|
PureBytes Links
Trading Reference Links
|
I certainly would not try to tell you not to use whatever works for you but
I can tell you some of the reasons why I rarely use a simple moving average.
I just applied a 10 bar simple moving average to the daily chart of Cisco,
The average turned down on the day of the recent crash, 10/27, then turned
sharply up 10 days later just as the price was dropping again. It did this
because the value of the price on the crash day ten days earlier happened
to drop out of the average at that point. A trading system based upon that
average would have said "buy" when clearly we should have been selling.
This happens because the simple moving average weights all days equally so
the effect of a day long gone can affect the average quite significantly.
The more sophisticated averages do not have this characteristic.
In addition, a simple moving average has no response to periodic signals
whose period is equal to the length of the average. (This is because the
new value added will be equal to the old value that drops off.) So for the
10 day average mentioned above, there would be no response to a periodic
signal component with a 10 bar period. Lots of stocks, including Cisco tend
to show a cyclic component at about that period so using that average would
tend to eliminate that component of price. This is OK if this is what you
want to do but would be bad if you wanted to trade this cycle.
Then there is the lag. A moving average of that type has a lag equal to
half of the length-plus-one. This is much longer than that of the more
sophisticated filters for comparable smoothing. Most people are looking for
good smoothing with as little lag as possible in order to act on some
characteristic change as soon as possible.
I use a simple average sometimes when I really want an average, such as:
Average(Range,5)
and sometimes for very long averages such as 50 day or 200 day averages
since a lot of people refer to these averages. In fact, I have seen trading
systems that try to take advantage of the behavior of a stock when its
price crosses one of these averages because so many people use these points
as triggers.
The science of designing digital filters (which is what moving averages
are) has advanced quite significantly since the early 70s when digital
signal processing began to be economically feasible.
The T3 filter in the recent article is one of the best I have seen
published. There are, of course, several very good moving average
commercial products that use proprietary algorithms.
I suggest that most traders could benefit for understanding them better.
There are lots of good books and articles in past issues of TASC (buy the
CD-ROM and do a search). Anthony Warren wrote quite a few.
In addition, the web pages of people selling some of the better moving
average products such as Mark Jurik (http://www.jurikres.com/) also contain
lots of good information.
One of the best books I have is "Digital Filter Design" by Parks & Burrus
(ISBN 0-471-82896-3).
I hope this has been helpful.
Bob Fulks
>On Sun, 14 Dec 1997 10:59:12 Bob Fulks wrote:
>snip<<
>I just received the January issue of Technical Analysis of Stocks &
>Commodities and noticed the article "Smoothing Techniques for More Accurate
>Signals", by Tim Tillson.
>>>snip
>
>Bob;
>Thank you for the work that you did on coding the T3 Averages from S&C. I
>offer the following comments with regard to the efficiency of this
>particular indicator.
>
>Some time ago I did quite a bit of work with moving averages (haven't we all
>at one time or another). In my application (S&P 1 minute) I found that the
>application of (High,10), (Low,8) and (Close,3) simple averages was the most
>effective in helping to determine short term market conditions, looking at
>the CloseAverage in relation to the other two averages.
>
>I dredged up my old indicator and compared it to the T3 averages on an S&P 1
>minute chart. I found that the T3 averages from S&C track the simple
>averages within a few percentage points with no noticable improvement in
>response time. Actually, I found that the T3 averages have a slightly
>poorer response time than the simple averages. Given the relatively large
>code for the T3 indicators I wonder where the benefit is.
>
>I have not investigated other markets or other data compressions other than
>taking a tertiary look-see. I have only tested the T3 indicator in my
>particular application and offer these comments as a general discussion.
>
>Regards
>Rod Wheeler
--
Bob Fulks
bfulks@xxxxxxxxxxx
|