Rounding of $2-10^{20}$ in IEEE double precision

How do we get the rounding of $$2-10^{20}$$ in IEEE double precision? The textbook says it is $$-10^{20}$$, but I do not know why. I think my textbook only explains the rule for rounding mantissa.

Determine the binary exponent of $$10^{20}$$. From that determine the value of the lowest bit of the mantissa of $$10^{20}$$. Then ask yourself: How large does a number x have to be at least so that round (x-$$10^{20}$$) is not $$-10^{20}$$?