Empirical Likelihood Methods Based on Characteristic Functions With Applications to Lévy Processes

Ngai Hang Chan, Song Xi Chen, Liang Peng, Cindy YuJournal of the American Statistical Association 2009 · DOI

TL;DR

This paper introduces a novel application of empirical likelihood methods to characteristic functions, enabling robust parameter estimation and hypothesis testing for Lévy processes, which are crucial for modeling financial asset prices with jumps. The proposed methods offer a powerful, self-studentizing approach to analyze complex financial dynamics without relying on explicit density function estimation.

What Problem Does This Paper Address?

Financial modeling often relies on stochastic processes to describe the evolution of asset prices over time. Lévy processes are particularly attractive because they can capture the discontinuous jumps observed in real-world financial markets, a feature absent in simpler models like the Black-Scholes model. However, analyzing these processes presents challenges: their underlying probability density functions are often intractable, and traditional statistical inference methods can be sensitive to model misspecification. This paper addresses the need for flexible and robust statistical tools to estimate parameters and test hypotheses about Lévy processes, specifically focusing on their characteristic functions, which are readily available and contain complete information about the process. The authors aim to develop methods that can handle the complexities of jump processes and avoid the pitfalls of density estimation.

Methods

The core contribution of this paper is the development and application of Empirical Likelihood (EL) methods to characteristic functions for inference on Lévy processes. EL is a powerful nonparametric technique that constructs likelihood ratios without assuming a specific parametric form for the underlying distribution.

Here's a breakdown of their approach:

  • Maximum Empirical Likelihood Estimator (MELE): The authors adapt the standard EL approach for estimating means to the context of characteristic functions. They define an empirical likelihood function based on the observed increments of the Lévy process and then maximize it to obtain parameter estimates. This method is "self-studentizing," meaning it inherently accounts for the variability of the data without needing to explicitly estimate the variance.
  • Goodness-of-Fit Tests: They propose EL-based tests for:
  • Parametric Class Specification: Testing whether the observed data's characteristic function belongs to a specific parametric family (e.g., a process with or without jumps).
  • Symmetry: Testing if the underlying distribution of the Lévy process increments is symmetric around an unknown mean.
  • Integrated Distance Approach: Instead of relying on a finite number of points, their tests are based on an integrated distance between the empirical characteristic function and the hypothesized parametric function. This is a more robust approach for continuous processes.
  • Bootstrap for Critical Values: Due to the complexity of the limiting distributions of their test statistics, the authors propose a bootstrap method to obtain critical values, making the tests practical.

Datasets

The paper uses simulated data from three common financial models for evaluation:

  1. Black-Scholes (BS) Model: A continuous-time model without jumps.
  2. Black-Scholes Model with Merton Jumps (BS-MJ): A finite-activity jump process.
  3. Variance Gamma (VG) Model: An infinite-activity jump process.

For the case study, they use continuously compounded returns of the S&P500 index from January 2, 1987, to December 31, 2007, sampled at a weekly frequency.

Key Insights and Ideas

  • EL as a Unified Framework: The paper demonstrates that empirical likelihood can serve as a unifying framework for both parameter estimation and hypothesis testing in the context of characteristic functions, offering a more elegant and robust approach than traditional methods.
  • Characteristic Functions as a Powerful Inference Tool: The study highlights the utility of characteristic functions, which are often analytically available for Lévy processes, as a direct and efficient route for statistical inference, bypassing the difficulties associated with estimating often intractable density functions.
  • Self-Studentizing Property: The EL approach's inherent self-studentizing nature is a significant advantage, as it eliminates the need for separate variance estimation, simplifying the inference process and reducing potential sources of error.
  • Flexibility for Jump Processes: The methods are shown to be effective in distinguishing between continuous processes and those with jumps, a critical capability for accurate financial modeling.

Key Results

  • Parameter Estimation: Simulation studies show that the proposed Maximum Empirical Likelihood Estimator (MELE) performs well, yielding estimates close to Maximum Likelihood Estimators (MLEs) for the Black-Scholes model and demonstrating good performance compared to Method of Moments estimators (MME) for the Variance Gamma model. The EL estimates are generally more efficient, especially for smaller sample sizes.
  • Goodness-of-Fit Tests:
    • Test 1 (BS vs. BS-MJ): The test for detecting jumps (distinguishing BS from BS-MJ) shows very high power (up to 97% for n=500), indicating its effectiveness in identifying jump components.
  • Test 2 (Symmetric VG vs. Asymmetric VG): The test for symmetry in the VG model shows moderate power (17-70%), suggesting it can detect asymmetry but might require larger sample sizes or more pronounced asymmetry.
  • Test 3 (Finite vs. Infinite Activity Jumps): The test distinguishing finite (BS-MJ) from infinite (VG) activity jumps also shows moderate power (17-71%).
  • Case Study (S&P500): The empirical analysis of S&P500 returns suggests that:
  • The classical Black-Scholes model is not a suitable fit.
  • A symmetric Variance Gamma model is rejected, indicating the presence of skewness.
  • Including jumps (both finite and infinite activity) significantly improves the model fit, suggesting that jump processes are essential for capturing the dynamics of S&P500 returns.

Limitations and Future Work

  • Tuning Parameter 'a': The proposed methods require specifying a tuning parameter 'a' which defines the range of the characteristic function to be considered. While the paper suggests that moderate values work well for many models and that the results are not overly sensitive to 'a', its optimal selection can still be a consideration.
  • Bootstrap Computational Cost: While the bootstrap method is effective for obtaining critical values, it can be computationally intensive, especially for large datasets or complex models.
  • Dependent Data: The current framework is primarily for independent and identically distributed (iid) data. The authors mention that their methods are also applicable to dependent data using blockwise empirical likelihood, but this will be dealt with in a separate paper. This suggests a clear avenue for future research.
  • Further Model Specifications: The paper focuses on specific tests (symmetry, jump presence). Future work could extend these methods to test for other specific characteristics of Lévy processes or more complex parametric families.

Summary updated

Abstract

Abstract Lévy processes have been receiving increasing attention in financial modeling. One distinctive feature of such models is that their characteristic functions are readily available. Inference based on characteristic functions is very useful for studying Lévy processes. By incorporating the recent advances in nonparametric approaches, empirical likelihood methods based on characteristic functions are developed in this paper for parameter estimation, testing a particular parametric class including the presence of a jump component in the Lévy process and testing for symmetry of a distribution. Simulation and case studies confirm the effectiveness of the proposed method. Keywords: : Characteristic functionEmpirical likelihoodGoodness-of-fit testLévy processes
BibTeX
@article{chan2009empirical,
  author = {Ngai Hang Chan and Song Xi Chen and Liang Peng and Cindy Yu},
  title = {{Empirical Likelihood Methods Based on Characteristic Functions With Applications to Lévy Processes}},
  journal = {Journal of the American Statistical Association},
  year = {2009},
  doi = {10.1198/jasa.2009.tm08349},
}
Figure