|
PureBytes Links
Trading Reference Links
|
I am looking for trading partner(s) in the Toronto area,
if interested please reply by e-mail stnahc@xxxxxxxxx
-Lawrence Chan
p.s. Barney (I am not sure if I have your name spelled right),
I have not talk to you for a while, so please call if you
see this message.From ???@??? Tue Apr 14 10:59:13 1998
X-POP3-Rcpt: neal@xxxxxxxxx
Return-Path: <omega-list-request@xxxxxxxxxx>
Received: from mx1.eskimo.com (mx1.eskimo.com [204.122.16.48])
by purebytes.com (8.8.4/8.8.4) with ESMTP
id KAA23925 for <neal@xxxxxxxxxxxxx>; Tue, 14 Apr 1998 10:28:57 -0700
Received: (from smartlst@xxxxxxxxx)
by mx1.eskimo.com (8.8.8/8.8.8) id KAA20812;
Tue, 14 Apr 1998 10:26:35 -0700
Resent-Date: Tue, 14 Apr 1998 10:26:35 -0700
Message-ID: <35339C4E.5798@xxxxxxxxxxxxxxxx>
Date: Tue, 14 Apr 1998 10:26:38 -0700
From: dcountach <countach@xxxxxxxxxxxxxxxx>
Organization: rich and famous
X-Mailer: Mozilla 2.01E-compaq (Win95; I)
MIME-Version: 1.0
To: omega-list@xxxxxxxxxx
CC: KKiely6335@xxxxxxx
Subject: Re: luck (or Indicator Validity)
References: <199804132158.OAA19538@xxxxxxxxxxxxxx>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Resent-Message-ID: <"ZEBqz3.0.r45.7nvCr"@mx1>
Resent-From: omega-list@xxxxxxxxxx
X-Mailing-List: <omega-list@xxxxxxxxxx> archive/latest/17128
X-Loop: omega-list@xxxxxxxxxx
Precedence: list
Resent-Sender: omega-list-request@xxxxxxxxxx
There is another thing to consider. Many indicators are not valid when they are first
displayed. This is a result of the dreaded MaxBarsBack phenomenon.
Any variable which is not based solely on the price (i.e., contains another variable) is not
calculated (or, perhaps, is not saved) during the MaxMarsBack period. An indicator based on
such a variable is calculated starting at Bar 1, without the benefit of previous variable
values, and will not be valid until BarNumber = MaxBarsBack. Any trades during this period are
suspect.
Yes, if you need 50 MBB, then that indicator is not valid for the first 50 bars of its display.
If you need 200 MBB, then ... This becomes a real problem when using long lengths, such as
200-day moving averages.
This is relatively easy to test. Use an indicator (say Volatility). Put it in two charts of
the same stock, with one chart having a month less data. Note the date and value of the
indicator at the first bar of the shorter data series. Then note the value at that same date on
the longer series. The value will be different.
One way to avoid this is to not plot or use an indicator until BarNumber > MaxBarsBack.
This subject was discussed thouroughly on the list about two years ago. You can probably find
the discussion in the list archives. (Where are those archives?) I think I used the term
"derivative variable" to describe the problem. You could search for that phrase.
donc
>Date: Mon, 13 Apr 1998 09:53:16 -0400
>From: Bob Fulks <bfulks@xxxxxxxxxxxx>
>To: KKiely6335 <KKiely6335@xxxxxxx>
>Cc: omega-list@xxxxxxxxxx
>Subject: Re: luck
>Message-Id: <v03102803b157c686805a@[205.161.35.232]>
>Content-Type: text/plain; charset="us-ascii"
>
>At 1:55 PM -0400 4/12/98, KKiely6335 wrote:
>
>>I have a question. If you are testing a system (say a 20 breakout
>>system) on 10,000 bars of data and you start @ bar 1 and end at 10,000 and get
>>a certain profit, then you should be able to test the same system starting on
>>bar 10 and ending on bar 10,000 and get the roughly same profit & loss. The
>>question i have is that if you start 10 days later will this start a chain
>>reaction because changing any given trade's market position (long, short,
>>flat) will affect all the subsequent trades in unexpected ways . Am i wrong
>>on this or am i right?
>
>
>It actually would depend upon the system in some cases. A 20 bar breakout system might
>actually end up with the same initial trade even if you started 10 days later.
>
>But I believe you are correct in principal. However, the effect would be
>extremely small on a test of 10000 bars. This is why you like to have a
>large number of trades to test a system.
>
>The purpose of the backtesting is to get an idea of how the system works on typical data. >You can get such an idea by using any part of the 10000 bars of available data. In fact, >most
people try their systems on different blocks of data to test its robustness in >different kinds
of markets.
>
>Bob Fulks
|