|Metrix » Tuning|
First, when we say that we exclude from tuning those months whose thresholds are below a minimum, it does *not* mean that they are disregarded or somehow ignored (though we'll be the first to admit to the understandable confusion by the term "exclude") from the baseline model. Here is why:
The full baseline model is "constant x # of days plus a number of terms in HDD and/or CDD and/or independent variables, plus (optionally) modification factors".
The next question is: "given this form of a baseline model, what is the best way to determine its parameters?". Here is where regression, bill-matching algorithms, replication algorithms, etc. come in. The regression is only the first step of the full model development. To "exclude months with CDD or HDD below threshold from the regression" is not the same as saying to exclude such months from the model.
To illustrate this point, take a look at the tuning report, specifically at the monthly deviations and the annual deviation (the mean bias). These are the true criteria of the goodness of the model, not the R2 value. Note that all monthly deviations are used to determine the mean bias, regardless of whether a bill is included in the regression or not.
So, the full baseline model must be applied to all future months. The same model must, by definition, be applied to all tuning months, especially where the calculation of Net Bias and Mean Monthly Error are concerned. Both statements are implicitly warranted in Metrix, but, again, we are the first to admit that some of our own past statements and literature have contributed to confusion.
To answer this user's question directly: all CDD and HDD values during the performance period must be considered, regardless of threshold. Similarly, all CDD and HDD values during the tuning period must be (and are) considered by Metrix in calculating the net mean bias and the monthly mean error (see tuning report). The only exclusion, if any, is from the regression, and there only as a means to ensure more robust regression results.
Otherwise feel free to contact our Tech Support staff at (805) 329-6565, or via email at firstname.lastname@example.org.