|
PureBytes Links
Trading Reference Links
|
My comments interspersed below and noted with a **.
Steven Buss
sbuss@xxxxxxxxxxx
Walnut Creek, CA, USA
-----Original Message-----
From: Gary Fritz <fritz@xxxxxxxx>
To: Steven Buss <sbuss@xxxxxxxxxxx>
Date: Friday, July 31, 1998 8:34 AM
Subject: Re: Timeframes are not created Equal
> - The same system backtested on different timeframes will have
> varying results partially because the system will turn out to be
> better or worse at leveraging the opportunities implicit in a
> specific timeframes bar characteristics/expression list.
I'd agree with that. I know of several systems that perform great on
30-min bars but much worse on 60-min, and vice versa.
> - What are the specific mathematical expressions that, taken as a set,
> account for the difference in "look" across timeframes?
> Untested First Cut at the Expression List:
> - Average True Range of timeframe 1 for last x number of bars /
> Average range of price intersection at timeframe 1 from bar to bar
> for last x number of bars
Didn't you intend to include Timeframe 2 in here somewhere?
What exactly do you mean by "price interseaction at timeframe 1" ?
** No. Draft expression #2 attempts to get at multi-timeframe
relationships. I was thinking after I sent the note that the text in the
paragraph above wasn't clear. Let's see if I can state it clearly this
time.
** I think the numerator in the above paragraph is clear right? But in the
denominator, by "intersection" I mean overlap. So, for example, if the true
range h and l of price bar "1" are 83.6 and 81.2 respectively and the true
range h and l of price bar "2" are 86.4 and 82.7 then the "range overlap" is
.9 (i.e., 83.6 minus 82.7).
I'm not certain that you can get equivalent results in 2 timeframes
just by tweaking a few inputs with expressions like this.
** The purpose of getting clear about the expressions in the expression list
was not to try to tweak trading signals to make a system tradeable in any
timeframe. Instead, it was to try to specify the differences between
timeframes in some quantifiable way.
It may
very well be that tradeable trends in a short timeframe disappear
into the bars in a long timeframe, or tradeable trends in a long
timeframe are swamped out by noise in a short timeframe. I suspect
that many markets will have an optimum timeframe for a particular
strategy.
** I think (but don't have empirical evidence) that this is right. I also
suspect that there is no single "optimal" timeframe per strategy. That is,
out of say 5,10,15,30,60 min and daily, weekly timeframes a strategy may be
"optimal" at 2 out of the total of 7 of these timeframes. I saw the process
of getting empirical evidence as consisting of the following steps:
** 1 Specify the mathematical expressions that account for differences
in the "look" of bars for timeframes (say 5, 10, 15, 30, 60 min, daily, and
weekly).
** 2 Determine which timeframes are the most similar in terms of the
data resulting from applying the expressions in the expression list.
** 3 Run multiple trading systems of different types on these different
timeframes. Determine which timeframes are the most similar in terms of the
kinds of results from the application of these trading systems. (This may
not be easy to do. I haven't thought through how one would do it but can
think of a couple of problems one would have to overcome to do it.)
** 4 THE PURPOSE OF THIS COMPLETE EXERCISE is to provide an understanding
of what specific timeframes a certain TYPE of trading system may be most
profitable in. (I'd like eventually to understand that if I have a system
that has solid backtesting results eod it will also probably be solid at xyz
timeframe but not at abc timeframe or def timeframe.)
** My note initiating this thread was really to determine if anyone had
attempted to get these questions before and already had the answer. But I
have found that I get much better responses to questions if I pose initial
answers myself and hence, hopefully, provide value to others. Truth be
known, I don't have much time myself to work through all four steps above.
|