[Paper Review] Typical Algorithms for Estimating Hurst Exponent of Time Sequence: A Data Analyst's Perspective

Hongyan Zhang, Zhiqiang Feng+2 more|arXiv (Cornell University)|Oct 29, 2023

Fractal and DNA sequence analysis4 citations

TL;DR

This paper presents a comprehensive survey of 13 dominant algorithms for estimating the Hurst exponent in time series, emphasizing practical implementation through detailed pseudo-codes and algorithmic frameworks. It demonstrates that spectrum-domain methods outperform time-domain methods in accuracy, with relative errors below 6%, and provides recommendations for both offline and real-time applications based on extensive simulations using fractional Brownian noise and human behavioral data.

ABSTRACT

The Hurst exponent is a significant metric for characterizing time sequences with long-term memory property and it arises in many fields. The available methods for estimating the Hurst exponent can be categorized into time-domain and spectrum-domain methods. Although there are various estimation methods for the Hurst exponent, there are still some disadvantages that should be overcome: firstly, the estimation methods are mathematics-oriented instead of engineering-oriented; secondly, the accuracy and effectiveness of the estimation algorithms are inadequately assessed; thirdly, the framework of classification for the estimation methods are insufficient; and lastly there is a lack of clear guidance for selecting proper estimation in practical problems involved in data analysis. The contributions of this paper lie in four aspects: 1) the optimal sequence partition method is proposed for designing the estimation algorithms for Hurst exponent; 2) the algorithmic pseudo-codes are adopted to describe the estimation algorithms, which improves the understandability and usability of the estimation methods and also reduces the difficulty of implementation with computer programming languages; 3) the performance assessment is carried for the typical estimation algorithms via the ideal time sequence with given Hurst exponent and the practical time sequence captured in applications; 4) the guidance for selecting proper algorithms for estimating the Hurst exponent is presented and discussed. It is expected that the systematic survey of available estimation algorithms could help the users to understand the principles and the assessment of the various estimation methods could help the users to select, implement and apply the estimation algorithms of interest in practical situations in an easy way.

Motivation & Objective

• To address the lack of accessible, code-ready implementations of Hurst exponent estimation algorithms in existing literature.
• To evaluate and compare the accuracy, robustness, and computational efficiency of 13 major estimation methods across diverse time series types.
• To provide a unified framework for sequence partitioning and parameter estimation that enhances algorithmic precision and reproducibility.
• To guide researchers in selecting appropriate methods based on data characteristics and application context—especially for offline vs. real-time use.
• To validate the effectiveness of the methods using synthetic fractional Gaussian noise sequences and real human behavioral reaction time data.

Proposed method

• Proposes a general sequence partitioning framework for time-domain methods based on optimal segment length and feasible grouping strategies to improve estimation stability.
• Introduces fixed-point iteration, local minimum search, and ℓ₁-norm-based linear regression to refine parameter estimation and enhance accuracy.
• Classifies estimation methods into two perspectives: sequence representation (time-domain vs. spectrum-domain) and parameter estimation (linear regression, Bayesian, etc.).
• Employs fractional Gaussian noise (FGN) to generate ideal time series with known Hurst exponents for controlled simulation experiments.
• Uses real human behavioral reaction time data from public repositories to validate method performance on empirical, non-ideal data.
• Provides detailed flowcharts and pseudo-codes for R/S and DFA methods to aid implementation and conceptual understanding.

Figure 1: Principle and implementation of optimal sequence partition

Experimental results

Research questions

• RQ1 Which estimation methods for the Hurst exponent offer the highest accuracy and robustness across diverse time series types?
• RQ2 How do spectrum-domain methods compare to time-domain methods in terms of estimation error and computational complexity?
• RQ3 What impact do different norms (ℓ₁ vs. ℓ₂) have on the accuracy of linear regression-based estimation in Hurst exponent computation?
• RQ4 How do estimation methods perform on real-world behavioral data with inherent noise and non-stationarity?
• RQ5 What are the most suitable methods for real-time versus offline analysis of long-memory time series?

Key findings

• Spectrum-domain methods consistently outperform time-domain methods, achieving a relative estimation error of less than 6% in most cases.
• For small Hurst exponents (H < 0.35), the R/S and LW methods exhibit significantly higher relative errors, exceeding 5%.
• The choice between ℓ₁-norm and ℓ₂-norm has minimal impact on estimation accuracy across tested methods.
• All 13 methods successfully detect long-term memory features in real human behavioral reaction time data, confirming their empirical applicability.
• For offline applications, the TTA method is recommended among time-domain methods, LSSD and LSV among Bayesian methods, and PM among spectrum-domain methods.
• For real-time applications, the R/S method is preferred due to its iterative computation of range and standard deviation, minimizing computational overhead.

Figure 2: Typical Algorithms for Estimating Hurst Exponent