Post New

I combined 2 screens {N1Mo} and {N1HL} using a few methods. These screens use price momentum to pick from Nasdaq 100 stocks. The momentum measures here are:
PHL252lagged: Price 3 days ago / average of 1-year low and high prices
Mo: 1-year total price return

The screen {picktally_N1Mo_N1HL} uses picktally to find stocks that appear in both screens. To get an average near 3, I set the screen depths to 5. The number of passing stocks varies by month with a range of 0 to 5 picks.

The screen {rank_N1Mo_N1HL} uses the rank function, with an almost identical result as picktally. The 2 screens are merged into 1 screen, and the following steps are added:
rankPHL: rank of [PHL252lagged]
rankMo: rank of [Mo]
rankPHL <= 5
rankMo <= 5

The {Overlaps_N1Mo_N1HL} screen uses a [Overlaps] field to find the stocks in both base screens. The number of passing stocks varies from 3 to 7.
Overlaps: if(rankPHL<=3,if(rankMo<=3,0,100),100)
rankSum: plus(Overlaps,rankPHL,rankMo)
rankSum bottom 3

The {OverlapsMin_N1Mo_N1HL} screen has a different rankSum function. The number of passing stocks varies from 3 to 4.
rankSum: plus(Overlaps,min(rankPHL,rankMo))
rankSum bottom 3

The {OverlapsNoTies_N1Mo_N1HL } screen uses weights in the sum-of-ranks to avoid ties:
rankSum: linear(1,Overlaps,0.999,rankPHL,1,rankMo)
rankSum bottom 3

The average depth of these screens is about 3. {N1HL} has the highest CAGR, followed closely by {OverlapsMin_N1Mo_N1HL}. Results from 19850201 to 20251128 (monthly trading, no friction):

         Screen           CAGR  GSD  Sharpe  Depth            notes
          N1Mo            31.4  49    0.80    3.0               -
          N1HL            34.6  45    0.89    3.0               -
  picktally_N1Mo_N1HL     31.9  47    0.82    3.1   screen overlap at 5 depth
     rank_N1Mo_N1HL       31.9  47    0.82    3.1   screen overlap at 5 depth
   Overlaps_N1Mo_N1HL     33.1  46    0.86    3.2               -
  OverlapsMin_N1Mo_N1HL   33.7  47    0.86    3.3               -
OverlapsNoTies_N1Mo_N1HL  32.6  46    0.85    3.0               -

The number of passing stocks varies depending on which method is used to combine the 2 base screens. Using picktally or rank has a wide range of passing stocks, and sometimes picks no stocks. Using the Overlaps and sumranks fields always picks at least 3 stocks. Using a small weight in sumranks to break ties always picked 3 stocks in this backtest. The number of passing stocks (depth) with the various methods in these backtests was (as a percentage):

depth  picktally  rank   Overlaps  OverlapsMin  OverlapsNoTies
  0      1.2%     1.2%     0.0%        0.0%          0.0%
  1        6%       6%      0%         0%             0%
  2       19%      19%      0%         0%             0%
  3       37%      36%     82%         66%           100%
  4       31%      31%     15%         34%            0%
  5        6%       6%      2%         0%             0%
  6      0.0%     0.0%     0.2%        0.0%          0.0%
  7      0.00%    0.00%    0.02%      0.00%         0.00%

=== links ===
N1Mo is https://gtr1.net/2013/?~N1Mo::nas100.a:et1:dspo%28...
N1HL is https://gtr1.net/2013/?~N1HL::nas100.a:et1:dspo%28...
picktally_N1Mo_N1HL is https://gtr1.net/2013/?~picktally_N1Mo_N1HL::pickt...
rank_N1Mo_N1HL is https://gtr1.net/2013/?~rank_N1Mo_N1HL::nas100.a:e...
Overlaps_N1Mo_N1HL is https://gtr1.net/2013/?~Overlaps_N1Mo_N1HL::nas100...
OverlapsMin_N1Mo_N1HL is https://gtr1.net/2013/?~OverlapsMin_N1Mo_N1HL::nas...
OverlapsNoTies_N1Mo_N1HL is https://gtr1.net/2013/?~OverlapsNoTies_N1Mo_N1HL::...

Having thought about this for a while, here is the method I settled on for implementing the overlap picks.

As a result of generating the picks for the RS and PHL screens, I have the top 12 of each, in rank order. (Both those screens are 5-HTD-12.) It would be easy enough to do a ranked sum as a figure of merit, but I really like the idea of taking the true overlaps, the stocks that are in the top 6 of both screens.
This gives you anywhere from 0 to 6 stocks. The 6 could be any number. Even up to 12 to catch all the common stocks.

Various reported methods of picking overlaps seem to average about 3 stocks, with a significant number of months coming up with 0 or 1 stock. That's too few for me; just too risky.

If that is too few stocks for you, fill in with stocks from each screen alternately. As these are in rank order, it naturally takes them in order. This was easy to do in my script, just 2 greps and a merge. This will result in a final list of 12 to 24 stocks.
Take the top N as the overlap screen list.

Since I view this list as a conglomeration of the two momentum screens I feel more comfortable with more then 5 stocks, so I arbitrarily chose N to be 7. Taking the top 8 or even 10 would be reasonable. No HTD for this screen.

I am operating kind of blind here, because I don't know how this could be backtested with GTR1. Which is why I chose a larger number than 5 for the number of stocks to hold.

This method takes the stocks in the order that each screen sees as best, regardless of where the stock is ranked by the other screen. Using a ranked sum method penalizes any stock that appears in one screen but not the other. Which does not seem reasonable to me. (Remember, I only have the top 12 of each, not the entire 100.)

I think {OverlapsMin_N1Mo_N1HL} is close to what you want. Is the 20251128 result different than your method?

You can check the gtr1 fields:
Click Signal Values
Run Screener (no password needed because no .s fields are used)
Download Report

Symbol  [PHL252lagged]  [Mo]  [rankPHL]  [rankMo]  [Overlaps]  [rankSum]
  WBD        0.72       131       1          3         0           1
 AVGO        0.70       147       3          2         0           2
 PLTR        0.60       161       19         1         100        101
  MU         0.71       127       2          4         100        102
GOOGL        0.69        92       4          6         100        104
 LRCX        0.67       114       7          5         100        105
 GOOG        0.69        90       5          7         100        105
 IDXX        0.68        76       6         10         100        106
  APP        0.59        88       23         8         100        108
 AMAT        0.64        46       8         15         100        108
 KLAC        0.63        84       9          9         100        109
  EA         0.63        24       10        28         100        110
 REGN        0.62         5       11        47         100        111
 INTC        0.60        63       20        11         100        111
 MNST        0.62        36       12        20         100        112
 ASML        0.60        56       17        12         100        112

The {OverlapsMin_N1Mo_N1HL} screen first adds the overlaps for the top 3 in each screen. On 20251128, this was 2 stocks: WBD and AVGO. Then stocks are added based on the lowest rank:
PLTR rank 1 in N1Mo
MU rank 2 in N1HL
GOOGL rank 4 in N1HL
LRCX rank 5 in N1Mo
GOOG rank 5 in N1HL
IDXX rank 6 in N1HL
APP rank 8 in N1Mo

https://gtr1.net/2013/?~OverlapsMin_N1Mo_N1HL::nas...

Ahhh, I see, that is not quite what you want.

For example, with a 12 depth, say the overlaps were [rankPHL] #1 to #6, and [rankMo] #7 to #12. {OverlapsMin_N1Mo_N1HL} will pad with [rankMo] #1 to #6. But you want to pad with [rankPHL] #7 to #9, and [rankMo] #1 to #3.

Is the 20251128 result different than your method?

Note: I remove GOOG when GOOGL is also present.

My screens in rank order:

        --PHL--    --RS--
1         WBD       PLTR
2         AVGO      AVGO
3         MU        WBD
4         LRCX      MU
5         GOOGL     LRCX
6         IDXX      GOOGL
7         AMAT      APP
8         KLAC      KLAC
9         EA        IDXX
10        CEG       ASML
11        AMD       AMD
12        APP       INTC
13        ASML      GILD

My script gets:
For top 7 commons
AVGO, WBD, MU, LRCX, GOOGL, idxx, pltr
Lower case ones are fillins after the stocks in common.
ranked-picks.txt
1 AVGO
2 WBD
3 MU
4 LRCX
5 GOOGL
6 IDXX
7 PLTR
8 AMAT
9 APP
10 KLAC
11 EA
12 CEG
13 ASML
14 AMD
15 INTC
16 GILD

For top 3 commons
AVGO, WBD, mu, pltr, lrcx, googl, idxx
Lower case ones are fillins after the stocks in common.
ranked-picks.txt
1 AVGO
2 WBD
3 MU
4 PLTR
5 LRCX
6 GOOGL
7 IDXX
8 AMAT
9 APP
10 KLAC
11 EA
12 CEG
13 ASML
14 AMD
15 INTC
16 GILD

Interesting how the order shifts depending on the actual number of stocks in common.

The final list, yours (from post) and mine (from script):

	Yours	My with top 3 
1	WBD	AVGO
2	AVGO	WBD
3	PLTR	MU
4	MU	PLTR
5	GOOGL	LRCX
6	LRCX	GOOGL
7	GOOG	IDXX
8	IDXX	AMAT
9	APP	APP
10	AMAT	KLAC
11	KLAC	EA
12	EA	CEG
13	REGN	ASML
14	INTC	AMD
15	MNST	INTC
16	ASML	GILD

For investing in the top 7 and omitting GOOG, the same stocks are picked. There are more differences near the bottom.

Based on just this one comparison, it would appear that this {OverlapsMin_N1Mo_N1HL} is a good enough backtest to validate my script's method of generating the overlap screen.

I do not believe the 32% CAGR, though.

Thanks for the discussions, your efforts and sharing.

Do you ever eliminate any stocks for non-mechanical reasons? Like WBD is based on a buyout so if it were trading near $31 would you eliminate it?

I actually own positions in the top 9 except for LRCX, WBD and IDXX. I added LRCX today but held off on WBD and IDXX since I felt I already owned enough of the others (7 of 9).

I do not believe the 32% CAGR, though.

I wanna believe!

Thanks again

Do you ever eliminate any stocks for non-mechanical reasons? Like WBD is based on a buyout so if it were trading near $31 would you eliminate it?

Every time that thought crosses my mind, I steel myself to remember some quotes
"Discretion over buy-and-sell decisions in aggregate can turn a model that generates a market beating return into a sub-par return. IOW, follow the signals religiously!"

and
"(In a proven investment trading strategy...)
Do not -- repeat DO NOT -- pay much attention to the effect of every trade. The majority of individual trades will be irrelevant to your long-term results.

You should be concerned about portfolio results, not position results.

You need to trade like a casino – probability based, with no particular outcome required of any one trade.
1. Anything can happen
2. You don’t need to know what’s going to happen next in order to make money.
3. Wins and losses are random – You will never know when a trade will be a winner in advance, only that the conditions that define your edge are present.
4. Your edge is nothing more than a higher probability of one thing happening over another. Your edge is no guarantee of a winning trade, just of winning over time.
-- http://www.roguetraderette.com/2011/04/01/the-art-..."

and
"The historical evidence is clear that the winning strategy in both stocks and bonds is to ignore all forecasts and stick to a well-developed plan. -- Larry Swedroe"

So, yeah, no.
I held onto WBD until it dropped off the screen. In a moment of weakness, early on I set a GTC sell at 30.99 just in case it got there.

---------------
For the sake of completeness I will say that I changed the way to pick overlaps. It is trickier than it appears at first blush. Thinking about it overnight I was unhappy about the way I was doing it. Remember, this is all in a bash script -- so no handwaving allowed. It was a bit of a challenge to get it working right, especially making sure that some stocks were not included twice in the final overlap picks. And then to get the non-common stocks in the "proper" order. Trick here is to settle on what you think is the proper order.

So my current method is:
* The stocks that are common in the top 3 of both screens (PHL and RS).
* The stocks that haven't been picked that are in the top 6 of both screens.
* The remaining stocks that haven't been picked, alternately from each screen, in rank order.
* Invest in the top N stocks. My current thinking is N=7.
* The source lists are top 12-15 of each screen, in rank order, irrespective of how many stocks that screen invests in.

My convenient data source for the RS screen data gives the top 20, which is plenty.

Post New

Mechanical Investing FAQ

---

Contact Shrewd'm
Contact the developer of these message boards.