Random Walk

A random walk is a mathematical process where a value changes by random, independent steps over time. The simplest example: start at position 0 on a number line. At each time step, flip a fair coin: heads moves you +1, tails moves you -1. The sequence of positions {0, 1, 0, -1, -2, -1, ...} is a simple random walk.

The key property is independence of increments: knowing the entire history of past steps tells you nothing about the next step. The walk has no memory.

In finance, the random walk hypothesis states that stock price changes are like random steps — each day's return is independent of previous returns. This was popularized by Burton Malkiel's 1973 book "A Random Walk Down Wall Street" and is closely related to the Efficient Market Hypothesis. If prices truly follow a random walk, then technical analysis (chart patterns, trends, support/resistance levels) cannot predict future prices.

The random walk model is also the discrete-time precursor to Brownian motion — as the step size and time interval both shrink to zero, the random walk converges to a continuous Brownian motion path.

The simple symmetric random walk (equal probability of +1 and -1 steps) has several important properties:

• Expected position: E[S_n] = S₀ for all n. The expected position is always the starting point — the walk has no drift.
• Variance: Var(S_n) = n. The variance grows linearly with the number of steps, so the standard deviation grows as √n. After 100 steps, you're on average about 10 steps from the origin.
• Recurrence (1D): A one-dimensional random walk returns to the origin with probability 1 — no matter how far it wanders, it always comes back. (This is not true in 3 or more dimensions.)
• Arcsine law: The fraction of time a random walk spends positive is NOT uniformly distributed — it follows the arcsine distribution. Counterintuitively, the walk spends most of its time either mostly positive or mostly negative, not split evenly.
• Markov property: The random walk is a Markov chain — the next step depends only on the current position, not the path taken.

For a biased random walk (probability p > 0.5 of going up), there is a positive drift: E[S_n] = S₀ + n(2p - 1). This models a stock market with a positive expected return.

Problem: A gambler starts with $50 and makes $1 fair bets (50% chance of +$1, 50% chance of -$1). The game ends when the gambler reaches $100 (wins) or $0 (goes broke). What is the probability of winning?

This is the classic gambler's ruin problem. For a fair random walk:

P(reaching $100 before $0 | starting at $50) = 50/100 = 50%

The result is intuitive for a fair game: starting halfway to the target gives a 50% chance. But consider the expected duration — it takes, on average, 50 × 50 = 2,500 bets for the game to end. Random walks wander extensively before being absorbed.

With a slight edge: If the gambler wins each bet with probability 51% instead of 50%, the probability of reaching $100 before $0 is approximately:

P = (1 - (49/51)⁵⁰) / (1 - (49/51)¹⁰⁰) ≈ 88%

A tiny 1% edge transforms a coin-flip outcome into near-certainty over 100 dollars. This demonstrates the power of a small positive edge over many trials — the mathematical foundation of quant trading.

The random walk hypothesis has profound implications for quantitative finance:

• If prices are a pure random walk: No trading strategy can consistently outperform the market (after costs). Technical analysis is useless. Index investing is optimal. This is the implication of the Efficient Market Hypothesis in its strong form.
• Reality is more nuanced: Decades of research have shown that stock prices are approximately random walks but not perfectly so. Small departures from randomness — mean reversion at short horizons, momentum at intermediate horizons, value effects over long horizons — represent the alpha that quant firms exploit.
• Testing for randomness: Statistical tests (variance ratio tests, autocorrelation tests, runs tests) evaluate whether a price series is a random walk. Rejecting the random walk hypothesis for a particular instrument or time horizon suggests that predictable patterns exist — and potentially tradable.
• Modeling foundation: Even though prices aren't perfect random walks, the random walk / Brownian motion model remains the starting point for most financial models. Departures from the random walk (fat tails, volatility clustering, predictable patterns) are modeled as modifications to the basic framework.

The practical takeaway for aspiring quants: understanding the random walk is essential both as a baseline model and as the null hypothesis against which all alpha signals are measured.