Location: L'ville, GA
Here's an out of the box idea: use volume as a metric to provide discrete dynamic-volume pseudo-bars (dvpb's), from which ATR (or whatever) can be calculated.
This would be *different* than NTB "Volume Bars" in that the dpb-discretization would be based on a moving average of volume over the past N minutes. Standard NTB vol bars use a fixed input number of shares to define a bar - which doesn't work well over a very long time period since there can be major overall characteristic increases or reductions in volume as a symbol matures.
A dvpb's "bucket volume" would be dynamic: sum up the tick-volume (or could be a lower timeframe n-min bar) over some fairly long time period (related to a multiple of the fractal you've discussed), then divide by that multiple to get an average. Once the HRE has accumulated a complete new internet multiple of volume, call that the end of the bar and move on to the determination of the next bar, with an updated baseline average volume.
That is, do a standard NTB Vol Bar build process, but with a dynamically adjustable volume-bar "bucket" size.
Then, once those bars are defined, store the OHLC prices for that dvpb, and use standard ATR (or other indicator) formulae to calc this.
Fair warning: this is a complex programming process, since it cannot make use of N native "static" NTB bars. I've coded a full suite of NTB bars before, complete with plotting - it's definitely not a small task.
But you asked for out of the box thinking. There ya go.
[Edited by Jim Dean on 7/5/2017 7:21 AM]