Dealing with idiosyncratic cross-correlation when constructing confidence regions for PC factors