Data Models - Linear vs. Autoregressive

Our forecasting module uses one of two different models to accurately forecast expenditure for your selected set of datasources, accounts, and campaigns. The model applied depends on the data provided.

AUTOREGRESSIVE FORECASTING

This is the most accurate model; it takes into account both recent history and seasonal trends for its prediction.

However, it also has certain requirements for the data that will be used to fit the model. In order to use autoregressive forecasting with seasonality, the selected dataset has to fulfill two requirements:

• More than 30 days of data in the recent past.
• The daily mean of cost for the last 30 days is more than 1.

LINEAR FORECASTING

This is a more simplistic model, and will only be used when a dataset does not satisfy the criteria for Autoregressive forecasting. Linear forecasting is based solely on a mean average calculation from the available historical data. Since seasonality is not taken into account, the future prediction will be one constant number for each day, representing the average of previous spending amount / number of days.