Handling Important Variables in Econometric Modeling

When there are too many variables, a variable master list is prepared to move forward with. To make this list many statistical methods are applied. And few variables are included if they make strong business sense. However these variables may not necessarily be good enough to be included. This discovery is made at much later stage, specially when the model is built and is problematic with high multicollinearity and/or over fitting. I have found such variables creating lots of problems to the modeler. I came across such problem once and here is my suggestion for checking such variable beforehand.

Make a table of variables with a business interpretation variable with value 1 if the variable is expected to have a positive effect and 0 otherwise. Here 1 means variable with positive correlation (intuitive) with the DV and 0 means negative correlation.This table can be merged with the correlation table of all IDVs with DV and eliminate all variables which are found counter intuitive before actual model building process starts.

This is a one time work which will save huge time in model building process. Moreover this table can be made by thorough discussion, with business experts if needed. Thus, this may result to be better that the modeler's personal judgment.