On the variance of the error associated to the squared return as proxy of volatility