I've been playing around with bollinger bands, and came up with something that yields outstanding backtest results...really too good to be true. Do bollinger bands repaint or do something quirky so that it creates backtest trades that cannot happen in real time?
There are a lot of different things to consider when looking at "feel good back tests" . First , which program are you using or are you doing it manually by hand? If using a program, you need to make sure it's not "peeking into the future" that it is entering on the close of the bar and not the open. Next , did you factor in commissions and slippage? How long did you backtest it and in how many markets did it work? Have you used out of sample data with the same parameters? How many paremeters are there, if there are many , chances are it is curve fit. The above are just some , "checklist" things to look think about or look over.
been a long time, but back then we would run backtests, and if anything produced outsized profits or losses it got tossed. We looked for something that consistently broke even. Then you knew you had a good one. But yes, those Bollinger Bands are very quirky. I don't put much stock in any indicator, but I always liked those BB and still do to this day. If I was starting all over, I would remove clear fundamental moves, like when the fed speaks and major announcements and so forth. Usually that is about 1 to 3 hours of data. This is especially important if you are using BB.
Yes, they recalculate their values ( and therefore their lines ) on-the-fly as market trading ranges and price scale value areas change. If you print out a hard copy of the chart, and then look at the same study bands at a later period in time, you will see what I mean. I personally would prefer a more "static" study with more stationary results to backtest - like pivot points. Whether it's BBs or Channels or PPs you are essentially performing a variation on a trading range study. There are some technical studies that are a serious challenge to realistically backtest and achieve some sort of real world results in my opinion. Don't get me started about wave counts...
So you mean to say that a BB chart of this week when looked at the end of the next week will be different? Could you kindly back this up with some real chart example.
Yes if the trading range is substantial enough. You don't have to wait for weekly data - try it with a 5 minute chart with ES during an economic release or news and you will see what I mean exactly. Print one chart out for a snapshot, and then let the other one work on-the-fly. The forward BB curves will change substantially in some cases.
True but that is only for the current bar which is still in the play. For the bars that have hit their close price, the BB graph should not change. Agree?