Variance Gamma Model

The variance gamma model is an option pricing framework based on a pure-jump Lévy process that avoids the geometric Brownian motion’s implicit symmetry assumption. Instead of modeling stock returns as continuous (even if volatile), it treats them as bundles of discrete jumps with independent controls over skewness and tail weight—matching the leptokurtotic, negatively skewed returns observed in real equity markets.

Why geometric Brownian motion misses the empirical reality

The classic Black-Scholes model assumes stock prices follow a continuous, normally distributed process. This choice is convenient for math—it yields closed-form option prices and simple Greeks—but it fails to match what investors observe. Real equity returns show pronounced negative skewness (large down-moves more frequent than up-moves of equal size) and excess kurtosis (fatter tails than a normal distribution). These are not defects that disappear with more data; they are structural properties of financial returns.

Early attempts to patch this used stochastic volatility (Heston model) or jump-diffusion hybrids (Merton). Both add parameters and computational burden. The variance gamma model takes a different turn: it abandons diffusion entirely and builds the entire return process from jumps—a pure-jump Lévy process.

The mechanics of the pure-jump process

Under the variance gamma model, the log stock price evolves via a subordinated Brownian motion. At each “event” (subordination time), a jump occurs. The size and direction of the jump are independent of the volatility of the jump: you can have a large jump with low local volatility, or a tiny jump with high local volatility.

Formally, if S(t) is the stock price at time t, then log S(t) = log S(0) + X(t), where X(t) is a variance gamma process. The VG process is constructed as a Brownian motion evaluated at a random time (gamma-subordinated time), which generates the three key degrees of freedom:

• Volatility σ: controls the typical magnitude of moves.
• Skewness ν: when non-zero, biases jumps in one direction (typically downward, since equity returns are negatively skewed).
• Kurtosis shape κ: thickens the tails; larger κ means more extreme outliers.

This parametrisation is elegant because each lever operates independently. A trader can fit the at-the-money volatility via σ, then use ν to nail the volatility smile skew at the wings, and κ to control the height of deep out-of-the-money prices.

Fitting the volatility smile and skew

The Black-Scholes model produces a flat implied volatility surface—all strike prices have the same implied volatility. In reality, put options (especially those protecting against large downside) trade at higher implied volatility than calls of the same distance from the money. This mismatch—the volatility smile or skew—plagued practitioners for decades.

The variance gamma model naturally generates a skewed smile because skewness and kurtosis are built into the return distribution. When ν < 0 (downside skew), out-of-the-money puts become relatively expensive compared to calls, matching observed prices. The excess kurtosis (fat tails) lifts the prices of far out-of-the-money options, again matching the data.

Fitting an equity index to a variance gamma model typically yields:

• σ ≈ 15–25% (annualized volatility)
• ν ≈ −0.1 to −0.2 (negative skew reflecting crash risk)
• κ ≈ 0.05–0.2 (kurtosis shape; smaller values mean heavier tails)

These parameters remain relatively stable across short time windows, making the model practical for risk management.

Pricing options in the variance gamma framework

Unlike Black-Scholes, there is no closed-form formula for variance gamma option prices. Instead, practitioners use numerical methods: Fourier transforms (FFT), Monte Carlo simulation, or recombining lattices. The Fourier approach is popular because it is fast and accurate for European-style options. For American options (which allow early exercise), a lattice or finite-difference scheme becomes necessary.

The variance gamma model also yields exact closed-form expressions for the characteristic function of the log stock price, which enables efficient numerical option pricing via inversion methods. This is a significant practical advantage: you can price a large book of options across many strikes and maturities in seconds.

Historical adoption and present use

The variance gamma model was introduced by Dilip Madan and Frank Milne in the mid-1990s and gained traction among academics and quantitative traders. Banks including JPMorgan and Goldman Sachs built variance gamma pricing libraries. However, it never achieved the dominance of Heston or jump-diffusion models, partly because it requires computational overhead and partly because simpler models often suffice for liquid traded options.

Today, variance gamma serves a niche: volatility traders fitting exotic equity options, hedge funds running systematic options strategies, and researchers testing option-pricing theory. It appears in some FX exotics pricing libraries because currency returns often exhibit pronounced skewness and kurtosis.

Strengths and limitations

The model’s chief strength is parsimony with realism: three intuitive parameters capture skewness, kurtosis, and volatility without forcing a choice between a jump-diffusion and a pure-diffusion model. The independence of these parameters makes calibration straightforward and interpretation clear.

The limitations are computational (no closed-form prices for American or exotic options) and psychological. The pure-jump assumption, though empirically sound, conflicts with the intuition many traders build over years of Black-Scholes. Switching toolkits requires retraining and validation. As a result, many firms use variance gamma only for specific high-value trades (e.g., long-dated exotics) and default to Heston or simpler jump-diffusion for vanilla options.

See also