Nostalgia to me a year ago reading the forums and seeing wave types and wondering “how will I ever make that?”

Edit:

I’ve decided to add an additional category, FTrend, or fake trend. I defined it to be anytime the third leg does not exceed the first leg, or other wise known as the additional logic of 3>1 to be a Trend, and 1>3 as FTrend.

If there is an assumption that there’s a difference between Trend and Ftrend on future development, the the left section is the correct one to be reading. if they are assumed to be the same, the right one is correct. Either way, it’s really showing the strength of trend following, which honestly is a bit strange to me. It makes sense, but you would think that if it was so clear and so well taught, more people would be successful using it.

Likely scenario:

Next, I want to take a dig into the Trend Spotting thread that was linked in the Similarity thread. Interesting that Mr. Patrick was active in that thread when it was created in 2010, would love to know some of your thoughts on it if you played around with it (:

### Like this:

Like Loading...

*Related*

Patrick M. WhiteHi LG,

I’ve been meaning to get around to doing this sort of study in a more meaningful and structured way. I’ve previously run stats for 1-3 prior swings and intended to get back to application of this idea, but I didn’t have this structure worked out when I ran the initial tests. I really like this structure and your pictures make it easy to see how it all fits together. Thanks for making it easier for me to conceptualize and test! This should help to make turning point predictions more accurate as per the Ultimate Truth and its offshoot thread.

By the way, I presume this study was done with price ranges? Have you looked at computing the stats for bar counts using this same structure? If you’re familiar with the Ultimate Truth thread you can probably see where I’m going with this.

SCPatrick,

Talking about the offshoot thread, are you hinting at constructing fractal patterns in a much different way? Every error analysis points to issues with fractal patterns, meaning they are too short or long or just absent in a specific point in time, in addition the near-perfect IS-performance magnifies this tendency even more. GG seems to get away with using the shorter ones. Essentially if the algo smooths the length according to ATR, then it should “smooth” the fractal pattern as well. I know, this just my theory, but it should improve accuracy.

Patrick M. WhiteHi SC,

I’ll try to make this brief and understandable, though I could probably write dozens of pages on this subject (and have). I hadn’t thought about this in terms of fractal patterns. Going back to the original algos, there are basically 3 logic rules. Each of the 3 could be thought of as a single probability calculation. If you look at the stats presented by LG in this blog under the 1 leg section it could be simplified to say that there are basically 2 probabilities that favor longer next zz leg range, and 2 probabilities that favor shorter next zz range with accuracy between ~61% and ~74%. So an additional probability (4th logic condition?) could be calculated based on zz patterns that might help to override signals that come “too soon”. In like manner the problem of “late or missing signals” could be addressed by this 4th calculation which could be used when the likelihood is high for a smaller next zz leg.

By the way, and for what it is worth, gg53 early on hinted that he was doing something like what I describe above but later said he wasn’t using it.

Regarding the 2nd point/paragraphs in my post above, by calculating an additional set of bar length probabilities (as well as range probabilities, which I’m assuming is what LG had in mind in the calcs shown above) you could have a logic gate 4 and 5 that might help tilt the balance either by speeding up or slowing down the arrow signals when the probabilities align. And maybe it would help improve overall model accuracy for those of us who haven’t successfully mind melded with gg53.

Relativity*giggles* ;d

SCPatrick,

Thank you for the answer! I got you loud and clear. I wouldn’t call it a 4th logic, essentially this is about “enhancing” the second one of the 3 (pattern -> length -> next bar).

Splendid idea, you got me thinking. Thanks!

lgtraderPost authorI have not tried to do much work on bar counts.. I assume you mean it the literal sense. If that is the case then the only stat I have available is the one I did here with timing between transient bars. I actually think it is not bad for the time frame that it is working on, but of course more work will need to go into it. I kind of see bar counts as a type of filter, among the dozens and dozens that I think could potentially work..It will have to go on the list of things I try. The wave counts in this stat were slightly better than I expected, but I think I want to find some way to account for the up/down type and “longevity” of a trend before I add those elements in.

I’m glad I was able to give you some ideas on how to pursue your own work (:

Patrick M. WhiteYes timing between transient bars is about the same idea. One thing I’m working through is that depending on how you actually define a transient bar, you may/may not get a tip transient bar at each zig-zag high/low. If you measure the distance (in # of bars between alternating tip transients you *should* get about the same as if you measure the # of bars between zig-zag highs and lows. In most cases when you have a greater # of bars you will have a greater range than the previous range. But in some cases you can get a larger range with fewer bars or a smaller range with a larger # of bars.

Patrick M. WhiteHi LG,

I’ve been studying this some more and doing some of my own experiments. In your “projection.png” graphic under the 1 Leg / Post Flat section it appears there is a mis-categorization. In general terms, it should be logically impossible to have an Expand pattern following a Flat pattern because a Flat pattern is essentially two sequential decreases, while an Expand is essentially two consecutive increases. Also impossible is a Spring after a Flat pattern because Spring requires the middle leg to be larger, which is not possible coming from a Flat. Both these comments exclude the possibility of “ties” which may allow for a few outliers in each category depending on which category gets the equal sign. I didn’t do my equal signs exactly the same as you.

Here are my numbers for EU M15 based on bar counts for comparison:

Stats for next 1 leg of pattern

Trend : Greater: 38.9% Less: 59.3% Equal: 1.9%

Trend: 3.5% Flat: 0.0% Expand: 35.4% Spring: 61.1%

Flat : Greater: 67.5% Less: 28.1% Equal: 4.4%

Trend: 71.9% Flat: 28.1% Expand: 0.0% Spring: 0.0%

Expand: Greater: 21.4% Less: 76.9% Equal: 1.7%

Trend: 0.0% Flat: 0.0% Expand: 21.4% Spring: 78.6%

Spring: Greater: 56.0% Less: 41.1% Equal: 2.9%

Trend: 58.7% Flat: 39.6% Expand: 0.0% Spring: 1.7%

Trend and Expand patterns (which have a larger final leg) tend toward smaller next legs while Spring and Flat patterns (which have a smaller final leg) tend toward larger next legs. Flat and Expand seem to be the exception / minority patterns while Trend and Spring seem to be the most common / majority patterns.

Relativity+1

=)

Patrick M. Whitety! 😉

Patrick M. WhiteThe formatting of my stats didn’t really work out but basically there 4 categories of two lines each that equate to your: “Post Trend”, “Post Flat”, “Post Expand”, and “Post Spring” lines for 1 Leg patterns respectively.

lgtraderPost authorEditted! with some additional comments

(or rather at the time of posting this, editting)