My apologies if this has been answered elsewhere but I can’t seem to find the answer.
I use Bollinger Bands in my live trading and am always looking to improve.
I trade the bounce from the bottom to the top of the bands and I have an edge doing this.
One concept I struggle to really comprehend is from rule14. from John Bollinger which states:
“Make no statistical assumptions based on the use of the standard deviation calculation in the construction of the bands. The distribution of security prices is non-normal and the typical sample size in most deployments of Bollinger Bands is too small for statistical significance. (In practice we typically find 90%, not 95%, of the data inside Bollinger Bands with the default parameters)“
Why does John say that the data the Bollinger Bands use is “too small for statistical significance” when he himself uses them, advocates for their efficacy in trading, and discusses their uses in trading various patterns?
Surely if the data the bands were drawing from were not large enough to have statistical significance, then the theory behind the Bollinger Bands would be useless and the bands themselves wouldn’t be of use at all.
If someone who is more knowledgeable than me explain what is meant by rule 14, and then how this relates to using the bands in trading, that would be hugely helpful.
Volatility and price tend to follow it, but never perfectly, and often skewed.
For example, selling the BollingerBand(20) break of the top band to buy back at the bottom of the band… statistics say that should work out 50% of the time with a 1R target… but it can fail 3 times in a row. What we didn’t see was in the past, it worked 3 times in a row and this is the equalizer. Inversely, it can lose 3 times in a row because it’s going to work 3 times in a row in the future.
I’ve been able to run millions of back tests across all major pairs, all BB(20-30) at 2.0-3.0SD, all timeframes. There is no statistical edge using any R-multiple on any of them. All follow Normal Distribution with times of Non-Normal distribution laced throughout them. Unpredictable as to when the distribution is Normal/Non-Normal until afterwards.
The best outcomes were buying at relative ‘lows’ and selling at relative ‘highs’
That is a great explanation. Thankyou. I can’t go into details since I read this a long time ago, but I read The (Mis)behaviour of markets by Benoit Mandelbrot (fractals), a summary of which is presented at this link. Look at fat tails and long periods of dullness with short periods of highly volatile activity, to explain why the bell curve is a poor fit for real markets. I would not lose sleep over this. If you have an edge using BBs, carry on doing so.
Just enter at close candle, such as pinbar or dojo. Lots is intuitive such as the subtle differences in the bands reacting to price action and judging the correct market structure to trade. Only enter when entry signal cuts through the 2.5SD band and take profit at 1.5Sd opposing band. I don’t add or take anything away but I do move my stop up slightly when certain milestones in the trade have been reached. Thats on fx pairs. Hope that helps
Entry at close of candle or entry at candle break of BB. Both yielded similar results. Entry at close had little less drawdown on average, similar Monte Carlo chart returns.
Exit based on R-multiples, exit based on central line cross, exit based on opposing trading signal. None changed the outcome either. Stop Loss 40-200 pips, TP 30-100 pips. Best (low DD, steady outcomes) seemed to be 40 pip TP, 50 pip SL… on Eur/Usd.
So, you made millions back tests with enter at market order with different constant SL and TP, but you forgot something. You can use pending orders, stop and limit on different levels, with SL/ TP based on ATR, percent price, S/R area, indicators levels and exit types like trailing stop, trailing stop activation, time exit and so on. Your tests are really tiny part from whole tests which you can do, so don’t make that kind of statement that there is no statistical edge just because you didn’t find it.
