Random models of evolution are instrumental in extracting rates of microscopic evolutionary mechanisms from empirical observations on genetic variation in genome sequences. In this context it is necessary to know the statistical properties of empirical observables (such as the local homozygosity, for instance). Previous work relies on numerical results or assumes Gaussian approximations for the corresponding distributions. In this paper we give an analytical derivation of the statistical properties of the local homozygosity and other empirical observables assuming selective neutrality. We find that such distributions can be very non-Gaussian.