Ken,
You are more than welcome! The RSI is one of my favorite indicators.
It took us a bit to realize the differences between the two smoothing
methods. I have to laugh everytime I read that someone is using
shorter lookback periods with success. Seems Wilder knew that already.
While I have your attention and since you mentioned recussion, let me
ask for your help. A few days ago we were discussing the Rainbow/RMO
indicator. The method used to smooth the indicator was a recursive
smoothing. Normally this wouldn't be so tedious but in the case of
the Rainbow/RMO it is a moving average of a moving average..... ..and
on. We can code the indicator but not in a way that allows for any
lookback changes without reproducing the complete indicator,
basically its hardwired. Just wondering if you might have some
thoughts on ways to simplify the indicator.
Preston
--- In equismetastock@ yahoogroups. com, "antoinesax" <kennethmetz@ ...>
wrote:
>
> Thanks, Preston.
>
> I had just come to a similar conclusion.
>
> After reviewing Wilder's original publication, I recognized that he
> used a form of exponential smoothing with factors 1/N and 1 -
(1/N),
> where N is the RSI period. In contrast, the usual EMA uses factors
2/
> (M + 1) and 1 - 2/(M+1), where M is the EMA period. Equating the
> corresponding factors produces the results that M = 2*N -1 and N =
(M
> + 1)/2.
>
> Using Mov(C, M, E) instead of Mov(C, N, E) in my original code
> renders identical results to MS's RSI.
>
> Ken
>
>
> --- In equismetastock@ yahoogroups. com, pumrysh <no_reply@> wrote:
> >
> > Kenneth,
> >
> > I can verify that Wilders used Wilders smoothing when he wrote
the
> > RSI. Wilders is a variant form of exponential smoothing.
> >
> > The difference is in the lookback periods.
> >
> > Here's a quicky formula to confirm this. Not precise but very
close:
> >
> > N:=Input("Periods" ,1,250,25) ;
> > R:=(N/2)+1;
> > Mov(CLOSE,N, E);
> > Wilders(CLOSE, R);
> >
> > Notice that the Wilders formula uses a different lookback period
> that
> > is calculated in the "R" array…basically half plus one. If you
> change
> > the lookback period you will notice a larger variance at shorter
> > periods. It is due to the initial seeding at the beginning of the
> > data.
> >
> >
> > Roy Larsen has discussed this many times before and again a
simple
> > search of our archived messages will yield a number of emails to
> this
> > very point. His website is :
> >
> > www.metastocktips. co.nz
> >
> > There you will find a number of indicators that he has written.
His
> > link along with others that will help you can be found in our
links
> > section. A RSI written by Roy is listed below.
> >
> >
> > {RSI Indicator}
> > N:=Input("Periods" ,2,99,10) ;
> > A:=Close;
> > B:=Ref(A,-1) ;
> > U:=Wilders(If( A>B,A-B,0) ,N);
> > D:=Wilders(If( A<B,B-A,0) ,N);
> > 100-(100/(1+ (U/D)));
> >
> >
> > Hope this helps,
> >
> >
> > Preston
> >
> >
> >
> >
> > --- In equismetastock@ yahoogroups. com, "antoinesax"
<kennethmetz@ >
> > wrote:
> > >
> > > Preston, Thanks for your reply.
> > >
> > > I'm not sure which method Wilder used. But it seems that MS
> > actually
> > > does use EMA smoothing in computing the RSI. I state this for
the
> > > following reason.
> > >
> > > When I directly compare my spreadsheet calculations of RSI
(where
> > > EMAs are used but are computed recursively) with MS's results
for
> > SPY
> > > from 1993 through the present (using a 14 bar period), the
> greatest
> > > percentage error is 0.0003%. On that basis, it most definitely
> > > appears that MS also uses EMAs in its internal calculation.
> > >
> > > I'm wondering if anyone among the MS development community is
> > > participating in this group who can definitively state which
> method
> > > is used?
> > >
> > > KM
> > >
> > >
> > > --- In equismetastock@ yahoogroups. com, pumrysh <no_reply@>
wrote:
> > > >
> > > > Kenneth,
> > > >
> > > > The major flaw that I see with your code is the method of
> > smoothing
> > > > which should be Wilder's not Exponential. There are literally
> > > dozens
> > > > of ways to write the code. You can find some here:
> > > > http://trader. online.pl/ MSZ/e-0-tytulowa -r.html
> > > > You may also search our archived messages and find a few that
> are
> > > not
> > > > there and may work even better.
> > > >
> > > > Preston
> > > >
> > > >
> > > >
> > > > --- In equismetastock@ yahoogroups. com, "antoinesax"
> > <kennethmetz@ >
> > > > wrote:
> > > > >
> > > > > Prior to exploring some variations for RSI, I thought it
best
> > to
> > > > first
> > > > > verify my own coding for this indicator in comparison with
> > > > MetaStock's
> > > > > calculation.
> > > > >
> > > > > Much to my surprise, the results differed widely when
applied
> > to
> > > > SPY
> > > > > (from first data in 1993 through present date). Although
> > > generally
> > > > > zigging and zagging in the same direction, the two differ
> > > > > significantly with ratio (my own code vs MS) varying from
> > around
> > > > 0.5
> > > > > to 1.3.
> > > > >
> > > > > On the other hand, a spreadsheet version nearly matches MS,
> > > > suggestion
> > > > > a problem in the MS coding. The only difference in the two
> > > versions
> > > > is
> > > > > that I used recursive formulas in the spreadsheet to
> implement
> > > the
> > > > > EMA, whereas I simply used the Mov(C, N, E) function in MS.
> > > > >
> > > > > Can someone point out the error in the code shown below?
> > > > >
> > > > >
> > > > > Thanks,
> > > > >
> > > > > KM
> > > > >
> > > > > PS. Please note that the code uses the most logical
> definition
> > of
> > > > RSI,
> > > > > which is mathematcially equivalent to the standard version
> > using
> > > RS
> > > > > that was (maybe) computationally simpler three decades ago
> when
> > > > this
> > > > > indicator was invented...
> > > > >
> > > > >
> > > > > ============ ========= ========= ========= ====
> > > > >
> > > > > Npds := Input("Periods" ,1,5000,14) ;
> > > > > Chng := C - Ref(C,-1);
> > > > > Adv := If(Chng>0,Chng, 0);
> > > > > Dec := If(Chng<0,-Chng, 0);
> > > > > AvgAdv := Mov(Adv,Npds, E);
> > > > > AvgDec := Mov(Dec,Npds, E);
> > > > > RSItest := 100*AvgAdv/( AvgAdv+AvgDec) ;
> > > > >
> > > > >
> > > > > RSItest
> > > > >
> > > >
> > >
> >
>