Merton's Jump-Diffusion Model

The The Black-Scholes Model model assumes prices move continuously — like a particle jiggling, never teleporting. But real markets gap. A stock closes at $100 and opens at $85 after a bad earnings call; an index drops 7% before you can blink. Those are jumps, and a smooth diffusion can't produce them. In 1976 Robert Merton fixed this by adding a jump term to geometric Brownian motion: most of the time the price diffuses gently, but every so often it leaps by a random amount, with the leaps arriving at random times.

The result explains two things Black-Scholes gets wrong. Real return distributions have fat tails — extreme moves happen far more often than a bell curve allows — and option markets show a volatility smile, where out-of-the-money options trade richer than a single flat volatility would justify. Jumps put weight back in the tails, and that weight is exactly what the smile is pricing.

The model

Merton's price follows the usual diffusion plus a jump process:

$$\frac{dS_t}{S_t} = (\mu - \lambda k)\,dt + \sigma\,dW_t + (J - 1)\,dN_t.$$

Reading it left to right: $\sigma\,dW_t$ is the ordinary Black-Scholes wiggle, the Brownian diffusion with volatility $\sigma$. The new piece $(J-1)\,dN_t$ is the jump: $dN_t$ counts jumps arriving as a Poisson process with intensity $\lambda$ (the average number of jumps per year), and $J$ is the random multiplier when one hits — $J = 0.9$ means a 10% drop, $J = 1.05$ a 5% pop. The term $-\lambda k\,dt$, where $k = E[J-1]$ is the average jump size, is a bookkeeping correction that offsets the jumps' average drift so the discounted price still behaves like a fair game.

How it prices options

Merton's insight makes the option price surprisingly clean. Condition on how many jumps happen before expiry. If exactly $n$ jumps occur, the path is just a Black-Scholes path with a tweaked volatility and drift, and its price is the ordinary Black-Scholes formula. Since the number of jumps is Poisson-distributed, the true price is a probability-weighted average of Black-Scholes prices, one for each possible number of jumps:

$$C = \sum_{n=0}^{\infty} \frac{e^{-\lambda' T}(\lambda' T)^n}{n!}\; C_{\text{BS}}(\sigma_n, r_n).$$

The Poisson weights $\frac{e^{-\lambda' T}(\lambda' T)^n}{n!}$ are just the chance of seeing $n$ jumps in time $T$; $C_{\text{BS}}$ is Black-Scholes with the volatility and rate adjusted for those $n$ jumps. Most of the weight sits on zero or one jump, but the rare many-jump terms fatten the tails and lift the price of out-of-the-money protection.

Worked example

Say jumps arrive on average once a year, so $\lambda = 1$, and you're pricing a one-month option, $T = \tfrac{1}{12} \approx 0.083$ years. The chance of a calm month with no jump at all is the Poisson zero-count probability:

$$P(\text{0 jumps}) = e^{-\lambda T} = e^{-0.083} \approx 0.92 .$$

So about 92% of months are pure diffusion, and roughly 8% of months contain at least one jump. That 8% is small, but it is exactly the slice Black-Scholes throws away. Now suppose jumps are on average down — say the typical jump is $-5\%$, so $k = -0.05$. Because crashes are more likely than melt-ups, the extra tail risk lands disproportionately on the downside. Out-of-the-money puts must therefore cost more than a single flat volatility implies, and when you back out each strike's implied volatility you see it rising toward the low strikes — the volatility skew. The jump parameters $\lambda$, the mean jump, and its spread are what let Merton's model bend a flat Black-Scholes line into the observed smile.

Where it misleads

• You cannot hedge a jump away. Delta hedging assumes you can rebalance continuously as the price inches along. A jump moves the price between one tick and the next, so no amount of rebalancing catches it. Jump risk stays on your book, and the market charges a premium for bearing it (this feeds the variance risk premium).
• The market is no longer complete. With jumps there's no single self-financing hedge that replicates the option, so the "risk-neutral" price isn't unique — it depends on assumptions about how jump risk is priced. Merton sidestepped this by assuming jump risk is diversifiable; that assumption is debatable for market-wide crashes.
• Parameters are hard to pin down. Rare jumps mean sparse data. Estimating how often crashes come and how big they are from a short history is notoriously unstable, so calibrated jump parameters can swing wildly.

Merton's model is the simplest step beyond Black-Scholes and the gateway to a whole family of jump and stochastic-volatility models built to reproduce the smile. Its lasting lesson is blunt: the tails are thicker than a normal distribution, and pricing anything that pays off in a crash — exotic options, deep puts, tail hedges — starts with taking jumps seriously.