[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Stop Placement Concept Question



PureBytes Links

Trading Reference Links

Conrad wrote:
>
>
>Cynthia Kase in "Trading with the odds" describes (partially) a
>statistically derived stop.  Her stop is placed at a distance to
>withstand "2-bar reversals".  She measures a mean value and the standard
>deviation and places the stop at (mean + n*sigma) away fro the price
>action.  However the exact formulas are not given.  I'm not sure if she
>is measuring historical reversals and averaging them or just saying a
>2-bar reversal is 2 times the current avg-true-range.  If the latter,
>you would just get the average and sigma of ATR and then you could
>calculate the stop.  (This was an interesting book, but i had the same
>"complaint" in various places: while the general idea was explained,
>actually formulas or procedures were rare; i guess to have that you have
>to buy her software.  Maybe that's only fair, but....)  In any event,
>she recommends using 3sigma for staying with a trend and 1 sigma if
>warning signs of the end of the trend exist.  
I have coded up versions of her stop code, the slowk permission and the 
oscillators if you're interested.  While probably not the exact formulas 
she uses (for example the stop doesn't use the fat tail on the high side 
she talks about), they're good enough for pratical purposes.  As you say, 
it required a fair amount of reading between the lines to figure some of 
these out.  email me direct if you're interested.

>I have to wonder if it's worth the trouble to do statistical analysis to
>the point of finding sigmas.  Sure, you are finding the distribution of
>the last 15 or 30 days or whatever but the next ten days may be a
>different "population".  Certainly basing stops on current volatility
>makes sense.  But is calculating the stops to be a distance of  avgrng +
>n*sigma away, any better than just using n*avgrng?
The best results seem to be using longer time frames (100 days+) so that 
you have a good working sample. A statiscition could probably come up 
with an "optimal" value for the number of days.

>							Conrad Bowrers
>

Jerry