Statistical skorohod embedding problem and its generalizations

Denis Belomestny, John Schoenmakers

TL;DR

This paper tackles the challenging problem of inferring the distribution of a random time $T$ when we only observe a time-changed Lévy process. The authors introduce a novel statistical Skorohod embedding problem and develop a method using Mellin and Laplace transforms to estimate the distribution of $T$, proving that their method achieves optimal convergence rates. This work has implications for understanding complex stochastic processes and has applications in financial modeling and time-series analysis.

What Problem Does This Paper Address?

The core problem this paper addresses is the statistical Skorohod embedding problem (SSE). In essence, imagine you have a random process (like a Lévy process, a generalization of Brownian motion) and a random "clock" that dictates how long the process runs. We can't directly observe the clock ($T$), but we can observe the process at the time the clock stops ($L_T$). The goal is to figure out the distribution of the clock ($T$) just by looking at many independent observations of the stopped process ($L_T$).

The classical Skorohod embedding problem deals with deterministic processes and finding a stopping time. However, the statistical version, where we have noisy observations and a random underlying process, is much harder. Existing methods often rely on specific properties of Brownian motion, which don't generalize to Lévy processes. This paper aims to bridge this gap by developing a statistically sound method for inferring the distribution of the random time $T$ for general Lévy processes.

Methods

The authors' approach is rooted in the powerful tools of Mellin and Laplace transforms. Here's a breakdown of their key techniques:

  • Mellin Transform: This is a generalization of the Fourier transform, particularly useful for problems involving products of random variables. Since the observed data $X = L_T$ can be thought of as a product-like structure (where $L$ has a scaling property related to $T$), the Mellin transform becomes a natural tool. The paper defines and utilizes the Mellin transform of probability densities to relate the distribution of $L_T$ to the distribution of $T$.
  • Laplace Transform: This transform is also employed, especially when generalizing to time-changed Lévy processes. It helps in characterizing the behavior of the underlying Lévy process.
  • Multiplicative Deconvolution: The SSE problem is framed as a multiplicative deconvolution problem. The Mellin transform is instrumental in solving such problems, allowing the authors to "unmix" the contributions of the Lévy process and the random time.
  • Consistent Estimator: They propose a specific estimator for the density of $T$, denoted as $p_{T,n}(x)$. This estimator is constructed using the empirical Mellin transform of the observed data.
  • Regularization: A crucial step involves regularizing the inverse Mellin transform. This is necessary because directly using the empirical Mellin transform can lead to unstable estimates. They introduce a kernel-based approach for this regularization.
  • Convergence Rates and Optimality: A significant contribution is the derivation of the convergence rates of their estimator. They prove that these rates are "optimal in the minimax sense," meaning no other estimator can perform better asymptotically, up to logarithmic factors. The rates depend on the decay properties of the Mellin transform of $T$.
  • Generalization to Lévy Processes: The framework is extended beyond Brownian motion to general Lévy processes, which requires a more sophisticated combination of Mellin and Laplace transforms.

Datasets

Not discussed in the paper. The paper focuses on theoretical development and uses simulated data for illustrative examples.

Key Insights and Ideas

  • Mellin Transform as the Key: The paper highlights the Mellin transform as a superior tool compared to the Fourier transform for problems involving multiplicative deconvolution, which is central to the Skorohod embedding problem.
  • Optimality of Rates: The demonstration that their estimator achieves minimax optimal rates is a significant theoretical result, suggesting that their approach is as efficient as possible.
  • Dependence on Mellin Transform Decay: A crucial insight is that the speed at which the estimator converges depends heavily on how quickly the Mellin transform of the random time $T$ decays. Faster decay leads to faster convergence.
  • Generalization is Feasible: The successful extension of the method to general Lévy processes, which lack the simple scaling properties of Brownian motion, is a major advancement.
  • Applications Beyond Theory: The paper explicitly connects these theoretical results to practical problems like estimating variance-mean mixture models and analyzing time-changed Lévy processes, demonstrating the broad applicability of their framework.

Key Results

  • Consistent Estimator: The paper proposes a consistent estimator for the density of the random time $T$.
  • Optimal Convergence Rates: For the standard Skorohod embedding problem with Brownian motion, the authors derive polynomial convergence rates for their estimator. They show these rates are optimal in the minimax sense.
  • Generalization to Lévy Processes: For generalized Skorohod embedding with Lévy processes, they obtain similar optimal convergence rates, demonstrating the robustness of their approach.
  • Examples: The paper illustrates the methodology with examples from Gamma densities and specific variance-mean mixture models, providing numerical evidence of the estimator's performance.

Limitations and Future Work

  • Logarithmic Factors: While the rates are shown to be optimal in a minimax sense, they often coincide up to a logarithmic factor. The paper doesn't explicitly state whether these logarithmic factors can be removed.
  • Assumptions on Mellin Transform: The convergence rates depend on the decay of the Mellin transform of $T$. The paper defines classes of functions based on these decay properties (e.g., $C(\beta, \gamma, L)$ and $D(\beta, \gamma, L)$), and the performance is analyzed within these classes. Extending the results to broader classes of distributions might be a future direction.
  • Dependence in Time-Changed Processes: While the paper mentions generalizing to dependent data in time-changed Lévy processes, the detailed analysis of such dependent cases is not fully presented.
  • Practical Implementation: While the theoretical framework is strong, practical implementation details and performance on real-world, noisy data would be valuable future work.

The authors suggest that their results can be applied to various statistical inference problems involving complex stochastic processes, indicating a rich area for future research.

Summary updated

Abstract

Given a Levy process L, we consider the so-called statistical Skorohod embedding problem of recovering the distribution of an independent random time T based on i.i.d. sample from LT : Our approach is based on the genuine use of the Mellin and Laplace transforms. We propose a consistent estimator for the density of T; derive its convergence rates and prove their optimality. It turns out that the convergence rates heavily depend on the decay of the Mellin transform of T: We also consider the application of our results to the problem of statistical inference for variance-mean mixture models and for time-changed Levy processes.
BibTeX
@inproceedings{belomestny2014statistical,
  author = {Denis Belomestny and John Schoenmakers},
  title = {{Statistical skorohod embedding problem and its generalizations}},
  booktitle = {Technische},
  year = {2014},
  doi = {10.17877/de290r-15565},
}

Loading related content...

Figure