|
PureBytes Links
Trading Reference Links
|
How about brushing your geometry books?
Once you have a collection of points (the closes on
the chart), you can find a line (straight line) where
the addition of each one of the distances from each
close to the corresponding point in the line is
smallest in the last X bars.
Then you can find the slope of the line. And your
decision rule would simply be how much you expect the
market to rise (or fall) in average for the market to
be considered to be trending in the last X bars -
geometry code name: slope. For example, if the market
goes up more than 2 points every bar for the last 20
bars then its trending.
What you have to define is:
X - The number of bars
Slope - How much does the market increase (or decrese)
each day to consider the market trending
This (loosly) is the definition of the least squares
method of calculating a linear regression.
In EasyLanguage all you need to use is
value1 = LinearRegSlope(Close, X);
Then you look at value1 and if it is greater than your
magic number then the market is trending. And finding
the right slope and X is what backtestin and
optimization are there to do...
Its similar to C > C[X] (and as simple) only it takes
out the posibilities of a one bar spike causing the
whole thing to give a false signal.
H
--- TradeWynne@xxxxxxx wrote:
> Hans,
>
> My purpose was not to refute your MA definition of
> trend (your post did jog
> my memory though). Indeed, I use MA's in some
> systems to define trend. I
> wanted to put out an extremely simple definition of
> trend based simply on
> price. But, BTW I have used the C > C[x] definition
> of trend in certain other
> systems and I can tell you it can be a good tool as
> well.
>
> Bill Wynne
> http://smarttrades.com/
>
>
> In a message dated 2/23/00 9:15:14 AM Pacific
> Standard Time,
> he96@xxxxxxxxxxxxxx writes:
>
> >
> > > C > C[x] then {trend is up};
> > > C < C[x] then {trend is down};
> > >
> > > Not so vague once you decide what "x" should
> be.
> >
> > this is totally bound by 2 single days.
> >
> > a single drop X days ago will tell you TODAY is
> UPTREND - even if the
> whole
> > period is flat at best.
> >
> > same for a single spike TODAY against X days ago.
> A "trend" is certainly
> the
> >
> > behaviour of 2 single days out of a group.
> >
> > A simple MA or a regression line should give
> better results :-)
> >
> > rgds hans
> >
>
>
|