Core Concepts
This paper develops a fast and exact algorithm for simulating the first passage time, undershoot, and overshoot of a tempered stable subordinator over an arbitrary non-increasing absolutely continuous function. The algorithm has finite exponential moments and its expected running time is bounded explicitly in terms of the model parameters.
Abstract
The paper presents an algorithm called TSFFP-Alg that can efficiently and exactly simulate the first passage time (τb), undershoot (Sτb-), and overshoot (Sτb) of a tempered stable subordinator S over an arbitrary non-increasing absolutely continuous function b(t).
The key steps are:
- TSFFP-Alg reduces the problem to the stable case (q=0) using a rejection sampling approach based on the Esscher transform.
- The stable case is handled by SFP-Alg, which first samples the crossing time τb conditional on τb ≤ t*, then determines if the crossing occurred via a jump or by creeping.
- The most technically challenging part is the simulation of the undershoot Sτb- in the jump case, which is achieved by SU-Alg. SU-Alg samples from the undershoot density by breaking up the state space and using tailored accept-reject algorithms on each subinterval.
The authors provide a detailed complexity analysis, showing that the expected running time of TSFFP-Alg has finite exponential moments and grows at most cubically in the stability parameter 1/α as it approaches 0 or 1. The running time is also linear in the tempering parameter q and the initial value of the function b(0).
Numerical experiments demonstrate the good agreement between the theoretical complexity bounds and the observed performance of the implemented algorithms.
Stats
The paper does not contain any explicit numerical data or statistics to support the key claims. The analysis focuses on the theoretical complexity of the algorithms.
Quotes
"The expected running time grows at most cubically in the stability parameter (as it approaches either 0 or 1) and is linear in the tempering parameter and the initial value of the function."
"The running time of our algorithm has finite exponential moments and provide bounds on its expected running time with explicit dependence on the characteristics of the process and the initial value of the function."