Forums - too dumb

Forums (http://www.elitetrader.com/vb/index.php)
- Strategy (System) Design (http://www.elitetrader.com/vb/forumdisplay.php?forumid=34)
-- too dumb (http://www.elitetrader.com/vb/showthread.php?threadid=15896)

Posted by man on 04-03-03 04:35 PM:

too dumb

to figure that out. I build three time series for daily data on the spx. I calculate ln(Ht0/Ot0), ln(Lt0/Ot0) and finally ln(Cto/Ot0). Thus I have three percentages each day for the High, Low and Close, all three referenced to the open.

Now when I build correlations among them I find high correlations between high/open:close/open as well as for
low/open:close/open. They are between 0.7 and 0.8 and I do not understand why they exist. Why is the relation between High and Open highly correlated to the relation between close and open, plus the same for Low to open???

It must be wrong and stupid and I do not see why? Anyone an idea, since I really seem to have a blackout on this.

peace

__________________
do it now.

Posted by theTaoTrader on 04-03-03 05:07 PM:

The problem lies the use of logarithm (I assume your "Ln(Ht0/Ot0)" means you take log of the ratio of Ht0/Ot0). Since Ln(Ht0/Ot0)=Ln(Ht0) - Ln(Ot0), the reference point of opening price has been canceled out. So these correlations are saying if today's high (or log of high, to be more precise) is higher than yesterday's high then 70% to 80% the time today's low is also higher than yesterdays' low, which makes sense. Try re-run the correlations using Ht0/Ot0 and Lt0/Ot0 without log. Hope this helps.

__________________
The most important questions of life are, for the most part, really only problems of probability. - Pierre Simon de Laplace

Posted by BKuerbs on 04-03-03 06:26 PM:

Have a look at the range: That is the difference between High and Low, R = High - Low.

When the range is large, chances are LN(H/O) is large ditto for the other relations and when the range is low all these relations have a small value. That's all.

I have not looked at these relations too closely, the only thing I looked at was the relation H/C during up and down trends.

Regards

Bernd Kuerbs

Posted by man on 04-04-03 07:20 AM:

Tao
I will try as you say (I am currently just not at the computer I use to do this). But hence that your final explanation can hardly be valid, since I do not have any reference to yesterday at all. I took the log, since I am used to and with this small numbers it usually does not make any difference.
It is still not completely clear to me, but thanks for posting. I come back once I know.

peace

__________________
do it now.

Posted by man on 04-04-03 07:22 AM:

Bernd
I am currently making quite different things in that field. range depending on trend, daily intradaychange, overnight change - it is rather interesting stuff.
I am not sure with your explanation - the correlations are much too high. I still think I did some really stupid mistake that I am too blind to see.

peace

__________________
do it now.

Posted by man on 04-04-03 10:14 AM:

Tao
using percentage instead of log doesn't change the picture.

peace

__________________
do it now.

Posted by BKuerbs on 04-04-03 10:29 AM:

Quote from man:

..I am not sure with your explanation - the correlations are much too high. I still think I did some really stupid mistake that I am too blind to see.

peace

Introduce the range as grouping variable for your H/O etc. numbers and do some ANOVA or better the non-parametric equivalent. I did not do it, but I'm confident it will be significant.

Regards

Bernd Kuerbs

Posted by man on 04-04-03 01:04 PM:

bernd
my current problem is too much of significance - not too little. thnx anyways.

peace

__________________
do it now.

All times are GMT. The time now is 06:45 PM.