Stable and tempered stable distributions and processes: an overview toward trajectory simulation
Jalal Taher · 2024
Read Stable and tempered stable distributions and processes: an overview toward trajectory simulation by Jalal Taher (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-06374.txt.
Introduction
Jalal Taher surveys a family of probability laws in a 2024 paper aimed at people who need to simulate them. Stable distributions show up in many fields, from finance to physics. Their jump-heavy paths make them hard to fake on a computer, since small jumps never stop coming.
The survey first reviews what makes these laws stable and gives sampling methods with worked number examples. It then turns to tempered versions, which tame the largest jumps. A known method for drawing their steps is laid out step by step. Cost checks across time scales show where the compute burden lies. The paper serves as a road map for users who must build such paths in practice.
Catalog record
| CREATOR | Jalal Taher |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 57 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-06374
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Stable and tempered stable distributions and processes: an overview toward trajectory simulation Jalal Taher Abstract Stable distributions are a celebrated class of probability laws used in various fields. The α \alpha -stable process, and its exponentially tempered counterpart, the Classical Tempered Stable (CTS) process, are also prominent examples of Lévy processes. Simulating these processes is critical for many applications, yet it remains computationally challenging, due to their infinite jump activity. This survey provides an overview of the key properties of these objects offering a roadmap for practitioners. The first part is a review of the stability property, sampling algorithms are provided along with numerical illustrations. Then CTS processes are presented, with the Baeumer-Meerschaert algorithm [ 3 ] for increment simulation, and a computational analysis is provided with numerical illustrations across different time scales. Keywords: Stable distribution; Tempered stable distribution; Lévy processes; Process simulation. Introduction Among the many examples of Lévy processes, the \alpha -stable process stands out as a central example of processes with an infinite jump activity. Understanding these processes reduces to the study of stable distributions, a topic extensively covered in the literature. Stability was first characterized in 1925 by Levy in [ 35 ] while exploring the properties of the sum of i.i.d. random variables. Gnedenko and Kolmogorov later emphasized the potential applications of stable distributions in [ 21 ] , stating that it deserves ” the most serious attention. It is probable that the scope of applied problems in which they play an essential role will become in due course rather wide ”. There are many reasons why stable distribution and processes are attractive for applications. They encompass a wide range of behaviors; e.g. they include the Cauchy and the Gaussian distributions as special cases. They also arise in the generalized Central Limit Theorem as the only limit of normalized sums of i.i.d. random variables. Additionally, their heavy tails align with studies suggesting the importance of extreme events, providing an alternative to Gaussian models. Therefore, stable distributions and processes have found applications across various fields, including behavior studies [ 55 , 48 , 5 ] , signal processing [ 44 ] , computer science [ 54 ] . It also holds a particular place in finance since Mandelbrot in [ 38 ] discarded the historical Gaussian assumptions for asset returns and inaugurated the use of stable distributions. It has been central in pricing options [ 30 ] and modeling the price of commodities (e.g. electricity [ 58 ] , agricultural [ 28 ] ). There also has been a renewed interest in heavy-tail distributions in more recent topics in deep-learning [ 24 , 29 ] . For further details on applications, refer to [ 43 , 60 , 11 , 47 ] . However, the infinite variance of non-Gaussian stable distributions (see ( 8 )) and the inadequacy of this assumption in various models (see [ 47 ] ), led to considering alternative distributions. Methods to circumvent this issue have historically originated in physics, trying to maintain a stable-like distribution in some central region while lightening the tails so the variance is finite. Mantegna and Stanley [ 39 ] introduced the truncated Lévy flight, cutting the density function f of a stable distribution; f_{T}(x)\propto f(x)\mathbf{1}_{|x|\leq T}. While this modification yields finite moments, and resembles stable distribution for large T , it looses the infinite divisibility property (see Remark 2 ) which is a central tool for computations. Koponen in [ 32 ] solved this issue with a smooth exponential cutoff resulting in the Classical Tempered Stable (CTS) distribution. They have been discovered multiple times resulting in various terminology; smoothly truncated Lévy flight (STLF) [ 32 , 22 ] , Tempered stable (TS) [ 11 , 34 , 43 ] , KoBol in [ 4 ] , CGMY (for the symmetric version) [ 8 ] , exponentially tilted stable distributions in the sampling literature [ 27 , 13 ] . Notable limit cases include the bilateral Gamma [ 33 ] and the Variance Gamma [ 36 ] distributions. In this article, we adopt the term Classical Tempered Stable (CTS) following [ 47 ] . Finally, in [ 51 ] , Rosinski generalized the CTS distribution with different tempering functions and in higher dimensions; these are referred to as Tempered Stable distribution. The CTS discussed in this article can be seen as a specific subclass in dimension d=1 with exponential tempering function (see Chapter 6 of [ 22 ] ). The CTS distributions and processes are also widely applied in fields including actuarial science [ 23 , 25 ] cellular biology [ 46 ] , computer science [ 6 ] . But it has been extended and considerably developed in financial applications (see [ 22 , 47 ] for an overview). Simulating trajectories of stable and CTS processes is crucial for their use in applications. These processes exemplify infinite activity Lévy processes and are used in many inference models [ 59 , 41 ] . They also serve as numerical validation proxies of statistical methods [ 18 , 37 , 15 ] . In practice, once the simulation of stable and CTS distribution is achieved, a trajectory approximation of the corresponding process can be derived thanks to the independent and stationary increments property. This incorporates a temporal aspect to the problem through the sampling rate \Delta>0 which can impact the computational efficiency of algorithms. For the stable distributions, direct sampling algorithms relating to the work of Chambers, Mallows, Collin, and Stuck [ 9 ] are available and commonly used in practice [ 11 , 43 ] . Thanks to the selfsimilarity of \alpha -stable process (see 1.2 ), the sampling of X_{\Delta} can be reduced to that of X_{1} , hence the computational cost of trajectory simulation remains consistent across time scales: it is as costly in high (large \Delta ) of low (small \Delta ) frequency (see 1.3.1 ). For CTS distribution, simulation is more intricate. Lévy process methods include Compound Poisson approximation complemented with Gaussian approximation of the small jumps (see Section 6.3 of [ 11 ] , [ 2 , 7 ] for theoretical justifications, Section 4 of [ 31 ] for computational details), shot noise series representation (see [ 49 , 31 ] and Section 8.3.2 of [ 47 ] ), subordination in the symmetrical case (see Proposition 4.1 of [ 11 ] and Section 8.5.1 of [ 47 ] ). We refer to the textbook by Asmussen and Glynn [ 1 ] for a general overview of Lévy process simulation. Alternatively, rejection-based sampling algorithms can be derived from specific properties of CTS [ 12 , 3 , 14 , 27 ] . Kawai and Masuda [ 31 ] provide a comprehensive comparison of these methods, highlighting the Baeumer-Meerschaert algorithm [ 3 ] as the most efficient and practical to tune for small time increments. It is widely used for its simplicity in recent work involving CTS distributions as in [ 19 , 10 ] . This article aims to provide hands-on algorithms for simulating increments and trajectories of both \alpha -stable and CTS Lévy processes. It also serves as a road map for practitioners to navigate the interlaced concepts and algorithms provided for