Stationarity in Python

Most time-series models assume it, so test and difference first.

import numpy as np
from statsmodels.tsa.stattools import adfuller, kpss

rng = np.random.default_rng(12)
walk = np.cumsum(rng.normal(size=300))

statistic, pvalue, lags, nobs, crit, _ = adfuller(walk)
print(round(statistic, 3), round(pvalue, 4), "non-stationary" if pvalue > 0.05
      else "stationary")
print({k: round(v, 2) for k, v in crit.items()})

differenced = np.diff(walk)
print(round(adfuller(differenced)[1], 6))
print(round(kpss(differenced, nlags="auto")[1], 4))

How it works

1. adfuller tests the null that the series is non-stationary.
2. A trend makes the test fail; differencing usually fixes it.
3. kpss tests the opposite null, which is why people run both.

Keywords and builtins used here

• as
• else
• for
• if
• print
• round

The run, in numbers

Lines

14

Characters to type

470

Tokens

150

Three-star pace

115 tpm

At the three-star pace of 115 tokens a minute, this run takes about 78 seconds.

Type this snippet

Step 2 of 6 in Time series, step 14 of 19 in Statistics with statsmodels.

← Previous Next →