-wavelet regularized semismooth newton iteration for image restoration Mathematical Problems in Engineering 2015 2015 12 647254 10.1155/2015/647254 2-s2.0-84926320264 Ding L. Cao J. Electromagnetic nondestructive testing by perturbation homotopy method Mathematical Problems in Engineering 2014 2014 10 895159 10.1155/2014/895159 MR3179034 2-s2.0-84896962059 Baig A. M. Campillo M. Brenguier F. Denoising seismic noise cross correlations Journal of Geophysical Research: Solid Earth 2009 114 8 B08310 10.1029/2008jb006085 2-s2.0-72749095792 Hennenfent G. Herrmann F. J. Simply denoise: wavefield reconstruction via jittered undersampling Geophysics 2008 73 3 V19 V28 10.1190/1.2841038 2-s2.0-43549122094 Istepanian R. S. H. Petrosian A. A. Optimal zonal wavelet-based ECG data compression for a mobile telecardiology system IEEE Transactions on Information Technology in Biomedicine 2000 4 3 200 211 10.1109/4233.870030 2-s2.0-0034257893 Sameni R. Shamsollahi M. B. Jutten C. Clifford G. D. A nonlinear Bayesian filtering framework for ECG denoising IEEE Transactions on Biomedical Engineering 2007 54 12 2172 2185 10.1109/TBME.2007.897817 2-s2.0-36348946459 Engl H. W. Hanke M. Neubauer A. Regularization of Inverse Problems 1996 Dordrecht, The Netherlands Kluwer Academic 10.1007/978-94-009-1740-8 MR1408680 Hanke M. Neubauer A. Scherzer O. A convergence analysis of the Landweber iteration for nonlinear ill-posed problems Numerische Mathematik 1995 72 1 21 37 10.1007/s002110050158 MR1359706 ZBL0840.65049 Blaschke B. Neubauer A. Scherzer O. On convergence rates for the iteratively regularized Gauss-Newton method IMA Journal of Numerical Analysis 1997 17 3 421 436 10.1093/imanum/17.3.421 MR1459331 ZBL0881.65050 2-s2.0-0031532935 Burger M. Kaltenbacher B. Regularizing Newton-Kaczmarz methods for nonlinear ill-posed problems SIAM Journal on Numerical Analysis 2006 44 1 153 182 10.1137/040613779 MR2217377 2-s2.0-33847617979 Bredies K. Lorenz D. A. Maass P. Mathematical concepts of multiscale smoothing Applied and Computational Harmonic Analysis 2005 19 2 141 161 10.1016/j.acha.2005.02.007 MR2163076 2-s2.0-24044447221 Brune C. Sawatzky A. Burger M. Primal and dual Bregman methods with application to optical nanoscopy International Journal of Computer Vision 2011 92 2 211 229 10.1007/s11263-010-0339-5 MR2780767 ZBL1235.94016 2-s2.0-80052398615 Burger M. Resmerita E. He L. Error estimation for Bregman iterations and inverse scale space methods in image restoration Computing 2007 81 2-3 109 135 10.1007/s00607-007-0245-z MR2354192 2-s2.0-36549078210 Cai J. F. Osher S. Shen Z. Convergence of the linearized Bregman iteration for l1-norm minimization Mathematics of Computation 2009 78 2127 2136 10.1090/s0025-5718-09-02242-x Cai J.-F. Osher S. Shen Z. Linearized Bregman iterations for compressed sensing Mathematics of Computation 2009 78 267 1515 1536 10.1090/S0025-5718-08-02189-3 MR2501061 2-s2.0-67349244805 Geng F. Cui M. Homotopy perturbation-reproducing kernel method for nonlinear systems of second order boundary value problems Journal of Computational and Applied Mathematics 2011 235 8 2405 2411 10.1016/j.cam.2010.10.040 MR2763153 ZBL1209.65078 2-s2.0-79251597732 Jajarmi A. Ramezanpour H. Sargolzaei A. Shafaei P. Optimal control of nonlinear systems using the homotopy perturbation method: infinite horizon case International Journal of Digital Content Technology and Its Applications 2010 4 9 114 122 10.4156/jdcta.vol4.issue9.14 2-s2.0-78751509056 Ganji D. D. Rafei M. Vaseghi J. Application of homotopy-perturbation method for systems of nonlinear momentum and heat transfer equations Heat Transfer Research 2007 38 4 361 379 10.1615/HeatTransRes.v38.i4.70 2-s2.0-34347251140 Harris J. R. Todaro M. P. Migration, unemployment and development: a two-sector analysis American Economic Review 1970 60 1 126 142 Dou Y.-X. Han B. Total variation regularization for the reconstruction of a mountain topography Applied Numerical Mathematics 2012 62 1 1 20 10.1016/j.apnum.2011.09.004 MR2856059 2-s2.0-81055157693 Dou Y.-X. Han B. Reconstruction of a velocity field for a 3-D advection-diffusion equation International Journal of Thermal Sciences 2011 50 12 2340 2354 10.1016/j.ijthermalsci.2011.08.002 2-s2.0-80052693565