|
PureBytes Links
Trading Reference Links
|
At 12:14 AM 9/11/2005, Donald Beck wrote:
>Assuming something greater than 1 tick compression,
>lets say 100 tick bars; In all cases the signal will
>be fallaciously better if historic data is used.
>Assuming that there is a price gap that the signal
>trades through. No idea about 8 (my bet would be the
>issue still exists), but in 2000i, if you change
>anything that requires the chart to refresh, signals
>price points will move.
Think about this a minute. Your system reacts to each tick as they
come in. After the chart is refreshed, the system is working on only
the OHLC values for each bar. So it is IMPOSSIBLE for it to know the
order of the ticks from the four OHLC values. That has to be obvious
and it is true for any trading platform. The only way to get the exact
sequence of ticks is to test within-the-bar to the 1-tick level.
Anything else is an approximation.
It has been my experience that if you chose a time compression such
that you get at least ten to twenty bars in a typical trade, all of
these factors become insignificant.
>For me, the result becomes a cost. No different than
>asking the broker what the commission charge is or
>figuring in slippage.
True. That is why trading platforms allow you to add in a slippage
factor so that you can more realistically "simulate" the real world.
>For the SPY market I had created some custom
>software's that fully automated my trading. Risk,
>cash, orders as well as trades were managed. I also
>blended and created a synthetic data stream using a
>number of financial instruments. In all it was capable
>of placing more than 30 orders per minute. Net profit
>was to have been a mere .005 per share, per trade
>profit. With those very tight margin's everything
>becomes critical.
I have also tried those kinds of systems but they get killed by
real-world factors such as delays in the data feed, slippage, etc.
Bob Fulks
|