Autocorrelation

Repeatability can be measured
Autocorrelation is a tool that can be used to analyse both patterns and patterns of phenomena that repeat over time. It is a measure that tells us how closely values are correlated to data from previous observations.

We use it in time series. Autocorrelation allows us to describe to what extent, compared to the previous time series, a value is correlated with a value from the previous time series.

Obviously, time series are characterised by observations with a constant time shift. The unit of measurement can be, for example, one month, one year, etc.
When calculating the autocorrelation coefficient for the desired lag – apart from the number of observations and the values of consecutive observations – we also need the values of consecutive observations lagged by the lag value and the arithmetic mean for the observations.

This is how we investigate seasonality
In order to better understand what autocorrelation is and how to use it, it is necessary to introduce the concept of lag. By lag, we mean the number of observations that have been omitted. We can count the autocorrelation between the first and the second observation, as well as between the first and the third observation, the first and the fourth observation, etc.

An autocorrelation with a lag of ‘one’ is one that compares values with previous values.
By using autocorrelation, we can identify the repeatability of specific phenomena for every period, which is the period of the aforementioned lag.

The attractiveness of autocorrelation is undoubtedly determined by its practical application (e.g., in economic sciences in the study of the intensity of a phenomenon’s recurrence over time).