|
PureBytes Links
Trading Reference Links
|
Yuki,
Thanks for answering. There is no way to know the future, of course,
but (1) and (2) are totally different situations since I belive
results obtained with (1) are more likely to simulate real time
results in the future cause at least the signals are stable (they can
be checked afterwards) while backtesting results in (2) don't reflect
intraday false signals, therefore it is less likely that simulation
results will indicate the probable performance of the system.
I do belive there is a paradox since the more you approach (when
simulating systems) to the exact time of real trading the less likely
is that simulation results will represent the reality of your system.
In other words, the more you delay your entry the more likely your
results will represent a probable future. At least this is how I see
it.
Let me ask an specific question. Let's imagine I have a system
developed according to (1) with 0.5 cents of expectancy per dollar
risked and I have a system developed with (2) with 0.8 cents of
expectancy per dollar risked. Which one is best?. I am sure there is
no way of knowing it. Isnt it?
--- In amibroker@xxxxxxxxxxxxxxx, Yuki Taga <yukitaga@xxx> wrote:
>
> Hi cagigas00,
>
> There is no paradox, really. What you are looking for is commonly
> called the Holy Grail (you didn't specifically ask for that, but you
> want a system that doesn't cough up a hair ball now and then, and
> that is the same thing). There isn't one.
>
> Take your example (1). If the market goes down 5 percent the next
> day, you think that this should cancel your previous day's long
> signal. (Maybe it should, but sometimes a long signal on one issue
> on a particularly bad day is a very, very, good long signal.)
>
> But in any case, you cannot know tomorrow, much less its close, even
> when tomorrow is today. At what intraday point would you decide
that
> today's action invalidates yesterday's signal?
>
> I dare say you cannot.
>
> Perhaps, by way of filters, you might know when to *abort*
> yesterday's signal (taken, presumably, on today's open) by the end
of
> the day today. But even this much is likely only in the very short
> term.
>
> So, really, there is no paradox. In *neither* situation can you
know
> the future.
>
> What you *can* know is your system's expectancy, and your capital,
> and what is an intelligent risk based on both of those.
>
> Expected holding period means a lot in conjunction with my comments.
> Obviously, the longer time frames are easier to work with, but they
> have more lag. Also obviously, you want to enter when there are
> unexpected corrections in those longer time frames -- occurring of
> course in the short term. The short term time frames have very
little
> lag. But they turn on a dime, and give 9.9 cents change.
>
> Either you have a system you can trade, or you don't. It's actually
> that simple. And having a system you can trade means taking the
> losers with the winners, because you *know* you can stand the
losers.
> It means no opinion about the market. It means discipline, not
> prognosticative powers or reflection, or second-guessing. You put
> the orders in and let the devil take the hindmost. If your system
> degrades, it will do so over a long time frame, not in a single
day's
> action. (I am assuming proper back testing procedure, enough
sample,
> enough trades, walk forward, and some stats that make you say, "I
> *must* do this, in spite of the fact that I will have drawdowns once
> in a while.")
>
> If your system is good enough, you can survive the infrequent hair
> balls. If it isn't, you must look for another system. You
absolutely
> cannot "win them all", and how to do that is really the gist of your
> question. ^_^
>
> Yuki
>
> Tuesday, April 25, 2006, 6:16:11 PM, you wrote:
>
> c> Hello,
>
> c> I wanted to share my experience developing systems and there is
> c> something when implementing real-life that I have noticed that
you may
> c> want to comment/help me on. This is causing me a headache since
it is
> c> a kind of paradox I need to understand. Please tell my if I am
missing
> c> something here:
>
> c> When you develop a system you have 2 options.
>
> c> 1. Use all available information for today and trade tomorrow
>
> c> 2. Use all available information for today and trade today
>
> c> In 1. you will be able to know today tommorrows picks (this is a
great
> c> advantage if you want to publish your signals) but tommorrow may
be a
> c> day that is not good for your system (i.e. you have a signal to
go
> c> long tommorrow but tommorrow the market goes down 5%). Should you
take
> c> the signal?. Yes you should, otherwise you are not following your
> c> system. Although this is an extreme example I belive you
understand
> c> what I mean.
>
>
> c> In 2 you are using today's information to trade today. This is
good
> c> because you only enter if at the moment of the trade all
conditions
> c> are good. My experience is that i.e. adding a PDI(14) > MDI(14)
as a
> c> filter to my system (no tradedelays) results improve a lot in
> c> backtesting but this is not 100% real since the cross PDI-MDI
> c> sometimes activate my signal instead of the intended conditions.
When
> c> I trade with today's information I find a lot of false signals.
There
> c> is a long signal, I buy, conditions change (i.e. PDI moves below
MDI)
> c> and the trading signal dissapears from the scan and from the
chart.
> c> Despite this I get much better results in backtesting even with
big
> c> slippage compare to 1. In real life trading I am using 2 and I am
up
> c> +45% for the year despite of the problems mentioned.
>
> c> I understand that only in 1 the backtesting will provide me some
> c> insight into the future, but 2 is too appealing to ignore it and
I
> c> have not yet seen very good systems in 1 compared to 2. I belive
there
> c> is no way to backtest 2 to obtain real (achivable) results since
the
> c> backtester ignores signals that happened during the day but were
not
> c> true at the close.
>
> c> There is something in between (let's call it option 1.5) which is
to
> c> trade tommorrow if open > reference_price. But I have not tried
this
> c> yet.
>
> c> Have you had this same experience?. What option do you use and
what
> c> are your results?. Any comments about this will be welcome.
>
> c> Thanks
>
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
<*> To visit your group on the web, go to:
http://groups.yahoo.com/group/amibroker/
<*> To unsubscribe from this group, send an email to:
amibroker-unsubscribe@xxxxxxxxxxxxxxx
<*> Your use of Yahoo! Groups is subject to:
http://docs.yahoo.com/info/terms/
|