The hybrid exact scheme for the simulation of first-passage times of jump-diffusions with time-dependent thresholds
Sascha Desmettre, Devika Khurana, Amira Meddah · 2025
Read The hybrid exact scheme for the simulation of first-passage times of jump-diffusions with time-dependent thresholds by Sascha Desmettre, Devika Khurana, Amira Meddah (2025) 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-2511-00155.txt.
Introduction
Sascha Desmettre, Devika Khurana, and Amira Meddah develop a hybrid exact scheme for simulating first-passage times of jump-diffusion processes with time-dependent thresholds. First-passage time, when a process first hits a barrier, is central in finance and risk modeling. Exact simulation exists for constant thresholds, applying exact diffusion simulation between jumps, and an adaptation to moving thresholds was proposed in a more general setting.
They show that adaptation applies to jump-diffusion models by establishing a formal correspondence. Probability and finance readers get exact, bias-free simulation where thresholds move, as with time-varying barriers in options or default models. Moving thresholds are realistic, and exactness avoids discretization error.
Catalog record
| CREATOR | Sascha Desmettre, Devika Khurana, Amira Meddah |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 48 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-00155
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
The hybrid exact scheme for the simulation of first-passage times of jump-diffusions with time-dependent thresholds Sascha Desmettre, Devika Khurana, Amira Meddah Abstract The first-passage time is a key concept in stochastic modeling, representing the time at which a process first reaches a specified threshold. In this work, we consider a jump–diffusion (JD) model with a time-dependent threshold, providing a more flexible framework for describing stochastic dynamics. We are interested in the Exact simulation method developed for JD processes with constant thresholds, where the Exact method for pure diffusion is applied between jump intervals. An adaptation of this method to time-dependent thresholds has recently been proposed for a more general stochastic setting. We show that this adaptation can be applied to JD models by establishing a formal correspondence between the two frameworks. A comparative analysis is then performed between the proposed approach and the constant-threshold version in terms of algorithmic structure and computational efficiency. Finally, we show the applicability of the method by predicting neuronal spike times in a JD model driven by two independent Poisson jump mechanisms. Keywords— First-passage time, Exact simulation, Jump diffusion, Time- dependent thresholds, Hybrid rejection sampling, Spike times MSC 2010 Classification —- 37M05, 65C20, 60G05, 60H35, 92C20 1 Introduction Stochastic processes are widely used to model dynamical systems in various fields to capture inherent randomness and uncertainty. A key quantity of interest in such systems is the first-passage time (FPT), representing the time at which a process first reaches a specified threshold. This concept arises naturally in various applications, for instance, in finance, it appears in the pricing of barrier options barone2008barrier ; in population dynamics, it characterizes the time for a population to reach a critical level goel2013stochastic ; and in oncology, it is used to model disease progresses beyond a critical stage roman2021using . In neuroscience, stochastic modeling plays a fundamental role in describing the intrinsic variability of neural activity resulting from complex biophysical mechanisms. Such models have a wide range of applications, including the characterization of firing rates, interspike interval distributions, and neural coding variability. Among these, a diffusion process is often used to describe the evolution of the membrane potential of a neuron braun2015first ; ermentrout2010mathematical . When this potential reaches a certain threshold, a spike is generated, and the information is transmitted to the next neuron or target cell. The time at which the membrane potential first reaches this threshold, i.e., the spike time, can be mathematically formulated as the FPT of the stochastic process. To account for larger input contributions that cannot be adequately approximated within the diffusion framework, JD models were later introduced, lansky2008review ; giraudo2002effects ; giraudo1997jump . These models incorporate discontinuous jumps in addition to the continuous diffusion component, allowing for a more realistic representation of neuronal activity influenced by discrete input events. Analytical expressions for the FPT density of JD processes are generally unavailable, except in specific cases such as when the jump size follows a doubly exponential distribution kou2003first or is non-negative with the boundary located below the initial value blake2003level . The main challenge arises from the possibility of an overshoot when the process crosses the boundary. While the overshoot distribution is tractable for exponential jumps due to the memoryless property, it becomes difficult to handle for more general jump size distributions. When analytical results are unavailable, simulation-based approaches are often employed. One possibility is to simulate full sample paths using numerical methods such as Runge–Kutta or Euler-type schemes buckwar2011runge ; higham2005numerical , and compute the FPT as a by-product. Another approach avoids modeling jumps explicitly by considering a piecewise-defined threshold; in abundo2010first , the author shows that the hitting time in this setting satisfies a certain integral equation, which is then solved numerically. A different line of work, introduced in atiya2005efficient , proposes two Monte Carlo methods for Brownian motion. A key distinction of this method is that it requires only a few simulation points per iteration and avoids discretization bias. We find the exact simulation approach particularly appealing, as it does not involve time-discretization error. This method was originally developed for pure diffusion processes, cf. herrmann2019exact ; khurana2024exact . The main idea relies on Girsanov’s transformation, which ensures that the FPT of a diffusion process has the same law as the FPT of a standard Wiener process under a different probability measure. The algorithm samples candidate FPTs from the Wiener process and accepts or rejects them with a probability derived from Girsanov’s theorem. To adapt this method to the JD setting, an algorithmic structure was established in herrmann2023exact that applies the original idea between jumps, considering a constant threshold. In many practical settings, however, the boundary is time-dependent. This is particularly relevant in neuroscience, where adaptive thresholds provide more realistic firing dynamics. To address such cases, in desmettre2025first , the exact simulation framework was extended to piecewise diffusion Markov processes (PDifMPs) with time-dependent boundaries. In this work, we establish a correspondence between PDifMPs and JD processes, allowing the exact simulation scheme developed in desmettre2025first to be applied in the JD setting. The structure of the paper is as follows. Section 2 establishes the correspondence between the JD model and the PDifMP formulation. Section 3 presents the Exact simulation method applied from desmettre2025first to our setting. Section 4 analyzes the computational properties of the proposed approach and compares it with the constant-threshold case. Finally, Section 5 illustrates the applicability of the method through a stochastic neuron model with two independent sources of jumps. 2 Model structure and reformulation via PDifMPs In this section, we build upon the work presented in herrmann2023exact and reformulate their JD model within the PDifMP framework. While both settings feature a combination of continuous dynamics and discontinuous jumps, a PDifMP provides a more structured and hybrid representation, in which the continuous and discrete components evolve on well-separated time scales and are governed by distinct rules. In particular, the continuous dynamics evolve according to a stochastic differential equation (SDE) between random jump times, while the discrete changes are governed by a state-dependent jump rate and a transition kernel. This separation of dynamics allows the model to be interpreted as a hybrid Markov process with piecewise diffusion trajectories interrupted by stochastic resets. For a broader theoretical foundation of such hybrid representations and their applications, we refer the reader to meddah2024stochastic ; bect2007processus ; buckwar2023stochastic ; buckwar2025american ; buckwar2025numerical . This approach is particularly advantageous because it allows us to treat the continuous dynamics and the jump mechanisms on separate, independent intervals, thereby reducing the complexity of finding the FPT for the JD SDE and connecting it to the framework introduced in khurana2024exact ; desmettre2025first . We start by recalling the JD model introduced in herrmann2023exact . Let \big(\Omega,\mathcal{F},(\mathcal{F}_{t})_{t\geq 0},\mathbb{P}\big) be a filtered probability space satisfying the usual conditions and let \mathcal{E}\subset\mathbb{R}\setminus\{0\} denotes the mark space, that is, the set that defines the possible locations or outcomes of the jumps, bruti2007strong . Consider a Poisson random measure p_{\phi}(d\eta\times dt) on \mathcal{E}\times[0,\infty) with intensity measure \phi(d\eta)dt , where \phi is a finite, non-negative measure. The associated total jump intensity is \lambda=\phi(\mathcal{E}) . The dynamics of a JD process Z_{t} are governed by the following jump SDE dZ_{t}=\mu(t,Z_{t-})\,dt+\sigma(t,Z_{t-})\,dB_{t}+\int_{\mathcal{E}}j\big(t,\,Z_{t-},\,\eta\big)\,p_{\phi}(d\eta\times dt),\hskip 9.24994ptZ_{0}=z_{0}\in\mathbb{R}. (1) Here, B_{t} is