|
PureBytes Links
Trading Reference Links
|
At 04:30 PM 3/23/00 -0700, Gary Fritz wrote:
>> Mark Brown wrote:
>> one of the primary ways to defeat bad data is to use other than
>> high and low calculations in your code. the reasoning for this is
>> that a bad tick will most likely appear on a chart as the high or
>> the low of the intraday session.
>
>And you should realize that some indicators use High & Low, whether
>you know it or not. AccumDist, ADX, CCI, Stochastics, etc -- all
>refer to H & L. So if you trade with those indicators, your methods
>are sensitive to bad ticks.
>
>> all my methods must at least pass a flat score on random data.
>> that is they must at least break even, on random generated data
>> with a minimum of 17,500 data points.
>
>That's an interesting test, Mark. What kind of random data do you
>use? I would think that most good systems are looking for certain
>market behaviors, and if you remove those market behaviors, there's
>no guarantee the system will break even.
>
>Gary
>
Good questions Gary.
I think that Mark's suggestion of multiple data feeds which are
instantaneously compared is a good one. But very unrealistic for
the average TradeStation trader on this list.
Testing with random data is helpful but is it sufficient? Bad
data is not necessarily random. Perhaps certain types of bad
data should be searched for as well as random data.
These are important issues...
How does the average trader on this list build confidence in their system?
How can the trader filter for data anomalies?
What back-testing techniques can be used to incorporate bad data from
vendors?
Is there an effective way to put the breaks on a system if it is giving too
many bad signals?
Can that be done without missing "the big trade that makes up for prior
losses".
Should systems be designed to find bad data, in the same way that systems
are designed
to exploit good trading patterns?
Is too much emphasis placed on searching for profitable patterns, not
enough on
dealing with other trading problems like data errors?
Can anyone claim to have a good mechanical system without adequately
catering to the bad data issue?
This is an interesting area. Intellectually it's challenging, sure beats
watching TV. I haven't really given it much thought in the past, so I'd
like to hear what others think.
If any list-members are in Oakland this weekend, I'll be at the Coast
Investment Software booth
at the Online Traders World Expo. I think it's booth 618. Drop in, say
hello, it's always fun
to meet traders.
Good trading,
-Neal.
-----------------
Neal on the 'net.
Trade well. Train hard.
http://www.halcyon.com/neal/
|