non-linear regression, Cubic Spline and Least S

[cbahia]

thank you Arbuthnot......and what about non-linear regression and Least Squares Fitting ?

[arbuthnot]

thank you Arbuthnot......and what about non-linear regression and Least Squares Fitting ?

Carlos - just to say very quickly that:

non-linear regression = quadratic, cubic, etc., regression.

I didn't state this explicitly. Sorry.

On the other hand:

least squares fit = linear regression

With this, end points of the set are as valid as any other. This method has great value statistically but I'm not sure how valuable in trading.

A huge problem, as with all indicators, is to determine over what period you're working.

I really urge you, Carlos, to look at these things (just google them) to get an idea of what you're dealing with.

More next time!

[koganam]

...
A huge problem, as with all indicators, is to determine over what period you're working.

What period? Use the period of the dominant cycle. See how by clicking on my signature.

[arbuthnot]

thank you Arbuthnot......and what about non-linear regression and Least Squares Fitting ?

Hi again Carlos

Hopefully you've understood that it's just about essential to have a good idea of what's behind the indicators (i.e. the math) you're interested in.

As final 'proof' that non-linear regression can't help you, please see the attached image from Wolfram Alpha, a wonderful online math (and information) site.

http://www.wolframalpha.com/

At this site, you can enter data points for all types of regression: they call it 'fit'. So linear regression = linear fit.

The image is of a quadratic fit of data based on a real Forex sample.

It simply won't work - because the end points of the fit wizz off towards plus or minus infinity!

[arbuthnot]

What period? Use the period of the dominant cycle. See how by clicking on my signature.

Hi again Koganam.

Thanks for your posts #12 & 18. It's always fascinating and motivating to read what you have to say.

Price Action (PA): I agree with you totally the term is highly nebulous - and an incorrect understanding of what PA is or meant to be will 'cloud' your judgement. (Sorry for the appalling pun. It's my English sense of non-humor...)

Your 'candle' illustration is exactly how traders can misunderstand PA.

The LineAlert indicator (which it isn't - it's basically a movable trendline you can trade off) for Ninja is a wonderful example of how effective PA ideas can be - and it's so simple: especially, by the way, with Renko bars, so I've found.

All indicators lag, they must do, because they all in some way average out previous price bars - but some lag much less than others. There must be some theoretical and maybe actual indicator which has the minimum lag over the universe of all possible indicators for any given period.

I've had a good read of your site and what you say about the period of the dominant cycle is extremely interesting and I don't doubt that your approach is highly effective.

[koganam]

... All indicators lag, they must do, because they all in some way average out previous price bars - but some lag much less than others. There must be some theoretical and maybe actual indicator which has the minimum lag over the universe of all possible indicators for any given period. ...

"All indicators" as used here, may make that too definitive a statement. While true of just about any indicator that uses an average, (after all, you must gather the data before you can take an average, and that gathering takes time), it is not necessarily true of indicators which are differencing, or calculating using the present bar relative to other data. Such indicators are often co-incident.

Many people assert that differencing indicators can even be leading. I dispute the existence of leading indicators (and am willing to be taught otherwise), for the very reasons that I state here, as on our website: "As an example, look at any momentum indicator, and you will see that momentum wanes before the market turns. Does this provide predictive power? Of course not. How many times have you taken a position based on momentum reversal, only to have the market rip back in your face? The key to understanding why momentum wanes before a turn, is to realize that momentum measures the speed at which the market is moving. Just as with anything, the market usually slows down before it turns. Hence you see the reversal of momentum first. ..."

Talking about co-incident indicators, take for example any indicator that shows candlestick patterns. It will always be a co-incident indicator. The pattern is either complete, or it is not. If it is complete, it is marked at completion. e.g., a hammer is a hammer. It will be marked as soon as it is complete.

The other class of often co-incident indicators, are those that are measured by using the current candle as the measure against anything. The Stochastics class of indicators are such. They are based on the current candle's relation to the overall range of a previous number of candles. This is especially true, if such an indicator is tied to the dominant cycle at all times. That means of course that the indicator is not using a fixed period, but a period that is continuously adjusted to take advantage of the current instantaneous cycle length.

As an example, I submit the first picture below. How many of the market swing points coincide with the extreme of the indicators. Deliberately, the first 2 are actually the same indicator, one tuned to the half-cycle, the other to the quarter-cycle, both of which, by analogy to a pure sine wave should mark a turn. (Remember that we are adjusting the cycle at all times to be the same period as the dominant wave, which by definition is a sine wave). The third indicator, is actually set to the full-wave, again to remove the idea of selection bias, and is an attempt to do two things: bound the Fisher Transform, turning into a bounded-oscillator, and hence answering the key question, "how far is high or low value"; adjust that oscillator to use the dominant cycle as its period. Notice how, because the Fisher Transform is an averaging indicator, it does show a lag over the Stochastics indicator, even when adjusted for cycle duration.

The second picture is actually the same, just with a change to a minute based period, as many, including me, are suspect of things on a Renko style chart (because of its fudged opening). Truth to tell, overlapping Renko bars, such as UniRenko, or what I am using here, do ameliorate the issue somewhat.

[cbahia]

Nice indicator.
Is this available to share ?

thank you

[cbahia]

Koganam,

You have a nice indicator.
Is this available to share ?

thank you

[koganam]

Koganam,

You have a nice indicator.
Is this available to share ?

thank you

I am not sure what you mean by "share" in this context. If you want a 14-day Trial of any indicator on our site, send me a PM here, or message from the site, using one of the provided forms. Specify what indicator you want to try out.