|
PureBytes Links
Trading Reference Links
|
PPMig@xxxxxxx wrote:
>
> In an effort to expand my horizons I am interested in learning more about
> adding time as a component to my trading plan/system. There seems to be
> several varieties of *time* out there ranging from Fdates, astro, dynamic
> cycles etc.
I have spent the last three years researching and testing various
time/turn day forecasters. I have had mixed success. I have found
quite often that the 'turn day' must first be qualified in magnitude.
Are you looking for trend reversals or short term (1-3 td) reactions?
Will a two day rally high that resumes a down trend suffice? (I have
found those 'trend joiners' to be some of the most profitable.) For me
the turn day must be mathematically significant, predict the defined
turn more than random, on several degrees of turns before I want to
include it in my trading. To test a turn day I first define several
degrees of turn to be tested. ie A 10 td extreme followed by a 10%
change in price, or a 30 td price extreme followed by at least 30 days
in the opposite direction etc. Then I decide how accurate the turn day
should be. I use the turn day and the day after. Once these parameters
are established, I determine what % of random days would produce a turn
of the chosen degree by dividing the turns of this degree times two by
the number of trading days in the sample. I then see how the turn day
does over the same sample. If the turn day does 20% better across 3-4
different degrees, I will consider it a non-random turn indicator. This
is a difficult process, but I believe necessary before putting money any
type of turn indicator.
I do have some questions for those with a more advanced statistical
background. How much better should the turn day be than the random days
to make it significant? How many test days? Should I not include the
turn days in the random set? Should the sets be the turn days vs. non
turn days?
|