Statistically 3 Standard Deviations would put prices within the band 99% of time but more than a few times prices just touch the band limits and keep of going in same direction which I call 'Sticking to the Bitch' movement. Bollinger recommended using 2 Standard Deviations for 95% probability but even 2.5 or 3 do not work..........when the price wants to stick it will no matter where the stupid bands are. Any advises on how to play the bollinger bands which boast of statistical chances in your favor? Cheers!

depends - are you using the touch of the upper band as a buy or a sell signal? You seem to be complaining about the fact that when prices touch the upper band they often keep going.... but isn't that a good long signal?

If you are looking for a trend (directional) change, I recently read that if one uses four standard deviations the probability of a reaction increases to 99%+....do not know if this works but perhaps you could backtest and let us all know if the author was correct.

I use the ADX with BB's. Low and non trending ADX readings should give you nice reversals off the band touches. High and trending ADX readings should lead to as you say "sticking to the bitch" moves.

I'm prolly not telling anyone anything new but its worth bearing in mind that BollingerBands are entirely based on historical events (like any indicator). So even though 95% of prices were within the bands its not to say that prices are conforming to the bands but rather the bands are conforming to the price series. Perhaps the real question is what is the percent likelyhood that the price at tomorrows close will be within the bands?

ADX and DMI are probably the two powerful indicators, one should play them along with their favorite indicator and immediately see 10% jump in the winner trades. No wonder Linda Raschke terms her holy grail system after ADX. While not a mathematician (even forgotten what little was learned) I am hopeful that some system can be developed where by it would become mathematically or geomatrically impossible for prices to keep going i.e. REVERSAL for sure scenario. Cheers! PS: Regarding some reply on using 4 STD DEV. for Bollinger, I think you will get only one trade a year and then no guarantee that sticking to bands will not take place.

IMHO, Bollinger Bands are worthless in predicting the future movement of stock prices, except perhaps if you are trying to play retracements. "Sticking to the band" gets it backwards. It would be more accurate to say that "the band sticks to the prices". Bollinger Bands have one useful predictive function IMO, which is to predict volatility. They can be very helpful in keeping your money from sitting still. Also, I suppose they might be useful as a screening device. But as a predictor of future prices? Pfft.

I think a system called "abberation" actually uses this type of stuff to identify strong trend instead of fading it. In terms of statistics, I don't think you can apply the numbers: 1) Standard BB uses sample size 20 for the moving average etc... not statistically significant enough. 2) Usually statistical techniques should be applied on the returns and not directly on price data itself. It is just plain weird to apply it on the price data. However I read somewhere there could be mathematical relationship (some ratio) between the statistics of the two populations (returns and price) given certain assumption on the populations (log-normal). So don't throw BB into the garbage just yet. 3) The assumption is that the population is "NORMAL" (bell curve) for the data in the selected timeframe. The academics found them to be fat-tailed distribution instead of normal distribution. Now you can throw it out or adjust your own expectation on how the price data should disperse. I think Kevin Haggerty has this volatility band technique for intraday trading using implied volatility of the options. I have not used it. He also showed you how to calculate it by hand and not paying for his expensive service. His technique probably makes more sense since the implied volatility is being voted on the returns of the index?? But he still makes the same erroneous population assumption (being normal) for the probability. I think both techniques can still be valuable (a tree branch is better than no ruler at all). I am not very good at math... Please correct me and give your thoughts. Thanks.