Levy Processes and Jump Diffusions

In the classical model of finance, asset prices are often assumed to follow a continuous path, described by Brownian motion. Yet real markets experience sudden, discontinuous shocks—crashes, rallies, or news events that cause prices to jump. Levy processes provide the essential mathematical framework to model such behavior, generalizing Brownian motion to include jumps while retaining analytical tractability. Understanding these processes is key to modern quantitative finance, insurance risk modeling, and any field where phenomena exhibit both continuous evolution and sudden, random discontinuities.

From Brownian Motion to General Levy Processes

A Levy process is a stochastic process that generalizes the familiar Wiener process (Brownian motion). Formally, a stochastic process $\{X_t{\}}_{t\ge 0}$ is a Levy process if it satisfies four key properties:

1. Independent Increments: The increments $X_t-X_s$ are independent of the history of the process up to time $s$.
2. Stationary Increments: The distribution of the increment $X_{t+s}-X_s$ depends only on the length of the interval $t$, not on the starting time $s$.
3. Stochastic Continuity: For any $ϵ>0$, $P(|X_t+h-X_t|\ge ϵ)\to 0$ as $h\to 0$. This means jumps occur at random times, not at fixed times.
4. Starts at Zero: $X_0=0$ almost surely.

Brownian motion is the quintessential example of a continuous Levy process. The defining innovation of the general Levy process is the relaxation of path continuity, allowing the process to exhibit jumps. This seemingly small change unlocks a vast universe of behaviors, including the heavy-tailed returns and sudden movements observed in financial markets. The power of the theory lies in a fundamental decomposition theorem, which states that any Levy process can be split into three independent components.

The Levy-Ito Decomposition and the Levy-Khintchine Formula

The Levy-Ito decomposition provides a precise blueprint for constructing any Levy process. It states that any Levy process $X_t$ can be represented as: $$X_t=\mu t+\sigma W_t+J_t$$ where:

• $\mu t$ is a deterministic linear drift.
• $\sigma W_t$ is a scaled Brownian motion (the continuous diffusion component).
• $J_t$ is a pure jump process.

This jump component $J_t$ is itself a complex object, encapsulating all discontinuous movements. It can be further decomposed into two parts: the sum of "large" jumps (those exceeding a certain size) modeled by a compound Poisson process, and a limit of "small" jumps, which is a square-integrable martingale. This decomposition is intrinsically linked to the Levy-Khintchine formula, the cornerstone of Levy process theory.

The Levy-Khintchine formula characterizes the process not by its paths, but by its characteristic function (the Fourier transform of its distribution). For a Levy process $X_t$, we have: $$E[e^{iu{X}_t}]=e^{t\psi (u)},\text{where}\psi (u)=i\mu u-\frac{1}{2}{\sigma }^2{u}^2+{\int }_{\mathbb{R}\setminus \{0\}}(e^{iuy}-1-iuy{1}_{|y|<1})\nu (dy).$$ This is the Levy-Khintchine formula. The triplet $(\mu ,{\sigma }^2,\nu )$ completely defines the law of the Levy process. Here, $\nu$ is the Levy measure, which encodes the jump structure. The value $\nu (A)$ describes the expected number of jumps, per unit time, whose sizes fall in the set $A$. The integrand's compensating term $-iuy{1}_{|y|<1})$ ensures convergence for processes with many small jumps, a necessity for processes with infinite jump activity.

Key Examples: Compound Poisson and Stable Processes

Two fundamental examples illustrate the spectrum of behaviors Levy processes can capture. The first is the compound Poisson process. This is the simplest jump process, defined as $J_t={\sum }_{i=1}^{N_t}{Y}_i$, where $N_t$ is a Poisson process with rate $\lambda$ (counting the number of jumps by time $t$) and $\{Y_i\}$ are i.i.d. random variables representing the jump sizes, independent of $N_t$. Its Levy measure is simply $\nu (dy)=\lambda f_Y(dy)$, where $f_Y$ is the distribution of the jump sizes. It models isolated, infrequent jumps of varying magnitude.

At the other end of the spectrum lie stable distributions and the corresponding stable Levy processes. These are defined by a stability property: if $X_1$ and $X_2$ are independent copies, then $a{X}_1+b{X}_2$ has the same distribution as $cX+d$ for some constants $a,b,c,d$. Except in the Gaussian case (which is the stable process with continuous paths), stable processes have infinite variance and exhibit power-law (heavy-tailed) behavior. Their Levy measure for an $\alpha$-stable process, with $\alpha \in (0,2)$, is given by: $$\nu (dy)=\left(\frac{C_1}{y^{1+\alpha }}{1}_{y>0}+\frac{C_2}{|y{|}^{1+\alpha }}{1}_{y<0}\right)dy.$$ The parameter $\alpha$ is the tail index; smaller $\alpha$ means heavier tails and more extreme jumps. These processes are crucial for modeling phenomena with extreme events, such as financial crashes or large insurance claims.

Application to Financial Modeling

The primary application driving the study of Levy processes is in financial modeling. The classic Black-Scholes model, driven solely by Brownian motion, fails to capture the "smile" or "skew" in implied volatility and underestimates the probability of large price moves. Levy processes directly address these shortcomings.

By replacing the Brownian driver in an asset price model $S_t=S_0{e}^{X_t}$ with a general Levy process, one can build models that incorporate jumps. For example, the Merton jump-diffusion model uses a process defined as $X_t=\mu t+\sigma W_t+{\sum }_{i=1}^{N_t}{Y}_i$, combining a diffusion with a compound Poisson jump component. This explicitly captures sudden market movements like those caused by earnings announcements or macroeconomic news. More sophisticated models use processes with infinite jump activity (like the Variance Gamma or Normal Inverse Gaussian processes) to generate the desired continuous-looking yet jagged paths and heavy-tailed return distributions observed empirically.

Pricing options under such models typically requires using the Levy-Khintchine exponent $\psi (u)$ in Fourier-based pricing methods, as the closed-form Black-Scholes formula no longer applies. The great advantage is that the independent and stationary increments property is preserved, keeping many calculations analytically tractable within a more realistic framework.

Common Pitfalls

1. Confusing Jump Activity with Jump Size: A common error is to assume a process with a finite Levy measure (like compound Poisson) cannot have frequent jumps. The frequency is governed by $\nu (\mathbb{R})$, while the size distribution is governed by the shape of $\nu$. A compound Poisson process can have a very high jump rate $\lambda$, leading to many small jumps, which might visually resemble diffusion.
2. Misapplying the Levy-Khintchine Formula for Simulation: The formula provides the characteristic function, not a direct path-wise simulation algorithm. Attempting to simulate paths by directly inverting the characteristic function at each step is computationally prohibitive. Correct simulation relies on the Levy-Ito decomposition, using methods like Poisson thinning for the jump component.
3. Overlooking the Compensator in the Small-Jump Integral: In the Levy-Khintchine integrand $(e^{iuy}-1-iuy{1}_{|y|<1})$, the term $-iuy{1}_{|y|<1}$ is a compensator. It ensures the integral converges for processes where ${\int }_{|y|<1}|y|\nu (dy)=\infty$ (infinite variation small jumps). Ignoring its role leads to a misunderstanding of how processes like the Variance Gamma are defined and how their martingale property is established.
4. Assuming All Heavy-Tailed Models are Stable: While stable distributions are the canonical heavy-tailed models, many other Levy processes (e.g., tempered stable processes) also exhibit heavy tails but with better mathematical properties, such finite moments of all orders, which are often required for financial applications. It is a mistake to equate "heavy-tailed" exclusively with "stable."

Summary

• Levy processes generalize Brownian motion by allowing for discontinuous jumps while retaining the crucial properties of independent and stationary increments and stochastic continuity.
• The Levy-Ito decomposition provides a path-wise construction of any Levy process as the sum of a drift, a Brownian motion, and a pure jump process.
• The Levy-Khintchine formula, characterized by the triplet $(\mu ,{\sigma }^2,\nu )$, provides a complete description of a Levy process's distribution via its characteristic function, with the Levy measure $\nu$ detailing the jump structure.
• Key examples include the compound Poisson process for modeling isolated jumps and processes with stable distributions for modeling systems with infinite variance and heavy-tailed returns.
• In finance, these processes are foundational for building asset price models that realistically capture sudden market movements and the observed statistical properties of returns, moving beyond the limitations of pure diffusion models.