Spectral estimation of the fractional order of a Lévy process

Denis BelomestnyThe Annals of Statistics 2009 · DOI

TL;DR

This paper introduces a unified spectral estimation method for determining the "fractional order" (a key parameter governing jump activity) of Lévy processes, using either historical price data or option prices. The method offers theoretical guarantees on accuracy and is demonstrated to be effective through simulations, providing a robust tool for financial modeling.

What Problem Does This Paper Address?

Financial time series often exhibit "stylized facts" like heavy tails and sudden jumps, which are well-captured by Lévy processes. A crucial parameter in these models is the "fractional order" or Blumenthal-Getoor index ($\alpha$), which quantifies the intensity and frequency of small jumps. Estimating this parameter is vital for accurate financial modeling, especially for option pricing, but it's challenging due to the complexity of Lévy processes and the nature of financial data (often low-frequency and noisy). Existing methods often struggle with unified treatment of estimation from historical data and calibration from option prices, and lack comprehensive theoretical analysis. This paper aims to bridge this gap by developing a single, theoretically sound, and practically applicable method for estimating $\alpha$.

Methods

The core of the proposed method is Spectral Estimation Method (SEM), which leverages the spectral domain (frequency domain) properties of Lévy processes. The approach involves two main steps:

  1. Estimation of the characteristic function: This step adapts to the data source. For historical data (under measure P), it uses the empirical characteristic function. For option data (under risk-neutral measure Q), it employs a Fourier technique based on option prices.
  2. Weighted least squares estimation of the fractional order: This step is unified for both estimation and calibration. It involves a transformation of the estimated characteristic function that linearizes the relationship with $\alpha$. A weighted least squares approach is then applied in the spectral domain, incorporating a "spectral cut-off regularization" to handle noise and high-frequency components effectively.

The paper also develops an adaptive procedure for choosing the regularization parameter, combining multiple testing and aggregation ideas to improve robustness. Theoretical analysis includes deriving minimax rates of convergence and proving asymptotic normality of the estimator.

Datasets

The paper does not use real-world datasets. Instead, it relies on simulated data to evaluate the performance of the proposed estimation and calibration procedures. The simulations involve generating data from generalized hyperbolic (GH) Lévy models under different parameter settings (including varying sample sizes, noise levels, and model parameters like $\kappa$ and $\delta$) for both historical and risk-neutral scenarios.

Key Insights and Ideas

  • Unified Framework: The spectral approach elegantly unifies the estimation of the fractional order from both historical price data and option prices, a significant advantage over methods that treat these separately.
  • Spectral Domain Power: By transforming the problem into the spectral domain, the authors can exploit the inherent structure of Lévy processes to simplify estimation and derive strong theoretical results.
  • Regularization is Key: The "spectral cut-off regularization" is crucial for handling noisy, low-frequency data and preventing overfitting to high-frequency noise, which is a common challenge in financial econometrics.
  • Adaptive Procedure: The proposed adaptive procedure for selecting regularization parameters is a practical advancement, making the method less sensitive to subjective choices and more applicable in real-world scenarios.
  • Ill-Posedness: The paper highlights that estimating the fractional order is an inherently "ill-posed" problem, meaning small errors in data can lead to large errors in the estimate. The derived minimax rates reflect this difficulty, but the proposed method achieves the best possible rates within the considered model class.

Key Results

  • Theoretical Guarantees: The paper provides theoretical guarantees for the proposed estimator, including derived minimax rates of convergence and proofs of asymptotic normality.
  • Effective Estimation: Simulation studies demonstrate that the spectral estimation method performs well in estimating the fractional order $\alpha$ under various conditions, both for historical data (under P) and option data (under Q). The estimates show good accuracy, especially with larger sample sizes.
  • Adaptive Procedure Performance: The adaptive procedure for choosing cut-off parameters is shown to be effective, reducing the dependence on subjective choices and improving the practical performance of the estimator.
  • Model Class Analysis: The paper analyzes the estimation problem for a broad class of "regular Lévy processes of exponential type" (RLE) and even extends the results to models with a non-zero diffusion part.

Limitations and Future Work

  • Ill-Posedness: As mentioned, the estimation problem is severely ill-posed. This means that achieving high accuracy requires a large amount of data, which might be a practical limitation.
  • Subjectivity in Parameters: While the adaptive procedure helps, some parameters (like the choice of loss function power $r$ and confidence level $\gamma$ in the adaptive procedure) remain subjective and depend on the specific application.
  • Asymptotic Results: The theoretical results are primarily asymptotic, meaning their performance for small sample sizes might differ.
  • Future Work: The paper suggests exploring more general models and potentially developing methods that are less sensitive to the ill-posed nature of the problem. Further investigation into the practical implementation and calibration of the adaptive procedure for real financial data would also be valuable.

Summary updated

Abstract

We consider the problem of estimating the fractional order of a Lévy process from low frequency historical and options data. An estimation methodology is developed which allows us to treat both estimation and calibration problems in a unified way. The corresponding procedure consists of two steps: the estimation of a conditional characteristic function and the weighted least squares estimation of the fractional order in spectral domain. While the second step is identical for both calibration and estimation, the first one depends on the problem at hand. Minimax rates of convergence for the fractional order estimate are derived, the asymptotic normality is proved and a data-driven algorithm based on aggregation is proposed. The performance of the estimator in both estimation and calibration setups is illustrated by a simulation study.
BibTeX
@article{belomestny2009spectral,
  author = {Denis Belomestny},
  title = {{Spectral estimation of the fractional order of a Lévy process}},
  journal = {The Annals of Statistics},
  year = {2009},
  doi = {10.1214/09-aos715},
}
Figure