Most methods for a time series forecast rely on the fact that the series is stationary. This makes sense, since this increases the probability of repeating a certain behavior in the future and makes the prediction easier.
How can we know whether a given time series is stationary or not? There are formal, statistical methods to measure this, but we can also look at some properties of the series. There are three main checks of stationarity in practice:
- The mean value is constant (does not depend on time).
- The variance is constant.
- The covariance of the elements i and i+m is constant.
In our previous example, in the Modeling and visualizing time series section, we plotted the moving average (mean) and variance. If you revisit the diagram, you will see that none of them is constant with time, hence the series is non-stationary, and we...
