One problem is that of retaining ROBUSTNESS whilst achieving better targeting precision
 
 
- better targeting performance in a historic context seldom translates into better predictions in a forecasting context
- many methods offer performance benefits based on old data that  do not always work when applied in prediction mode