Consider a simple model 

              y_i = β₂x_2i + β₃x_3i + u_i ; i = 1,2,......,n

and the variables are in deviation form. The disturbance term u_i satisfies all the classical assumptions. Suppose x_2i or x_3i are multicollinear. Should you drop either x_2i or x_3i to have precise estimates of the remaining parameters ? If so, under what condition are you permitted to do so ? (15 Marks)