Hello, Anant, thanks for reply!!
Actually, this is not exactly what i was looking for, but nevertheless what you described is extremely interesting. These bands are similar to bollinger bands (especially in case of linear regression) - if i understand you correctly, we just use Linear Regression Indicator instead of moving average - so that it should look like in the attachment.
In metastock language it will look like:
LinearReg(C,13)+Stdev(C,13);
LinearReg(C,13)+2*Stdev(C,13);
LinearReg(C,13)+3*Stdev(C,13);
LinearReg(C,13)-Stdev(C,13);
LinearReg(C,13)-2*Stdev(C,13);
LinearReg(C,13)-3*Stdev(C,13);
Am I right?
By the way, about the probabilities: Price stays within middle bands in 95% of cases but it slides along the band quite often. So the probability of price bounce (going down of up from the band) is much less then 95%. Otherwise it would have been too easy))
But definitely it looks much better then bollinger bands. this is because linear regression is much more adoptive to rapid price change then MA.
I think it's valuable thing to use it to define OB/OS levels or estimate potential movement size to define TakeProfit levels or manage risks.
Last time i was thinking of applying polynomial linear regression (eg for some stock its parameters could be indices, related commodities etc.) but i's so lazy). did you try something about it?
Thanks for good idea, i also have good idea for you!))