Prophet Modeling (2): How to Organize Variables?

During my discussion with Mr. L & Ms. H on Prophet modeling, I briefly touched about how to organize variables more efficiently into various Variable Groups (from A to Z).

In the existing reserving workspace and simplified asset share workspace (deterministic models) which set up by Ms. H & I (also Ms. B) previously, we also split the Variable Groups into “core variable groups” and “input variable groups”, which is similar to the standard Conventional Library. The major different between our approach and the standard Conventional Library is we have the same type of classifications for both core variables & input variables – i.e. A-M for core variables, and N-Z for input variables.

For example, we have created Variable Groups for decrements like mortality rates – say we used Variable Group “D” (the 4th alphabet starting from A) for core variables (D – Decrements (C)), and similarly we use Variable Group “Q” (the 4th alphabet starting from N) for input variables (Q – Decrements (I)). This is because for the same types of variables, they can appear as either core variables or input variables:

1. The variable used to define the valuation mortality table needs to be an input variable because different valuation mortality tables may be used for different products.
   
2. The variables used to calculate monthly mortality rate (i.e. 1-(1-qx)^(1/12)) needs to be a core variable because the same formula is used regardless of the valuation mortality table used (i.e. generic for all products).

Of course, we also use different way classifying the variables – which we think to be more helpful in term of managing the variables systematically. For example, we have a variable for policy profile which most of the variables are read from model points – such as entry age, gender and sum assumed. “Systematic” is very important especially once we have more and more variables in our library.