Don't Use Moving Averages

Interesting, but a difficulty I see with the suggested binomial moving average is that it is two-sided -- it uses data both before and after a given time to smooth that observation. In online smoothing you cannot use a two-sided average.