Skip to content
Tutorials Center
My learning

Topics

Lesson 30 of 36 · Random Walks and Brownian Motion

Brownian Motion Basics

Get Brownian motion as the scaling limit of a random walk. Learn its defining properties, covariance min(s,t), quadratic variation, and simulate it.

Zoom out on a random walk: make the steps smaller and faster. The rough, jagged path settles into a continuous curve with a life of its own, called Brownian motion or the Wiener process. It is named after the botanist Robert Brown, who observed pollen grains jittering in water, and after Norbert Wiener, who put it on a rigorous footing. It is the basic building block of continuous-time random models in physics and finance.

From random walk to Brownian motion

Take the symmetric random walk Sₖ with ±1 steps. Speed up time by a factor n and shrink space by √n:

Text
W^(n)(t) = S_{⌊nt⌋} / √n

At a fixed t, S_{nt} has variance nt, so W^(n)(t) has variance t. By the central limit theorem its law tends to Normal(0, t). In fact the whole path converges (this is Donsker's theorem). The √n scaling is the only one that keeps variance finite and non-zero.

Python A random walk rescaled toward Brownian motion

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

As n grows the staircase becomes a continuous but very rough curve.

Definition

A process W(t), t ≥ 0, is a standard Brownian motion if:

  1. W(0) = 0.
  2. Independent increments: for 0 ≤ t₀ < t₁ < … < tₖ, the increments W(t₁) - W(t₀), …, W(tₖ) - W(tₖ₋₁) are independent.
  3. Gaussian increments: W(t) - W(s) ~ Normal(0, t - s) for s < t.
  4. Sample paths are continuous.

Compare with a Poisson process: the same independent, stationary increments, but with normal increments and continuous paths in place of Poisson counts and jumps.

Basic consequences

  • E[W(t)] = 0 and Var[W(t)] = t.
  • Covariance: Cov(W(s), W(t)) = min(s, t). For s < t, write W(t) = W(s) + (W(t) - W(s)); the second piece is independent of W(s), so the covariance is Var W(s) = s.
  • Scaling: W(ct)/√c is again a standard Brownian motion. Zooming in by c in time requires zooming by √c in space to see the same picture (a self-similar fractal).
  • Symmetry and restart: -W is a Brownian motion; starting afresh at any fixed time s gives a new Brownian motion W(s+t) - W(s).
  • With drift μ and volatility σ: X(t) = μt + σW(t) has mean μt and variance σ²t.

Roughness: quadratic variation

Brownian paths are continuous but nowhere differentiable. A quantitative mark of this is the quadratic variation: partition [0, T] into n pieces of width Δ = T/n and sum the squared increments:

Text
Σ (W(t_{k+1}) - W(t_k))² → T       as n → ∞

Each squared increment has mean Δ, so the sum has mean T, and its variance 2TΔ → 0. A smooth function would give quadratic variation 0, because (f' Δ)² summed over n pieces is of order Δ. This non-zero limit is why ordinary calculus rules fail for Brownian motion and why Itô calculus (with its extra ½σ² terms) exists.

Python Simulate Brownian motion and check its properties

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The estimated variance, covariance and quadratic variation match the theory. Notice the quadratic variation stays near 1 as the partition becomes finer, whereas for a smooth curve it would fall toward 0.

Hitting times and the maximum

Passing to the limit in the reflection formula gives P(max over [0,t] of W ≥ a) = 2 P(W(t) ≥ a) = 2 (1 - Φ(a/√t)) where Φ is the standard normal cumulative distribution function. So Brownian motion eventually reaches any level, but the expected time to do so is infinite.

The most common mistake is to treat dW like an ordinary small change. Because (dW)² behaves like dt rather than being negligible, products of increments matter. The next lessons use this when simulating prices.

Check your understanding

Quick check · 4 questions

of 4 answered You scored / 4 · perfect!

  1. 1. What are the mean and variance of W(t) for a standard Brownian motion?

    W(t) ~ Normal(0, t).
  2. 2. What is Cov(W(2), W(5))?

    Cov(W(s), W(t)) = min(s, t) = 2.
  3. 3. What does the quadratic variation of Brownian motion over [0, T] converge to?

    The sum of squared increments over finer and finer partitions tends to T.
  4. 4. Why is the random walk rescaled by 1/√n, not 1/n?

    S_{nt} has variance nt, so dividing by √n yields variance t.

Summary

  • Brownian motion is the scaling limit of the symmetric random walk with space scaled by 1/√n.
  • Defined by W(0) = 0, independent Normal(0, t − s) increments, and continuous paths.
  • Var W(t) = t, Cov(W(s), W(t)) = min(s, t), and the process is self-similar.
  • Quadratic variation over [0, T] is T, so paths are rough and calculus changes.
  • Continuous-time models built on it appear in the stock price project in Chapter 9.

Next: Simulating Continuous-Time Chains with Gillespie's Method