1. PlanIQ
  2. Create a forecast model
  3. Understand how forecast horizons are calculated

PlanIQ calculates a forecast horizon based on your historical data and your selected algorithm. 

Historical data helps PlanIQ work out what time intervals can apply to the forecast horizon. PlanIQ calculates if forecasts can be predicted per week, month, or year.  

How far users can predict, the forecast horizon, is primarily based on the data provided, and secondly on the the algorithm. Some algorithms can predict for the same length as the historical data timeline, whereas others only predict for one third of the historical data timeline. For the former, even if the algorithm can predict a forecast horizon of 12 months, this can still be capped based on the historical data. 

Example:

Based on the historical data and the selected algorithm, PlanIQ calculates it can generate predictions on a weekly basis for a total period of 6 months. The total period is the forecast horizon. 

ETS, ARIMA, and Prophet algorithms

The ETS, ARIMA, and Prophet algorithms let you forecast for an extended horizon, so for nearly the entire timeline of your historical data. 

The supported forecast horizon is equal to the length of historical data minus one period. 

Example:

  • If you have 24 months of history, these algorithms allow a forecast of up to 23 months 
  • If you have 52 weeks of data, these algorithms allow a forecast of up to 51 weeks

Set a shorter forecast horizon

Even though you can have a longer term forecast, we recommend you set it for a shorter time period:

  • Set your forecast horizon to cover half as much time as your historical data covers
  • This generates higher quality, more accurate predictions than a longer horizon would produce

Example:

For 24 months of history, we recommend you forecast up to 12 months ahead.

CNN-QR and DeepAR+ algorithms

The neural network algorithms, CNN-QR, and DeepAR+ have specific requirements:

  • They only let you forecast up to a third of your historical timeline
  • Your historical dataset must have at least 300 data points

The ultimate maximum supported horizon depends on your historical data's time period and the number of data points it contains. 

Example:

  • If you have 24 months of history, these algorithms let you forecast up to 8 months, provided you have enough data points (a minimum of 300)
  • If the dataset has a low number of items (below 300 data points), or it's too sparse, then the supported forecast horizons are shorter, say 4-6 months 

Data points 

PlanIQ keeps track of the number of data points you provide. This ensures it always identifies the correct supported forecast horizons.

Amazon Forecast AutoML algorithm

The Amazon Forecast AutoML algorithm has the same properties and behavior as neural network algorithms.