The Properties of Fractional Gaussian Process and Their Applications

Yong Chen, Ying LiarXiv 2023 · DOI · arXiv

TL;DR

This paper provides a rigorous, unifying definition for the "fractional Gaussian process" (fGp), a term previously used loosely in stochastic literature. By anchoring the definition to the covariance structure of fractional Brownian motion (fBm), the authors establish a framework that encompasses seven existing types of Gaussian processes and prove that these processes obey fundamental limit theorems and statistical estimation bounds.

What Problem Does This Paper Address?

In the world of stochastic processes, fractional Brownian motion (fBm) is a superstar—it’s used to model everything from financial market volatility to the firing of neurons because it captures "long-range dependence" (where what happens now depends on what happened long ago). However, researchers often encounter "fractional-like" processes—such as sub-fractional or bi-fractional Brownian motion—that share some properties with fBm but aren't quite the same.

Until now, the term "fractional Gaussian process" was used as a catch-all without a precise mathematical definition. This lack of rigor made it difficult to generalize results across different models. The authors set out to create a formal definition for fGps that is broad enough to include existing models but specific enough to allow for powerful mathematical analysis, particularly in the realm of statistical estimation.

Methods

The authors define an fGp based on its covariance function $R(s, t)$. Specifically, a process is an fGp if the difference between its covariance and the standard fBm covariance is "smooth" enough—meaning its first-order partial derivative is a normalized function of bounded variation.

To explore the properties of this new class, the authors employ several high-level mathematical tools:

  • Reproducing Kernel Hilbert Spaces (RKHS): They quantify the relationship between the "function spaces" associated with fGps and fBm.
  • Lebesgue-Stieltjes Measures: They use measure theory to handle cases where the process might have jumps or non-continuous derivatives.
  • Malliavin Calculus & The Fourth Moment Theorem: To derive Berry-Esséen bounds (which tell us how fast a statistical estimator converges to a normal distribution), they use modern techniques in stochastic analysis that relate high-order moments of a process to its convergence rate.

Datasets

As this is a theoretical mathematics paper focused on proving theorems and establishing definitions, no empirical datasets were used. The "data" consists of mathematical proofs and the "evaluation" is the verification of these proofs against existing stochastic models (like the bi-fractional Brownian motion).

Key Insights and Ideas

  • The "Common Term" Anchor: The authors realized that most fractional-like processes in literature share a common term in their covariance: $-\frac{1}{2}|t-s|^{2H}$. By focusing on the difference between a new process and this benchmark, they created a universal yardstick for the entire class.
  • Unification of Seven Models: The paper demonstrates that seven distinct Gaussian processes previously studied in isolation (including sub-fractional, bi-fractional, and generalized fractional Brownian motions) all fall under this new fGp definition. This turns a fragmented field into a cohesive one.
  • Relaxing Continuity: A major "aha moment" is the realization that Gladyshev’s theorem (a classic result for identifying the parameters of a process) can hold even if the mixed second-order partial derivatives of the covariance are not continuous everywhere. This allows the theory to handle more "rugged" or discontinuous processes.

Key Results

  • RKHS Equivalence: The authors provide an explicit formula (Theorem 2.1) relating the inner products of the Hilbert spaces of fGps and fBm, which is crucial for signal processing and machine learning applications.
  • Gladyshev’s Theorem: They prove that for any fGp, one can perfectly recover the Hurst parameter $H$ (which measures long-term memory) from a single path of the process as the sampling frequency goes to infinity.
  • Optimal Convergence Rates: For the fractional Ornstein-Uhlenbeck process (a common model for mean-reverting systems), the authors show that the Least Squares Estimator (LSE) and the moment estimator converge to the true parameter at a rate of $O(1/\sqrt{T})$. This is the "gold standard" rate for statistical consistency.

Limitations and Future Work

The authors acknowledge that while their definition is broad, it is still "linked" to the structure of fractional Brownian motion. In Section 6, they briefly discuss a variant where the covariance is not linked to fBm, suggesting that even broader classes of Gaussian processes could be studied using similar techniques.

Additionally, they note that for certain highly discontinuous examples (like Example 1.3), the moment estimator does not perform as well as the Least Squares Estimator. Investigating why some estimators are more robust to discontinuities than others remains an open path for future research.

Summary updated

Abstract

The process $(G_t)_{t\in[0,T]}$ is referred to as a fractional Gaussian process if the first-order partial derivative of the difference between its covariance function and that of the fractional Brownian motion $(B^H_t)_{t\in[0,T ]}$ is a normalized bounded variation function. We quantify the relation between the associated reproducing kernel Hilbert space of $(G)$ and that of $(B^H)$. Seven types of Gaussian processes with non-stationary increments in the literature belong to it. In the context of applications, we demonstrate that the Gladyshev's theorem holds for this process, and we provide Berry-Esséen upper bounds associated with the statistical estimations of the ergodic fractional Ornstein-Uhlenbeck process driven by it. The second application partially builds upon the idea introduced in \cite{BBES 23}, where they assume that $(G)$ has stationary increments. Additionally, we briefly discuss a variant of this process where the covariance structure is not entirely linked to that of the fractional Brownian motion.
BibTeX
@misc{chen2023properties,
  author = {Yong Chen and Ying Li},
  title = {{The Properties of Fractional Gaussian Process and Their Applications}},
  year = {2023},
  eprint = {2309.10415},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2309.10415},
}

Loading related content...

Figure