Skip to content
ML Academy

Introduced on day 22

White noise

A series of uncorrelated values with constant mean and variance; the benchmark that model residuals should resemble if no linear structure is left.

In detail

White noise has zero autocorrelation at every lag. Forecast residuals that are not white noise contain information a better model could use. White noise is not necessarily Gaussian and says nothing about whether the model will forecast well out of sample or after a regime change.