Univariate and multivariate forecasts

Univariate forecasts

A univariate forecast predicts the future values of a single variable based on historical data, using the Holt-Winters model.

Multivariate forecasts

A multivariate forecast takes into consideration multiple time-dependent variables. The forecast model looks not only at historical data, but also the dependencies between the specified variables to create a forecast. The selected variables play a significant role in driving the forecast results. Therefore, it is important to select variables that have high correlation with the historical data.

Multivariate forecasting uses one of two models, vector auto-regressive (VAR) and auto-regressive integrated moving average (ARIMAX). If the historical data consists of less than 64 data points, VAR is used. If future values are included in the forecast, ARIMAX is used. If the historical data consists of more than 63 data points, but future values are not included in the forecast, then the optimal model is used. You have the option to include or exclude future values when you configure your forecast.

The variables for a multivariate forecast can be selected from different cubes on the same Planning Analytics database, as long as both the historical cube and the variable cube use the same time dimensions. Time dimensions are the dimensions placed in the columns position of a view to represent a continuous set of data points.

Note that neither multivariate models are seasonal. If your data is highly seasonal, then univariate forecasting is recommended.