Generate Statistical Forecasts using PREDICT feature

Statistical forecasting uses historical data to identify trends, seasonality, and other statistical patterns to generate forecasts automatically without requiring manual calculations. The Predict feature enables forward-looking analysis by estimating future values from historical observations.

It supports configuration of confidence interval, seasonality, growth factor, algorithm selection, and hierarchy evaluation methods (Top-Down or Bottom-Up), allowing forecasts to be generated across planning hierarchies. Forecast results can be reviewed in both graphical and tabular formats.

How forecasting works#

The Predict feature uses the selected cell of the measure as a time series. It analyzes historical data within the selected historical date range and generates estimated values for the selected forecast period.

During forecasting, the system considers the following characteristics of the time series:

• Level: The underlying baseline value of the series.
• Trend: A sustained upward or downward movement over time.
• Seasonality: A repeating pattern that occurs at regular calendar or operational intervals.
• Past-value relationships: The extent to which previous observations help explain future values.
• Past-error relationships: The extent to which previous forecast errors improve future predictions.

Prerequisites#

Before running a statistical forecast, ensure the following:

• A forecast measure exists with either blank initial values or prepopulated values.
• The planning sheet contains sufficient historical data.
  • For monthly data, at least 24 months of historical data are required.
  • For quarterly data, at least 24 quarters of historical data are required.
  • Some forecasting algorithms require 36 or 48 historical periods to produce more reliable forecast results.

Configure statistical forecasts#

1. Create a planning model by assigning row fields, column fields, and measures under the value field.

Consider the model in the following image.

2. Create a new forecast measure (for example, 2026 Forecast) by selecting the required forecast period through December 2026.

3. Select the All Row Total cell of the 2026 Forecast measure.
4. Navigate to Model, Select Predict.

Row Selected represents the hierarchy member for which the forecast will be generated. Measure Selected is used as the historical input series, and forecast values are written into this measure.

These fields are automatically populated based on the selected planning sheet cell. Selecting another visible cell updates the selected row and measure in the Predict panel.

Lock the selection to prevent choosing another cell from changing the selected row or measure.

5. Specify the date ranges used for forecasting.

Historic Data Range specifies the historical periods used to build the statistical model. By default, all closed periods are selected. Forecast Date Range specifies the future periods for which forecasts are generated. By default, all open periods are selected.

You can modify both ranges to suit your business requirements.

6. Select an existing forecast profile or create a new one by selecting More Options (⋯).

A Profile is a saved collection of forecast settings that helps standardize forecast execution. Profile includes:

• Confidence (%).
• Growth Factor (%).
• Hierarchy evaluation.
• Negative-value handling.
• Algorithm choices.
• Seasonality choices.

You can create, duplicate, delete, or reset profiles. Use descriptive profile names so forecast configurations can be easily reproduced.

7. Set the Confidence (%).

Confidence (%) controls the requested width of the forecast confidence interval. It represents the probability that the actual value will fall within the predicted range. Typical confidence% choice include:

• 80%: Produces a narrower confidence interval suitable for central planning.
• 90%: A balanced confidence level for most business forecasts.
• 95%: Produces a wider, more conservative confidence interval.

8. Set the Growth Factor (%).

Growth Factor (%) applies a business adjustment to the statistical forecast. It also defines the expected rate at which the forecast values increase or decrease over time. Examples:

• 5% increases forecast values to reflect an expected business uplift.
• -5% decreases forecast values.
• 0% preserves the statistical forecast without adjustment.

9. Assign the Evaluation method to.

Bottom Up generates forecasts at detailed hierarchy levels and aggregates them into higher-level totals.

Top Down generates forecasts at higher hierarchy levels and distributes them into lower levels.

10. Set Round all negative values to zero as

Yes, when the forecasted measure cannot be negative. No, when negative values are valid for the business scenario.

11. Choose a forecasting algorithm that best matches the characteristics of your historical data.

The following forecasting algorithms are available:

• Trend Decomposition with MSTL - Breaks down complex data with multiple repeating patterns into simple parts so they are easier to handle.
• Exponential Smoothing - Focuses on smooth trends and clear, single-cycle seasons, giving more weight to recent data.
• ARIMA - Looks at recent spikes, drops, and lags to project where the trend is heading next.

For more information about forecasting algorithms, statistical models, and model orders, refer here.

12. Seasonality is a repeating pattern in data that occurs at regular time intervals, such as yearly or quarterly or both. Choose Set Seasonality as: Year, when the pattern repeats on an annual cycle. Quater, when the pattern repeats on an quarterly cycle. Both when the pattern repeats on annual and quarterly cycle.

13. Select Run Forecast to generate forecast values.

14. The forecast preview is available in both graphical and tabular formats:

• Historical values are displayed in grey.
• Forecast values are displayed in green.
• The shaded green band represents the selected confidence interval.

15. If necessary, select Reconfigure or Re-run Forecast to adjust forecasting parameters and improve forecast accuracy. When you are satisfied with the forecast, select Save Forecast.

16. You can also apply the forecast values to a planning scenario or export it to a CSV file.

17. Forecasts are generated for the selected measure and row category. As the Bottom-Up approach is used, values are automatically aggregated to the parent levels.

Choosing the right forecasting algorithm

Refer the below table for a summary or when you have a question as "Which algorithm to choose?"