I'd also be very interested in Grey1's approach to volatility measurement and use - I suspect it might have a significant bit of math associated with it, but maybe not.
In the interim...
I have been researching this subject myself as my own approach to factoring in 'current' volatility had been largely visual rather than strictly mathematical. I look at the 'twitch factor' for the past N periods. N being defined by similar levels of 'twitch'. Need to explain further I think.
Twitch is noise - to me. Other traders are very probably trading this 'noise', but on my 5min charts, to me, it's noise I want to ignore.
If you look back on any instrument on a 5min chart over a reasonable period of time, you'll see varying levels of noise/twitch/volatility. I'm NOT talking about trends, advances, declines, consolidations i.e. the bread & butter moves - I'm talking about how much fuzz there is while they're doing whatever they're doing. What you will notice as you get to the lower levels of granularity is that at different times, there'll be different sizes of twitch. But here's the really interesting bit - these periods of similar twitch tend to cluster.
So for the last 4 hours of trading for instance, a stock will have advanced from $25 to $27.50 maybe with a pullback or two, doesn't matter. But look at the tick data for that same period. You'll notice there are periods within that 4hour period where the price twitches around maybe +/-5 cents. Another period where it's barely moving more than +/-2cents. Then maybe +/-10cents. The things is, these periods of specific size do tend to cluster. The +/-5cents period may be for an hour, the +/-10cents period for 90 minutes. The +/-2cents may only be for 30minutes. And this is why I said above N periods. I let the stock action determine the N periods I'm going to look at.
Anyway, that is the way I currently calculate Volatility. Fairly dynamically and not mathematically precise in any way. I'll take whatever the twitch factor currently is (as defined by my ruggedly unscientific measurements) and factor that into my risk calculations. These go into the pot along with Support & Resistance levels and all those other good things.
However, getting back to the point of this post (anyone still listening?
) is that I was impressed with a fairly rigidly mathematical approach to this by Ehler (Rocket Science for Traders). While I think there are limitations in attempting to do a complete X-over from the field of Digital Signal Processing to Trading
in general - I do think there may be some merit in considering treating volatility in stocks as noise in the signal (fluctuations in the price action from a trader's perspective).
I've been looking at the ratio of the standard deviation of closes (of the 5min bars) of the last N bars (where N is from my flaky definition above) to the standard deviation of closes over the last M bars - where M is N*3.33 and using that as my twitch factor.
I haven't traded this in real-time yet as I don't rush anything into the 'production line', but today might be a good day to give this some active research.
If I do get round to it, I'll post an update on my findings. If anyone's interested. Actually, probably even if they're not. Why else am I here?
Iraj...sorry...you were about to tell us about your volatility calcs...