Time series are one of the most basic and versatile mathematical tools used in business ​​to support statistical models and consist of a series of data points linked to a particular point in time. Time series are frequently used to model anything from the evolution of a company’s sales, product prices and lead times on a yearly, monthly, daily or even hourly basis.

Within the terminology of a time series forecast, there is a primary level called the baseline, a long-term evolution called the trend, cyclical or periodic variations called seasonality, and other random variations we call noise. This allows data variations linked to regular cycles to be distinguished from an underlying decreasing or increasing trend.

However, as time series are a very simplified depiction of reality, there are frequent misinterpretations of data. For example, a calendar month is a somewhat arbitrary way of sectioning time and one shouldn’t be under the illusion that our months are homogeneous from a business perspective. Unequal numbers of days and weekends in a month can provide an explanation for what could first appear as discrepancies in the data.

Knowing how data is collected, and being aware of the limitation of time series forecasts is essential when choosing the right forecasting method.