|
PureBytes Links
Trading Reference Links
|
You're asking the same question I posed in offering up the "Dinner in San
Francisco Challenge."
I said I'd buy dinner in San Francisco for anyone who could "replicate the
weekly period DJIA 5,3,3
stochastic from daily data..."
I've realized that San Francisco is too far away for most. I'm coming up
with alternative prizes.
The two individuals who have actually expended effort on the task have said
that they can get close but not replicate it exactly. I'll award prizes for
close enough, but I'm in the middle of figuring out what "close enough"
means.
If you have a specific contest entry, I'd be happy to consider it. <G>
Have questions about what all the fuss is about? Look at the S&P500 chart I
sent out yesterday with accompanying descriptive text. I'm arguing that the
same kind of structured relationships you see there between the weekly and
daily stochastic are the RULE for how stochastics work across multiple time
periods regardless of issue, time periods chosen, or market and not the
exception (although there are exceptions). If that's true...
I'm really trying to get this over with quickly though by taking the
conversation offline.
Steven Buss
Walnut Creek, CA
sbuss@xxxxxxxxxxx
-----Original Message-----
From: Rick Roegner <oiis@xxxxxxxxxx>
To: metastock-list@xxxxxxxxxxxxx <metastock-list@xxxxxxxxxxxxx>
Date: Tuesday, January 20, 1998 12:23 PM
Subject: Re: Stochastic brain teaser
>Please forgive me, oh masters.
>
>Couldn't you use something like the zig-zag function to break a price
>series into definite trends (up, sideways, down ?), then find something
>like an EMA to describe & relate to these various trends. Then look for
>the average or mean behavior of say, stochastics during these different
>"discrete" types of trends, then use the EMA (or something else) to signal
>a gradual change in say the stochastic parameters or adjust the SD? Has
>anyone tried this type of backtesting?
>
>The optimizer can be a hinderance and sometimes, unfortunately, you have to
>get out and walk? Or break up backtesting into segments?
>
>I'm not trying to tell you what to do or how to think, likely more or less
>how stupid I am. There has to be something in Metastock that will adjust
>these indicators to reality. A hammer?
>
>I'm over my head here, but aren't we all? Its a lot easier to talk about
>TA, than to do it?
>
>
>
>At 08:11 PM 1/19/98 -0000, you wrote:
>>Let me be a bolder grasshopper,
>>
>>1. I believe that Elder has provided a significant conceptual framework
for
>>thinking about integrating trend following and momentum indicators using
>>multiple periods. In addition to a couple other differences I have with
>>some specifics of his, I don't believe that he written his book with the
>>purpose of specifying that integrated view throughout his book. You'll
>>notice that in the section you reference with a page number, he speaks
only
>>of a single time-frame. It's as if his discussion of "Triple Screen" is
>>inserted as an afterthought...
>>
>>2. I don't think you've made the case for statistically adjusting the
>>stochastic and williams%r. These indicators are statistically adjusted
>>already for data only within the lookback period, i.e., 0-100 for both.
How
>>the world would you adjust them further.
>>
>>3. I do understand the important point you're making for other momentum
>>oscillators like MACD however and appreciate the tip very much. (Did you
>>mention what kind of wine you liked? <G>)
>>
>>4. A last point about the stochastic: It is called a momentum indicator
or
>>oscillator by many, maybe most. That's right, but it has become one of my
>>most important "trend" indicators. That is, the weekly stochastic depicts
>>the "trend" for prices at lower time frames than itself. The daily
>>stochastic depicts the "trend" for time frames lower than itself, etc.
I'm
>>making a general point that has important caveats related to matching up
>>timeframes (along the lines of elder's "factor of five" but not the same
>>thing), but if you doubt that this general point is true, look at the
weekly
>>period stochastic for 4/97 through 7/97 and the way that it indicated the
>>"trend" for daily price bars. Then note that when the weekly stochastic
>>broke down late July early August, the daily price bar pattern also
changed.
>>I believe that an exhaustive, empirical analysis of this kind of
>>relationship among multiple stochastics based on different time periods
>>would show that this is the RULE of how they can be analyzed together for
>>profit and not the exception...I am NOT, of course, advocating that one
>>focus on the stochastic alone.
>>
>>5. You write that "it's only natural they confirm each other, even across
>>multiple time frames". My view of what the data actually shows is that,
>>most of the time, stochastics based on multiple time frames do NOT confirm
>>each other. They are in different phases of their run up and down most of
>>the time. I've looked at lots of data intra-day and my view is that the
>>biggest and fastest runs up or down begin when the daily, 60, 30, 10, 5,
and
>>1 minute stochastic are all in the same place and begin to move. There
are
>>structured relationships among the stochastics for various periods that
can
>>be recognized and analyzed. I've begun my list and descriptions of these.
>>(e.g., I think now about hesitation patterns mostly in terms of direction
>>divergences among the various appropriate period multi-period stochastics:
>>For certain values and slopes of the 10, 30, and 60 minute stochastic,
>>ROC(5,3), and MACD(9,18,4) it is relatively simple to predict whether a 5
>>minute stochastic breakout will be "significant" with a degree of
certainly
>>that is surprising. This is not the same thing as being able to trade it
>>successfully. <G>)
>>
>>Steven Buss
>>Walnut Creek, CA
>>sbuss@xxxxxxxxxxx
>>
>>-----Original Message-----
>>From: Rick Mortellra <rmjapan@xxxxxxxxxxxxx>
>>To: Steven Buss <sbuss@xxxxxxxxxxx>
>>Cc: MetaStock List <metastock-list@xxxxxxxxxxxxx>
>>Date: Tuesday, January 20, 1998 2:54 AM
>>Subject: Re: Stochastic brain teaser
>>
>>
>>>Ahhhh Grasshopper,
>>>
>>>It's not surprising that you see such good things in your charts.
Afterall,
>>>markets are fractally symmetrical. A chart generally appears the same
>>across
>>>all time frames. All your indicators reflect price momentum in one way or
>>>another, so it's only natural they confirm each other, even across
multiple
>>>time frames.
>>>
>>>To understand what I mean, have someone make several charts of the same
>>>security using different time frames. Print them out and have them cut
off
>>>the X-axis showing time period. Bet you'll have a hard time figuring out
>>>which is which.
>>>
>>>It's these repeating market rhythms that make Elliott Wave and Fibs so
>>>enticing. My main problem with EW is that I CAN'T SEE THE DAMN WAVES!
>>>Doesn't sit well with my mechanical system ideas either.
>>>
>>>Fibs on the otherhand, seem to work quite well, especially in determining
>>>various lookback lengths. Haven't really come across any convincing
>>argument
>>>as to why they do, but they do.
>>>
>>>As for the stochastic brain teaser, I'll point you to Elder p.142 last
>>>paragraph. Implies the stoch is impossible to test historically over any
>>>significant length of time. Good indicator in the present but not for
>>>building a robust system on. His 5% comment provides a good clue though
as
>>>to what is needed to make it more robust.
>>>
>>>FYI, a good public primer on the weaknesses of traditional momentum
>>>indicators is Tushar Chande's work with the CMO, stochRSI, and DMI.
>>However,
>>>he still doesn't give you a mathematically precise way of determining
where
>>>to place OVERBOUGHT and OVERSOLD cutoffs.
>>>
>>>That's why you need to substitute a STATISTICAL measure of trend, so
>>cutoffs
>>>can be placed 2 standard deviations (5%) away from the mean on a
historical
>>>basis. To this end, the Random Walk Index seems well suited, but details
>>>about it's actual construction are a little hazy as it's author never
>>really
>>>spells it out. Metastock is even vague.
>>>
>>>My quant friends here say they've solved it, but won't share the
solution.
>>>Secrets like these is what gets them the 7 figure incomes, I guess ;-).
>>>
>>>As for how to tell what's tradable and what's not, generally the upside
>>>should be at least 3 times the downside. You can use average true range,
>>>price channels, or minimum favorable excursion to calculate this, for
>>>example.
>>>
>>>All this posting is making me hungry and thirsty,
>>>Rick
>>>Tokyo, Japan
>>>
>>>-----Original Message-----
>>>From: Steven Buss <sbuss@xxxxxxxxxxx>
>>>To: Rick Mortellra <rmjapan@xxxxxxxxxxxxx>; MetaStock List
>>><metastock-list@xxxxxxxxxxxxx>
>>>Date: Tuesday, January 20, 1998 9:07 AM
>>>Subject: Re: Stochastic brain teaser
>>>
>>>
>>>>Speaking of the stochastic specifically:
>>>>
>>>>A few weeks ago I asked about the same issue in a different manner. I
>>>asked
>>>>about how one could distinguish (or test) between stochastic
>>>>breakouts/breakdowns that were tradeable vs. untradeable for profit.
The
>>>>only answer I remember getting back was that price itself was the
>>>>confirmation. This makes sense and can be useful I think.
>>>>
>>>>The only problem is that sometimes you get a stochastic
breakout/breakdown
>>>>with an accompanying movement in price only to find later that it
doesn't
>>>>get you very far. (I think this is the point you're making Rick.)
That's
>>>>why I, and I believe many others, have found the stochastic to be
alluring
>>>>yet in the end disappointing.
>>>>
>>>>But if you look carefully at the S&P500 chart I sent out earlier today,
>>>>you'll see that there is another potential confirmation indicator for a
>>>>stochastic breakout/breakdown in the topmost indicator section shown in
>>the
>>>>2nd indicator section. Follow each stochastic breakout/breakdown in the
>>>>topmost indicator section, compare to price value and direction, and
then
>>>to
>>>>the stochastic value and direction in the 2nd indicator section. If I'm
>>>not
>>>>mistaken, the stochastic value and direction shown in indicator section
2
>>>>would have been an incredibly accurate indicator for assessing the
>>>"quality"
>>>>of the stochastic breakout/breakdown in indicator section 1.
>>>>
>>>>I could have sent hundreds of charts of varying multi-periods that
>>>>illustrate the same thing. If I seem to have been foaming at the mouth
>>>>about this, it's because I have found hundreds of examples of this. I'm
>>>NOT
>>>>saying it provides 100% confirmation. I am saying that the kinds of
>>>>relationships you see in that graphic are the standard ways these
>>>>relationships work not the exceptional ways.
>>>>
>>>>So, Rick, I disagree that testing the stochastic specifically over
>>>>historical data is problematic because of what we might call price/value
>>>>differences among lookback periods for the same kind of oscillator
>>>>breakout/breakdown signal. The problem is that we haven't had the
>>>"concept"
>>>>or the software to test indicators at multiple periods within the same
>>>test.
>>>>
>>>>Steven Buss
>>>>Walnut Creek, CA
>>>>sbuss@xxxxxxxxxxx
>>>
>>>
>>
>>
>>
>R. Roegner
>
|