Integration

Understanding integration

In the context of time series data, integrating means taking the difference of observations so the data can become stationary. In practice, we replace each value with the difference between the current value and the preceding one.

Usually, integrating once is enough, and we rarely need to integrate more than twice. Although integrating might make the data stationary, sometimes it might not be enough, so we should check that the resulting time series after integration is stationary.

How to add integration to forecasts

Integrations are combined with models such as AR and MA, but let's first look at how they behave on their own.