A general continuous time Markov chain approximation for multi-asset option pricing with systems of correlated diffusions

https://doi.org/10.1016/j.amc.2020.125472Get rights and content

Highlights

  • This paper develops a general methodology for modeling and pricing financial derivatives which depend on systems of stochastic diffusion processes.

  • Weak convergence of the approximation is demonstrated, with second order convergence in space.

  • Numerical experiments demonstrate the accuracy and efficiency of the method for various European and early-exercise options in two and three dimensions

Abstract

Continuous time Markov Chain (CTMC) approximation techniques have received increasing attention in the option pricing literature, due to their ability to solve complex pricing problems, although existing approaches are mostly limited to one or two dimensions. This paper develops a general methodology for modeling and pricing financial derivatives which depend on systems of stochastic diffusion processes. This is accomplished with a general decorrelation procedure, which reduces the system of correlated diffusions to an uncorrelated system. This enables simple and efficient approximation of the driving processes by univariate CTMC approximations. Weak convergence of the approximation is demonstrated, with second order convergence in space. Numerical experiments demonstrate the accuracy and efficiency of the method for various European and early-exercise options in two and three dimensions.

Introduction

Option pricing for multi-asset systems is an especially challenging problem in quantitative finance, with numerous practical applications. Contracts such as spread options [1] are very common in energy and power markets, where crack spreads (oil) and spark spreads (electricity) are actively traded on major financial exchanges. In energy markets, various forms of averages are traded, often on multiple underlying futures, usually a set of consecutive maturities. Basket options, mountain range options, and best-of options are other common examples.

While the literature on single-asset option pricing is quite dense, the problem of multi-asset pricing remains relatively underdeveloped. The most commonly used approach is Monte Carlo and its many variations [2], [3], [4], [5], [6]. Techniques such as GPU acceleration [7], grid bundling [8], and quasi-Monte Carlo [9] have been proposed to enhance the efficiency of Monte Carlo. Memory reduction methods [10] have also been proposed for high dimensionality. For lower dimensional problems, partial differential equation (PDE) methods such as operator splitting are often applied, see Ikonen and Toivanen [11], and Clift and Forsyth [12] for an implicit method in two dimensions. Other PDE approaches include [13], [14]. Decorrelation techniques for removing the mixed derivatives in PDEs has been used by Company et al. [15], Egorova et al. [16]. Coordinate transformations with sparse grids are applied in [17] to price basket options by PDE, for up to five underlying assets. Radial basis techniques have also been applied [18], [19] in higher dimensional problems, as well as hexaonomial lattices [20], and finite element methods [21].

In two dimensions, Fourier transform techniques offer highly efficient alternatives, such as the CONV method [22], the COS method, [23], the SWIFT method [24], [25], and the PROJ method [26], [27]. Efficient specialized methods exist for certain payoffs, such as spread options [28], [29]. Closed form approximations have been provided for spread options [30], and basket options [31].

A more recent option pricing approach is provided by Markov chain approximation. In an early work, [32] consider the problem of barrier option pricing in one dimension. Later extensions include [33], [34], [35]. For a review of recent work in the area of CTMC approximation techniques for financial modeling, see [36], and rigorous error analysis can be found in [35], [37] for problems in one dimension.

The application of CTMC approximation in higher dimensions is still in its early stages of developement. For example, the case of stochastic volatility with two correlated diffusions is developed in [34], [38], [39], [40] using a combination of Markov chain approximation, regime-switching, and Fourier transform techniques. They then provided a two-layer CTMC approximation method which extends to the case of stochastic local volatility in [41], as well as one and two-layer approximations for the case of time-changed Markov processes in [42]. The use of hybrid Markov models, such as CTMC-Heston, is proposed in [43]. More recently, Xi et al. [44] provides an approach which can handle general two dimensional systems, based on matching correlations of the driving processes. However, the construction of the CTMC process proposed in [44] is implicit and it is not clear to extend their method to higher dimensional cases. In higher dimensions, the most important feature of the model is the correlation between various assets. In this work, we provide a CTMC approximation framework which can handle n-dimensional systems of diffusions with constant coefficients, allowing for arbitrary correlation between assets. The method relies on a novel decoupling transformation, proposed in [15], which allows us to de-correlate the diffusion system and approximate each dimension with a univariate CTMC. The final step is to re-correlate the CTMCs, and conduct pricing. Most importantly, the CTMC scheme is proved to weakly converge to the continuous diffusion in the second order. The methodology is quite general, and allows for European and exotic option pricing through simple matrix products, and is thus well-suited for practical applications. Extensive numerical examples not only confirm that our method is stable, fast as compared to the existing approach such as [23] but also validate the second order convergence of the CTMC scheme.

The rest of the paper is organized as follows: Section 2 introduces the correlated model and a general method to remove the correlations among the governing Brownian motions. In Section 3, we propose a general continuous time Markov chain framework to approximate the model introduced in Section 2. The weak convergence of the continuous time Markov chain approximation is considered in Section 3.2.1. Numerical examples in two and three dimensions are provided in Section 4. Section 5 concludes the paper with some future research directions.

Section snippets

Diffusion system

The multi-asset model we consider is the system of diffusionsdSi(t)=(μiqi)Si(t)dt+σiSi(t)dWi(t),i=1,2,,n,where Si is the ith underlying asset having expected return μiR, and continuous dividend of qiR, and volatility of σi > 0. The Brownian processes are correlated with ρijdt=E[dWi(t)dWj(t)],1i,jn, where ρij[1,1]. Note that the model (1) was considered in [28] and [23].

Let S=S1×S2××Sn be the state space of (Si(t))1 ≤ i ≤ n, and C0(S) denote the set of functions on S that vanish at

Continuous time Markov chain approximation

After de-correlating the diffusion system, the next step in our methodology is to obtain a Markov chain approximation to the system (11), which will result in closed-form value approximations. For notational simplification, let θi=j=1ncij(μjqjσj2/2)σj and νi=dii, then the system in (11) is reduced todYi(t)=θidt+νidWi*(t),i=1,2,,n,For each i1,n¯, let Gi,N={y1(i),y2(i),,yN(i)}Si. We consider a general n-dimensional continuous time Markov chain (CTMC)Y(N)(t)={(Y1(N)(t),Y2(N)(t),,Yn(N)(t))}t

Numerical examples

In this section we conduct numerical experiments demonstrating the versatility of the proposed N-Dimensional CTMC method (N-CTMC), and we verify the theoretical second order convergence proved in Theorem 3. Experiments are provided for European, Bermudan, and barrier options, with a variety of different payoffs and parameter settings. All experiments are conducted in Matlab 8.5 on a personal computer with Intel(R) Core(TM) i7-6700 CPU 3.40GHz.

Conclusion

We consider a general Markov chain approximation framework for correlated diffusion systems with constant coefficients. We prove that the continuous time Markov chain approximation leads to the second order convergence in n-dimensions, which is confirmed by extensive numerical examples. European, Bermudan, and barrier contracts are all considered in this work. The proposed scheme is stable, accurate, and fast as compared to the other existing methods. It would be interesting to extend the

References (62)

  • V.N. Egorova et al.

    A local radial basis function method for high-dimensional American option pricing problems

    Math. Model. Anal.

    (2018)
  • P. Kovalov et al.

    Pricing multi-asset American options: a finite element method-of-lines with smooth penalty

    J. Sci. Comput.

    (2007)
  • D.-M. Dang et al.

    A dimension reduction Shannon-wavelet based method for option pricing

    J. Sci. Comput.

    (2018)
  • L. Li et al.

    Error analysis of finite difference and Markov chain approximations for option pricing

    Math. Finance

    (2018)
  • Z. Cui et al.

    A general framework for discretely sampled realized variance derivatives in stochastic volatility models with jumps

    Eur. J. Oper. Res.

    (2017)
  • Z. Cui et al.

    A general valuation framework for SABR and stochastic local volatility models

    SIAM J. Financ. Math.

    (2018)
  • Á. Leitao et al.

    The CTMC-Heston Model: Calibration and Exotic Option Pricing with SWIFT

    (2019)
  • J.L. Kirkby

    An efficient transform method for Asian option pricing

    SIAM J. Financ. Math.

    (2016)
  • R. Carmona et al.

    Pricing and hedging spread options

    SIAM Rev.

    (2003)
  • L. Andersen et al.

    Primal-dual simulation algorithm for pricing multidimensional American options

    Manage. Sci.

    (2004)
  • D. Belomestny et al.

    Regression-based complexity reduction of the nested Monte Carlo methods

    SIAM J. Financ. Math.

    (2018)
  • P.P. Boyle et al.

    Numerical evaluation of multivariate contingent claims

    Rev. Financ. Stud.

    (1989)
  • J. Lelong

    Dual pricing of American options by wiener chaos expansion

    SIAM J. Financ. Math.

    (2018)
  • Á. Leitao et al.

    Gpu acceleration of the stochastic grid bundling method for early-exercise options

    Int. J. Comput. Math.

    (2015)
  • V.N. Egorova et al.

    Computing Stable Numerical Solutions for Multidimensional American Option Pricing Problems: A Semi-Discretization Approach

    (2017)
  • C. Leentvaar et al.

    On coordinate transformation and grid stretching for sparse grid pricing of basket options

    J. Comput. Appl. Math.

    (2008)
  • W.-H. Kao et al.

    The hexanomial lattice for pricing multi-asset options

    Appl. Math. Comput.

    (2014)
  • C. Leentvaar et al.

    Multi-asset option pricing using a parallel Fourier-based technique

    J. Comput. Finance

    (2008)
  • M.J. Ruijter et al.

    Two-dimensional Fourier cosine series expansion method for pricing financial options

    SIAM J. Sci. Comput.

    (2012)
  • G. Colldeforns-Papiol et al.

    Two-dimensional Shannon wavelet inverse Fourier technique for pricing European options

    Appl. Numer. Math.

    (2017)
  • J. Zhao et al.

    Efficient pricing of European options on two underlying assets by frame duality

    J. Math. Anal. Appl.

    (2020)
  • Cited by (20)

    • Simulation of multidimensional diffusions with sticky boundaries via Markov chain approximation

      2023, European Journal of Operational Research
      Citation Excerpt :

      Kushner & Dupuis (2001) present a comprehensive study on constructing controlled discrete-time Markov chains for controlled diffusions without boundaries or with reflecting boundaries to solve stochastic control problems. Kirkby, Nguyen, & Nguyen (2020) consider pricing multi-asset financial options under the multidimensional geometric Brownian motion (GBM) model. They apply a suitable change of variables to decorrelate the multidimensional Brownian motion and construct a CTMC to approximate each coordinate, which is independent of each other after the transform.

    • Two-factor Heston model equipped with regime-switching: American option pricing and model calibration by Levenberg–Marquardt optimization algorithm

      2023, Mathematics and Computers in Simulation
      Citation Excerpt :

      From [10], the MRS models can be investigated as extensions of hidden Markov models. For the applications of hidden Markov models to finance and insurance, see [20,37,50,60] and [57]. As is well known, the mean-reverting models are used in many stochastic volatility models.

    • Pricing of spread and exchange options in a rough jump–diffusion market

      2023, Journal of Computational and Applied Mathematics
      Citation Excerpt :

      Hurd and Zhou [13] propose a new formula for general spread option pricing based on Fourier analysis of the payoff function. Kirkby et al. [14] develop a continuous time Markov chain approximation for pricing financial derivatives, such as spread or rainbow options, which depend on systems of stochastic diffusion processes. Alfeus and Schlögl [15] compare the Hurd and Zhou [13] approach to the lower bound approximation formula of Caldana and Fusai [16].

    • Maximum likelihood estimation of diffusions by continuous time Markov chain

      2022, Computational Statistics and Data Analysis
      Citation Excerpt :

      The breadth of processes that are well approximated is large, including those encountered in finance, economics, and the physical sciences. The present work focuses on the important case of univariate diffusions, with extensions to the multivariate case left for future analysis (see for example Kirkby et al., 2020). Because the approach is immune to time-discretization bias, it can be safely applied in situations with infrequent time-sampling, such as weekly, monthly, quarterly, or yearly sampled time-series, which are all common in econometric data.

    View all citing articles on Scopus
    View full text