Simple Testing of Bollinger Bands

[BlueHorseshoe 10]

Bollinger Bands are currently being discussed on another thread. This post is not a comment on that thread, but a quick look at "fading the bands" for anyone who might be interested. I am not suggesting that this or similar approaches are either viable or otherwise.

 

The attached chart shows the results of fading any close outside of the Bollinger Bands in the ES over the last 10 years, using a single contract. Entries are on the close, and exits are on the subsequent close after a position has been held for one day (no other exit type or stoploss was used). No slippage or commission has been deducted. These results were obtained by optimising the lookback length for the indicator. A graph is also attached, showing how these results would have varied for different lookback lengths.

 

BlueHorseshoe

Bollinger Band Test.pdf

Bollinger Band Equity.pdf

[steve46 12]

Thanks, thats very nice of you to show it in a graphical fashion...I hope you won't mind if I say that my own research shows similar results for INTRADAY data....the only disclaimer is that I have only done research using my own method for entry and exit..

 

Thanks again, makes my task a lot easier.

[steve46 12]

Gotta say one more thing about this....since I have been here (a few years) this is one of the only times that anyone has taken a moment to do some of their own research (I am sure the fact that you decided to do this on Bollinger Bands is just coincidence)...

 

Nevertheless, people should make note....this is (in my opinion) the single most productive thing that they can do to learn about a subject and (if we can take this one step further) find ways to use available indicators (technology in general) to advance their understanding of "how to trade successfully"...

 

If someone (anyone) were to take a moment to look at a daily chart, re-read Blue Horseshoe's comment and really think about it, they might see something similar to what I saw when I first started to test BB's. As it turns out, there are a number of ways to use Bollinger Bands, some which John Bollinger anticipated, some that he may not have thought of....

 

To declare (as Predictor did) that Bollinger's invention was "trivial" is in my view an example of stunning, enduring ignorance...People should take a moment to read about the guy...

 

Thanks again BH for your comment.

 

Steve

[BlueHorseshoe 10]

Gotta say one more thing about this....since I have been here (a few years) this is one of the only times that anyone has taken a moment to do some of their own research (I am sure the fact that you decided to do this on Bollinger Bands is just coincidence)...

 

Steve

 

Really? I assume everyone on here does something similar to this. I can't really imagine what it would be like to try and trade without undertaking research like this first, but maybe that's just me. If it helps to illustrate what you're trying to explain in your other thread then that's good - there's not point dismissing Bollinger Bands out of hand.

 

Of course, far more interesting than some curve-fitted return is the fact that no tested parameter produced a negative return. Rather than telling us something trivial about Bollinger Bands, this edges towards telling us something much more fundamental about the general behaviour of this market over the test period. Whether Bollinger Bands are the best tool for exploiting such behaviour is open to debate.

 

As for using this intraday . . . If you exit at the close of the next bar as this test did, then your maximum dollar profit would be limited to the movement within that single bar. If you take a five minute chart then the average range of a bar is somewhere around 5 ticks. This means that your profit per trade is limited to $62.50. You either have to use limit orders and risk not getting filled (happens far more often than most would imagine, and only affects would-be winning trades - losers are always filled), or use market orders and pay the spread. Assume you have a 100% win rate. If you have pay the spread then you can deduct $12.50 in, $12.50 out, a bit more for slippage, and then your commissions, exchange fees . . . Let's call it $30. Now adjust the win rate to something more realistic, and suddenly there's nothing left.

 

To make this work in lower timeframes a trader would either need a much more accurate entry and exit method than the simple one I tested, or superior execution capabilities.

 

BlueHorseshoe

[TheDude 10]

What interested me, was that I would hae thought a close outside the BB would have been an indication of continuation - does the reversion/correction just occur on the next day?

 

Perhaps if you were to trade in the direction of the first close when you get a day with a close in the same direction again (suggesting the correction is over), you could have the beginnings of a CTA businness ??

[BlueHorseshoe 10]

What interested me, was that I would hae thought a close outside the BB would have been an indication of continuation - does the reversion/correction just occur on the next day?

 

Perhaps if you were to trade in the direction of the first close when you get a day with a close in the same direction again (suggesting the correction is over), you could have the beginnings of a CTA businness ??

 

That's certainly the way that I view it.

 

I don't know what CTA types do nowadays, as they've supposedly become a little smarter (although most of those quants at the trend-following funds are probably set to work on risk management/money mananagement/diversification/order execution type problems), but thirty years ago a Bollinger Band breakout strategy would have been the height of sophistication. Nor would trend following CTAs have bothered waiting for continuation after a pull back into the bands as you suggest - they would just have bought on the close outside the bands, then weathered the adverse excursion. Breakout systems with Bollinger Bands are described in Curtis Faith's 'Way of the Turtle', and also the Emilio Tomasini book 'Trading Systems'.

 

An example of how the two approaches interact (but with Donchian Channel breakouts rather than Bollinger Bands) would be the 'Turtle Soup' strategy described in 'Street Smarts'. Most trend following entries result in small losses, so for someone taking the other side of the CTA's entry and only looking for a small profit, there is high probability trade.

 

I'll try and run a similar test of trading Bollinger Band breakouts if I get time.

 

BlueHorseshoe

[Predictor 11]

Just as an aside, Blue the type of testing you are doing now is empirical in nature. I use this type of testing extensively when developing my trading systems. This sort of testing is much different then relying on statistical laws/properties for known distributions. Most of all, the statistical properties of price series often change dramatically in the future rendering profitable historical rules unprofitable.

 

The point I was making earlier wasn't that standard deviation couldn't be used. Notice, I said as much that some traders might find them useful. My point was that because price series don't have a known distribution and don't have the properties that standard deviation assumes that they can't be used correctly/formally/etc. This concept applies to most statistical measures that I know of.

 

You may want to look up descriptive vs predictive statistics.

[steve46 12]

Like others here I am always amazed at how much you have to say about subjects (you say) you know so little about...

 

discontinous data sets offer challenges to be sure...your problem is that you can't find a solution to that obstacle by doing a couple of hours research on the Internet....

Edited August 26, 2012 by steve46

[steve46 12]

Really? I assume everyone on here does something similar to this. I can't really imagine what it would be like to try and trade without undertaking research like this first, but maybe that's just me. If it helps to illustrate what you're trying to explain in your other thread then that's good - there's not point dismissing Bollinger Bands out of hand.

 

Of course, far more interesting than some curve-fitted return is the fact that no tested parameter produced a negative return. Rather than telling us something trivial about Bollinger Bands, this edges towards telling us something much more fundamental about the general behaviour of this market over the test period. Whether Bollinger Bands are the best tool for exploiting such behaviour is open to debate.

 

As for using this intraday . . . If you exit at the close of the next bar as this test did, then your maximum dollar profit would be limited to the movement within that single bar. If you take a five minute chart then the average range of a bar is somewhere around 5 ticks. This means that your profit per trade is limited to $62.50. You either have to use limit orders and risk not getting filled (happens far more often than most would imagine, and only affects would-be winning trades - losers are always filled), or use market orders and pay the spread. Assume you have a 100% win rate. If you have pay the spread then you can deduct $12.50 in, $12.50 out, a bit more for slippage, and then your commissions, exchange fees . . . Let's call it $30. Now adjust the win rate to something more realistic, and suddenly there's nothing left.

 

To make this work in lower timeframes a trader would either need a much more accurate entry and exit method than the simple one I tested, or superior execution capabilities.

 

BlueHorseshoe

 

 

Well it took a long while to get there, but you have arrived at a conclusion that is clear...you can't use BB's as the centerpiece of a trading system....and thats why in my system they are called "training wheels"....

[BlueHorseshoe 10]

Well it took a long while to get there, but you have arrived at a conclusion that is clear...you can't use BB's as the centerpiece of a trading system....and thats why in my system they are called "training wheels"....

 

Hi Steve,

 

I didn't mean to give the impression that I had arrived at any clear conclusion.

 

Nor did I mean to imply that Bollinger Bands were the sole component of your trading system - that's why I brought this discussion over to another thread so as not to confuse whatever concept you're trying to develop on the other thread. I literally intended this thread to be a footnote pointing out that Bollinger Bands can be a useful aid . . .

 

Bollinger Bands could, possibly, be used as the centrepiece for a trading system in a higher timeframe (whether "swing trading" rejections from the bands, or looking for breakouts entries as a trend-follower might). In much lower timeframes though, I think that one would need other tricks up ones sleeve which, given that you never implied that the Bollinger Bands were anything other than visual "training wheels" on your charts, I'm assuming that you have.

 

Cheers,

 

BlueHorseshoe

[BlueHorseshoe 10]

discontinous data sets offer challenges to be sure...

 

If you can find a function to accurately describe behaviour around the limit, you can wave the magic wand of the statistician and treat the discontinuity as a jump process within a gaussian distibution . . .

 

Should you choose to do this, then you will need to be both very good at maths, and pretty damn stupid!

 

BlueHorseshoe

Edited August 26, 2012 by BlueHorseshoe

[BlueHorseshoe 10]

The point I was making earlier wasn't that standard deviation couldn't be used. Notice, I said as much that some traders might find them useful. My point was that because price series don't have a known distribution and don't have the properties that standard deviation assumes that they can't be used correctly/formally/etc. This concept applies to most statistical measures that I know of.

 

I think that you're right - most statistical measures can only be applied to the market in a very general way, and we shouldn't be at all suprised when they fail to perform as expected unless we have very conservative expectations.

 

I also probably overstated my case somewhat. If I'd run the same test on something like Orange Juice futures, or even the Euro/Usd currency pair, I wouldn't have got the same results. The ES has shown a pronounced tendency for mean-reversion that is far less reliable in other instruments. My only apology for this is that many traders on TL primarily trade the ES.

 

I also think there's an important difference between each of the following three statements:

 

a) Markets have a known distribution

b) Markets don't have a known distribution

c) Markets behave as though they have a known distribution most of the time.

 

The third would be my choice of description, and I think it is one that can be useful.

 

BlueHorseshoe

[steve46 12]

I think that you're right - most statistical measures can only be applied to the market in a very general way, and we shouldn't be at all suprised when they fail to perform as expected unless we have very conservative expectations.

 

I also probably overstated my case somewhat. If I'd run the same test on something like Orange Juice futures, or even the Euro/Usd currency pair, I wouldn't have got the same results. The ES has shown a pronounced tendency for mean-reversion that is far less reliable in other instruments. My only apology for this is that many traders on TL primarily trade the ES.

 

I also think there's an important difference between each of the following three statements:

 

a) Markets have a known distribution

b) Markets don't have a known distribution

c) Markets behave as though they have a known distribution most of the time.

 

The third would be my choice of description, and I think it is one that can be useful.

 

BlueHorseshoe

 

Remarkable..so here is a comment that I feel obligated to make...nothing personal about it. Nothing controversial about it as follows

 

If by "statistical measures" you mean the basic tools that skilled operators can use (both parametric and non) then it is no different than asking a skilled carpenter to build something for you....In either case a skilled, experienced person will know what tools to choose, and how to use them to reach the desired result...

Edited August 27, 2012 by steve46

[BlueHorseshoe 10]

Remarkable..

 

Sorry, Steve, but I'm not going to "take sides" in yours and Predictor's ongoing quarrels. As such, I am likely to agree with anything that either of you says if I think there is some truth in it. Of course, if you start to become rude to me again, then I shall revert to treating you as I did before.

 

If by "statistical measures" you mean the basic tools that skilled operators can use (both parametric and non) then it is no different than asking a skilled carpenter to build something for you....In either case a skilled, experienced person will know what tools to choose, and how to use them to reach the desired result...

 

What you say is true - a skilled carpenter will select the correct tools for the job. A plane to smooth a surface, say. However, no matter how skilled he may be, and no matter how deftly he planes, he will never entirely eliminate the grain from the surface of the material he works. The wood will always have a grain, no matter how much he smooths it.

 

Price data will always contain series that cannot be 'smoothed away' with statistical procedures, even the 'correct ones', wielded by an adept.

 

As I pointed out to Predictor, I don't think this matters. Your Bollinger Bands, and whatever else you use, only needs to be accurate enough to make you money. The fact that they will never perfectly describe price distributions matters no more than 100th of a millimeter grain in the surface of the table I write at. It's all good enough.

 

That's enough with the extended metaphors!

 

BlueHorseshoe