|
PureBytes Links
Trading Reference Links
|
Subject: Re: System results changing at reboot?
I have experienced the same problems in realtime and backtesting particulary
as I receive Signal (satellite) in the UK and TradeCenter in receives
BMI(Cable). Signal appears to average and skips ticks sending Roughly 2/3
of the ticks of BMI - greatly affecting results.
However these problems did not exist before I added (exit) signals to entry
signals working in conjunction with mm stops - its like we have different
systems. Would anyone know if "begining" and "ending" sections of code (a
la LBRaschke) gives a more consistent result across data vendors and if so
how and where does one begin and end and why ?
Alternatively would it make any sense to "include" all signals as seperate
systems ? If so theres precious little in the TS4 manual ( and from
technical support) on include systems (which use 10-12 functions and
variables for each signal). I was simply told TS5 will have no limit -
that there won't be the error message "exe code" too large. Altering the
time frame doesn't help either.
Ive missed 3 great trades - its driving me and TradeCenter nuts - does
anyone have any thoughts? remedies ideas? abuse?
Michael
Subject: Re: System results changing at reboot?
>In message <199806111853.UAA09556@xxxxxxxxxxxxxxxxxxxxx> hans wrote:
>> The problem is "bouncing ticks" - see also online help/manual
>
>Yes, I know about the "bouncing ticks" problem but I've never really
>understood it. Let's say the manual is not extremely thorough
>or clear. :-\ The online help is better but still pretty vague.
>
>I thought that, unless you had the "update every tick" box checked,
>the system is not run until the bar closed. At which point,
>I would think, it "knows" everything about the price action
>(basically just OHLC) that it would know in historical testing.
>
>> So it might hit a stop or profit target first -
>
>Hmm, I suppose that could be it. That could cause different behavior
>in realtime vs. backtesting. I.e. if you're long and the previous bar
>set up a "sell stop" AND an "exitlong stop", then it might have hit
>the "sell" in realtime ticks but the "exitlong" in "bouncing ticks"
>simulation.
>
>> Best way around is to use as short as possible time bars, i.e. try to
CHANGE
>> the code of your system to run on 1min bars instead on 5min or 60min -
often
>> thats possible, but not allways.
>
>Yes, I've done things like that, but it's a pain. It also limits
>your tests to 30day periods, which is often not acceptable.
>
>> .....and its not a bug - its a "feature" <g>
>
>Ya right. Yet another high-quality Omega "feature."
>
>I understand why Omega did this -- it saves a lot of time over
>testing every tick when you go through a system test. But if the
>bars in the chart were constructed from tick data, then it seems
>to me there's an obvious, fairly simple, and CORRECT solution:
>use only the bar's OHLC until you get to a bar that hits your
>signal prices. If the bar's H/L encompass an entry/exit price,
>THEN you can do the tick-by-tick analysis for that bar only.
>
>Ohwell, supposedly TS5 fixes this and all other problems.
>(And, if history is any guide, it will introduce 1000 new ones...)
>
>Gary
>
|