|
PureBytes Links
Trading Reference Links
|
Dimitri,
What you describe is not what I meant.. The rsi formula
rsi = 100-(100/(1+(up/down)).
is the fundamental problem.
As up tends to zero, up/down tends to zero and rsi tends to 0
as down tends to zero, up/down tends to infinity and rsi tends to 1.
The characteristic of these reciprocal type curves is long, unresponsive tails that lead to the rsi being unresponsive as the
numerator or denominator tends to zero. All the sensitivity is at the 50% mark but you are trying to make decisions in the rapidly
declining sensitivty region.
I suggest the nicety of limit bands be discarded entirely and the rsi be structered as something else entirely, so that:
it has no limits,
crosses zero
and can be equally positive and negative.
For example, and I make this up in my head without checking first,
modrsi = up(last 4)/down (last15) - down(last4)/up(last15)
or closer to the original which I think will tone it down some. (if not reverse the order - put the last14/last3)
modrsi = 100-(100/(1+(up/down)(last3)/(up/down)(last14))).
I am suggesting to be radical and not simply rearrage or blend outcomes from various established indicators, but break them open
completely an re-cast their principles.
P
----- Original Message -----
From: "DIMITRIS TSOKAKIS" <TSOKAKIS@xxxx>
To: <amibroker@xxxxxxxxxxxxxxx>
Sent: Wednesday, October 10, 2001 3:55 PM
Subject: [amibroker] Re: Modified RSI
> I have always the alternative of Normalization, exposed here in some
> earlier post, ie
> NormRSI=100*(RSI-RSImin)//RSImax-RSImin)
> which fit better in [0,100] band.
> I think that this new idea of Modified RSI is a better approach.
> I will revert after more tests.
> Thank you for hint.
> Dimitris Tsokakis
> --- In amibroker@xxxx, "server not recognized" <winchp@xxxx> wrote:
> > Dimitri,
> >
> > Constance Brown in her book "Technical Analysis for the Trading
> Professional", advocates that the 30/70 is not appropriate as
> > constants, and should be modified depending upon recent history.
> For example after an RSI > 70 period then a decline to 40 is
> > significant and vice versa, i.e 30/70 shpould be applied with
> judgement.
> >
> > Secondly the non zeroing of RSI and its lack of sensitivity beyond
> the 70/30 is a fault with its ideal of normalising. Constances
> > suggested opened the door to eve better understanding for me. I
> would suggest that rather than using the RSI() supplied that you
> > play around with actual formula and try and achieve the following:
> >
> > no upper/lower limit,
> > able to cross zero
> > can have negative numbers
> >
> > I have done this with ADX with some satisfaction. I would also
> suggest that additonal terms can be included inside the RSI formula
> > to increase sensivity to certain events when they are present. I
> would also suggest that where ever possible ema and ma not be used
> > except when trying to establish long term baselines. You may not
> like the jerkiness without them, but the peak values take on
> > significance in magnitude and timing after you use the
> reconstructed formulas that ema and ma do their best to blurr.
> >
> > Both RSI() and ADX() in my view are horses of different colours
> because both are associated with volatility breakout. The results of
> > one are similar to the other.
> >
> > I hope this helps with your search.
> >
> > P
>
>
>
>
>
> Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
>
>
>
|