|
PureBytes Links
Trading Reference Links
|
Keeping copies of code updated is a real pain. You do not suggest using
functions, but they would help to a bit if you did. I guess there is a
reason, so I wondered why you avoid functions. I realise they cost overhead,
I also know they can not use all order position info, but some of it. Just
curious for the reasons.
> -----Original Message-----
> From: Bob Fulks [mailto:bfulks@xxxxxxxxxxxx]
> Sent: den 2 augusti 2001 17:47
> To: jonmac
> Cc: omega
> Subject: Re: TS Precision...The Lessons, wrap for Beginners
>
>
> At 2:06 PM +1000 8/2/01, jonmac wrote:
>
> >As a very green programmer with lesser knowlegde of maths;
> >
> >I been building a system based on use of COT interaction as
> a filter. I
> >find it not only interesting, but figure I've almost a good start on
> >the concepts visualised in the indicators I've built. Taken
> me forever.
> >Next stage is to start on entry & exits. I'll probably be
> dead before
> >I've got it together, but nevertheless....
> >
> >NOW after all this I'm wondering how these indicators would come
> >together in respect of our lengthy discussion on precision.
> >
> >I wonder if its not too much to ask our best for, some very
> elementary
> >detail (FOR NOVICES) on what to be wary of and workarounds, please?
> >
> >(Bob, I have your summary and noted the likes of "if Mod(currentbar,
> >100) = 1". It appears a very simple fix, and from the way I
> read it, I
> >presume it could always be used in place of "if currentbar = 1"? My
> >request here is related to what could be used as a very
> basic list of
> >scenarios/fixes).
>
> This usually works well for cases where there is a different
> calculation on CurrentBar = 1. As I mentioned, a few
> functions (such as RSI) initialize with a different algorithm
> than they use for succeeding bars because that is the way it
> was defined originally. It should not be done in cases where
> certain variables are initialized on CurrentBar = 1 because
> you will want them to retain their values from bar to bar.
>
> I often separate the things I want to do once from those I
> want to do every 100 bars as follows:
>
> Vars: Init1(TRUE), Init2(TRUE);
>
> if Init1 then begin
> <initialize variables one time only>
> Init1 = FALSE;
> end;
>
> if Init2 or Mod(CurrentBar, 100) = 1 then begin
> <complete calculations to be repeated every 100 bars>
> Init2 = FALSE;
> end else begin
> <calculations to be done the other 99 bars>
> end;
>
> I use the variable "Init1" and "Init2" rather than
> "CurrentBar = 1" in functions so that code will run on the
> first bar it is executed in case that is not CurrentBar = 1
> (if it is a "Simple Function" enclosed in some "if ... then"
> statement.)
>
> I usually want to know about "divide-by-zero" and
> "subtract-near-equal-number" errors so often use the follow code:
>
> if Denom <> 0
> then Result = Numer / Denom
> else Print(Date:7:0, Time:5:0, "Error XX");
>
> where XX is some number. The Print Log then records cases
> where you get this error. If it is just some weird
> combination of values that occur occasionally then you can
> usually forget it and "Result" will retain the value from the
> previous bar. But sometimes you find major errors that need
> to be handled by different code or algorithms.
>
> The subtract-near-equal-number" errors can be detected by
> code similar to:
>
> If Term1 is always > 0 (often the case) you can use:
>
> Result = Term1 - Term2;
> if AbsValue(Result) < 0.0001 * Term1 then
> Print(Date:7:0, Time:5:0, "Error XX");
>
> If the terms can be either positive or negative you would
> need to check additions also. This adds lots of calculations
> to the program so I would only use it when the terms are big
> and complex, such as in some statistical routines, and would
> comment it out after I was satisfied that there was seldom a
> problem with that calculation.
>
> As Dennis mentioned, it is always wise to build an indicator
> with exactly the same code as your system to plot the key
> values. I ALWAYS do this, even for the simplest systems. It
> is amazing what you find! Something like the following code:
>
> Inputs: IMode(2), ......
>
> Vars: Init1(TRUE), Init2(TRUE, ....);
>
> if Init1 then begin
> <initialize variables one time only>
> Init1 = FALSE;
> end;
>
> if Init2 or Mod(CurrentBar, 100) = 1 then begin
> <complete calculation to be repeated every 100 bars>
> Init2 = FALSE;
> end;
>
> <calculate the values of all variables here>
>
> {=================System-only code========================}
>
> <put Buy/Sell/Exits here>
>
> {===============Indicator-only code=======================}
>
> if IMode = 1 then begin {Subgraph1 Plots}
> Plot1(Variable1, "1");
> Plot2(Variable2, "2");
> if FALSE then Plot3(0, "3");
> if FALSE then Plot4(0, "4");
> end;
>
> if IMode = 2 then begin {Subgraph2 Plots}
> Plot1(Variable3, "1");
> Plot2(Variable4, "2");
> if FALSE then Plot3(0, "3");
> Plot4(0, "4");
> end;
>
> if IMode = 3 then begin {Subgraph3 Plots}
> Plot1(Variable5, "1");
> Plot2(Variable6, "2");
> Plot3(Variable7, "3");
> Plot4(0, "4");
> end;
>
> {===============End Indicator-only code===================}
>
> Then copy/paste the EXACT SAME CODE in both the System (or
> "Signal" in TS2000i) and Indicator. Comment-out the
> System-only code in the Indicator and comment out the
> Indicator-only code in the system. Then apply both the system
> and as many copies of the Indicator code (with different
> values of IMode) as needed to cover all the key variables.
> Look at the values of the indicators while the system is
> running to confirm that they look "reasonable". With the Data
> Window, you can read the actual values of the key variables.
>
> Make sure the MaxBarsBack setting is the same for both the
> System and all copies of the Indicators so all copies of the
> code see the exactly the same values. When you run an
> optimization on the system ("Strategy" on TS2000i), you have
> to change the values of the inputs on all of the copies of
> the Indicator so that they are the same (a royal pain).
>
> (There are more sophisticated versions of this technique that
> use Global Variables to allow the Indicator values to
> actually be those calculated in the System code but they
> require extra code modules.)
>
> Many errors will become readily apparent on the plots. You
> can then look into why each one occurs and take appropriate actions.
>
> Of course, for simple things that you are just "researching",
> you can be less demanding and just leave out the checks. But
> if you do not take steps like this for a system you are going
> to trade, then you are really asking for trouble. These
> techniques catch the TradeStation precision errors we have
> been discussing and also give you a much better idea of how
> your system is actually working on real data.
>
> I will post a chart illustrating this in the next message (to
> avoid the size limit on messages.)
>
> Bob Fulks
>
>
>
> Bob Fulks
>
This message contains information that may be privileged or confidential and is the property of the Cap Gemini Ernst & Young Group. It is intended only for the person to whom it is addressed. If you are not the intended recipient, you are not authorized to read, print, retain, copy, disseminate, distribute, or use this message or any part thereof. If you receive this message in error, please notify the sender immediately and delete all copies of this message.
|