Help choosing the optimal time series analysis package

Question

I am developing an app for time series analysis that should support the following:

Are you familiar with open source stat packages other than R, that supports these models? (Python?)

Add features that test the normality of the errors which would include 1) Intervention Detection to disclose Pulses/Level Shifts/Seasonal Pulses 2) non-constant variance of the errors due to changes in parameters over time or due to break-points in variance suggesting the need for power transformations and/or Garch .You should also be looking for automatic detection of lead and lag structureSs (PDL) around user-specified causal/regressor series.Additionally being able to sort out the need for ARIMA versus fixed predictors e.g.day-of-the-week would along with optimal ARIMA structure. — IrishStat, Commented Mar 24, 2012 at 19:58
@lilile Yes AUTOBOX . Available from Automatic Forecasting Systems .They have 30 day demo available at autobox.com/30day.exe . — IrishStat, Commented Sep 26, 2012 at 19:03
@lililie I should have mentioned I am one of the developers of this program. — IrishStat, Commented Sep 26, 2012 at 20:10

Galen · Accepted Answer · 2021-07-16 06:49:36Z

1

pmdarima gives you (S)ARIMA models, with orders chosen by minimizing AIC. You don't want to do the Box-Jenkins approach, information criteria are superior. statsmodels gives you multiple regression, which in turn can fit curves. I don't know of a module that does exponential smoothing, Croston's method or Bass models.

Galen

9,6803 gold badges27 silver badges61 bronze badges

answered Jun 8, 2021 at 14:24

Stephan Kolassa

131k22 gold badges264 silver badges497 bronze badges

Add a comment |

1 Answer 1