Its is due to roundoff of single precision floating point where there are only 24 bits to represent the mantissa (or fraction) and 8 bits for the exponent. Thus 8 bits of the mantissa are alloted for the sign bit and the 123. The remaining 16 bits must represent the 0.9881 in powers of two. Thus if you do 0.9881*2^16 = 64756.1216 and use the whole part that only give you 64756/2^16 = 0.988099 as the closest representable value.