-- THIS IS AN ADVANCED TOPIC ---
The 2 snapshots above give the 3 available algorithms showing good potential in capturing market dynamics by way of a no-lag realtime smoothing algorithm. Our Swing indicator uses the Lag128 algorithm, despite Lag128T being 'theoretically' the most accurate one statistically. One possible reason is stats are often simply wrong when it comes to time series. One could debate about this non-stationarity, fat-tail distributions etc, but that is besides this topic and could be a little difficult to grasp. In any case, the two calculations may sometimes come pretty close. I would certainly prefer Lag128T to be as good as Lag128, as it calculates much much faster.
Obviously the optimised algorithm in Lag128 has been tweaked for time series, in particular to favour a low error level on the most recent bars (in other words, mistakes should rather belong to the past and be ignored). Insisting on a good model over recent bars provides far better patterns overall.
The last algorithm LombSin is very very processing intensive and should be used with caution. It goes further than the other 2 in extracting salient cycles along with the phase, thus allowing a price projection over a few bars. That would seem great if only such cycles had any chance to 'survive' over time. A lot of serious physicists will indeed tell you that cycles are very elusive or show you an instantaneous picture of what cycles have just looked like, and have no predictive value whatsoever. I would tend to agree.
However, i have noticed that when the LombSin error level falls close to the Lag128 error level (which is close to optimal), one can fit cycles correctly and predict at least a few bars ahead. Having said that, one may also just as well get that information from other indicators (MM, Fib etc) in our toolset.
