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

Re: [EquisMetaStock Group] Weekly MACD histogram



PureBytes Links

Trading Reference Links

Roy,

As always words of wisdom. Appears to be a good chapter for the book.

Preston

--- In equismetastock@xxxxxxxxxxxxxxx, "Roy Larsen" <rlarsen@xxxx> 
wrote:
> Hi Alan
> 
> > Thanks for your excellent advice. As a newbie, it's difficult for 
me
> > to know what to believe and what not to believe on the internet.
> > That's why I wish I could analyse any formula that I find - such 
as
> > the weekly MACD in my post.
> 
> > I found the weekly MACD histogram formula at:
> > 
http://www.paritech.com/education/technical/custom/indicators/macd.asp
> >
> > I have already posted your reply on their Metastock forum, so
> > hopefully they will remove their formula from their site, or at 
least
> > try to justify it.
> 
> That will be interesting. I don't really want to get into an 
argument with Paritech but when I
> plotted the formulas you posted the weekly one was so far off it 
was a joke
> 
> > There's a fiendishly complicated formula for weekly MACD at:
> > www.users.bigpond.com/prominex/MetaStock/Roy-63.txt
> 
> This is my formula :-) , hence the name "Roy"
> Yes it does look feindishly complicated but quite a bit of the 
formula has to do with making
> allowances for things that can be wrong with the data so that it 
will exactly emulate the MetaStock
> MACD values as plotted on a weekly chart. Try plotting my formula 
on a weekly chart and then plot
> the MS drop-down indicator over the top of it. There will possibly 
be some differences at the left
> side of the chart but values should agree to  4 decimal places at 
the right unless you only have a
> small amount of history.
> 
> My weekly formulas attempt to create exact weekly data from the 
daily chart and are very accurate in
> doing so. Most attempts at weekly data emulation are done on 
a "near enough is good enough" basis
> but in my view that's not good enough. When you have thinly traded 
issues, suspensions,
> reconstructions, Easter or Christmas breaks, and various other 
situations that cause gaps in the
> data, great care has to be taken to ensure that the right days are 
included in the right weeks.
> 
> I could create simpler weekly indicators but for some issues they 
would be less accurate. Unless
> they are accurate 100% of the time how do you know when they will 
be off? The thing with a weekly
> formula (for daily charts) is that it must plot the same values 
that you would see for the standard
> MS formula on a weekly periodicy chart. That's the only standard it 
can be judged by as far as I'm
> concerned.
> 
> I can take you through a complete weekly formula and explain every 
detail if you wish. It would take
> you a while to absorb but you would learn quite a lot.
> 
> > I suspect that the Paritech formula is trying to calculate MACD 
for
> > Friday's closing prices only. I verified this by plotting all the
> > individual bits and pieces as separate indicators, because I'm too
> > inexperienced to understand the formula.  (If you know anyone who
> > understands the bit between the {start} and {end}, could you ask 
them
> > to explain it.  It doesn't matter whether they agree with it or 
not.
> > At least I'll learn something.)
> 
> I could do this but my thinking is your time and mine would be 
better spent on studying something
> that has underlying principles that give accurate results. I didn't 
attempt to study the Paritech
> code (I may have if I'd known where it came from) but I plotted it 
alongside code I knew to be
> accurate. It was so far off I didn't see any reason to investigate 
further.
> 
> > I suppose it makes some sense:  the
> > daily MACD formula is based on daily closing prices, so the weekly
> > MACD formula is based on Friday's closing prices.  Presumably, if
> > there were a monthly MACD histogram, it would be based on monthly
> > closing prices.
> 
> Exactly right. I could knock up a monthly MACD and it would only 
use the CLOSE for each month, just
> as MetaStock would on a monthly periodicy chart. The plot that 
doesn't change for several bars may
> look a little odd but when you think about it, it can't be any 
other way if it is a true
> representation. A weekly MACD doesn't change value on Wenesday then 
again on Friday.
> 
> > I agree with your comment that the Paritech daily MACD 
is "slightly
> > off".  I found this description of the daily MACD histogram in a 
book
> > by someone called Dr. Alexander Elder (why do all conmen and 
shysters
> > call themselves "Dr"):
> 
> I don't know that I'd call Dr Elder a conman. He is widely 
respected in the trading industry and is
> the author of a couple of "must have" books.
> 
> The reason why the Paritech MACD is slightly off is because they 
are using the convventional 12/26
> "periods" to create the two exponential moving averages. This is 
widely accepted as normal so in one
> sense these are "more" accurate than the MS version. You need to 
understand how an EMA is
> constructed to understand why these differences show up.
> 
> An EMA adds a proportion of new data for evey bar, and it also 
retains a portion of old data
> (previous value). If the proportion of new data added is 20% then 
the proportion of old data
> retained is 80%. Can you see that? A standard EMA calculates the 
amount of new data to add like
> this - 2/(1+periods). The actual construction of an EMA follows so 
that you can check the code for
> yourself.
> 
>   {Exponential Moving Average}
> n:=Input("Periods",1,999,10);
> R:=2/(n+1); {ratio of new data added each bar}
> M:=If(Cum(1)=1,C,PREV*(1-R)+C*R);
> M;
> 
> So the factor determining how much new data to add and how much old 
data to retain for a 12 period
> and 26 period EMA is 2/13 and 2/27 respectively. In decimal terms 
this is 0.153846 and 0.074074.
> Metastock actually uses values of 0.15 and 0.075 rather than basing 
on 12 and 26 periods, and that's
> why the MetaStock MACD is slightly of the Paritech code.
> 
> > Page 129:
> > 1.  Calculate the 12-day and 26-day exponential moving averages of
> > closing prices.
> > 2.  Subtract the 26-day EMA from the 12-day EMA to obtain the fast
> > MACD line.
> > 3.  Calculate a 9-day EMA of the fast MACD line to obtain the slow
> > Signal line. Plot both lines to obtain the classic MACD indicator.
> > 4.  Subtract the Signal line from the MACD line to obtain an MACD-
> > Histogram.
> 
> This is the accepted norm but I think may be only only a close 
approximation of what the creator
> intended. I honestly don't know which is the "authentic" version. 
At the end of the day they're only
> numbers that we interpret how we want. Anyway I have no argument 
with the there being two different
> versions because I understand the slightly different values that 
each version is based on.
> 
> > The "MetaStock canned version", as you call it, is bound to be
> > superior, because at least I've heard of Equis :-), but I've never
> > heard of Dr. Alexander Elder.
> 
> You have now :-)  Don't write him off. There are many "experts" out 
there but just take what you can
> use and leave the rest. My comment in my earlier reply "If this 
works for you then great, but a
> weekly MACD it is not." applies across the board. If anything is 
suspect as to name or accuracy but
> still works for you then don't throw it away, just make sure you 
understand the differences. At the
> end of the day you have to use what works for you, not what some 
guru tells you works for everyone
> else.
> 
> > Thanks for your help.
> 
> No problem. Remember that a fair bit of the above comments are my 
opinions only, and may not be
> shared by others.
> 
> I meant what I said about being willing to walk you through any of 
my code if you want to learn.
> 
> Good luck.
> 
> Regards
> 
> Roy



 
Yahoo! Groups Links

<*> To visit your group on the web, go to:
     http://groups.yahoo.com/group/equismetastock/

<*> To unsubscribe from this group, send an email to:
     equismetastock-unsubscribe@xxxxxxxxxxxxxxx

<*> Your use of Yahoo! Groups is subject to:
     http://docs.yahoo.com/info/terms/