|
PureBytes Links
Trading Reference Links
|
If we were to review all the problems with system test that don't click with
reality, their number would be legion. But forget about the buy and hold
reading. It starts reading from the first data point. While I can't imagine
a buy and hold, think what we are saying the "investor" decides to buy on
that date, there is no logic to that.
Then remember that It is money over time. Always note how long your capital
was at risk. Duration of trade is important. A stock shows a 150% increase
over 5 years, buy and hold. Your system shows 35% with total IN duration of
less than a year. Your system wins in my book.
Metastock gives us points to compare between systems. That is the reason for
the test. Not to recreate reality. MSWIN could really be a show stopper if
they provided system test over muti-stocks rather than the one at a time
test that has no meaning in today's market. I think you need a minimum of 5
years data, more is better and you need to look at a minimum of 50 stocks
similar to what you trade. It would mean more if you selected the stocks at
random. Those futures guys have an advantage, stocks have 10,000 markets,
they only look at a few.
Richard Estes
-----Original Message-----
From: rudolf stricker <rst@xxxxxxxxxxx>
To: metastock@xxxxxxxxxxxxx <metastock@xxxxxxxxxxxxx>
Date: Sunday, August 02, 1998 3:39 PM
Subject: Re: System test w/o "re-investment": How?
>
>Jan,
>
>On Sat, 1 Aug 1998 12:51:13 +0200, you wrote:
>
>>If the System Tester would also note the % increase/decrease your problems
>>should be solved.
>
>That would work for me, because that's the same like having a sequence
>of trades with a _constant_ investment for every single trade.
>
>A step into this direction might be MS' "semi-log scale", but this
>scale is not used explicitly in system test's rating system for the
>tested trades, is it?
>
>Besides, MS' "semi-log scale" works unreliably in many situations,
>e.g. when it comes to large and small numbers.
>
>>I have another question regarding the System Tester. Almost every
strategie
>>I test with the System Tester losses against the Buy and Hold strategie.
>>It's not that my strategies are bad, but because a lot of my systems
require
>>a lot of data. Therefore the first buy is at a much higher price than my
>>first datapoint, which is used for the Buy and Hold strategie. Is there an
>>option/workaround to set the date at which the Buy and Hold strategie
should
>>start?
>
>You can e.g.
>
>set the x-axis to start at a certain day, where you want to start the
>b&h strategy
>
>or
>
>define a b&h system test which includes a condition like
>"year()>=1998" to start your b&h with the beginning of 1998.
>
>mfg rudolf stricker
>| Disclaimer: The views of this user are strictly his own.
|