From a difference equation to an SDE
In ordinary calculus an equation like dX/dt = f(X) is shorthand for a recipe: at each instant, the rate of change of X is f(X). A stochastic differential equation is the same idea with one extra term — a kick of randomness driven by the Brownian motion B(t) you met earlier in this rung. We write it as dX = a(X, t) dt + b(X, t) dB. Read it out loud: over a tiny time dt, X moves by a smooth, predictable amount a(X, t) dt — the drift — plus a random jolt b(X, t) dB whose size scales with the diffusion coefficient b(X, t) and whose direction is the unpredictable increment of Brownian motion.
The honest way to read dX = a dt + b dB is not as a derivative — Brownian paths are nowhere differentiable, so dB/dt does not exist as an ordinary slope. It is shorthand for an integral equation: X(t) = X(0) + the integral of a ds plus the Ito integral of b dB, taken from 0 to t. That second piece is exactly the Ito integral you built in guide 4, and everything that makes SDEs strange traces back to it. An SDE is a statement about accumulated change, written in differential clothing.
Solving the stock-price SDE: geometric Brownian motion
A stock price should never go negative, and a fixed-dollar wobble means more for a $10 stock than a $1000 one — what stays roughly steady is the percentage move. So the natural model is dS = mu * S dt + sigma * S dB: both the drift and the noise scale with the price S itself. Here mu is the expected growth rate and sigma is the volatility. This is geometric Brownian motion, the SDE at the heart of mathematical finance, and we can solve it exactly with one application of Ito's lemma.
The trick is to look at the logarithm, because for ordinary exponential growth the log grows linearly. Let Y = ln(S) and apply Ito's lemma to f(S) = ln(S). Ordinary calculus would give dY = (1/S) dS, but Ito's lemma adds the extra second-order term (1/2) f''(S) times the quadratic variation of S, which here is (sigma * S)^2 dt. Since f''(S) = -1/S^2, that correction is (1/2)(-1/S^2)(sigma^2 S^2) dt = -(1/2) sigma^2 dt. Substituting dS = mu S dt + sigma S dB and simplifying, the S cancels beautifully and we get dY = (mu - (1/2) sigma^2) dt + sigma dB.
Now dY has constant coefficients, so it integrates directly: Y(t) = Y(0) + (mu - (1/2) sigma^2) t + sigma B(t). Exponentiating back, S(t) = S(0) * exp[(mu - (1/2) sigma^2) t + sigma B(t)]. Because B(t) is normal, the exponent is normal, so S(t) is lognormal — a clean fact you can build straight on the lognormal distribution. Notice the haunting -(1/2) sigma^2 inside: that is the Ito correction surviving into the answer, and it is the difference between a sloppy guess and the truth.
Pricing an option: cancel the randomness
A European call option is the right (not obligation) to buy the stock for a fixed strike price K at a future expiry T. Its payoff at T is max(S(T) - K, 0): worthless if the stock ends below K, otherwise the gap above K. The question Black, Scholes, and Merton answered is: what should this contract cost today? The genius move is not to forecast S(T) — it is to build a portfolio whose randomness exactly cancels the option's, so that no luck is left to be rewarded.
- Let V(S, t) be the option's value, an unknown function of price and time. Apply Ito's lemma to V: dV = (V_t + mu S V_S + (1/2) sigma^2 S^2 V_SS) dt + sigma S V_S dB, where subscripts are partial derivatives. The random dB rides entirely on the term sigma S V_S.
- Form a hedged portfolio: hold one option and short V_S shares of the stock, so its value is P = V - V_S * S. Its change is dP = dV - V_S dS. Substitute both Ito expansions.
- Watch the magic: the dB terms are sigma S V_S dB from the option and -V_S * sigma S dB from the shorted stock. They cancel exactly. The portfolio P has no random term left — over the next instant it is risk-free.
- A risk-free portfolio must earn the risk-free interest rate r, or arbitrage would exist: dP = r * P dt. Equate this with the surviving (non-random) drift of dP and the mu, the dB, and the gambler's optimism all vanish.
Carrying out the cancellation in step 4 leaves a clean partial differential equation with no randomness at all: V_t + r S V_S + (1/2) sigma^2 S^2 V_SS = r V. This is the Black-Scholes equation. Two things deserve a pause. First, mu — the stock's expected return — has completely disappeared; the price of the option does not depend on whether you are bullish or bearish, only on the volatility sigma and the rate r. Second, this PDE is the heat equation in disguise, which is no accident: heat diffuses exactly the way Brownian motion spreads.
The formula, and what it really says
Solving the Black-Scholes equation with the boundary condition V(S, T) = max(S - K, 0) gives the famous closed form for a European call. There is also a beautiful shortcut to the same answer: the Feynman-Kac formula says the solution of such a PDE equals an expectation of the payoff — but taken under a 'risk-neutral' world where the stock drifts at the risk-free rate r instead of mu. So the price is just a discounted expected payoff, which is exactly why mu dropped out: in the risk-neutral world it was never there.
European call price: C = S * N(d1) - K * e^(-r*T) * N(d2)
d1 = [ ln(S/K) + (r + (1/2)*sigma^2) * T ] / (sigma * sqrt(T))
d2 = d1 - sigma * sqrt(T)
N(.) = standard normal CDF (the bell curve's running total)
S = price now, K = strike, r = risk-free rate, T = time to expiry, sigma = volatility
Tiny example: S = 100, K = 100, r = 0, sigma = 0.20, T = 1
d1 = (0 + 0.02)/0.20 = 0.10, d2 = -0.10
N(0.10) approx 0.5398, N(-0.10) approx 0.4602
C approx 100*0.5398 - 100*0.4602 = 7.97 (about $7.97)Read the formula as a story. The term S * N(d1) is what you expect to collect from owning the stock in the scenarios where the option finishes in the money; the term K * e^(-r*T) * N(d2) is the discounted strike you expect to pay, weighted by N(d2), the risk-neutral probability that S(T) ends above K. The whole price is benefit minus cost, averaged over the lognormal future the SDE handed us. Every symbol traces back to a piece of this rung: sigma * sqrt(T) is the standard deviation of ln(S(T)), straight from geometric Brownian motion.
Honest limits, and where SDEs go next
Be honest about the model's assumptions, because real markets break them. Black-Scholes assumes constant volatility, continuous trading with no transaction costs, no jumps in the price, and lognormal returns. Reality has volatility that itself wobbles (volatility clustering), occasional crashes that are sudden jumps no diffusion can make, and fatter tails than the lognormal predicts. The clearest fingerprint of the mismatch is the 'volatility smile': if the formula were perfect, the sigma implied by market prices would be the same for every strike, but it bends into a smile. The model is a brilliant first approximation, not the final word.
SDEs are far bigger than finance, and you have already met the other landmark example in this rung's vocabulary: the Ornstein-Uhlenbeck process, dX = -theta (X - m) dt + sigma dB. Its drift pulls X back toward a mean m — a 'mean-reverting' spring fighting the noise — so unlike a free Brownian motion it settles into a stable normal distribution rather than wandering off forever. The same dt-plus-dB grammar models neuron voltages, particle velocities, interest rates, and population sizes. Learn to read drift and diffusion and you can read all of them.
Step back and see the whole arc of this rung. You built Brownian motion as the random walk's continuous limit, marveled at its broken-everywhere paths, discovered that those paths force a nonzero quadratic variation, and turned that fact into the Ito integral and Ito's lemma. This final guide spent all of it at once: an SDE is Brownian motion steered by drift and diffusion, Ito's lemma is the gearbox that transforms it, and Black-Scholes is what you can engineer when you grasp how randomness accumulates. From a coin flip in the very first guide of this ladder to pricing a derivative — that is the reach of probability done carefully.