Suppose we have far more variables than observations, so that K>N. We can include at most N-1 variables, at which point the R2 is 1. That suggests that we must select at most N-2 variables, if not N-30 variables. What rules might we use to decide which variables to include?