|
PureBytes Links
Trading Reference Links
|
System
design should follow some logic: evaluate your "signal strength" first.
Please
know I work in the Min or smaller timeframe my experience is limited and
very one-sided. I do things "my way" and try to ignore tradition. The only
things I have learned from "other systems" are coding techniques. You won't find
the perfect system or the HG anywhere. Sorry to burst your bubble :-) In trading
you have to do your own work or work in a group. Whichever way, to
attain above average results takes hard and innovative work. The following
is jmo.
Since you may end up testing hundreds or even
thousands (yes!) of ideas you need to be efficient. Getting drawn
into advanced system concepts, fancy stops, money
management and creating fancy charts before you have a validated the
basic signals is a waste of time.
By
fiddling with all the above you increase the chance of 1) curve fitting and
2) drawing profits from what are essentially Random signals. It is actually
possible to take random signals and add so many "shock-breakers" to it that it
becomes profitable (Tharp?). This, to me, is a totally "non-fun" way of
developing systems, all you are doing is increasing your statistical odds to
draw a little money out of the market.
The
Popularization and Commercial Exploitation of TA encourages this
approach. Looking at fancy and colorful chart gives us ( I've been
there! ) a feeling of accomplishing something important. But are we
really? Just go to your local bookstore and browse through
some books or browse the web for TA software and you'll see what i
mean. Tons of stuff out there that looks impressive but won't make you any
money.
Try to
focus on what is important. The backtests I perform to validate price
patterns are:
- Exit at the Entry bar's Close (one-bar trade) or at the Next Bar's
Close (two bar trade).
- If the system makes money I
start testing intrabar exits during those two exit
bars.
- Test 1-10 minute time
frames.
- If the system looks
profitable I add some code and run a Short
test.
- Run it over a watchlist, how many stocks are
profitable characterizes the system as selective or
common
- Any kind of fancy
stuff, like stops and MM "masks" the true signal performance, make it a very
last step.
I have developed some standard code modules and placed
them in my indicator space so that i can drag them onto my code under
test and to plot trade prices, signals and equity. With this
setup you can test hundreds of ideas a day.
Always look at the equity curve and try to get as many
trades as possible - thousands if possible. Numerical Stats
don't mean a lot since they can be distorted by a single big loss or profit. If
this happens and the equity curve is good otherwise you can study the
single event and add code to exploit, or protect against,
it.
Best regards,
herman
Herman
Your replies and experience are
appreciated.
So one could look at placing a time stop into the system,
rather than leaving this to discretion. Previously you said: "the significance
of a signal fades quickly" : interesting thought: in designing a system then I
would need to work out my preferred time frame, expected duration of the
effect of the signal and then factor in an appropriate time stop.
It is
valuable to see "why" and "how" others' trading systems are designed,
philosophically: to see what is capable of being done, and then seeing if
there are parts of others systems or ideas that converge with one's own ideas.
Better than starting the whole learning curve from scratch and so that is why
appreciate your input and that of others in this forum.
Long way to go
still.
Regards
ChrisB
Herman van den Bergen
<psytek@xxxxxxxx> wrote:
I
simply try to trade price patterns based on a handfull of minute
bars... I expect profitable conditions to exist early in the trade, not
near the end. If a trade doesn't behave as expected soon i feel like
gambling, or trying to get lucky, and I rather get out and wait
for another opportunity.
While this may sound easy it isn't and a lot of work goes into
peripheral code to trade patterns automatically.
best regards,
herman
Herman,
thanks
for the reply.
Presumably one would have a high win ratio (% of
winners) with these short term systems, and smaller pay off
ratios?
When exits are based on price action or price targets
alone, and they do not reach the profit target, would you use a time based
stop?
If not then one could assume we could use oscillators that
will continue to oscillate and return a sell signal, even if price does
not reach the target?
Or indicators that have a time based decay
such as parabolic SAR?
ChrisB
Herman van den Bergen
<psytek@xxxxxxxx> wrote:
Hi Chris,
Yes i like to develop short term systems: the shorter the
better, usually 1-10 minute trades. imo, The significance of a signal
fades quickly. And yes, I meant ApplyStop() type stops where
you set the position to close at a given % trade DD. Invariably maxloss
stops make me "lock in Losses" and the price goes my way after
I close the position. I don't use complicated stops and perhaps that is
the problem. I find it better to develop system without and kind of
profit/loss stops, just the basic system working on signals only. Almost
always after i have that working adding profit stops will increase
profits and adding Max Loss stops decrease profits.
best regards,
herman
Herman
Could I trouble you to
expand on that briefly?:
by "max stop loss" I presume you mean
an initial capital protection as
per
Applystop(stoptypeloss,...,......) or similar.
or do you
mean trailing stops as in
Applystop(stoptypetrailing,..., ..., ...
?
I appreciate your systems may be shorter term, rather than
longer.
Only ask because this w/e I have been reviewing all my
trades since Jan 2004 (ASX markets, stocks, long only, trend
following) and found that it is my trailing stops (whatever volatility
parameter), that curtailed my results, (together with emotions etc but
that is another story). I might look at locking in the stop at
breakeven, then only trailing when there is a
pattern/retracement/consolidation above each successive R-multiple
profit level, starting at 3R. This would have served me far better in
the trending market we have experienced over the last few years.
I am beginning to think that for shorter term systems, initial
Capital protection stops may prematurely halt the cyclical nature of
whatever is causing the system to work, but need to get to grips with
more coding and backtesting skills to confim this.
Your
comments would be most
appreciated.
ChrisB
Herman van den Bergen
<psytek@xxxxxxxx> wrote:
stocks have "character" some work Long and some work short
and some work both ways. Same wrt rhythmic and other
characteristics...imho there is no reason why we should assume any
characteristic to be permanent or common to a large
population.
When designing a system I try to find similar
performance for Long and Short, this gives me more confidence that I
won't go broke in a strong trend. Systems that only work Long or
Short make me nervous as I worry that they will stop working
abruptly. Sometimes, most of the time I
should say... it is necessary to adjust parameters
individually for long and short. I try to develop systems
that give me thousands of trades (minute time frame) and produce a
nice smooth surfaces on the 3D charts. I never trust systems that
give me more than one significant hotspot.
wrt indicator, I don't use any. I trade only very short term
patterns - I am a skeptic on the use of traditional indicators.
Never got any to work well - probably because I don't have the
patience (or nerve) to sit through long trades and through major
DDs.
tips? don't use any max loss stops, imho they kill systems.
Use profit stops instead. Design both Entries and exits
individually, only rarely will an entry rule give same performance
as an exit rule. The exception to this may be high speed
automated reversal trading systems (50-100 trades a day) that are in
the market full time
jmo... from a developer's viewpoint, I enjoy development more
than trading :-)
herman
I have
some nice, well-tested long systems in place. I was surprised
when testing my discretionary systems, to find that none of my
short
signals performed nearly as well as the long signals, in
the
optimization/backtest/monte carlo simulations.
Is
this common?
In addition, I am looking for some ideas
around what indicators to use
as the foundation for building
an adequate short system. Any ideas? I
did some searches on
previous messages here, and did not find anything
of value.
General rules of thumb, and bits of experiential wisdom, are
also welcome -- as they apply to short systems.
Thanks
in
advance,
Brian
Yahoo! Mail
Use
Photomail to share photos without annoying attachments.
Please note that
this group is for discussion between users only.
To get support
from AmiBroker please send an e-mail directly to
SUPPORT {at}
amibroker.com
For other support material please check
also:
http://www.amibroker.com/support.html
YAHOO! GROUPS LINKS
Yahoo! Mail
Use
Photomail to share photos without annoying attachments.
YAHOO! GROUPS LINKS
Yahoo! Mail
Bring photos to life! New
PhotoMail makes sharing a breeze.
Please note that this group is for discussion between users only.
To get support from AmiBroker please send an e-mail directly to
SUPPORT {at} amibroker.com
For other support material please check also:
http://www.amibroker.com/support.html
YAHOO! GROUPS LINKS
|