Zeev's Turnips Patch-No Politics (ZEEV)

[kevinb]

Zeev

(Or anybody else). I have a simple question relating to statistics which I hope I can get your help with.

I am studying a simple trading system based on a daily pivot. In an attempt to improve entries I am looking at several aspects to see if I can get a basis for a daily bias.

Simple things I can look at are whether yesterday was a long or a short day, and/or whether yesterday’s close was less than the pivot figure for today. In the back test sample I have 509 short days out of 1030 - Short days being those where the close was less than the pivot figure. On 300 of those days the previous day was also a short day. That says to me that if yesterday was short there is a 60% likelihood that today will also be short. Is that a correct interpretation?

The figures for where the previous close was less than the pivot figure were 525 (out of 1030 days). Of those 332 were short days. First question do I divide this 332 by 525 (# of pivot > prev close days) or 509 (# of short days in backtest) to arrive at a probability for predicting whether a day with a previous close less than the pivot will be a short day.

This gets even more complicated when I try to combine these stats. If I take the set of all days where yesterday was short, and the close was less than today’s pivot figure. I have a total of 375 days.

If I take the number of those days that were short days in the backtest, I get a total of 241 days.

Now my question is if I want to draw probability statistics from this data do I divide this 241 days by:

a) The total number of short days that were previously short (300 )
b) The total number of short days where the pivot was greater than the previous close. (332)
c) Or the total number of days where the previous day was a short day and the close was less than the pivot (375)

The last would seem the most logical but I can make a case for the others as well. Is there a rule to follow here that defines which one to use?

Also I notice that almost all the statistical relationships I have found so far are clustered around the 6n% area. This seems to be more than accidental but I don’t know what significance to attach to it. Does it suggest anything to you.

Thanks for all your help




Kevin