Applies to: @RISK 6.x/7.x, Industrial Edition

"Auto Detect" button in Time Series Fitting

Background:

Auto-detecting time-series transformations rely on heuristics.

You must review and adjust results based on your knowledge of the data.

 

Method used by @RISK for Autodetect

@RISK does not use KPSS or Dickey–Fuller tests.

1. KPSS isn’t suitable because @RISK doesn’t support trend-stationary series.
2. DF tests have limitations and assumptions (such as requiring seasonality and other sources of non-stationarity to be removed first).

Instead, @RISK uses a method adapted from electronic signal processing:

It analyzes small data windows, compares their statistics over time, and tests which combination of transformations (functional, differencing, and deseasonalizing) produces stable, acceptable results.

Proprietary techniques help handle issues like avoiding over-differencing and determining seasonal periods.