Double exponential smoothing for unevenly spaced series?

drgs

New Member
#1
I have so far been using this formula (on page 10 in document below), to calculate simple exponential moving average:
http://www.eckner.com/papers/Algorithms for Unevenly Spaced Time Series.pdf

time_interval = (now - last_update) / tau
w = exp(-time_interval)
w2 = (1 - w) / time_interval
new_ema = old_ema * w + current_value * (1 - w2) + previous_value * (w2 - w)
If time interval is very long, this method assumes that the value changes linearly from previous_value to current_value.

If I was to extend this method to double exponential smoothing, how would I do that?

Standard double EMA:
double_ema_new = double_ema_old * alpha + (current_value + previous_trend) * (1-alpha)
current_trend = old_trend * beta + (double_ema_new - double_ema_old) * (1-beta)