|
PureBytes Links
Trading Reference Links
|
First MG, you definitely have my endorsement on your retirement, the
trend strength indicator is another issue.
I've never gotten much out of Active Trader Magazine, and while this
particular system is better than a lot of the junk they test, it's got
too many issues for me to feel comfortable with it.
It's very similar to Guppy's multiple moving average system. I like
Guppy's as a visual aid in identifying trends, and the use of mulitple
moving averages reduces whipsaws to some extent.
Both systems work better on futures than on stocks. The two big
problems with this system and other systems based on similar
principles are the huge drawdowns and the long periods of many loses
in a row. The average trader is never going to stick to a system with
those kinds of problems.
I like the quote in the magazine that says the system recovers from
it's 20% drawdowns pretty quickly. This is an example of how looking
at a graph is far different than looking at your wallet. The pretty
picture says "ah, my pretty litle red chart, you look so good."
Looking in your wallet says "I feel like I need to throw up, you got a
bag I could use."
Part of the issue can be avoided by staying out of the market when
it's in a downtrend. That can be seen by applying the same indicator
to the indexes and then adding in an advancing/declining issues
indicator. The two of them together pretty much define the trend.
However, even if a trader does that, they would be out of the market
for long periods of time.
The counter to this is "Well, you should simply short when the market
is going down." Okay, short away.
The code is fun to play with but it's a bit of overkill. Many
indicators, especially smoothed indicators like the Inverse Fisher
Transform or the Laguerre Filter perform just as well or better. The
drawdowns are smaller and the periods out of the market are shorter.
A lot of the really technical indicators that are smoothed can be
simulated to within a small fraction of the more complex. For example,
this simple Laguerre Transform
g:=Input("Alpha",0.1,0.9,0.8);
L0:=((1-g)*MP()) + (g*PREV);
L1:=(-g*L0) + Ref(L0,-1) + (g*PREV);
L2:=(-g*L1) + Ref(L1,-1) + (g*PREV);
L3:=(-g*L2) + Ref(L2,-1) + (g*PREV);
(L0 + (2*L1) + (2*L2) + L3)/6
can be approximated very closely by
Mov(Mov(C,Period1,E),Period2,S)
To trade this you pick the two periods for a buy with the closing
price of the stock crossing over the function and for the exit you
pick another set of periods for the close to cross as a sell. Longer
periods for Period1 and shorter periods for Period2 will probably work
the best.
If you want to see a function like this really perform, rather than
trying to trade it on the entire market, use it on the ETFs only or
even better apply it to a list of prescreened stocks like the those
from Value Line, IBD, or the Stock Scouter.
You can optimize it against those lists if you can find a few older
lists to work with. You'll need to use a date filter so you can load
enough bars into the tester for the function to work properly and
still allow you to limit the date range of the entry signals.
The combination of those two things will outperform the results in
Active Trader by a wide margin, without the huge drawdowns.
I can endorse that method and it will make you rich. Well, it will
make some people rich! But please feel free to retire anyway.
--- In equismetastock@xxxxxxxxxxxxxxx, mgf_za_1999 <no_reply@xxxx> wrote:
> Wow, I was catching up a bit during the weekend, stumbled on wabbit's
> initial modifications, and, after having thought you could speed up
> the original, posted my own variation. Then I noted the replies to it
> and now I feel as if I've walked straight into a trap! Anyhow, see my
> post as an illustration of how one could extend the system and modify
> the code when taking it outside of MSFL to speed it up in preparation
> for use with say an optimiser or Monte Carlo simulator.
>
> Anyhow, I am glad to hear that magazine found it works well. Note
> that there are many combinations, such as if the 10 day MA is above
> the 20 day MA, the 30 day MA and so on. Then if the 20 day MA is
> above the 30 day MA, the 40 day MA and so on. What we do with one
> neural net, is to calculate something like the following inputs
>
> Input 1 = C / 10 day MA - 1
> Input 2 = C / 20 day MA - 1
> Input 3 = 10 day / 20 day MA - 1
>
> etc.
>
> In the end, the close is compared to all moving averages and all
> moving averages are compared to one another. We feed this to the net
> that then tries to extract any historical patterns and voila! you have
> a trading model! I have no idea how this would perform on its own, as
> it forms part of a larger model, but it is encouraging if this type of
> indicator was found useful by others.
>
> Regards
> MG Ferreira
> TsaTsa EOD Programmer and trading model builder
> http://www.ferra4models.com
> http://fun.ferra4models.com
>
> --- In equismetastock@xxxxxxxxxxxxxxx, mgf_za_1999 <no_reply@xxxx>
wrote:
> > Now, if you want to build a model using this indicator, you don't want
> > to calculate all those moving averages, as it will make any
> > optimisation process very slow. MSFL is slow by default, so you want
> > to speed up things anyhow, even if you are not optimising. So you
> > calculate just one indicator, the cumulative sum of the closes, and
> > use only that.
> >
> > You also want to test how many of these steps you should perform.
> > Should you use 5, 10, 15 or whatever. Also, you rebase the indicator
> > to somewhere between -1 and +1, so that it remains the same for all
> > combinations. You also rewrite it a bit so that divisions, which take
> > longer to do, become multiplications, and the new 'thing' becomes
> >
> > ----8<------------------
> >
> > {Trend Strength Indicator}
> > {for metastock, coded by P Umrysh}
> > {from the August 2005 issue of Active Trader Magazine}
> > {modified by wabbit 08Jul05}
> > {modified by MG Ferreira 09 Jul 05}
> >
> > PRD := INPUT("Enter periods",1,100,10);
> > STEP := INPUT("Enter step size",1,50,10);
> > NSTEP := INPUT("Enter number of steps",1,20,10);
> >
> > XX := Cum(C);
> >
> > SS :=
> > (C*PRD > (XX-Ref(XX,-PRD ))) +
> > (C*(PRD+ STEP) > (XX-Ref(XX,-PRD- STEP))) * (NSTEP> 1) +
> > (C*(PRD+ 2*STEP) > (XX-Ref(XX,-PRD- 2*STEP))) * (NSTEP> 2) +
> > (C*(PRD+ 3*STEP) > (XX-Ref(XX,-PRD- 3*STEP))) * (NSTEP> 3) +
> > (C*(PRD+ 4*STEP) > (XX-Ref(XX,-PRD- 4*STEP))) * (NSTEP> 4) +
> > (C*(PRD+ 5*STEP) > (XX-Ref(XX,-PRD- 5*STEP))) * (NSTEP> 5) +
> > (C*(PRD+ 6*STEP) > (XX-Ref(XX,-PRD- 6*STEP))) * (NSTEP> 6) +
> > (C*(PRD+ 7*STEP) > (XX-Ref(XX,-PRD- 7*STEP))) * (NSTEP> 7) +
> > (C*(PRD+ 8*STEP) > (XX-Ref(XX,-PRD- 8*STEP))) * (NSTEP> 8) +
> > (C*(PRD+ 9*STEP) > (XX-Ref(XX,-PRD- 9*STEP))) * (NSTEP> 9) +
> > (C*(PRD+10*STEP) > (XX-Ref(XX,-PRD-10*STEP))) * (NSTEP>10) +
> > (C*(PRD+11*STEP) > (XX-Ref(XX,-PRD-11*STEP))) * (NSTEP>11) +
> > (C*(PRD+12*STEP) > (XX-Ref(XX,-PRD-12*STEP))) * (NSTEP>12) +
> > (C*(PRD+13*STEP) > (XX-Ref(XX,-PRD-13*STEP))) * (NSTEP>13) +
> > (C*(PRD+14*STEP) > (XX-Ref(XX,-PRD-14*STEP))) * (NSTEP>14) +
> > (C*(PRD+15*STEP) > (XX-Ref(XX,-PRD-15*STEP))) * (NSTEP>15) +
> > (C*(PRD+16*STEP) > (XX-Ref(XX,-PRD-16*STEP))) * (NSTEP>16) +
> > (C*(PRD+17*STEP) > (XX-Ref(XX,-PRD-17*STEP))) * (NSTEP>17) +
> > (C*(PRD+18*STEP) > (XX-Ref(XX,-PRD-18*STEP))) * (NSTEP>18) +
> > (C*(PRD+19*STEP) > (XX-Ref(XX,-PRD-19*STEP))) * (NSTEP>19);
> >
> > ( 2*SS - NSTEP ) / NSTEP
> >
> > ----8<------------------
> >
> >
> > Now, this looks bad but is easily done using any programming language
> > where loops are allowed. All those lines that form part of the sum SS
> > are done inside this loop as a fairly simple addition. The loop ends
> > at the right spot, so that the multipliation with NSTEP > xx falls
away.
> >
> > I also am not sure about MSFL, but this function is supposed to be
> > very fast. In a normal programming language, at each step, you just
> > update xx as XX = XX + C. XX starts at 0. MSFL does not allow this,
> > so the one difference between this function and the original is that
> > it starts one observation later than the original. Again, in a real
> > programming language, this would not be the case.
> >
> > Now, if only super can endorse the trading qualities of this new
> > indicator I can sell it to the world and retire early.....
> >
> > Regards
> > MG Ferreira
> > TsaTsa EOD Programmer and trading model builder
> > http://www.ferra4models.com
> > http://fun.ferra4models.com
> >
> >
> >
> >
> > --- In equismetastock@xxxxxxxxxxxxxxx, "bellamy_29m"
> > <bellamy_29m@xxxx> wrote:
> > > Similar results (0-10) instead of -100 to 100 with simpler code:
> > >
> > > {Trend Strength Indicator}
> > > {for metastock, coded by P Umrysh}
> > > {from the August 2005 issue of Active Trader Magazine}
> > > {modified by wabbit 08Jul05}
> > >
> > > PRD:= Input("ENTER PERIODS FOR SMA",1,100,10);
> > > STEP:= Input("ENTER SMA STEPS",1,50,10);
> > >
> > > (C>Mov(C,PRD+(STEP*0),S))+
> > > (C>Mov(C,PRD+(STEP*1),S))+
> > > (C>Mov(C,PRD+(STEP*2),S))+
> > > (C>Mov(C,PRD+(STEP*3),S))+
> > > (C>Mov(C,PRD+(STEP*4),S))+
> > > (C>Mov(C,PRD+(STEP*5),S))+
> > > (C>Mov(C,PRD+(STEP*6),S))+
> > > (C>Mov(C,PRD+(STEP*7),S))+
> > > (C>Mov(C,PRD+(STEP*8),S))+
> > > (C>Mov(C,PRD+(STEP*9),S))
> > >
> > > Hpe this helps.
> > >
> > > wabbit :D
> > >
> > > <snip>
------------------------ Yahoo! Groups Sponsor --------------------~-->
Try Online Currency Trading with GFT. Free 50K Demo. Trade
24 Hours. Commission-Free.
http://us.click.yahoo.com/DldnlA/9M2KAA/U1CZAA/BefplB/TM
--------------------------------------------------------------------~->
Yahoo! Groups Links
<*> To visit your group on the web, go to:
http://groups.yahoo.com/group/equismetastock/
<*> To unsubscribe from this group, send an email to:
equismetastock-unsubscribe@xxxxxxxxxxxxxxx
<*> Your use of Yahoo! Groups is subject to:
http://docs.yahoo.com/info/terms/
|