|
PureBytes Links
Trading Reference Links
|
Thx Bilo for a very useful overview!
My comments that follow apply to intraday data:
Where I can get stuck is in what I believe you categorized as the "second
step", whereas, once you have a workable system in your head, and in code,
that can recognize and capture a trend (for instance), then you wish to
allow your system to recognize mathematically what your eyes are seeing (in
my case, that the market is in consolidation or trendless, and to avoid
wasteful whipsaws), it becomes quite a headgame to attempt to imagine any
lagging indicator that doesn't have the effect of "not turning back on" fast
enough to hurt the bottom line, whereas with your eyes, and minderlists, and
futures charts, you can see the "threshold" of the entire market breakout in
ways (and seemingly endless variations) that can be difficult to qualify and
even harder... to quantify.
In spite of this, the present appears firmly routed in the past, and
seemingly in gratifyingly narrow windows at that.
The debate in my head, again, is whether the math has a merely statistical
basis (playing the odds) or an ongoing analytical basis that can conform to
certain parameters of price vs. time. The statistical approach seems to be
winning the battle. Sort of like Fib targets and NTU's... they just appear
out of the noise as a concrete description of seemingly random action.
I'm also surprised at how many exceptions appear to be just part of the
noise. Bad news, Fed announcements... the whole gammit. To my surprise, one
of the daytrader's credos appears upon testing to be completely false, and
that's the fear of holding positions overnight. I was amazed how often a
system can guess correctly about an overnight gap than incorrectly, and how
less profitable a system becomes without incorporating these gaps.
At any rate, I look forward Bilo to Episode II... if I miss it, I'll set my
TIVO for a "season's pass".
regards,
Gene
----- Original Message -----
From: "Bilo Selhi" <biloselhi@xxxxxxxxxxx>
To: "David Rosenthal" <davidrnews@xxxxxxxxxxxxx>; "Omega List"
<omega-list@xxxxxxxxxx>
Sent: Monday, April 30, 2001 5:12 PM
Subject: Re: Trading Systems
> obviously there is no single contained field
> of mathematics that you can apply to time series
> forecasting.
> on top of that you have to separate the math
> into math you use for modeling
> 1. profit/loss expectation, 2. supply/demand, 3. price/volume
> process from math you use
> for price forecasting.
> again since we do really know what p&l expectations
> of all participants are, therefore do we can not
> know the supply/demand as a result of actions based
> on those expectations. so we only know the result
> product of the process namely price and volume.
> so we can only model process in theory for which
> the math is separate and requires NLD and regression
> analysis and other kinks.
> once you passed this theoretical phase you
> will learn how to bypass 1 and 2 and work only with 3
> but still utilizing the theoretical laws from 1 and 2.
> these are the limitation of the holy grail, which is we
> don't know 1 and 2 and we can only know the process
> and the law that result in 3. so given those limitations
> you learn how to work with 3 only.
> *** price forecasting or prediction is basically comes
> down to finding short term vs long term dominant
> profit and loss expectations which in tern will result
> in action ( buying and selling or supply / demand ) which will produce
price
> and
> volume curve as the realized dominant price expectation.
> loop closed.
> if you can mathematically find what the majority of participants
> will do next ( how they will react ), then you got it....
> mostly it's non traditional regression analysis since all traditional
> regression techniques will not work in this case.
> noise is very important part of the equation but i hate calling it
> noise since there is no noise in price really.
> so you might want to take a look at the NDT.
> decision making process in system is pretty straight forward
> so there is no need to get into that.
> i only use and believe in univariate approach ( only the tape tells you
the
> truth )
> so i don't go out there looking for external things that move price.
> if price goes for who cares what reason, it goes and the system will pick
> that up.
> adaptive statistics is a major chuck of math but it's not too complex.
> you can't use straightforward stat analysis since data is non stationary.
> all typical stat bs will not work.
>
> you gotta think adaptive which is not the hardest part of the system math.
>
> and after all of the muffled stuff above and to clear things up
> you need to basically know three things:
> - probable signal ( noise vs signal , tradable vs untradable )
> - probable potential for the trade. ( min predictive delta )
> - those two have to be nailed 6-7 out of 10 times.
> if you get probable signal and the trade has potential then
> you go for it.
> as a result the tricks are: how do you pick probable
> signals adaptively and the most important and the hardest
> is how to do calculate the min potential of the trade
> and how do you do that with over 60% accuracy :-)
> it's hard not only mathematically but mainly conceptually.
>
> other thing is how do you properly correct the errors.
> on top of that there are a ton of other things.
>
> i always recommend to start from head to tail.
> picture how the system should trade
> where the signals should be ( dont be greedy and
> don't play "i want to sell the top and buy the bottom game )
> run the logic in your mind.
> picture the winners and the losers.
> lay out system signal logic next for winners and losers.
> find vars that are missing and those that you need for
> the logic to work.
> formulate the problem in math of how to find those vars.
> solve for those vars one step at a time.
> plug those into the logic and check how the system
> works.
>
> signals then logic then vars ( deduction )
>
> if you can get to the part where you can formulate the
> math problem then you are half way there :-)
> finding the solution is just a matter of time to apply proper math.
>
> curiously most people do the opposite ( induction ):
> they play with vars and the logic and then examine how
> the signals look :-) and if they look bad then
> they try some other vars or and other logic and
> there goes the never ending trial and errors approach.
> cart in the front of the horse.
> that is the key... if you have no idea how the system
> should trade and where the signal should be located
> on the chart you will never find anything that works
> because you don't know what to look for.
>
> the first approach makes you concentrate on
> proper solution and forces you on that narrow path...
> whereas the second approach forces you to run
> around flailing like a mad man looking for that road
> that you don't even know if it's the road that you
> should take.
>
> regards.
> bilo.
>
>
>
> ----- Original Message -----
> From: "David Rosenthal" <davidrnews@xxxxxxxxxxxxx>
> To: "Bilo Selhi" <citadel@xxxxxxxxxxxx>; "Omega List"
> <omega-list@xxxxxxxxxx>
> Sent: Sunday, April 29, 2001 6:46 PM
> Subject: RE: Trading Systems
>
>
> Bilo,
>
> Thank you for your thoughtful post, and congratulations on reaping the
> benefits of
> your years of hard work.
>
> Would you be willing to share what areas of mathematics have been most
> critical to
> your study?
>
> David
>
>
>
|