Did you www.talkstats.com ? Forum ? Statistical Research ? Statistical Research? ?
Because you intend to know the variance, I feel that is very inaccurate to get Confidence Intervals from it. The extreme quantiles, 0.025, 0.975 or 0.005 0.995 are the only that provide the *central* (in Probability) alpha 5% and 1% respectively, based on yours 400´000 values. In what concerns the skew and kurtosis estimation I would carefully chose them, I mean!