Higher-order ATM asymptotics for the CGMY model via the characteristic function

This paper derives short-time at-the-money call-price asymptotics for the exponential CGMY model with activity parameter $Y\in(1,2)$ by utilizing the characteristic function and the Lipton--Lewis formula to obtain a higher-order expansion with explicit coefficients and a controlled remainder, all without relying on closed-form expressions for the underlying density.

The Gist

Imagine you are trying to predict the weather, but instead of rain and wind, you are trying to predict the price of a stock option. Specifically, you want to know the price of a "bet" (an option) that is perfectly balanced right now—where the strike price is exactly the same as the current stock price. This is called an At-The-Money (ATM) option.

In the world of finance, models like CGMY are used to describe how stock prices jump around. These jumps aren't smooth like a car driving down a highway; they are more like a swarm of bees buzzing erratically. Some bees are tiny, some are medium, and occasionally, a giant bee crashes into the hive.

This paper is about figuring out exactly how much that "balanced bet" costs when you are looking at a very, very short time frame (like a split second).

Here is the breakdown of what the authors did, using some everyday analogies:

1. The Problem: The "Zoom Lens" Issue

When you look at a stock price over a long time (like a year), the tiny, chaotic jumps average out, and the price looks smooth. But when you zoom in to a tiny fraction of a second, those tiny jumps become the main event.

Previous mathematicians had figured out the first level of approximation (the "rough sketch" of the price). They knew the price behaves like $t^{1/Y}$ (where $t$ is time). But they wanted to know the next level of detail—the "second-order" correction. It's like knowing a car is going 60 mph, but you also want to know if it's accelerating or braking slightly, because that changes the exact arrival time.

2. The Tool: The "Magic X-Ray" (Characteristic Function)

Usually, to understand these jumps, mathematicians try to reconstruct the entire history of the stock's movement (the "density" of where the price might be). This is like trying to rebuild a shattered vase to see what it looked like before it broke. It's hard and messy.

The authors of this paper decided to use a different tool: the Characteristic Function. Think of this as an X-ray or a DNA test of the stock price. Instead of looking at the messy physical shape of the price movements, this function captures the "genetic code" of the jumps.

The authors used a special formula (the Lipton–Lewis formula) that lets you calculate the option price directly from this "DNA" without ever having to rebuild the shattered vase.

3. The Discovery: Finding the Hidden Terms

By using this X-ray view and zooming in on the tiny time scales, they derived a new, more precise formula.

• The First Term ($d_1$): This is the "Big Picture." It tells you the price is mostly driven by the sheer number of tiny jumps. This was already known.
• The Second Term ($d_2$): This is the "Fine Tuning." It accounts for how the jumps are slightly "tempered" (damped) at the edges. The authors found a specific integral (a fancy math sum) that calculates this correction perfectly.
• The Surprise (Higher Orders): They didn't stop there. They kept digging and found even smaller corrections (third, fourth, and fifth order).

4. The "Ghost" Jumps (The Cubic Drift)

One of the most interesting findings in the paper is about a specific type of mathematical term called the "cubic drift."

Imagine you are walking in a straight line, but you take a step left, then a step right, then a step left again. If you do this perfectly symmetrically, you end up exactly where you started. The math showed that a specific type of "jitter" (the cubic drift) is purely imaginary—meaning it's like a ghost. It exists in the math, but when you calculate the actual price, it cancels itself out completely.

Because this "ghost" term vanishes, the authors realized that the "next" important term isn't what everyone expected. It shifts the entire timeline of which corrections matter most. It's like thinking the next stop on a train is Station C, but because Station B is closed, the train skips straight to Station D.

5. The "Dynamic Cutoff" Strategy

To prove their math was right, they had to be very careful. They split the problem into three zones:

1. The Inner Core: Where the jumps are small and predictable.
2. The Tail: Where the jumps are huge and rare.
3. The Middle: The tricky transition zone.

They used a "dynamic cutoff" (like a sliding door that moves as time shrinks) to separate these zones. This allowed them to calculate the price in the core and the tail separately and then stitch them together without the math falling apart.

The Bottom Line

The authors successfully created a high-definition map of option prices for very short times.

• Why it matters: Traders and risk managers need to know the exact price of options, especially for high-frequency trading where milliseconds matter.
• The Innovation: They did this using only the "DNA" (characteristic function) of the model, avoiding the messy reconstruction of the price path.
• The Result: They provided a new, precise formula that includes a "ghost" term that disappears and a new "quartic" term that was previously missed.

In short, they took a blurry, low-resolution photo of short-term option prices and turned it into a crystal-clear, high-definition image, revealing hidden details that previous models missed.

Technical Summary

1. Problem Statement

The paper addresses the derivation of short-time at-the-money (ATM) call price asymptotics for the CGMY model (Carr-Geman-Madan-Yor), a pure-jump Lévy process widely used in financial mathematics.

• Context: The CGMY model is defined by four parameters: jump intensity $C$, exponential damping rates $G$ and $M$, and an activity parameter $Y \in (0, 2)$. The paper focuses on the regime $1 < Y < 2$, where the process has infinite activity and infinite variation.
• Gap in Literature: While first-order asymptotics (scaling as $t^{1/Y}$) were well-established, and second-order expansions existed via measure transformations and stable density expansions (e.g., Figueroa-López et al.), a rigorous derivation of higher-order terms (third order and beyond) directly from the characteristic function was lacking. Previous attempts using the Lipton–Lewis (LL) formula had gaps in rigor regarding the interchange of limits and integrals.
• Objective: To derive a complete, rigorous expansion of the normalized ATM call price $c(t, 0)$ up to higher orders using only the characteristic exponent and the Lipton–Lewis representation, avoiding complex measure changes or density expansions.

2. Methodology

The authors employ a Fourier-based approach centered on the Lipton–Lewis (LL) formula, which expresses option prices as integrals involving the characteristic exponent $\Psi(u)$.

• Lipton–Lewis Representation: The normalized ATM call price is expressed as:
  $$c(t, 0) = \frac{1}{\pi} \Re \int_0^\infty \frac{1 - e^{t\Psi(u - i/2)}}{u^2 + 1/4} du$$
• Rescaling: To capture the stable domain of attraction, the authors perform a change of variables $u = v/t^{1/Y}$. This rescales the integral to isolate the leading stable behavior ($t^{1/Y}$) from the tempering corrections.
• Combined-Integrand Approach: A critical methodological innovation is keeping the integrand terms combined rather than splitting them. Splitting the integral leads to non-uniform convergence near zero and incorrect coefficients. By keeping the full integrand intact, the authors preserve necessary cancellations.
• Dynamic Cutoff Strategy: To rigorously prove the expansion, the integration domain is partitioned into three regions using a dynamic cutoff:
  1. Inner Region ($[0, \Lambda]$): Where Taylor expansions of the exponent are valid.
  2. Core Region ($[\Lambda, t^{-\rho}]$): Where the interplay between the Taylor expansion and the stable exponential decay is analyzed using Laplace-type integrals.
  3. Tail Region ($[t^{-\rho}, \infty)$): Where the stable exponential decay dominates, ensuring the remainder is negligible.
• Asymptotic Analysis: The authors expand the characteristic exponent $\Psi(u - i/2)$ in powers of $t$, identifying contributions from the drift term ($\tilde{b}$) and the binomial corrections from the tempering parameters ($G, M$).

3. Key Contributions and Results

A. Second-Order Expansion

The paper rigorously derives the second-order expansion:
$$c(t, 0) = d_1 t^{1/Y} + d_2 t + o(t)$$

• $d_1$: The known first-order coefficient derived from the symmetric $Y$-stable limit.
• $d_2$: A new explicit integral formula involving the characteristic exponent and the limiting stable exponent. The authors verify numerically that this integral matches the closed-form expression derived previously via measure transformation, confirming the validity of the Fourier-only approach.

B. Higher-Order Expansion (Five-Term)

The authors extend the analysis to extract closed-form coefficients for higher-order terms, resulting in the following expansion:
$$c(t, 0) = d_1 t^{1/Y} + d_2 t + a_{2,1} t^{2-1/Y} + a_{4,1} t^{4-3/Y} + a_{1,2} t^{2/Y} + o(t^{\max(2-1/Y, 2/Y)})$$

Key Structural Findings:

1. Vanishing Odd Drift Powers: A significant structural discovery is that odd powers of the drift (e.g., $(i\tilde{b}w)^3$) are purely imaginary. Since the call price is a real quantity, these terms vanish upon taking the real part.

   • Consequence: The exponent $3 - 2/Y$ (associated with the cubic drift) is absent from the expansion.
   • Impact: This shifts the effective bifurcation point. While a naive lattice analysis might suggest a bifurcation at $Y=4/3$ (where $3-2/Y$ crosses $2/Y$), the actual bifurcation occurs at $Y = 5/4$ (where the quartic drift $4-3/Y$ crosses $2/Y$).

2. Explicit Coefficients:

   • $a_{2,1}$ (Drift-squared): Proportional to $\tilde{b}^2$. Matches existing results from [11] but derived independently via Laplace integrals against $e^{-\sigma_Y t w^Y}$.
   • $a_{1,2}$ (Binomial correction): Proportional to the real part of the first binomial correction $\beta_1$. This is a new closed-form result involving $\Gamma(1-2/Y)$.
   • $a_{4,1}$ (Quartic drift): Proportional to $\tilde{b}^4$. This term is relevant for $Y < 5/4$.

3. Bifurcation Structure:

   • For $1 < Y < 5/4$: The order is $t^{1/Y} \to t \to t^{2-1/Y} \to t^{4-3/Y} \to t^{2/Y}$.
   • For $5/4 < Y < 3/2$: The order is $t^{1/Y} \to t \to t^{2-1/Y} \to t^{2/Y}$.
   • For $Y > 3/2$: The order is $t^{1/Y} \to t \to t^{2/Y} \to t^{2-1/Y}$.
   • At $Y=3/2$, the exponents $2-1/Y$ and $2/Y$ coalesce at $4/3$.

C. Numerical Verification

The authors provide high-precision numerical verification:

• The integral formula for $d_2$ matches the closed-form expression from [10] to within $10^{-7}$.
• The coefficients $a_{2,1}$ and $a_{1,2}$ are verified by evaluating the Laplace integrals directly and observing convergence to the theoretical values as $t \to 0$.
• The absence of the cubic drift term is confirmed by showing the remainder divided by $t^{3-2/Y}$ converges to zero.

4. Significance

• Methodological Advancement: The paper demonstrates that high-order asymptotics can be derived entirely within the Fourier framework using the characteristic function, without resorting to Esscher transforms or stable density expansions. This simplifies the theoretical machinery required for such derivations.
• Rigorous Justification: It resolves technical gaps in previous applications of the Lipton–Lewis formula, specifically regarding the uniformity of limits and the handling of the integrand near zero.
• New Analytical Insights: The discovery that odd drift powers vanish and the subsequent correction of the bifurcation point from $Y=4/3$ to $Y=5/4$ provides a more accurate understanding of the CGMY model's behavior.
• Generalizability: The "dynamic cutoff" and "combined-integrand" techniques are applicable to other tempered stable models and potentially to the "close-to-the-money" regime, offering a robust toolkit for future research in option pricing asymptotics.

In summary, this work provides a complete, rigorous, and computationally verified higher-order expansion for ATM options in the CGMY model, refining the theoretical understanding of how jump activity and tempering parameters interact in the short-time limit.