04-13-2015 11:22 AM
I am new to SAS. I am starting to learn and apply proc UCM, and have a couple questions about it.
Could you help me understand how can I get the numerical fitted values on the training set (or on the historical record)? For example, if n0 and n1 are the first and last value of actual observation in the historical record separately, how can I get the numerical fitted values after estimation? Are those values stored under "S_NOIRREG" after using OUTFOR?
I am also confused by the ACF and PACF plots generated by SAS. Do they reflect ACF and PACF of predicted errors on the "Out of sample set" (or test set)? I re-plot the "Residuals" column by R, but found that they are not the same.....
Could you help me to understand these? I appreciate your help so much!
04-14-2015 08:26 AM
You can use the skipfirst= and back= options in the ESTIMATE statement to control the span of measurements used for parameter estimation (ESTIMATION SPAN). The (one-step-ahead) residual diagnostics (such as ACF and PACF plots) are based on the ESTIMATION span.
Similarly the skipfirst= and back= options in the FORECAST statement control the span of measurements used for producing forecasts. Miultistep-ahead-forecasts are produced based on the end of forecast span.
I do not know R implementation of the UCM models and cannot comment. I think the UCM doc is reasonably complete and explains the meaning of all component estimates (filtered and smoothed) as well as the residual diagnostics.