Stochastic Calculus: Calculus for Things That Never Sit Still

A stock price never moves in a smooth curve. Zoom in as far as you like — a minute, a second, a millisecond — and it's still jagged, still unpredictable at that scale. Ordinary calculus was built for smooth things: curves with a well-defined slope at every point. Stochastic calculus exists because a huge number of real quantities — stock prices, particle positions, noisy sensor readings — simply aren't smooth, and yet we still need to do calculus on them: integrate them, differentiate them, build models out of them.

The building block: Brownian motion

The basic random process underneath stochastic calculus is called Brownian motion, usually written W(t). It has three defining properties: it starts at zero, its increments over non-overlapping time intervals are independent of each other, and each increment is normally distributed with a variance equal to the length of time that's passed. Put together, those three simple rules produce a path that is continuous everywhere but differentiable nowhere — no matter how far you zoom in, it never looks like a straight line.

That last property is the whole problem. Ordinary calculus is built on the idea that df/dt exists — that a function has a genuine, well-defined rate of change. For Brownian motion, dW/dt simply doesn't exist in that sense. You can't differentiate it the normal way. So a new toolkit is needed for functions that depend on it.

Itô's Lemma: the stochastic chain rule

In ordinary calculus, the chain rule for a function f(W(t)) would just be f'(W) dW. Itô's Lemma says that's not quite right when W is Brownian motion — there's an extra term:

df = f'(W) dW + ½f''(W) dt

That second term is the strange, defining feature of stochastic calculus, and it comes from a genuinely odd fact: for ordinary smooth functions, (dt)² is so small it can be ignored, but for Brownian motion, (dW)² behaves like dt itself, not like zero. It doesn't vanish in the limit the way it would for a smooth path. Once you accept that one strange fact, the rest of Itô's Lemma follows from ordinary Taylor expansion — it's not a different kind of mathematics, just ordinary reasoning applied carefully to a process that behaves unusually at small scales.

The intuition worth keeping: Brownian motion is so jagged that its squared wiggle accumulates at a steady rate over time — that's the (dW)² ≈ dt fact — and that steady accumulation is exactly what shows up as the extra term in Itô's Lemma.

A concrete example: modelling a stock price

The standard model for a stock price S(t) is geometric Brownian motion, written as the stochastic differential equation:

dS = μS dt + σS dW

In words: the stock drifts upward on average at rate μ (the dt term), while being buffeted by random noise scaled by its own value and a volatility σ (the dW term) — a bigger stock price means bigger absolute swings, which matches how real prices behave. Applying Itô's Lemma to ln(S) — not ordinary calculus, which would give the wrong answer here — produces the famous result that S(t) is log-normally distributed. That single equation, and the machinery needed to solve it correctly, is the direct ancestor of the Black–Scholes option-pricing model.

Why it matters beyond finance

The same toolkit — a drift term plus a noise term, differentiated correctly with Itô's Lemma — describes a particle diffusing through a fluid in physics, electrical noise in a circuit, and population sizes fluctuating under random birth and death events in biology. More recently, it's also the mathematics running underneath diffusion-based generative AI models: they train a network to reverse a stochastic differential equation that gradually turns a real image into pure noise, then run that reverse process to generate a new one from scratch.

Stochastic calculus has a reputation for being one of the harder corners of a maths degree, mostly because it's usually taught as a wall of measure-theoretic definitions before the actual intuition — a random process that's too jagged for ordinary calculus, and one extra term that accounts for it — ever gets a chance to land. Once that intuition is in place first, the formal machinery stops feeling arbitrary. If postgraduate or final-year applied maths is where you or your student is headed, that's exactly the order we teach it in — see our postgraduate tutoring or the full learning pathway here.