Properties of time-series data are:
These properties are useful in order to understand which forecasting model to use when dealing with time-series analysis.
Trend
A trend refers to the overall direction of the data of a time-series over time (if it’s increasing or decreasing).
Seasonality
A time-series has seasonality if there is a periodic behavior over time which is predictable.
Stationarity
A time-series is called stationary if the mean and the standard deviations are constant over time, and if there is no seasonality.
We can turn a time-series data from non-stationary to stationary by:
- Removing the trend with a detrending operation, which will cause the mean to be constant over time.
- todo
Note
If a time-series is white-noise, then it’s also stationary, but the vice versa is true only if the mean is equal to 0.
We can check stationarity using one of these methods:
- Visually
- Global vs Local test
- With ADF - Augmented Dickey-Fuller Test