BookmarkSubscribeRSS Feed

New Contributor


Jan 17, 2018

Time Series - Prewhiten

Hello Everyone,


My question pertains to understanding what the terminology in JMP's prewhiten platform means.

Also, understanding how that helps me determine an appropriate Transfer Function that I can use to test whether x causes y.


JMP's prewhiten output shows three different ACF's labeled "Prewhiten Corr", "Noise AutoCorr", & "Noise Partial AutoCorr".

What is the Prewhiten Corr, Noise AutoCorr, & Noise Partial AutoCorr refer to?

      Is the prewhiten corr the correlation between the residuals of the x's sarima model & filtered y response?

      Is Noise AutoCorr & Noise Partial AutoCorr the noise of the y filtered variable?


Do I use the Noise AutoCorr & Noise Partial AutoCorr to specify the y sarima portion of the Transfer function & use the sarima model for the x input during the prewhiten phase? 



I am interested in utilizing JMP's prewhitening tools to help identify if one of two time series variables are 'causing / changing' the other time series variable.


My class is using R. We fit essentially the sarima model to the x 'input' time series and save the residuals.

Meanwhile, the y 'output' is filtered based on the sarima model fit to x.


Finally, we plot the residauls of x's sarima model versus the filtered y.


I am really trying to figure out how do I get this graph of the x's residuals versus filtered y.

Then, determine if x causes y.


Thank you in advance for your help in this matter,