Merton Jump-Diffusion European Call Price
A stock follows jump-diffusion dynamics under the risk-neutral measure:
$dS_t = S_t\bigl[(r - \lambda\kappa)\,dt + \sigma\,dW_t\bigr] + S_{t^-}(J - 1)\,dN_t$
where $r$ is the risk-free rate, $\sigma$ is the diffusion volatility, $N_t$ is a Poisson process with intensity $\lambda$, and each jump multiplier $J$ is drawn independently from $\ln J \sim N(\mu_J,\, \sigma_J^2)$. Define $\kappa = E[J - 1] = e^{\mu_J + \sigma_J^2/2} - 1$, so the drift compensates for the expected jump.
- Derive the European call price $C(S_0, K, T)$ as a Poisson-weighted infinite series of Black-Scholes prices. Write the full series representation.
- For each term in the series, specify the adjusted risk-free rate $r_n$ and adjusted volatility $\sigma_n$ that feed into the $n$-th Black-Scholes component.
- Explain the economic intuition: why does the price decompose as a mixture over the number of jumps?
Hints
- Condition on the number of jumps $N_T = n$. What does the terminal stock price distribution look like when you know exactly how many jumps occurred?
- A sum of $n$ independent $N(\mu_J, \sigma_J^2)$ random variables is $N(n\mu_J, n\sigma_J^2)$. Combined with the diffusion, the conditional log-price is Gaussian -- so you can apply Black-Scholes with adjusted parameters.
- Define $\sigma_n^2 = \sigma^2 + n\sigma_J^2/T$ and $r_n = r - \lambda\kappa + n\ln(1+\kappa)/T$. Average the resulting BS prices over $n$ with Poisson weights using intensity $\lambda' = \lambda(1+\kappa)$.
Worked Solution
How to Think About It: The Merton model adds random jumps on top of geometric Brownian motion. The key insight is that if you condition on exactly $n$ jumps occurring during $[0, T]$, the stock price is just a GBM with a modified drift and volatility -- so you can price it with plain Black-Scholes. The unconditional price is then a weighted average over all possible jump counts, with the weights being Poisson probabilities. This is a beautiful example of the law of total expectation: condition on the "hard" part (jumps), solve the "easy" part (GBM), then average.
Quick Sanity Checks:
- When $\lambda = 0$ (no jumps), only the $n = 0$ term survives, and you recover the standard Black-Scholes formula.
- As $\lambda \to \infty$ with small jumps, the jump component contributes additional variance, so the effective volatility should increase.
- Each Poisson weight $e^{-\lambda' T}(\lambda' T)^n / n!$ sums to 1, so the call price is a proper convex combination of BS prices -- it must lie between the cheapest and most expensive BS component.
Derivation:
The terminal stock price is:
$S_T = S_0 \exp\!\left[\left(r - \lambda\kappa - \tfrac{\sigma^2}{2}\right)T + \sigma W_T\right] \prod_{i=1}^{N_T} J_i$
Condition on $N_T = n$. The product of $n$ independent log-normal jumps gives:
$\prod_{i=1}^{n} J_i = \exp\!\left(\sum_{i=1}^{n} \ln J_i\right)$
where $\sum_{i=1}^{n} \ln J_i \sim N(n\mu_J,\, n\sigma_J^2)$. So, conditional on $n$ jumps, the log stock price is Gaussian:
$\ln S_T \mid N_T = n \;\sim\; N\!\left(\ln S_0 + \left(r - \lambda\kappa - \tfrac{\sigma^2}{2}\right)T + n\mu_J,\;\; \sigma^2 T + n\sigma_J^2\right)$
This is the same distribution as a GBM with adjusted parameters. Define:
$\sigma_n^2 = \sigma^2 + \frac{n\sigma_J^2}{T}$
$r_n = r - \lambda\kappa + \frac{n(\mu_J + \sigma_J^2/2)}{T} = r - \lambda\kappa + \frac{n\ln(1+\kappa)}{T}$
Note: the second equality uses $\mu_J + \sigma_J^2/2 = \ln E[J] = \ln(1+\kappa)$.
Conditional on $n$ jumps, the call price is just $C_{BS}(S_0, K, T, r_n, \sigma_n)$ -- the standard Black-Scholes formula evaluated at the adjusted rate and volatility.
The unconditional price uses the law of total expectation over $N_T \sim \text{Poisson}(\lambda T)$. However, there is a subtlety: we also adjust the Poisson intensity. Define $\lambda' = \lambda(1 + \kappa) = \lambda E[J]$. Then the Merton formula is:
$C = \sum_{n=0}^{\infty} \frac{e^{-\lambda' T}(\lambda' T)^n}{n!}\; C_{BS}(S_0, K, T, r_n, \sigma_n)$
The reason $\lambda$ shifts to $\lambda' = \lambda(1+\kappa)$ is that the risk-neutral drift compensation absorbs the expected jump: the factor $e^{-\lambda\kappa T}$ in the drift, combined with the jump size expectation, effectively changes the Poisson parameter from $\lambda T$ to $\lambda' T$ when you reorganize the series.
Practical Interpretation:
In practice, you truncate the series at $n = 10$-