I'd forgotten about this. As I recall, despite being logically more sound, the standard deviation method applied to %ATR did not work very well (as measured by backtest results). You ended up having to go to something like the 99.999999% (aka 10 std devs or whatever) level to approximate the more common simple ATR and discretionary multiplier (ex. 4x). I always wondered if the fact that the % method didn't work well (logically, it should) provided an indication that ATR methods of setting stops were potentially inherently less stable than they appear from backtests. No idea how to validate that assumption and maybe it doesn't really matter since some form of stop (even if not optimal) beats no stop. Regardless, it would be interesting to see academic types with more statistical firepower run some monte carlo type analysis on the concept (assuming the results remained in layman's terms vs becoming a Greek formula that was difficult to understand).
I'll follow up on this and on range based volatility measures in general, as there are more than ATR that can be used to take care of various peculiarities in the data.
Nice look at ATR. Bollingers do use standard deviation and indicators, like Donchian use price ranges to set levels, another way of looking at price volatility. Enjoy the ride!
I'd forgotten about this. As I recall, despite being logically more sound, the standard deviation method applied to %ATR did not work very well (as measured by backtest results). You ended up having to go to something like the 99.999999% (aka 10 std devs or whatever) level to approximate the more common simple ATR and discretionary multiplier (ex. 4x). I always wondered if the fact that the % method didn't work well (logically, it should) provided an indication that ATR methods of setting stops were potentially inherently less stable than they appear from backtests. No idea how to validate that assumption and maybe it doesn't really matter since some form of stop (even if not optimal) beats no stop. Regardless, it would be interesting to see academic types with more statistical firepower run some monte carlo type analysis on the concept (assuming the results remained in layman's terms vs becoming a Greek formula that was difficult to understand).
I'll follow up on this and on range based volatility measures in general, as there are more than ATR that can be used to take care of various peculiarities in the data.
Nice look at ATR. Bollingers do use standard deviation and indicators, like Donchian use price ranges to set levels, another way of looking at price volatility. Enjoy the ride!