|
PureBytes Links
Trading Reference Links
|
Herve:
I don't think that abs() and Prec() will get around the problem. You are
still dealing with only a few significant digits. One way might be to export
the data to Excel or another spreadsheet and do the calculations in the
spreadsheet.
Subtracting closely similar size numbers is unreliable. Whenever this occurs
we should try to find a different computation route so that we avoid this
problem (not always easy or feasible).
Metastock has a number of built in limitations that should be corrected.
Lionel Issen
lissen@xxxxxxxxxxxxxx
----- Original Message -----
From: "Herve" <genher@xxxxxxxxxxxxx>
To: <metastock@xxxxxxxxxxxxx>
Sent: Sunday, October 07, 2001 10:46 AM
Subject: RE: Calculation error in MS
> Very interesting posts!...need time to digest the lot. The idea of using
> function like abs(), prec() etc... sounds like a good one to go around
that
> problem.
>
> -----Original Message-----
> From: owner-metastock@xxxxxxxxxxxxx
> [mailto:owner-metastock@xxxxxxxxxxxxx]On Behalf Of Lance Collins
> Sent: Monday, 1 October 2001 8:50 PM
> To: metastock@xxxxxxxxxxxxx
> Subject: RE: Calculation error in MS
>
>
> Allen wrote:
> >An issue that also affects the database is that metastock stores data
> >in the old microsoft basic 32 bit precission format. Upon loading that
> >format get's converted to IEEE format single precission. We're probably
> >loosing bits just in this conversion. That is then compounded by
> >calculations being done in single precission.
> The same number of bits are used for exponent and mantissa in both IEEE
and
> the old MS format so there is no loss of precision in the conversion.
> There are 23 bits for the mantissa (approximately 6 significant decimal
> digits) which is more than enough to hold raw data values of price or
index
> in the data files.
>
> It is not easy to do extended calculations which avoid issues such as the
> problem of subtraction of similar size quantities. These things can
> produce nonsense results regardless of the precision available. Floating
> point arithmetic also makes the use of tests for equality problematic.
>
> The issue is not the use of single precision but the limitations of the
> formula language provided with Metastock. Some of you seem to be pushing
> the explorer way beyond it's capabilities. If Metastock used double
> precision it would merely mean that it took a little longer to get into
the
> same problems.
>
> What to do when you outgrow the explorer is another debate altogether.
>
> Lance
>
> p.s. I've always thought the financial world to be mesmerized by numbers
> but it's not numerate.
>
>
>
>
>
>
>
|