Hello Sheridan,

There's some code here:

http://www.packtpub.com/sites/default/files/5306OS-Chapter-3-Get-into-Terms-with-Commonly-Used-Functions.pdf

Also, I see that Bollinger Bands are simply:

• an N-period moving average (MA)
  
• an upper band at K times an N-period standard deviation above the moving average (MA + Kσ)
  
• a lower band at K times an N-period standard deviation below the moving average (MA − Kσ)

Ref.: http://en.wikipedia.org/wiki/Bollinger_Bands.

I don't have time now to tinker around with the code, but it should be straightforward. Note that there are some statistical tools in scipy. For example:

 prices_z = stats.zscore(prices, axis=0, ddof=1)  

Grant

Hi Sheridan,

I did some quick cleanup on the original bollinger band post and also added some record calls to plot the price and the bands. You can see the results in the attached backtest. You should be able to clone this algo and run it without any problems.

thanks,
fawce

Hello Sheridan & Fawce,

Here's some z-score code...seems like trading on z-score could be formulated to be equivalent to trading using Bollinger Bands, right? And if multiple securities are all normalized with a z-score, they can be visualized together on the same plot (see below).

Grant

Thanks for the help! I cloned the BB algo because I am more familiar with them than z-scores. I tried modifying the BB algo to accept multiple sids by modifying it slightly. I'd paste my Backtest but can't get it to compile so I figured I'd reach out for help. I replaced

context.sid = sid(4283)  

with
context.sids = [ sid(4922), sid(2), sid(679), sid(6653), sid(700), sid(698), sid(1267), sid(23112), sid(1900), sid(2119), sid(8347), sid(3149), sid(3735), sid(3496), sid(3951), sid(3766), sid(4151), sid(23146), sid(4707), sid(5029), sid(5061), sid(5923), sid(5938), sid(4283), sid(7560), sid(7883), sid(7792), sid(39702), sid(8229), sid(2190) ]

That obviously broke the next line:
context.add_daily = TemporalGrouper(context.sid)

so I tried commenting that out which subsequently broke
context.add_daily(data) in my handle_data block. Any ideas on how to get this to compile with multiple SIDs? It doesn't seem very "Quantopian" to just monitor 1 security... I could do that by hand!

Hello Sheridan,

A z-score is given by (see http://en.wikipedia.org/wiki/Standard_score):

z = (x-mu)/sigma

or re-writing,

x = mu + z*sigma

According to http://en.wikipedia.org/wiki/Bollinger_Bands, we have the upper and lower Bollinger Band limits, respectively:

UB = mu + k*sigma
LB = mu - k*sigma

So, UB is the z = k limit and LB is the z = - k limit.

The %b is:

%b = (x − LB) / (UB − LB) = (mu + z*sigma - mu + k*sigma)/(mu + k*sigma - mu + k*sigma) = (z + k) / 2k

The bandwidth (BW) is:

BW = (UB - LB) / mu = 2k / mu

Regarding the code Fawce posted above...not sure what's going on with multiple securities. In any case, it should be straightforward to implement Bollinger Bands in Quantopian for multiple securities...I may tinker with it when I get the chance, since it appears to be effectively z-score based trading, which I'd explored previously on Quantopian.

By the way, it seems that for Bollinger Bands to work really well, one needs to identify securities with prices that are effectively stationary in time (see http://en.wikipedia.org/wiki/Stationary_process) and normally distributed (perhaps not a strict requirement). Any ideas how to do efficient screening over a large number of securities?

Grant

Grant,

Great insight! Let's fork your code to a new thread and I'll chip in to calculate the bands with your approach. Should be way faster.

Sheridan,
Could you post your question to the original thread? Maybe the original author will be interested in making it multi. If not, I can take a look tonight.

thanks,
fawce

Hello Fawce,

I posted a new thread here:

https://www.quantopian.com/posts/z-score-route-to-bollinger-bands

In a day or two, I may have some time to tinker around with it. Feel free to do whatever you want with the code--maybe I'll learn a few tricks from you.

Grant