How accurate is the asymptotic approximation to the distribution of realised volatility?