[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: First calculation problem



PureBytes Links

Trading Reference Links

Vitaly,

Generally I would agree with your post but for one thing.  My calculation
didn't require any rounding.  It didn't require anything other than the
ability to subtract one 2 decimal number from another 2 decimal number and
give me an answer that was accurate to 2 decimals.  There should have been
no rounding involved.  I understand Ken's reasoning about single precision
and double precision arithmetic.  My point is that regardless of precision,
etc., MS should be able to perform simple addition and subtraction functions
and return accurate answers.  By accurate I mean answers that coincide with
generally accepted rules of arithmetic.  Now, I don't care about
multiplication and division as I already take care of rounding and precision
myself.  I shouldn't have to do this with addition and subtraction.  Again,
that's just a personal opinion.

Now we have specific, personal reasons for this.  We run a trading system
that has been developed over the last 40 to 50 years and has been run on
computers for 40 years.  We have run this system on mainframes, minis and
microcomputers as well as on timesharing in the days before micros.

We would like to be able to run our system in MS in order to take advantage
of its System Tester capabilities.  With this feature, we would be able to
automate some back testing for various things we'd like to try but are too
labor intensive to do manually.  We currently do all of our testing in
Excel.  In order to do this, I need to be able to duplicate our system's
results in MS.  Currently, MS is generating "noise" sporadically in my
calculations through addition and subtraction errors.

Now you may ask why is this important?  Since we have worked with some of
these calculations for over 40 years, we have developed a "feel" for these
numbers and have built trading rules over time.  Some of these rules require
accurate calculations.   For example, we have a number we call the
"Contrary" that we use to reflect the strength of the day's price activity,
either up or down.  When we receive a buy or sell signal, we use this
"Contrary" as a confirmation as to whether we should make that trade or not.
The number is important insofar as if the "Contrary" is > 35 meaning the
market was pretty strong that day, then we wouldn't sell short the next
morning, but would wait for a "Contrary" in our normal range of >= 12 and =<
35.  Now with this added "noise", our rule set is rendered meaningless.

You might ask why we can't just go back and redo our calculations over the
past 40 years and revise our trading rules?  The main reason we can't do
this is because MS randomly throws the "noise" into the calculations and you
can't allow for this in your rules.  I'll try to relate this to what a lot
of people use (we don't).  Let's take a look at the RSI.  If you use 20-80
as signal points and you have some "noise" floating around the calculations,
then you might have a 79 one day and not sell short while another
calculation might return an 80 and sell short.  It doesn't take much to foul
up a mechanical trading system.  If you can't have confidence in your
calculations, then how can you really trade, regardless of mechanical system
used?

Now if I had been further along in my calculations, I do take full
responsibility for my own rounding, precision, etc. since I am working with
division, moving averages, etc.  That code is already built into our
calculations.  The only place that I didn't build this type of protection in
was in my initial, simple addition and subtraction calculations.

Guy

" When I die, I want to go peacefully like my grandfather did, in his sleep.
Not yelling and screaming like the passengers in his car."

-----Original Message-----
From: owner-metastock@xxxxxxxxxxxxx [mailto:owner-metastock@xxxxxxxxxxxxx]On
Behalf Of Vitaly Larichev
Sent: Thursday, September 07, 2000 1:27 PM
To: metastock@xxxxxxxxxxxxx
Subject: Re: First calculation problem

With a risk to bore everybody even more, a comment on Lionel's recent post
...

Lionel Issen wrote:

> ....
> According to Guy's painstaking checks, Metastock cannot do single
precision
> calculations reliably...

It's hard for me to agree. What Guy's checks show is that Metastock produce
different results (don't laugh, I don't play semantics here<g>!). The source
of the
difference might be:

i)  a (stupid) error in coding . Then all that anger is valid;
ii) different algorithms for rounding off. Accumulated being able to spoil a
calculation.

Considering the second possibility, identical results provided by Excel,
Clipper, and
others do not testify to their accuracy, only that they might use the same
round-off
routines. Nothing more! Why Equis is so lonely vs. Excel, Clipper, and
others? Still
a good question, I agree. It would be nice to have an answer to it to make
sure
Metastock users could sleep tight, and Equis could sell even more Metastocks
<g>.

Cheers, Vitaly