|
PureBytes Links
Trading Reference Links
|
I'm escaping to the Japan Alps for a summer respite so I won't be able to
follow up for a few days. Briefly....
A system that generates wildly different results when a parameter is set at
8 or 10, for example, will not work in the real world. The system is too
finicky about its inputs. Parameters close together should generate similar
results.
Trade dependency occurs when the results of a trade are influenced by the
previous trade(s). You discover that 3 consecutive losses are always
followed by 2 consecutive wins for example. This is exploitable. A
statistical measure called the Runs Test can be used to find this out.
You "de-optimize" over your price history just like I described. You want
your system to work going forward, not backward. If a system is in fact
robust it will prove itself profitable on a forward testing basis. Forward
testing involves optimizing your system over a certain time segment, then
taking these optimal parameters and trading/testing them in the next period
of time to see how the system does. The idea is to mimic the real life
process of optimizing, finding the optimal parameters, and trading them into
the "future." When you've gone through the whole price history, your total
results from the trading/testing periods indicates how you would have done
in "real life." The parameter you end up with is your peak point.
Now, if you plot your test results with parameter values along the X axis
and trade returns along the Y, what you theoretically have is a smooth
domed-shaped curve. Of course its not really the case because in reality you
have a series of spurious peaks and valleys, but nevertheless, its overall
shape is that of a rough curve. Since parameters close together should give
similar results, identifying the peak gives you a "neighborhood" to
consider. Use one from this "neighborhood" that gives lower performance than
the parameters immediately before and after it. The bottom of a 'V' formed
between the highest and another nearby peak if you can imagine it. You don't
want the highest peak parameter because if performances are going to tend
toward each other then you are better off trading the parameter whose
results may tend higher, rather than lower. This is due to a phenomena
called entropy.
If your system can pass thru this routine, and only a very few due, then
you've got a good system that works over a variety of market conditions.
There will be no need to "re-optimize." FYI, of the 232 of systems I've
tested, only 6 made it thru.
Re-optimization sounds like a good idea but it's not. I've mentioned this
recently, but the statistical distribution of price changes or returns is
not "Normal" or stationary. It floats around and changes shape, hence all
robust systems go through down periods that are in no way positively
affected by changes in parameter values. Conditions haven't changed, the
system has just gotten temporarily out-of-sync with prices. Re-optimization
is akin to a dog chasing its tail. Avoid systems that require it.
sayonara,
Rick
-----Original Message-----
From: Roland Beauregard <beau@xxxxxxxxxxxx>
To: metastock@xxxxxxxxxxxxx <metastock@xxxxxxxxxxxxx>
Date: Friday, August 07, 1998 2:31 PM
Subject: Re: Optimizing
>Rick:
>
>Firstly, thanks for your contribution to this thread. I've been trading
>full-time privately for a few years now and still over half of my time is
>spent learning and testing.
>
>Elaboration on a couple of points would be helpful, when you have a moment.
>I assume this means a max of two or three optimized parameters.
>Insensitivity?? I have found that the range of results is very wide
>ie:highly sensitive. Most of the parameters in a test will exhaust capital
>long before current end data are reached. However, frequently a small
>percentage of the parameters will produce returns in the hundreds of
percent
>per year and higher.
>>2)Backtest over as much data as possible, but 5 years minimum is
>>recommended. You need a minimum 30 closed trades (buy & sells) to get
>>statistically valid results. Anything less over 5 years and you might as
>>well trade the 200 day SMA (or 233 for Fib fans!).
>
>No problem.
>
>>3)Check for dependency between trades. While rare, they can exploited.
>
>What is " . . .dependency between trades"?
>
>>4)Now FORWARD TEST. Divide up your data set into a minimum of 3 sections.
>>More is generally better, but the exact length of each section you choose
is
>>really just a function of your system itself. Then do the following:
>
>>Optimize over section 1. Trade over section 2.
>>Optimize over section 1+2. Trade over 3 etc, ad infinitum.
>
>How this done??
>
>>5)Don't trade the peak parameter, but choose one that has lower
performance
>>that the one immediately before and after it. Parameter values that are
>>close together tend to have similar performance (IF you followed Rule 1!).
>>You are better off trading the parameter whose performance will tend to go
>>higher than trading the peak, which tends lower. Don't fight the Second
Law
>>of Thermodynamics aka entropy!
>
>Not sure I understand why you wouldn't use the hishest performing
>combination of parameters and fine-tune [re-optimize] every other week or
>month to align the parameters with the evolving price and volume
>characteristics of the particular security.
>
>>6) IF the system passes successfully steps 1-5 then put this now very
robust
>>system in your toolbox and LEAVE IT ALONE. Build another system if the
>>"system tinkering bug" bites ya!
>
>Agreed!! Thanks again,
>
>Roland
>
>>Hope this helps,
>>Rick
>>Tokyo, Japan
>>
>>
>>-----Original Message-----
>>From: Roland Beauregard <beau@xxxxxxxxxxxx>
>>To: metastock@xxxxxxxxxxxxx <metastock@xxxxxxxxxxxxx>
>>Date: Friday, August 07, 1998 10:13 AM
>>Subject: Optimizing
>>
>>
>>
>>>However, I've been thinking along the same lines as John, that frequently
>>>re-optimized parameters logically should produce better results than any
>>>specific standard parameters, including Fibs. . . . and of course they do
>>>much better in system tests.
>>>
>>>I'd like to see more discussion on optimization. Has anyone done any
>>>extensive testing and validation? Is optimzation a significant factor in
>>>anyone's successful trading system? Is there something inherently
invalid
>>>with optimizing?
>>>
>>>Comments anyone??
>>>
>>>Roland
>>>
>>>At 10:26 AM 8/6/98 -0400, you wrote:
>>>>I am not sure if "trading with the wrong parameter" is correct, as you
>>>>stated. My belief is that each stk has its own trading characteristics.
>>>>Technical analyst in nothing more than measuring supply and demand of
>>price.
>>>>
>>>> And my premise is that the stk characteristic will remain in tack until
>>it
>>>>changes.
>>>>
>>>>Most complete trading sysyem have two major general characteristic. One
>>is
>>>>money management (i.e when to get in, get out and double up) and two, a
>>>>history of win vs lost and max drawdowns. And the reason by most people
>>do
>>>>not make "much money" on system trading is not having the trust when the
>>>>system turns against them.
>>>>
>>>>OminTrader assumes that the stk will be generating the same
charisteristic
>>>>and therefore the indicators can be trusted until the characteristics
>>>>changes. By being able to rerun the indicators against the stk on a
short
>>>>term basis allows re-optimizations of the indicators to the new
>>>>characteristic of the stk.
>>>>
>>>>This is not a guarrentee system. If it was, I would not be writing the
>>>>E-mail, but sunning somewhere in Europe counting my winnings. But,
>>>>combining with MetaStk to do additional analysis is not a bad way to
>>predict
>>>>the next short turn move
>>>>
>>>>regards
>>>>johnnie
>>>>-----Original Message-----
>>>>From: Jan Diederik <jddehaas@xxxxxxx>
>>>>To: metastock@xxxxxxxxxxxxx <metastock@xxxxxxxxxxxxx>
>>>>Date: Thursday, August 06, 1998 8:44 AM
>>>>Subject: RE: Omni Trader 3.5
>>>>
>>>>
>>>>>> -----Original Message-----
>>>>>> From: owner-metastock@xxxxxxxxxxxxx
>>>>>> [mailto:owner-metastock@xxxxxxxxxxxxx]On Behalf Of John Scudiero
>>>>>> Sent: Thursday, August 06, 1998 1:14 AM
>>>>>> To: metastock@xxxxxxxxxxxxx
>>>>>> Subject: Re: Omni Trader 3.5
>>>>>>
>>>>>> OT is the complete opposite of a trading system methodology. It
>>>>>> optimizes
>>>>>> indicators against a data range you chose, the intent is that the
>>price
>>>>>> trend of the immediate future is the same as the existing past,
thereby
>>>>>> giving you excellent readings based on the indicators. And so
>>>>>> many days in the future, you re-run the indicator test to reoptimize
>>>>>them.
>>>>>
>>>>>In my opinion this means that you have been trading with the wrong
>>>>>parameters for your indicators. Afterwards you can say which parameter
>>you
>>>>>should have taken. Pretty frustrating if you ask me ;-)
>>>>>
>>>>>Regards,
>>>>> Jan Diederik
>>>>>
>>>>>For the record: I never optimize; I usually take Fibonacci numbers.
>>>>>
>>>>
>>>>
>>>>
>>>
>>>
>>>
>>>
>>>
>>
>>
>>
>
>
>
>
>
|