It's not true that removing multicollinearity always improve your model, so I disagree. The other argument you used I already mentioned in the text.