|
PureBytes Links
Trading Reference Links
|
> From: Bob Fulks <bfulks@xxxxxxxxxxxx>
> At 08:09 PM 11/29/2004, Doug Tucker wrote:
>
> >If I display a five minute chart with volume and then write down the volume
> >at each bar, say 2500 tick volume in the morning when it is fairly active,
> >then if I later reload the data and look at the same bar it will show
> >something like 2540 volume, so I am not getting 40 ticks.
>
>
> 40 out of 2500 is still under 2% so I would be very suspicious of any system
> where a 2% difference in ticks could make much of a difference in the
> overall results.
>
> Bob Fulks
I wouldn't dismiss the difference so quickly. Variance in tick data can cause
systems built on tick data to vary widely. What you need to be aware of is tick
"phasing" and how and where the underlying tick count fills up the bar and how
that is interpreted by technical indicators.
Case in point: I helped a friend build a multiple period tick-based CCIB system
using multi-'timeframe' libraries. The concept is similar to multi-timeframe
minute bars, only on 50, 100, and 200 tick bars (as an example). Doing this
allowed us to line up signals on multiple tick 'time-frames', so-to-speak, just
like you might use 1 minute, 5 minute, and 30 minute time-frames together.
Our first finding was that MTF 200-tick signals didn't look like regular
200-tick signals. In fact MTF 200-tick bars didn't look quite like regular
200-tick bars. Why? The tick event on which the bar count starts determines
it's OHLC shape [that should start to cause alarm]. As a next step, I added
phase offset to realign MTF bars with regular system bars. This allows bars to
look similar and the indicators to produce simular behavior.
However, after refresh the TS tick count changed ... and shifted bars. The same
happened when todays ticks become yesterday's ticks. In other words, the N-tick
bars continually shifted and changed OHLC shape depending on what was in the TS
cache. When bar shapes shifted so did the indicators, and the CCIB is
especially sensitive to these kind of changes (as are many indicators). This
implied that any signal coming off those bars was fairly arbitrary.
Working MTF made the problem show up quickly and in a big way, but the issue is
still present even if you use a one chart. If you do use tick data with
indicators (I no longer do), I encourage you to look at where and when your
signals fire, and see if these are consistent when your cache is updated either
during the day or historically. It may be that you are just being fooled by
randomness.
Cheers,
Kevin
=====
Kevin Sven Berg To play the game
ksberg@xxxxxxxxx Get out of the bleachers
|