Subscribe Bookmark RSS Feed

Time Series Inconsistencies

I am running into major inconsistencies in Time Series Analysis, specifically ARIMA procedures, depending on how I load the data into JMP. In one method, I manually build a data table. In the other, I use JSL to create a data table and upload the data from a spreadsheet. The resulting data tables are identical. However, when running ARIMA procedures on those data tables, I sometimes get vastly different results/forecasts. It appears that the separate instances are following different paths to local solutions. I'd like to know how to force the same path, regardless of how the data is loaded. Does anyone know of a way to "seed" or initialize the ARIMA process so that it always reaches the same solution? Thanks for any help.



Jun 23, 2011

It's not my area, but I think ARIMA is deterministic. There must be some variation or a bug. Best to follow-up with Tech Support.