Validation of new data with predictive values of past datas

[jpprovost]

Hi there,

I really don't know if I am at the right place to ask this, but I  have datas (obs) from the past 5 years (each month = new data).

I want to use PROC ARIMA (I don't know how to use it very well, but I read it could be the good tool to do what I want) to know if the last data produced from last month (e.g. October) is "valid" within a forecast of past values.

By example, I have the number of obs (nobs) stocked in a dataset DATA1. I have 62 observations and I want to know if the 63rd is valid within the predicted values. My table DATA1 is updated monthly by adding a new line with the latest datas. The trend is that the nobs should always increase, but for quality control I'd like to validate that the new data is correct and within the "margins" of the past datas. If I'm not clear, do not hesitate since English is not my native language.

Is there a way to do that?

[rselukar · Accepted solution]

You can certainly "validate" the new observation in your time series by using PROC ARIMA or some other time series procedures (such as PROC ESM and PROC UCM) in SAS/ETS. Any of these procedures can be used to generate forecasts with confidence band around the forecast values.  If the new value lies between the confidence band you can consider it a valid value.  For example, based on your monthly time series of 62 observations you can generate forecast for the 63rd month and if the actual value of the 63rd month falls within the confidence band for the 63rd month's forecast you can call it valid.  Of course, this validation process works only if you have used some reasonable forecasting model (ARIMA, ESM, or UCM).  If you are unfamiliar with time series forecasting procedures in SAS/ETS, you can start with PROC ESM, which is very easy to use: see https://go.documentation.sas.com/?docsetId=etsug&docsetTarget=etsug_esm_examples01.htm&docsetVersion...   for an example.

Note that for monthly data WINTERS, assuming the observations are always positive, or ADDWINTERS, for both positive and negative values, are good model choices.  See https://go.documentation.sas.com/?docsetId=etsug&docsetTarget=etsug_esm_examples03.htm&docsetVersion...

[jpprovost]

Will give it a try for sure.
Thanks @rselukar for your time and help.