On the use of loss functions in the changepoint problem