On Time-subordinated Brownian Motion Processes for Financial Markets

Basic Information

• Paper ID: 2510.14108
• Title: On Time-subordinated Brownian Motion Processes for Financial Markets
• Authors: Rohan Shenoy (Imperial College London), Peter Kempthorne (MIT)
• Classification: q-fin.MF (Mathematical Finance), math.ST (Statistics Theory), stat.TH (Statistics Theory)
• Publication Date: October 17, 2025
• Paper Link: https://arxiv.org/abs/2510.14108

Abstract

The core objective of this paper is to propose Fourier methods for modeling stochastic time variation in time-subordinated Brownian motion models. The paper reviews Gaussian variance-mean mixture distributions and time-subordination models, using the Gamma process as a key example. A nonparametric characteristic function decomposition method for subordinated Brownian motion is proposed, enabling direct characterization and study of stochastic time variation from the complete process. Finally, an empirical decomposition example using S&P 500 log returns is provided, with the variance Gamma process serving as the central case study throughout.

Research Background and Motivation

Problem Background

1. Limitations of the Black-Scholes Model: The traditional geometric Brownian motion (GBM) model assumes that log returns of stock prices follow a normal distribution. However, empirical research demonstrates that actual financial data often exhibits "fat-tail" characteristics, inconsistent with the normality assumption.
2. Concept of Time Subordination: To improve the GBM model, researchers proposed using alternative time indicators (such as trading volume or transaction frequency) to replace calendar time, with this change in time scale referred to as "time subordination."
3. Advantages of Stochastic Time Subordination: Research by Madan and Milne (1991) and Madan et al. (1998) demonstrates that the variance Gamma model (a type of stochastic time-subordination model) outperforms geometric Brownian motion in option pricing.

Theoretical Foundation

• Skorokhod and Monroe Theorem: Any arbitrage-free model can be represented as Brownian motion with stochastic time variation, providing a solid theoretical foundation for studying such models.
• Dubins-Schwarz Theorem: Any continuous local martingale can be represented as a time-changed Brownian motion.

Research Motivation

Existing methods primarily study time-subordination processes through quadratic variation. This paper proposes a novel Fourier method that can directly decompose the subordination process from price observation data without computing quadratic variation.

Core Contributions

1. Proposed Variance-Mixture Transform: Defined a transformation method for directly extracting the stochastic variance distribution from Gaussian variance-mean mixture distributions.
2. Constructed Time-Change Transform: Developed a theoretical framework capable of directly converting time-subordinated Brownian motion processes into their subordination processes.
3. Established Nonparametric Decomposition Method: Provided a nonparametric method based on characteristic functions to characterize stochastic time variation without relying on specific parametric assumptions.
4. Proved Theoretical Existence of Transforms: Rigorously proved the mathematical existence of the proposed transforms through analytic characteristic function theory.
5. Provided Empirical Application: Verified the method's effectiveness on S&P 500 index data, demonstrating the characteristics of time-subordination processes under different holding periods.

Methodology Details

Task Definition

Given a time series of log returns $\{X_t\}$ for a financial asset, the objective is to identify and estimate the underlying subordination process $\{\tau_t\}$ such that: $X_t = \theta\tau_t + \sigma W_{\tau_t}$ where $W_t$ is a standard Wiener process, $\theta$ is the drift parameter, and $\sigma$ is the volatility parameter.

Core Theoretical Framework

1. Gaussian Variance-Mean Mixture Distribution

Definition 1: A Gaussian variance-mean mixture distribution is the increment distribution of a Brownian motion process over random time increments: $X = \theta V + \sigma\sqrt{V}Z$ where $V$ is a non-negative random variable (mixture distribution) and $Z \sim N(0,1)$ is a standard normal random variable.

Characteristic Function: $\psi_X(t) = E_V[e^{(it\theta-t^2\sigma^2/2)V}]$

2. Variance-Mixture Transform

Definition 2: For a Gaussian variance-mixture $X = \theta V + \sqrt{V}Z$, the variance-mixture transform is defined as: $V_\theta[X](\xi) := \psi^{-1}\{E[e^{(-\theta+\sqrt{\theta^2+2i\omega})X}]\}(\xi)$

This transform recovers the underlying variance distribution $V$ from the observed mixture distribution $X$.

3. Time-Subordinated Brownian Motion Process

Definition 5: A time-subordinated Brownian motion process is defined as: $X_t = b_{\tau_t} = \theta\tau_t + \sigma W_{\tau_t}$ where $\{\tau_t\}$ is an independent subordination process.

4. Time-Change Transform

Definition 6: For a time-subordinated Brownian motion $X_t = \theta\tau_t + W_{\tau_t}$, the time-change transform is defined as: $S_\theta[X](\xi,t) := \psi^{-1}\{E[e^{(-\theta+\sqrt{\theta^2+2i\omega})X_t}]\}(\xi)$

Theoretical Innovations

1. Fourier Domain Analysis: Established direct relationships between subordinated Brownian motion and its subordination process in Fourier space.
2. Analytic Characteristic Function Extension: Extended characteristic functions from the real axis to strip regions in the complex plane, proving the existence of complex domain integrals.
3. Nonparametric Identification: Directly inferred from data without assuming a specific distribution form for the subordination process.

Experimental Setup

Dataset

• Data Source: S&P 500 index daily closing price data
• Time Period: January 2022 to January 2024
• Data Processing: Converted to log return sequences
• Holding Periods: Analyzed returns over various holding periods (1, 5, 10, 15, 20 days, etc.)

Evaluation Methods

1. Distribution Fit Testing: Used Shapiro-Wilk normality test to assess normality of raw returns
2. Gamma Distribution Fitting: Tested whether inferred subordination processes conform to Gamma distribution
3. Lévy Process Property Verification: Verified whether subordination processes satisfy stationary independent increment properties

Implementation Details

• Employed empirical characteristic function estimation
• Performed characteristic function inversion via discrete Fourier transform (DFT)
• Set $\theta = 0$ (semimartingale assumption) for primary analysis

Experimental Results

Main Findings

1. Variance Distribution Identification: Through variance-mixture transform $V_0[X]$ analysis of S&P 500 data, the inferred stochastic variance distribution shows high consistency with the Gamma distribution (Figure 1).
2. Time-Subordination Process Characteristics:
   • Inferred subordination process distributions across different time increments approximately follow Gamma distribution (Figure 4)
   • As time increments increase, both mean and variance of the distribution increase correspondingly (Figures 5-6)
3. Deviations from Lévy Properties:
   • Theoretically, Lévy subordination processes should satisfy linear relationships between mean/variance and time
   • Empirical results show growth rates declining over time, suggesting simple Lévy subordination assumptions may be overly restrictive

Quantitative Results

• Gamma distribution parameter estimates across different holding periods show reasonable evolution patterns
• Mean parameters grow with time but at decreasing rates
• Variance parameters similarly exhibit nonlinear growth patterns

Case Analysis

Figure 2 presents a simulation example of the variance Gamma process, clearly illustrating:

• How time subordination causes time-axis compression of Brownian motion
• High volatility periods corresponding to time-axis compression and faster "time speed"
• Low volatility periods corresponding to time-axis expansion and slower "time speed"

Related Work

Development of Time-Subordination Models

1. Clark (1973): First proposed using trading volume as a time-subordination indicator
2. Madan & Milne (1991), Madan et al. (1998): Developed the variance Gamma model
3. Barndorff-Nielsen et al.: Studied infinite divisibility of normal variance-mean mixture distributions

Theoretical Foundation

• Skorokhod (1961), Monroe (1972): Established theoretical foundations for embedding arbitrary semimartingales in Brownian motion
• Dubins & Schwarz (1965): Time-change representation theorem for continuous local martingales

Innovations in This Paper

Compared to existing work, this paper's primary innovation lies in providing direct Fourier methods for decomposing time-subordination processes without computing quadratic variation.

Conclusions and Discussion

Main Conclusions

1. Method Effectiveness: The Fourier method effectively decomposes time-subordination processes from financial time series
2. Gamma Process Applicability: Empirical results support using Gamma processes as time-subordination processes for stock return rates
3. Model Limitations: Simple Lévy subordination assumptions may be overly restrictive for actual data

Limitations

1. Lévy Property Deviations: Empirical results suggest subordination processes may not strictly satisfy Lévy process properties
2. Parameter Selection: The choice of parameter $\theta$ significantly impacts results, requiring further theoretical guidance
3. Data Frequency Constraints: The method primarily applies to daily or lower frequency data

Future Directions

1. Parameter Optimization: Develop more systematic methods for $\theta$ parameter selection
2. Model Extension: Consider more complex models incorporating time series dependencies
3. High-Frequency Data: Extend the method to handle microstructure effects in high-frequency trading data

In-Depth Evaluation

Strengths

1. Theoretical Innovation:
   • Proposes a novel Fourier decomposition method, providing new tools for studying time-subordination processes
   • Rigorously proves existence of complex domain characteristic functions with solid theoretical foundations
2. Method Advantages:
   • Nonparametric approach avoids prior assumptions about subordination process distributions
   • Direct inference from price data without computing quadratic variation
   • Relatively simple computationally, suitable for practical applications
3. Empirical Verification:
   • Validates method effectiveness on real market data
   • Results align with theoretical expectations, supporting Gamma subordination process assumptions

Weaknesses

1. Theoretical Limitations:
   • Lacks systematic guidance for parameter selection when $\theta \neq 0$
   • Convergence conditions for complex domain characteristic functions require deeper analysis
2. Empirical Issues:
   • Lévy property deviations suggest model assumptions may be oversimplified
   • Verification limited to single market index; generalization capability requires further testing
3. Application Constraints:
   • Primarily applicable to low-frequency data; applicability to high-frequency data insufficiently explored
   • Lacks systematic comparison with other time-subordination models

Impact

1. Academic Contribution:
   • Provides new analytical tools for financial econometrics
   • Establishes theoretical foundations for Fourier methods in time-subordination models
2. Practical Value:
   • Applicable to model selection in risk management and derivative pricing
   • Offers new perspectives for volatility modeling in algorithmic trading
3. Reproducibility:
   • Clear method description with detailed theoretical derivations
   • Provides specific implementation steps and parameter settings

Applicable Scenarios

1. Risk Management: More accurately model probability of extreme market events
2. Option Pricing: Improve Black-Scholes model handling of fat-tail distributions
3. Portfolio Optimization: Better characterize dynamic features of asset returns
4. Algorithmic Trading: Provide more precise price models for high-frequency trading strategies

References

The paper cites 19 important references, primarily including:

• Black & Scholes (1973): Foundations of option pricing theory
• Madan & Milne (1991), Madan et al. (1998): Variance Gamma model
• Skorokhod (1961), Monroe (1972): Time-changed Brownian motion theory
• Applebaum (2004): Lévy process theory
• Lukacs (1970): Analytic characteristic function theory

Overall Assessment: This is a high-quality paper combining theory and empirical analysis, proposing innovative Fourier methods for studying time-subordinated Brownian motion models. Despite some theoretical and empirical limitations, it provides valuable new tools and insights for financial modeling.