I read in a bunch of papers where the authors applied genome-wide scans for selection (e.g. population branch statistics (PBS), integrated haplotype score (iHS) etc) that the significance threshold to consider a genomic window under selection was determined by simulating neutral genomic datasets for many repetitions (e.g. 1000) and then taking one value at the tail of the distribution as the threshold.

What I don't understand is how valid this approach is, since you'd never simulate the entire genome of you species for the sake of time, and even with the most sophisticated simulators like SLiM that allow you to add complexity to the model you may actually just lack several biological relevant information for your model to be realistic.

In case one wants to pursue this road, how to implement it?