Cherry picking loans that are an even multiple of $5,000 looks pretty good.
Bryce recently noted in another thread that univariate analysis alone was pretty worthless.
Of course you gotta do it in the first place to select a set of parameters for your model.
That said, I wonder if this $5k multiple parameter as a model factor would make any significant improvement to the model performance.
Perhaps some models yes and others no. Your mileage will vary.