In one sentence
A random walk says each price move is an unpredictable step up or down, and Brownian motion is the smooth, continuous-time version of the same idea.
How it works
Watch a team's Match Odds price in the hour before kick-off and it jiggles about: a tick in, two ticks out, one back in. A random walk treats each of those moves as a coin toss. The walk has no memory, so knowing the last five moves tells you nothing about the next one.
The key result is that spread grows with the square root of time. After 30 random one-tick steps the typical distance from the start is about 5.5 ticks, not 30. Brownian motion is the limit as the steps become tiny and frequent, and is the building block for most models of prices.
For traders, the random walk is the null hypothesis. If your chart pattern cannot beat what a coin-toss walk would produce, it is probably noise. But there is a trap: a random walk in ticks is not the same as a fair market in probability, as the example below shows.
The maths
- Xt: the price (or log-odds) at time t.
- μ: drift, the average direction of travel per unit time.
- σ: volatility, the size of the random wiggles.
- Wt: standard Brownian motion, pure random noise that builds up over time.
- n: the number of steps taken.
- p0, pup, pdown: implied probabilities now and at the two exit prices.
In plain English: prices wander by roughly σ × √n, and if implied probability is a fair game, the chance of hitting each exit depends only on how far away it is in probability.
Worked betting example
Pre-match, you back the away team in Match Odds for £100 at 5.0 and plan to exit at 6 ticks either way. Between 4 and 6 each tick is 0.1, so your exits are 4.4 (price shortened, good for you) and 5.6 (price drifted).
- Exit at 4.4: lay £100 × 5.0 ÷ 4.4 ≈ £113.64. Green profit ≈ £13.64.
- Exit at 5.6: lay £100 × 5.0 ÷ 5.6 ≈ £89.29. Loss ≈ £10.71.
Tick random walk view: equal chance of each exit. Expected value = 0.5 × 13.64 − 0.5 × 10.71 ≈ +£1.46. Looks like a free edge.
Fair-probability view: implied chances are 20% now, 22.73% at 4.4 and 17.86% at 5.6. Chance of hitting 4.4 first = (20 − 17.86) ÷ (22.73 − 17.86) = 44%. Expected value = 0.44 × 13.64 − 0.56 × 10.71 = £0.00.
The apparent edge came entirely from assuming a coin-toss in ticks. In an efficient market it disappears. With 2% commission taken off the winning exit, the fair-probability view is about −£0.12 per trade.
Where it's good
- A baseline for testing whether a trading signal beats pure chance.
- Setting realistic stop and target distances given how far prices typically wander.
- Simulating price paths to stress-test a trading bot.
- The basis for option-style pricing of cash-out and hedging values.
Limitations and pitfalls
- Football prices jump on team news, goals and red cards, which smooth Brownian motion cannot capture.
- Prices near kick-off often trend (a team backed in keeps shortening), breaking the no-memory assumption.
- Betfair's uneven ticks mean a walk in ticks and a walk in probability behave differently, as shown above.
- Volatility is not constant; it rises sharply around team news and kick-off, and in-play after every goal.
- Many losing trading systems are just random walks with a story attached; backtest carefully.
How to build it
- numpy can simulate millions of walks in seconds; scipy.stats provides the normal distribution for Brownian motion.
- Model log-odds or implied probability rather than raw odds, so moves are comparable across the ladder.
- Tip: run your strategy on simulated random-walk prices first; if it "profits" there, your backtest has a bug.
Related methods
- Jump-diffusion adds sudden jumps to the smooth random walk.
- Mean reversion and momentum are the two ways prices can differ from a pure walk.
- Tick size maths explains why tick walks and probability walks diverge.
- Market efficiency is the reason the random walk is a sensible default.