|
PureBytes Links
Trading Reference Links
|
Jeff et al,
thank you for pointing me to this article.
On Fri, 22 Feb 2002 10:55:07 -0800 (PST), you wrote:
>See the following pages at John Conover's site...
>he just posteed these within the last week or so.
>There's a lot of math used to derive his results, but
>feel free to skip to the "examples" given on these
>pages. The main point is that if you are interested
>in computing "risk", the normality assumption will
>severely underestimate your risk.
>
>http://www.johncon.com/john/correspondence/020213233852.26478.html
But as far as I can see, it deals not with my subject above, but
rather with the bet size problem, where of course _the details_ of
the probability distribution (like eg tail volume etc) become most
important.
The topic I tried to address is more basic. In (my) simple words:
Which given discreet (price) time series is well suited to derive
appropriate trading actions? What "necessary condition" has to be
fulfilled by a time series to be the basis for a successful trading
system? What may be a convenient measure to define this "tradeability"
of a time series? And as can be seen from my subject, "information
frequency" imo is a relevant parameter here.
My suggestion:
>From Fourier's theorem we know, that at least 4 points are needed to
>identify a "wave form" in a series of data vs time (like price data).
>Because this is valid independently from the kind of problem or the
>analysis method used, its also valid for "trading" and for any "TA
>technique". So based on eg day-to-day data, we can identify / predict
>only waveforms of at least (3 to) 4 days (to make some profit from
>them).
>
>If we look at the price data of our favorite security from this point
>of view, we can construct probability distributions for price changes
>for eg 1, 2, 4, 8 days (when working with day-to-day data) and compare
>eg the probability to loose / win more than xx% for these time
>intervals. To make it more simple, we can also calculate the standard
>deviation for these price changes (, assuming normal probability
>distributions).
BTW: Standard deviation and normal distribution may be sufficient
here, because only the _global content_ of the probability
distribution is relevant here (, and not the details like in the bet
size problem mentioned above).
The procedure to calculate "tradeability" for a given time series at a
given information frequency (eg day-to-day) would look like:
>If we read from these data, that e.g. the price changes for 1 and/or 2
>days are much higher than for 4 days, my thesis is, that we cannot be
>consistently successful with applying TA techniques to trade these
>securities. Or otherwise: The size of 1 and 2 day price changes
>divided by the 4 day price changes is a measure for the "general risk"
>or "non-tradeability" of this security.
Because this calculation procedure is rather simple, everyone who is
interested in the tradeability of his/her favorite security could
easily measure it, and I'd like to compare your results with my
results, eg in terms of
"non-tradeability" = max (stadev (price changes; 1 time step);
stadev (price changes; 2 time steps))
/ stadev (price changes; 4 time steps)
If this fraction approaches (or even exceeds) 1, the security is
untradeable for the information frequency at hand, and the only (*)
way out to make this security tradeable is to increase the information
frequency. [(*) other measures to be discussed]
>So my questions are:
How does your favorite security look like in this light, eg in terms
of "non-tradeability" defined above?
What amount of "non-tradeability" (eg 0.5) do you trade successfully
using TA techniques?
Anyone interested to exchange experience about this kind of
tradeability?
Maybe, we could save lots of work on definitely non-tradeable
securities (at a given information frequency) to concentrate our
effort on other, more successful trading projects by following this
line.
mfg rudolf stricker
| Disclaimer: The views of this user are strictly his own.
|