The Black–Scholes PDE

We now build the equation that started an industry. Let V(t, S) be the value of an option written on a stock whose price S_t follows geometric Brownian motion under the Black–Scholes model,

dS_t = \mu S_t\,dt + \sigma S_t\,dW_t.

The trick — due to Black, Scholes, and Merton — is not to price the option directly but to replicate it: build a portfolio of the stock and the option whose randomness cancels instant by instant. A portfolio with no risk must earn the risk-free rate r, or there is free money on the table. Force those two facts to agree and a single partial differential equation falls out — and, remarkably, the stock's expected return \mu vanishes along the way.

The hedging argument, line by line

Form a portfolio that is long one option and short \Delta units of the stock:

\Pi = V - \Delta\, S.

We will write down how \Pi moves over an instant, choose \Delta to kill the random part, then impose that a riskless portfolio earns r. Throughout we hold \Delta fixed across the instant (the self-financing assumption examined in replication).

Step 1 — Itô's lemma on V(t, S). Since V is a function of time and of the Itô process S, Itô's lemma gives dV = V_t\,dt + V_S\,dS + \tfrac12 V_{SS}\,(dS)^2. Substituting dS = \mu S\,dt + \sigma S\,dW and the box-algebra value (dS)^2 = \sigma^2 S^2\,dt and collecting the dt and dW pieces:

dV = \Big(V_t + \mu S\,V_S + \tfrac12\sigma^2 S^2\,V_{SS}\Big)\,dt + \sigma S\,V_S\,dW.

Step 2 — differentiate the portfolio. With \Delta held fixed over the instant, d\Pi = dV - \Delta\,dS. Substitute the dV just found and dS = \mu S\,dt + \sigma S\,dW:

d\Pi = \Big(V_t + \mu S\,V_S + \tfrac12\sigma^2 S^2\,V_{SS}\Big)\,dt + \sigma S\,V_S\,dW - \Delta\big(\mu S\,dt + \sigma S\,dW\big).

Step 3 — group the dt and dW terms. Collect the deterministic drift and the random kick separately:

d\Pi = \Big(V_t + \tfrac12\sigma^2 S^2\,V_{SS} + \mu S\,(V_S - \Delta)\Big)\,dt + \sigma S\,(V_S - \Delta)\,dW.

Step 4 — choose \Delta = V_S to cancel the randomness. The dW coefficient is \sigma S\,(V_S - \Delta); setting \Delta = V_S makes it zero. Notice the same choice annihilates the \mu S\,(V_S - \Delta) drift term — so the entire \mu contribution disappears at once:

d\Pi = \Big(V_t + \tfrac12\sigma^2 S^2\,V_{SS}\Big)\,dt.

This is the magic stroke: holding exactly V_S shares against the option leaves a portfolio with no dW — it is instantaneously riskless.

Step 5 — a riskless portfolio must earn r. If \Pi carried no risk yet grew faster or slower than the bank rate, a no-arbitrage trade would mint risk-free profit. So over the instant d\Pi = r\,\Pi\,dt, and with \Pi = V - \Delta S = V - S\,V_S:

d\Pi = r\Pi\,dt = r\big(V - S\,V_S\big)\,dt.

Step 6 — equate the two expressions for d\Pi. Steps 4 and 5 describe the same drift, so their dt coefficients match:

V_t + \tfrac12\sigma^2 S^2\,V_{SS} = r\big(V - S\,V_S\big).

Step 7 — rearrange into the standard form. Expand the right-hand side and move every term to the left:

V_t + \tfrac12\sigma^2 S^2\,V_{SS} + rS\,V_S - rV = 0.

There it is — the Black–Scholes partial differential equation. Every option on this stock, whatever its payoff, satisfies the very same equation; only the boundary and terminal conditions distinguish a call from a put from anything else.

Under the Black–Scholes model, the value V(t, S) of any derivative on a non-dividend stock obeys

\frac{\partial V}{\partial t} + \tfrac12\sigma^2 S^2\,\frac{\partial^2 V}{\partial S^2} + rS\,\frac{\partial V}{\partial S} - rV = 0,

a backward, second-order parabolic PDE, together with:

The most startling feature of the derivation is what is missing. The stock's expected return \mu entered in Step 1, survived into Step 3, then vanished completely the instant we set \Delta = V_S in Step 4. The final PDE contains r and \sigma but not \mu.

The interpretation is profound: an optimist who thinks the stock will soar and a pessimist who thinks it will languish must agree on the option's price, because both can hedge away the directional bet. Only the volatility \sigma — the size of the wiggle, not its direction — survives. This is precisely the seed of risk-neutral pricing: we may as well pretend the stock drifts at the risk-free rate r, and the answer is unchanged.

The PDE alone is solved by infinitely many functions; the conditions pick out the one that is the call. For a European call C(t, S) with strike K and expiry T:

Because the terminal condition is fixed at T and the PDE is solved backward in time, the value today is the smooth hockey stick relaxing into a curved surface as time to expiry grows — exactly the picture below.

Watch the value relax from the payoff

The solid curves are the call value C(t, S) against spot S, plotted at the chosen volatility for two times to expiry; the faint kinked line is the terminal payoff (S - K)^+ that the PDE relaxes back from as we step away from expiry. With more time or more volatility the curve lifts further above the kink — there is more chance of finishing in the money. Slide \sigma and the time to expiry and watch the hockey stick round off.

The slope of these curves is the hedge ratio \Delta = \partial C / \partial S — the number of shares to short — which we will study in the Greeks. That the price obeys a PDE at all is a fact we will need the theory of partial differential equations to solve in closed form.