Adaptive wavelet estimation of the diffusion coefficient under additive error measurements

M. Hoffmann, Axel Munk, Johannes Schmidt-Hieber

Journal masthead for Annales de l’Institut Henri Poincaré

Abstract

This study investigates the non-parametric estimation of the diffusion coefficient from discrete observations contaminated by additive noise. While such problems have gained prominence in high-frequency financial modeling over the last decade, research has primarily focused on parametric or semi-parametric approaches. This work addresses the estimation of the diffusion coefficient's trajectory—which may itself be stochastic—within a general framework. By integrating pre-averaging techniques with wavelet coefficient thresholding, we construct adaptive estimators that achieve quasi-optimal convergence rates across a broad spectrum of Besov-type regularity constraints. Furthermore, we propose a novel criterion to evaluate estimation quality for inherently random diffusion coefficients, recovering standard minimax theory in deterministic cases. Our approach leverages recent asymptotic equivalence results between Gaussian diffusion with additive noise and Gaussian white noise.

Introduction

In high-frequency data analysis, such as finance or functional MRI, researchers often observe a continuous-time process at discrete intervals. However, these observations are frequently obscured by "microstructure noise"—random errors arising from the observation process itself (e.g., bid-ask spreads). The challenge lies in accurately reconstructing the path of the diffusion coefficient $\sigma_t^2$, representing the intensity of random fluctuations over time. Unlike previous methods that estimate average volatility or assume a fixed parametric form, this research provides a non-parametric framework to recover the entire realized trajectory of $\sigma_t^2$, even when it is a stochastic process.

Methodology

The proposed estimation procedure follows a two-step approach designed to handle high-frequency noise and signal irregularity:

  1. Pre-averaging: To mitigate the impact of additive high-frequency noise, a local smoothing kernel is applied to the raw data. This "de-blurring" step yields a cleaner, albeit coarser, representation of the underlying process.
  2. Wavelet Representation: The diffusion coefficient is decomposed using a wavelet basis. Wavelets are localized in both time and frequency, making them uniquely suited for capturing signals with varying smoothness or abrupt transitions.
  3. Adaptive Hard Thresholding: To ensure the estimator is adaptive, a selection rule is applied where only wavelet coefficients exceeding a noise-dependent threshold are retained. This allows the model to automatically adjust to the signal's inherent smoothness without prior knowledge of its regularity.

To validate this method, the authors utilize the principle of asymptotic equivalence, demonstrating that the noisy diffusion model behaves mathematically like a simpler Gaussian white noise model. This allows for the establishment of theoretical lower bounds on estimation accuracy.

Key Results and Insights

  • The $n^{-1/4}$ Convergence Rate: While standard signal denoising often achieves an $n^{-1/2}$ rate, the presence of microstructure noise inherently slows the optimal rate to $n^{-1/4}$. The proposed wavelet estimator successfully reaches this nearly optimal benchmark.
  • Handling Stochastic Parameters: Traditional statistical theory often assumes the target parameter is a deterministic function. This work bridges the gap between classical minimax theory and stochastic volatility modeling by proposing a new criterion for achievable rates that accounts for the randomness of the diffusion coefficient.
  • Theoretical Sharpness: The estimator achieves a convergence rate of $n^{-\alpha/4}$ (where $\alpha$ relates to signal smoothness) in $L^p$ loss. These results are shown to be sharp across a wide range of Besov spaces, particularly when the smoothness $s$ satisfies specific technical conditions ($s - 1/\pi > 1.4$).

Conclusion and Future Directions

This research establishes a robust, adaptive framework for volatility path estimation under noise. However, some technical constraints remain; the lower bound proofs currently require specific relationships between smoothness and the $L^\pi$ norm which may be relaxed in future work. Additionally, while this paper focuses on the diffusion coefficient, estimating the "drift" (the underlying trend) in the presence of additive noise remains a significant challenge for future exploration in high-frequency statistics.

Summary updated

Abstract

On étudie l’estimation non-paramétrique du coefficient de diffusion à partir d’observations discrètes, lorsque les observations sont bruitées par un bruit additionnel. De tels problèmes se sont développés au cours des dix dernières années dans plusieurs champs d’application, en particuler pour la modélisation des données haute fréquence en finance, cependant plutôt d’un point de vue paramétrique ou semi-paramétrique. Ce travail concerne l’estimation de la trajectoire (éventuellement stochastique) du coefficient de diffusion dans un cadre relativement général. En développant des techniques de pré-moyennage combinées avec du seuillage des coefficients d’ondelettes, nous contruisons des estimateurs adaptatifs qui atteignent une vitesse quasi-optimale parmi une vaste échelle de contraintes de régularité de type Besov. Puisque le coefficient de diffusion est souvent intrinsèquement aléatoire, nous proposons un nouveau critère pour qualifier la qualité d’estimation ; nous retrouvons la théorie minimax usuelle lorsque cette approche est restreinte à un coefficient de diffusion déterministe. En particulier, on exploite les résultats récents de Reiß (Ann. Statist. 39 (2011) 772–802) de l’équivalence asymptotique entre une diffusion gaussienne avec un bruit additif et le bruit blanc gaussien.
BibTeX
@inproceedings{hoffmann2012adaptive,
  author = {M. Hoffmann and Axel Munk and Johannes Schmidt-Hieber},
  title = {{Adaptive wavelet estimation of the diffusion coefficient under additive error measurements}},
  booktitle = {Annales},
  year = {2012},
  doi = {10.1214/11-aihp472},
}
Figure