Option Pricing with Time-Changed Fractional Brownian Motion: A Fractional Variance Gamma Model

summary

The gist

Fractional Brownian motion (fBm) exhibits attractive features for financial modeling, including long-range dependence, path roughness…

In short

The paper introduces a Fractional Variance Gamma (fVG) model using a time-changed Fractional Brownian Motion to price options. This construction merges fBm's long-range dependence with the Variance Gamma model's structure, creating a semimartingale process compatible with arbitrage-free pricing. Estimation uses GMM to find parameters like the Hurst exponent, suggesting mild anti-persistence in S&P 500 returns.

Key concepts

Fractional Brownian Motion (fBm)
An extension of standard Brownian motion that captures long-range dependence in time series. It is attractive for financial modeling because it reflects the persistent nature of market movements over long periods, unlike standard models.
Time-Changed Process (X = BH(γ))
This construction modifies fBm by evaluating it at a stochastic gamma activity time ($\gamma$). This change preserves the fBm's core properties while ensuring the resulting process is a semimartingale, which is necessary for classical no-arbitrage pricing.
Fractional Variance Gamma (fVG) Model
This model extends the standard Variance Gamma model by substituting Brownian motion with fBm. It decomposes stock returns into four components, including an endogenous drift term driven by the process's dependence structure, allowing it to capture skewness and heavy tails.
Hurst Parameter (H)
A key parameter derived from GMM estimation that quantifies the scaling behavior of return moments. An estimate around 0.45 for the S&P 500 suggests mildly sublinear temporal scaling, implying mild anti-persistence in the returns.

Terminology used across episodes

This episode discusses

The paper

Option Pricing with Time-Changed Fractional Brownian Motion: A Fractional Variance Gamma Model · Read on arXiv

Samuel Curtis Johnson Graduate School of Management, Cornell University · Division of Finance, Nanyang Business School, Nanyang Technological University

Fractional Brownian motion (fBm) exhibits attractive features for financial modeling, including long-range dependence, path roughness, and anomalous diffusion. However, its non-semimartingale nature precludes the use of conventional no-arbitrage approaches to option pricing. We address this limitation by introducing a time-changed fBm, obtained by evaluating fBm at stochastic gamma activity time, where activity time represents cumulative executed trading time. The resulting process retains the defining properties of fBm while recovering the semimartingale structure. Building on this construction, we develop the fractional Variance Gamma (fVG) model and propose a generalized method of moments (GMM) estimation procedure for option pricing. An empirical analysis of the S&P 500 yields an estimated Hurst exponent of approximately 0.45, consistent with mildly sublinear temporal scaling of return moments.

DOI: 10.3934/fmf.2026.100088

Transcript

Introduction to the show: ident: Quantitative Finance. The day in quantitative finance: models of markets, risk and prices.

Ruth: We're looking at the paper: "Option Pricing with Time-Changed Fractional Brownian Motion".

Warren: Here's the gist of the paper. Fractional Brownian motion (fBm) exhibits attractive features for financial modeling, including long-range dependence, path roughness, and anomalous diffusion The resulting fractional Variance Gamma (fVG) model accommodates fractional dynamics in stock price evolution while preserving compatibility with the classical arbitrage-free pricing paradigm The Novel Process Construction The paper introduces a time-changed fBm process, defined as X = BH(γ), where the fBm BH is evaluated at stochastic gamma activity time γ This construction retains the defining properties of fBm while recovering the semimartingale structure The resulting process is a semimartingale, rendering it compatible with the classical no-arbitrage pricing framework based on equivalent martingale measures Building on this result, the authors introduce the fractional Variance Gamma (fVG) model which extends the Variance Gamma (VG) model of Madan and Seneta (1990) by replacing the Brownian motion component with fBm Model Structure and Components Under the fVG model, stock returns admit a decomposition into four distinct components: (i) compensation for the risk-free rate, (ii) a risk-premium, (iii) an endogenous drift generated by the dependence structure and historical trajectory of the process, and (iv) a martingale innovation that captures residual uncertainty The fVG stock price specification therefore contains a dependence-induced drift through which the current state of the process influences its future evolution Estimation Methodology For parameter estimation, the authors develop a generalized method of moments (GMM) procedure for the fVG model that exploits the unconditional moments of stock returns across multiple time horizons The GMM estimator identifies H primarily through the scaling behavior of return moments across investment horizons, that is, through the rate at which these moments accumulate with lag length The empirical analysis of the S&P 500 yields an estimated Hurst parameter H of approximately 0.45, consistent with mildly sublinear temporal scaling of return moments Key Findings from Empirical Analysis The GMM estimation procedure on the S&P 500 index suggests that the term structure of empirical moments exhibits mildly sub-linear temporal scaling, as the estimate H ≈ 0.45 implies this This result contrasts with studies reporting long range dependence in asset returns The estimated Hurst exponent is consistently below the Brownian benchmark of H = 0.5, suggesting mild anti-persistence in S&P 500 returns Model Equivalence and Tractability The time-changed fractional geometric Brownian motion (tfgBm) model is shown to be equivalent to the proposed fVG model up to a deterministic transformation of the jump amplitudes The fVG innovation process Y1 is obtained by applying the jump-amplitude transformation η(x):= ln(1 + x) to the jumps of the tfgBm innovation process Y2 The fVG model offers greater analytical tractability because log returns are linear in the driving innovation process Y The proposed model provides a parsimonious yet rich framework for capturing salient features of observed asset-price dynamics The authors hope that the results presented in this paper will stimulate further empirical and theoretical investigation, especially into the applications of asset pricing, risk management, and derivative valuation The model successfully captures skewness, heavy tails, and long-range dependence simultaneously The authors conclude that the accurate calibration of H requires that skewness and kurtosis are appropriately modeled.

Ruth: References Abry, P., & Veitch, D.

Warren: (1998).

Ruth: Wavelet Analysis of Long-Range-Dependent Traffic Bender, C., Sottinen, T., & Valkeila, E.

Title and authors: Warren: Okay, so assuming this fVG model is a good starting point, what do the authors suggest they should improve? They seem to be hinting that there’s more to say beyond just the basic structure of replacing Brownian motion with fBm.

Ruth: They are suggesting two main things. First, they show that the time-changed fractional geometric Brownian motion model is equivalent to their proposed fVG model, but they also introduce a way to get the innovation process Y1 from an alternative stock price specification called the time-changed fractional geometric Brownian motion, which involves applying a jump-amplitude transformation η of x equals log one plus x.

Iris: That transformation sounds like it’s handling the jumps in a very specific way. What does that do for the math? Does it simplify things enough to make estimation easier, or is it just a mathematical convenience?

Ruth: They argue it offers greater analytical tractability because when you use this transformation, log returns become linear in that driving innovation process Y2. That linearity is a big deal for computation and analysis because linear relationships are much easier to work with than non-linear ones.

Warren: From a trading perspective, if log returns are linear in the driving process, it means the dynamics of the asset price itself become more predictable when you look at the underlying innovation. That could potentially help in developing more robust strategies based on that structure. But what about risk management costs?

Ruth: The paper doesn't directly discuss strategy costs, but by making log returns linear in Y2, they are setting up a cleaner environment for applying simulation methodologies to evaluate the model. They construct a specific procedure for computing conditional expectations by first generating the discretized sample paths of both the fractional Brownian motion and the gamma processes independently.

Iris: I see they are separating the generation of paths to ensure accuracy by using a finer grid for the fractional Brownian motion than what is used for calculating those gamma activity times. That's a specific numerical trick to maintain precision in their simulation procedure, right?

Warren: Exactly. It’s about making sure you get high fidelity when you’re evaluating the model. It's not just throwing everything into one calculation and hoping for the best; they are being careful about the grid resolution here.

Ruth: So, the improvement seems to be moving from a theoretical equivalence to a practical computational method that handles the simulation step more carefully by separating how they generate those two key components of the model. This is where it gets less abstract and more about actually running a calculation.

Iris: Before we move on, I just want to check one identification issue again about this simulation setup. They are constructing the conditional expectations by generating paths independently and then evaluating them at the realized gamma activity times. Is that setup truly sufficient to isolate the dependence drift they are trying to measure?

Ruth: It’s designed that way to ensure accuracy, but as Iris pointed out, it’s a key assumption—it needs to be rigorously checked in practice. The paper doesn't prove it works perfectly under every condition.

Warren: If we think about the cost of running this simulation, generating those two sets of paths independently and then evaluating them at those activity times means the computational load is higher than just running a standard model. We have to factor that into any strategy cost analysis, right?

Iris: That’s fair. The paper doesn't quantify the exact computational overhead, but it does acknowledge that this simulation procedure is how they compute conditional expectations for the fVG model. It’s the mechanism they use to bridge the gap between their theoretical structure and a computable result.

Ruth: So, to summarize this improvement: they are moving towards a more rigorous simulation methodology that involves separating path generation for accuracy and ensuring log returns are linear in the driving process, which gives them better analytical tractability. This moves them from just proposing a model to providing a computational recipe.

The paper's summary: Warren: Alright, we’ve covered the theory, the estimation with GMM based on return moments across time horizons, and how they improved the simulation procedure for accuracy. Where do we land with these findings when it comes to real-world application? What’s the final word on this whole fVG model?

Ruth: The main implication is that this paper gives us a parsimonious but rich framework for capturing several important features simultaneously: skewness, heavy tails, and long-range dependence in asset prices. They successfully manage to capture all three of those characteristics at once.

Iris: I agree with that summary. It’s about achieving something rare in financial modeling—capturing dependence and fat tails without sacrificing the ability to model the drift structure needed for option pricing under market constraints, which is what they set out to do.

Ruth: Exactly. And they conclude that accurately calibrating the Hurst parameter H requires that those skewness and kurtosis are also appropriately modeled in the same framework, suggesting you can’t treat them as separate issues when modeling this process.

Warren: So, the final word is that this paper provides a workable model for asset-price dynamics that is both rich in features and still compatible with the classical no-arbitrage pricing paradigm because of its time-changed construction. It successfully captures skewness, heavy tails, and long-range dependence simultaneously.

Iris: And the limitation they state clearly is that the accuracy of calibrating H fundamentally depends on how well you model those skewness and kurtosis within their framework. If those moments are wrong, then the entire calibration fails to provide a reliable measure of persistence.

Ruth: That’s the final word for now. We hope this work stimulates further empirical and theoretical investigation into applications in asset pricing, risk management, and derivative valuation because it gives us a new way to look at these complex dynamics.

Warren: Yeah, that's the picture on the paper "Option Pricing with Time-Changed Fractional Brownian Motion: A Fractional Variance Gamma Model". We’re done for today.

The paper's improvements: Ruth: So we’ve seen how they built this model to handle those tricky features like long memory and fat tails in asset prices, now we look at how they improved it from a theoretical idea to something you can actually use.

Warren: Right, so they show that their time-changed fractional geometric Brownian motion is actually equivalent to the fractional variance gamma model after a simple transformation of the jump amplitudes. That’s a key step for me as someone thinking about how we can price stuff.

Iris: That transformation they introduce, which involves taking those jumps and applying this function where x equals log one plus x, it really seems to be what makes things more tractable for computation. It makes the log returns linear in that driving innovation process.

Ruth: Linear in that process is a big deal because it means the dynamics of the stock price itself become much easier to handle mathematically than if everything was tangled up nonlinearly. It simplifies how we calculate things, doesn't it?

Warren: It does, and from a trading side, if you can express log returns linearly in this driving process, your strategy parameters become more direct and less messy to tune. But I have to ask what the cost is for that simplicity.

Iris: The paper points out that they aren't just doing the math; they are building a specific simulation methodology for pricing. They separate how they generate paths for the fractional part of the motion from how they handle those gamma activity times, which is a bit of a numerical trick to keep things accurate.

Ruth: So it’s about making sure you get high fidelity when you’re running simulations, separating the generation steps to ensure precision in that conditional expectation calculation. It shows they’re thinking beyond just the equation and into how you actually run the numbers on your computer.

Warren: I see that as a way to bridge the gap between a pretty structure and a real-world computable result. But we have to be careful, Iris, because separating those paths doesn't magically solve all our market problems.

Iris: That’s true; it’s an important methodological step, but you still have to worry about the underlying assumptions about how the market behaves when you run those simulations. The paper notes that this approach is designed for better tractability, not necessarily perfect real-world replication of every tiny detail.

Ruth: Exactly. It’s a tool to get a workable framework, but we still need to be mindful of what this model assumes about the people—the market participants—when you change those underlying assumptions about the process itself.

Warren: So we’ve seen the structure, now we see how they're making it computationally practical by simplifying the math and setting up a better simulation pipeline. But that opens up a new question for me about risk management costs.

Iris: It does. And that leads us to where we need to look next—how this framework handles those skewed returns and heavy tails when you try to apply it in practice under real-world conditions.

Conclusion: Ruth: So we’ve walked through how they built this model to handle those tricky features like long memory and fat tails in asset prices, now we look at what this paper actually concludes for us as users of the model.

Warren: The main thing is that this fractional variance gamma model gives us a way to capture skewness, heavy tails, and long-range dependence all at once in stock price evolution. That’s a pretty comprehensive package for pricing options.

Iris: And they’re telling us that for accurate calibration of the Hurst parameter H, you absolutely have to make sure you are modeling the skewness and kurtosis properly within this framework. It’s not a separate issue anymore; they are linked.

Ruth: That means if you want to get a reliable measure of how persistent these returns are, you can't just look at one thing; it all has to be calibrated together in this model. It makes the whole calibration process more rigorous, I think.

Warren: From a trading perspective, it means the parameters you choose for your model have to reflect that complex interplay between dependence and those fat tails, which could really change how you set up your risk measures. You can't just use a simple random walk assumption anymore.

Iris: It definitely forces a more nuanced approach to risk management because the underlying dynamics are fundamentally different from what simpler models suggest. It’s about acknowledging that the market isn't behaving in one neat, predictable way.

Ruth: So, it’s a framework that is parsimonious—meaning it tries to keep things simple—but still rich enough to capture those complex features of real asset prices. That’s a big thing for practical application.

Warren: I think the main implication is that we can build pricing and risk models that are more realistic because they don't ignore the non-standard behaviors of stock returns. It gives us a richer description of what's actually happening in those markets.

Iris: I just want to stress that while this framework is rich, it does have its limitations, and you need to be aware of where it stops working perfectly, especially when you’re trying to apply it across different asset classes. It’s a good starting point, but not a universal solution.

Ruth: Exactly. It’s a really solid tool for getting started in modeling these kinds of complex dynamics without immediately jumping into the most computationally demanding territory. This paper is "Option Pricing with Time-Changed Fractional Brownian Motion: A Fractional Variance Gamma Model".

Warren: Alright, that wraps up our look at this work. It shows us how to combine fractional dynamics with variance gamma structure for a more complete picture of stock returns.

Iris: It’s definitely something worth further theoretical investigation, especially in how we can rigorously test those identification assumptions we discussed earlier. That’s where the real paper work is going next.

More episodes

← Home