I have a sample of n0 subjects, m0 are events, n0-m0 are non events. I have three continuous predictors, X1, X2, and X3, for the event.

Is it possible to use proc optmodel to answer this question:

Suppose I reduce my sample size from n0 to n1, what constrains do I put on X1, X2, and X3 so that I will minimize m1 (the number of events in n1)?

In context, I have a portfolio of 100 different stocks, 70 made me money in the past month, 30 lost me money in the past month. The predictors of a stock being able to make me money are others' demand, company monthly profit, and average daily traded volume. Let's say my portfolio can have as few as 80 stocks - what constraints do I put on my predictors so I can have as few losing stocks as possible? (i.e.Do I get rid of all stocks of companies whose monthly earning is <$1M and stocks whose average daily trade volume is >100,000?) And how do I mathematically determine these constraints?

I know in proc optmodel you specify the constraints and the proc will optimize your objective function, but what if you have an objective function and want the proc to determine constraints for you? ]]>