A time series is stationary if its statistical properties, such as its mean and variance, do not change over time. Many statistical tools assume stationarity, which is why quants usually model returns and spreads instead of raw prices.
Prices versus returns
| Price series | Return series | |
|---|---|---|
| Mean level | Wanders, with no fixed average | Roughly constant around a small mean |
| Typical behaviour | Trends and random-walk drifts | Fluctuates around a stable level |
| Stationary? | Usually not | Approximately |
Why it matters
If a series drifts, a model fitted to its past average describes a level that no longer applies. Regressing one trending price on another can show a strong relationship that is only a coincidence of both drifting, called spurious regression.
How it is tested
- Augmented Dickey-Fuller (ADF) test: the null hypothesis is that the series has a unit root (is non-stationary). A small p-value suggests stationarity.
- KPSS test: the opposite null, that the series is stationary.
Making a series stationary
- Take differences or percentage returns, instead of prices.
- Remove a trend or seasonal pattern.
- Use a combination of two prices that is stationary even though each price is not. This is cointegration, the basis of pairs trading.
Cautions
- Returns are not perfectly stationary: volatility changes over time. See GARCH models and market regimes.
- Over-differencing can remove useful information such as long-term level signals.
- Relationships that hold in one period may not hold in another. See model drift.
Frequently asked questions
What is stationarity?
It means a time series has a constant mean, variance and correlation structure over time.
Why are stock prices not stationary?
Because they wander with no fixed average level, behaving like a random walk with drift.
How do you test for stationarity?
With tests such as the Augmented Dickey-Fuller test, where the null hypothesis is a unit root, or the KPSS test, where the null is stationarity.
How do you make a price series stationary?
Usually by using returns or differences, or by combining prices in a cointegrated spread.