Rainier: Some thoughts on volatility-based indicators in general:
In a sense, the trend-delineated stuff I was working on might be of use here. I really do believe that noting changes in volatility of daily range, volatility of open to close, and volatility of yesterday's close to todays open (which gives complete coverage of the price activity day to day), and possibly adding volatility of close to close (which overlaps the above data), is very telling of what's to come price wise.
When these figures are confined to trend legs (i.e. between pivot points), and compared to the figures in other same-direction trend legs, an indication of "how normally" the current move is progressing can be had. The last run at a top typically shows an increase in volatility, and this needs to be seggregated by trend leg, rather than an arbitrary look-back period, to show any significance. It's not accurate (at least, I don't think it is) to include volatility info from prior DOWN TRENDLEGS in this calculation because a different sentiment is at work, and the hands in control are different (and therefore react differently).
A decline on increased volatility after the prior conditions are met seems to be a good way to confirm an imminent decline. I do believe there is value in the location of the close with respect to the max hi/lo range over some number of days (which is what stoc's are BASED on), but I don't believe multiple layers of smoothing and averaging is the way to exploit it (as is done with the canned stochastic indicator).
In other words, rather than smoothing OUT the "noise," we WANT the noise, and changes therein, to point us in the right direction (based on how the current noise correlates to noise from the past under similar conditions).
Also, I'm beginning to be biased toward volatility as measured by the range of actual prices (either as a percent or an absolute dollar value), as opposed to volatility as measured by standard deviation-based indicators. In other words, I'm leaning toward volatility based on actual trading points, rather than volatility based on the assumption of a known statistical distribution that presumably follows this distribution over all time periods, whether trending up, down, or sideways; or an arbitrarily-chosen multiplier of the variance in the data (e.g 2 standard deviations....why not 2.3, or 1.7, or ???).
I'm open to a convincing argument in favor of standard deviation as opposed to price range as a measure of volatility.
dh
P.S. I certainly don't have this "all figured out," so I'm open to any and all comments or rebuttals of my assumptions, or just general comments of what has worked in the past for anyone else out there. |
