|
PureBytes Links
Trading Reference Links
|
Chuck,
I have noticed this increase in volume over time and have been
trying to figure out the implications for testing for a while now. I
have not come to a firm conclusion. Actually I have a lot of
unanswered questions.
But it is not just the changes over time that concerns me. Consider
the following statistics that come from an eyeballing a list of
7,000 stocks sorted by average volume in 2002. The numbers are
rounded to nearest multiple of 5. I hope no one will worry that the
total is 95% - this is just an estimate and I consider the precision
of 5% intervals to be more than sufficient for my present purpose.
15% had average volumes under 5k
20% had average volumes of 5-20k
15% had average volumes of 20-50k
10% had average volumes of 50-100k
10% had average volumes of 100-200k
10% had average volumes of 200-500k
15% had average volumes of 500k and up
If minimum average volume for testing is set to 500k, that excludes
75% of stocks. Depending on how many or few candidates a stock
selection method generates, one might be left with too few to make a
good statistical conclusion about the method.
Just what is an appropriate minimum volume to set for trading? Each
of us will have a different answer to this question depending on the
size of our trading account, whether we enter with limit or market
orders, how much slippage below one's stop price one considers
tolerable, etc. These are all very important "practical"
considerations that must be taken into account at some point, but
they are "practical" considerations that usually do not relate to
the basic idea of the strategy being tested.
Thus, I work with the assumption that it is best to do 2 sets of
tests - one to determine if a trading concept is valid and a second
set to determine if it is tradable.
Thus, for initial testing of a trading concept or idea, I want to
include all the stocks that might behave in a similar fashion. Thus
I often test with a minimum volume of 100k. Using 100k still
excludes 50% of all stocks, but that is 3 times the number one would
have with a minimum volume of 500k. Dropping to 50k only adds
another 10% (actually increases the stocks in the test pool by 20%
since one is going from 50% to 60% - got to watch how those
percentages really work). I usually use 100k when testing stocks
during the past 5 years or so. However, now that Chuck has raised
the issue of low volumes for earlier years, and given the near
completion of a new database of 15 years data, I may consider going
to 50k as the minimum. Before deciding I should really do a cross
section study of the percentage of stocks in each volume grouping
for 5 years periods - say 1985, 1990, 1995, 2000.
Some may wonder why bother testing with a minimum of 100k if one is
going to use 500k when actually trading - would one not need to redo
the tests using 500k minimum just to be sure the method works with
the higher volumes as well it tested with the lower volumes? Of
course, one would need to retest because the larger number of stocks
with 100k-500k could mask under performance by the 500k group.
So why bother doing the initial testing with lower volumes? Because
using lower volumes gives a much larger universe of stocks and that
larger universe allows you to test dimensions (such as price
restrictions for stocks selected) that would not be possible with a
small group. This arises from what might be a personal hang up of
mine, but I will not even both recording on paper any test result
that has less than 7 stocks in a test bin or period. Test results
from a group of 5 stocks is just too prone to randomness to be of
much value for deciding if the basic concept of a method is sound.
Actually, I prefer to have 20 stocks in each result bin when testing
a concept. If the concept tests well, then I can later test to see
what the effect is of reducing the number to actually trade to 10 or
even 5. If the profit increases sufficiently to make up for
increased variability of returns, then one might decide to trade the
method just with 5. Not that I would ever have 20% of my capital in
any one stock. If the best way to trade a method is just to take the
top 5 stocks, then I might devote 25% of my funds to that method.
The rest would go either into other methods or into selections 6-20
or the first method. My preference is to have 20 stocks in 2 or 3
distinct methods than 20 stocks in one. But I am getting off focus
from the question of minimum volumes.
I am eager to learn what approaches others take to the question of
minimum average volumes. My ideas are still forming on this topic.
b
--- In amibroker@xxxxxxxxxxxxxxx, "Chuck Rademacher"
<chuck_rademacher@x> wrote:
> I was about to send this email to "b", but I would welcome
comments from
> anyone else interested in such historical work.
>
> At the risk of having some of you ask why it matters, my
backtesting
> generally goes back to 1985. Just yesterday, I posted a message
to this
> group saying that I always use one set of parameters across all
stocks and
> across all timeframes. One of the downsides of this approach
(perhaps) is
> that volume has changed over time. I suppose that one could
argue that
> volatility changes over time as well. Volatility, however, goes
through
> cycles and volume just keeps growing.
>
> The question that I have involves volume filtering. To me, it is
essential
> that volume filters be applied to actual volume and not
backadjusted volume.
> My concern, however, is that if I apply a filter requiring an
average of
> 300,000 shares, I don't get very many hits back in the late 80's
and early
> 90's.
>
> I have a solution in mind and would appreciate some input or
dialogue on the
> subject. It seems to me that volume filtering should be based
on some
> percentage of the total volume of all NYSE stocks (for
instance). I
> haven't done my homework yet, but let's say that the average
volume today is
> ten times more than it was in 1985. If I decide to filter today
at 300,000
> shares, wouldn't it make sense to filter based on 30,000 shares in
1985. I
> can probably answer that question myself by saying that I don't
think 30,000
> would be an adequate filter in 1985. But I could scale it from
100,000 to
> 300,000 progressively between 1985 and 2003 based on mathematical
equation.
>
> You may ask why backtesting to 1985 (or any other date) is
important.
> There are dozens of reasons, but the most important reason to me
is that
> prospective investors in any funds that I manage want to see how a
proposed
> system would have performed over a statistically meaningful period
of time.
> You can argue about the relevance of such information, but THEY
EXPECT TO
> SEE IT. For the record, I also think that it is very important.
>
> I welcome comments from anyone with an interest or knowledge in
this area.
------------------------ Yahoo! Groups Sponsor ---------------------~-->
Get A Free Psychic Reading! Your Online Answer To Life's Important Questions.
http://us.click.yahoo.com/Lj3uPC/Me7FAA/ySSFAA/GHeqlB/TM
---------------------------------------------------------------------~->
Send BUG REPORTS to bugs@xxxxxxxxxxxxx
Send SUGGESTIONS to suggest@xxxxxxxxxxxxx
-----------------------------------------
Post AmiQuote-related messages ONLY to: amiquote@xxxxxxxxxxxxxxx
(Web page: http://groups.yahoo.com/group/amiquote/messages/)
--------------------------------------------
Check group FAQ at: http://groups.yahoo.com/group/amibroker/files/groupfaq.html
Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
|