Registered User Joined: 6/30/2017 Posts: 1,227
|
Hi Bruce,
Hope you enjoyed your weekend on the yacht!
This should be easy peasy for a Ninja...
Here are the bullish PCFs. Even though stocks take the stairs up and the elevator down, to keep things simple we'll just talk about longs. I can always modify your PCFs for shorts.
-
Alert (price starting to swing up)
-
Price Closes above Time Series Forecast (20 period, 1 forecast bar)
-
Entry (enough slope to trade - flat slope sucks - and sufficiently strong trend)
-
Linear Regression Slope (20 period) greater than 0.13, and
-
Linear Regression R-Squared (20 period) greater than 0.20
-
Exits (there are 3 separate exits for now - will have to test if one is better than the others)
-
Price Closes below Time Series Forecast (20 period, 1 forecast bar)
-
Linear Regression Slope (20 period) less than 0.13
-
Linear Regression R-Squared (20 period) less than 0.20
I would've just uploaded a snap of the napkin but my handwriting sucks. And it's even worse when I'm overly caffeinated. :)
|
|
Registered User Joined: 9/17/2010 Posts: 484
|
Some thoughts, just to pass the time
Alert
XUp(FAVGC20, AVGC20)
Filter
6 * (FAVGC20 - AVGC20) / 19>0.13
((x + 1) / 2 * (FAVG(w, x) - AVG(w, x)) / SQR((x ^ 2 - 1) * (AVG((w ) ^ 2, x) - AVG(w, x) ^ 2) / 12)) ^ 2)
with X=20 and I assume w=C?
Exits
XDown(FAVGC20, AVGC20)
6 * (FAVGC20 - AVGC20) / 19<0.13
((x + 1) / 2 * (FAVG(w, x) - AVG(w, x)) / SQR((x ^ 2 - 1) * (AVG((w ) ^ 2, x) - AVG(w, x) ^ 2) / 12)) ^ 2)<0.20
These PCF's "work", in that they don't generate errors, but running the scan against US Common stocks returned no triggers. 276 stocks passed the filter conditions
70 stocks in US Common passed the alert without the filters
|
|
Registered User Joined: 9/17/2010 Posts: 484
|
FWIW, when I set the period of Time Series to 10,I got back 8 stocks from High Cap 1000 which passed the filters, and the trigger fired
|
|
Registered User Joined: 6/30/2017 Posts: 1,227
|
thanks b,
Let me refill my coffee and mull your response over.
Bruce,
Any thoughts?
|
|
Worden Trainer
Joined: 10/7/2004 Posts: 65,138
|
If the alert is price crossing up the forecast, I might change it to the following (but it is possible that bcochrane has done some simplification and I just didn't notice).
XUP(C, AVGC20 + 63 * (FAVGC20 - AVGC20) / 19)
Filling in the blanks gets us the following for the R-Squared portion.
((20 + 1) / 2 * (FAVGC20 - AVGC20) / SQR((20 ^ 2 - 1) * (AVG(C ^ 2, 20) - AVGC20 ^ 2) / 12)) ^ 2 > .2
So the exit versions of those would be:
XDOWN(C, AVGC20 + 63 * (FAVGC20 - AVGC20) / 19)
And:
((20 + 1) / 2 * (FAVGC20 - AVGC20) / SQR((20 ^ 2 - 1) * (AVG(C ^ 2, 20) - AVGC20 ^ 2) / 12)) ^ 2 < .2
-Bruce Personal Criteria Formulas TC2000 Support Articles
|
|
Registered User Joined: 6/30/2017 Posts: 1,227
|
b,
Still mulling. I might have fubared the parameters.
Bruce,
The alert isn't really essential (I think, but I've been wrong once or twice). I was just visualizing the lifecycle of a price swing/trend.
The first thing that happens (for longs) is price moving up over whatever moving average/linear regression/time series forecast parameters you choose.
Later is when things get interesting. Since flat-ish slopes aren't attractive the heart of this whole thing is the Linear Regression Slope.
Don't quote me on this but I think I got the 0.13 parameter from Trading Systems and Methods + Website (5th edition). I don't have the book in front of me. A guy YouTubed the basic idea here; not the greatest video but it's short and around the 7:30 mark he does show the general idea and uses the 0.13 slope as his entry/exit.
I haven't studied R-Squared sufficiently to make an educated statement, but my gut tells me as the slope rises so does R-Squared. Maybe not exactly (otherwise why would you need two measures?). Again, don't quote me. Traders who are not into statistics would just use ADX to tell them the same basic thing, ie, strength of the trend.
So,
To strip this down to the absolute basics, I would say,
-
Buy when the Linear Regression Slope > 0.13
-
Sell when the Linear Regression Slope < 0.13
-
Noodle around with the slope parameter
-
Noodle around with the lookback period
-
Noodle around with R-Squared
Nothing is ever written in stone, and even if it was, no two people trade the same way, so noodling is always recommeded. :)
20 minutes to the eclipse!! :)
|
|
Worden Trainer
Joined: 10/7/2004 Posts: 65,138
|
Have you thought about converting the linear regression slope to a percentage?
600 * (FAVGC20 - AVGC20) / 19 / AVGC20
The above is how much price moves up per bar as a percentage of the value of the simple moving average over the period (or the middle of the linear regression line).
-Bruce Personal Criteria Formulas TC2000 Support Articles
|
|
Registered User Joined: 6/30/2017 Posts: 1,227
|
Bruce, I can always count on you to take my genius ideas and come up with GENIUS ideas. :)
Will mull it over, after another coffee refill. I'm sure it's awesome.
But first,
10 minutes to the eclipse. Go outside!! Or at least get online!
|
|
Registered User Joined: 6/30/2017 Posts: 1,227
|
https://eclipse2017.nasa.gov/eclipse-live-stream
|
|
Registered User Joined: 6/30/2017 Posts: 1,227
|
The servers at that last link might be overloaded.
Here's NASA TV...
https://www.nasa.gov/eclipselive/#NASA+TV+Public+Channel
|
|
Worden Trainer
Joined: 10/7/2004 Posts: 65,138
|
Don't worry about me, I am using a laptop today so I can work outside (and still have about 8 minutes until it starts here).
It is pretty nice here other than being really hot out (some clouds but staying away from the sun so far). That might change a bit as the eclipse progresses. Not sure it will make enough difference for it to actually be comfortable.
-Bruce Personal Criteria Formulas TC2000 Support Articles
|
|
Registered User Joined: 1/28/2005 Posts: 6,049
|
((20 + 1) / 2 * (FAVG20 - AVGC20) / SQR((20 ^ 2 - 1) * (AVG(C ^ 2, 20) - AVG(C, 20) ^ 2) / 12)) ^ 2 > .2
Error message says missing "(" at position 21
Thanks
|
|
Registered User Joined: 6/30/2017 Posts: 1,227
|
Cloudy here @#$%^&*! ...
So I'm checking out the NASA livestream ... ironically it seems to be delayed a few minutes ... about to reach Oregon as I post this.
|
|
Registered User Joined: 6/30/2017 Posts: 1,227
|
QUOTE (Bruce_L)
Have you thought about converting the linear regression slope to a percentage?
600 * (FAVGC20 - AVGC20) / 19 / AVGC20
The above is how much price moves up per bar as a percentage of the value of the simple moving average over the period (or the middle of the linear regression line).
As I said, GENIUS!
|
|
Registered User Joined: 9/17/2010 Posts: 484
|
I had made an error in theTime Series calculations I had attempted, so thanks for the correction.
|
|
Registered User Joined: 6/30/2017 Posts: 1,227
|
Just an fyi, in the video above, the dude uses a 14-period LRS along with the 0.13 threshold.
As I mentioned, noodling is always recomended. :)
|
|
Registered User Joined: 6/30/2017 Posts: 1,227
|
Small sample size, I know, but if you stick with the video you see 7 trades. (YouTube dude missed one short trade so I used Snagit to point it out). I also ignored the left part of the chart, because frankly, this is just an example anyway. Needs more robust testing - but you knew that. :)
Like I said above, not the greatest video, but makes the point. Since the LRS is heading up in the video any self-respecting trader would only take the long signals.
Let's go to the video tape. What do you see?
-
Long (4 trades)
-
One teeny tiny winner (might not have even covered commissions. Can't win 'em all).
-
One good winner.
-
One gooder winner.
-
One great winner.
-
Shorts (3 trades - not recommend since they're going againt the trend, but hey - it's your money :)
-
One teeny winner
-
One tiny winner
-
One good winner
You'll notice the quality of the longs are better (read "more profitable") than the shorts).
Whether you trade my genius idea (which I "borrowed") or Bruce's GENIUS idea (see PCF above) this looks like a not bad approach considering it was born while scribbling on a Starbucks napkin. I've certainly seen worse.
Barista!
|
|
Registered User Joined: 6/30/2017 Posts: 1,227
|
Goodbye eclipse. Really cool.
I can't imagine how much our ancestors freaked out during one of these before they figured out the science behind an eclipse.
|
|
Registered User Joined: 6/30/2017 Posts: 1,227
|
Bruce, I think we can almost put a bow on this topic. I've been posting so much I think diceman had one question that's yet to be answered...
QUOTE (diceman)
((20 + 1) / 2 * (FAVG20 - AVGC20) / SQR((20 ^ 2 - 1) * (AVG(C ^ 2, 20) - AVG(C, 20) ^ 2) / 12)) ^ 2 > .2
Error message says missing "(" at position 21
Hope you got to enjoy the eclipse!
|
|
Worden Trainer
Joined: 10/7/2004 Posts: 65,138
|
This version seems to test fine in my copy of TC2000.
((20 + 1) / 2 * (FAVGC20 - AVGC20) / SQR((20 ^ 2 - 1) * (AVG(C ^ 2, 20) - AVGC20 ^ 2) / 12)) ^ 2 > .2
-Bruce Personal Criteria Formulas TC2000 Support Articles
|
|
Registered User Joined: 6/30/2017 Posts: 1,227
|
UPDATE:
-
I inadvertently posted some misinformation above. My bad, but I'm blaming too much caffeine and a cr*ppy YouTube video. :)
-
See today's "v17 PCFs for Kaufman's Original LinReg Slope and R-Squared Rules" post for the real scoop.
|
|
Registered User Joined: 5/30/2014 Posts: 53
|
Darn! I missed the official Worden forum eclipse thread!
Next time!
|
|
Registered User Joined: 6/30/2017 Posts: 1,227
|
Since Bruce brought his laptop to work outside that day I was waiting for him to livestream it, but he never did.:)
|
|
Worden Trainer
Joined: 10/7/2004 Posts: 65,138
|
My direct observations were entirely visual. The only photos and videos I took were of the interesting shadows, not the eclipse itself.
-Bruce Personal Criteria Formulas TC2000 Support Articles
|
|
Registered User Joined: 6/30/2017 Posts: 1,227
|
Stiil, a livestream would've been totally cool. :)
|
|
Guest-1 |