Course contentsShow
← Machine Learning and Deep Learning
Lesson 2389 of 3,53852. Time Series and ForecastingFree lesson

White Noise and Random Walks

Understanding white noise as unpredictable variation and random walks as integrated processes with no mean reversion.

White Noise and Random Walks

What you'll learn: You'll understand the fundamental building blocks of unpredictable time series: white noise (pure randomness) and random walks (accumulated randomness with no tendency to return to a mean).

What Is White Noise?

White noise is the purest form of unpredictability in time series. Imagine flipping a fair coin repeatedly and recording +1 for heads, -1 for tails. Each flip is completely independent of the last—no memory, no pattern, just pure randomness. In time series terms, white noise has:

  • Zero mean (centered around zero)
  • Constant variance (same spread over time)
  • No autocorrelation (past values don't predict future ones)

White noise is stationary because its statistical properties don't change over time. It's the irreducible random component you can't model or predict.

What Is a Random Walk?

A random walk is what happens when you accumulate white noise over time. Think of it as a drunk person taking steps: each step is random, but their position is the sum of all previous steps.

Mathematically: if today's value equals yesterday's value plus white noise, you have a random walk.

Key difference from white noise: Random walks are not stationary. They wander aimlessly with no tendency to return to a mean value. The variance grows over time—the longer the walk, the further it can drift from its starting point. This violates stationarity requirements.

Random walks appear in financial markets (stock prices) and other systems where changes are unpredictable but current levels depend on history.

Key Takeaway: White noise is pure, unpredictable noise (stationary), while random walks accumulate that noise into a non-stationary series that drifts without mean reversion—understanding this distinction is crucial before applying forecasting models that assume stationarity.