Hugh,
The longer the timeframe of your database, the less accurate this becomes
--you start removing legitimate data peaks along with bad ticks at anything
over a tick database. With 5 second database the effectiveness drops
noticeably. With 1 minute it is hopeless.
However, you can use an ATR based approach to remove the largest price
swings. For instance, if they exceed 5 ATRs knock it back to 3 ATRs. Opening
gaps would have to have more headroom though. You would have to experiment
with the limits. This could remove really wild bad ticks and not hit the real
data too much of the time.
Dennis
On Aug 31, 2007, at 10:18 AM, Humblybob wrote:
Would
be very useful if someone could modify this afl for 1-min data or 15-min
data. Or let us know how this can be done.
Regards
On 8/23/07,
Dennis Brown <see3d@xxxxxxxxcom> wrote:
Hugh,
Bad ticks are not so hard to remove if you
are running a tick
level
database.
However, the longer the time period of your database, the
more difficult it becomes. I have
worked on a method to reduce them
on 5 sec database (SPY is real bad for bad ticks on
eSignal). A lot
depends on
which stocks you want to remove
bad ticks from (different
algorithms work best of different
stocks depending on volatility) and
how critical the information be absolutely correct to your
purposes.
Do a search on the subject.
Dennis
On Aug 23, 2007, at 1:14 PM, huandy631 wrote:
>
Does anyone out there know how to filter out bad ticks? I
know the
> view is that they should be filtered out a vendor level,
but I don't
> think my vendor values my business enough to make an
effort.
>
> Possibly filter through excel, or run a code in AB
that deletes bad
> ticks in the past?
>
>
thanks
>
> Hugh
>
--
Join my
group http://groups.yahoo.com/humblybob/