Has anyone ever tried a machine learning approach to find a model the gives +/- the same shape as the non-causal smoothers (Hodrick-Prescott, SSA, ...). The inputs of the model would be a very good filter like the JMA, and also technical analysis indicators, or other variables that could possibly be related to the non-causal smoother, and the output would be a very smooth, low lag filter.
btw: before anyone mentions neural nets again: there's much better techniques these days!
I work with R; If you're interested, send me a pm
btw: before anyone mentions neural nets again: there's much better techniques these days!
I work with R; If you're interested, send me a pm