Said’s ranking method

[TrentRothall]

I thought i’d start a new topic so we don’t get the graveyard thread too off topic.

This is a quote of Said’s that started some good discussion/ findings

Quote:

there are two scenarios there are stocks that will achieve their performance and they have the price action all over the place and other that will move up steadily and slowly.

so what i am catching are the ones that are only going up slow and steady
the way to measure this is by using ATR as measure I know that the value of ATR is dependent on the price so the ATR of 5$ stock is not like the ATR of 100$ stock. the way to come over this is by dividing the ROC by ATR

so i am ranking using ROC(C, Duration)/ATR(Duration)

here is the code so that you can test
score = Ref(ROC(C,Duration)/ATR(Duration),-1);
score = iif(LE,score,0);
positionscore = iif(year()>=1985 AND Month() != ref(Month(),-1),score,scoreNoRotate);

also i am using volatility based position sizing

give it a try and tell me your finidings

I have only really just started looking into rotational systems and am trying to understand how the calculation actually works. The ROC(C,duration)/ATR(durtion) calculation.

Instead of just using ATR shouldn’t we have to normalize the ATR to the closing price?
i.e work out the ATR as a % of the price (ATR/C*100)

consider the example using the current formula.
xyz stock has a close of $100 a ATR of 1 and a ROC of 10
abc stock has a close of $10 a AtR of 0.1 and a ROC of 10

xyz = 10/1 = 10
abc = 10/0.1 = 100

abc will have a greater positionScore for our portfolio

in both examples the ATR is 1% of the closing price and they have the same ROC so shouldn’t they be equally qualified for our portfolio?

or have i stuffed something up?

[RobGiles]

hi Trent
is your aim here to risk the same % of investment capital per trade, so that on more volatile equities, your stops are wider / posn size is smaller, and with low vol equities your stops are tighter, but the position size is larger (but both have the same risk if the initial stop is hit)?

[TrentRothall]

Hi Rob,

No, this is more a continuation on the discussion in the “Trading System graveyard” topic. It’s to do with rotational systems and trying to capture the smoothest trends instead of stocks that might rocket up but might not sustain that momentum.

But some people are using a volatility measure to make sure that each position will impact their portfolio roughly the same amount each day.

[SaidBitar]

as you mentioned it is 100% accurate you need to divide the ATR by the Close to get a normalized value which will be 10% in the example that you provided for both
then you divide the ROC by it then it will be 1 for both of them
i believe i tested like this and the results were not as good

[TrentRothall]

Yeah ok, it just doesn’t make sense to me how it works. It clearly does though maybe that’s all that matters lol

i found the same thing, but still playing around because i have some issues

[SaidBitar]

just to be sure i tested again with the ranking as following:
score = Ref(ROC(Close, ROCDuration)/(ATR(ROCDuration)/Close),-1);

---

Backtest started.
Profit = 1012559.15 (1012.56%), CAR = 14.92%, MaxSysDD = -95546.15 (-26.85%), CAR/MDD = 0.56, # winners = 195 (58.91%), # losers = 136 (41.09%)
Completed in 27.79 seconds. Number of rows: 331
( Timings: data: 1.34, setup: 0.00, afl: 211.03, job: 1.56, lock: 0.00, pbt: 0.92, UI thread: 2.27, worker threads: 212.59/212.59 )

---

so two things we have over here the winning percentage and the CAR/MDD
the winning percentage indicated that the strategy is profitable especially when the average winner is higher than the average loser.
the CAR/MDD it is saying that we need to check some stuff around position sizing to make it better

[Anonymous]

I had similar hesitation to Trent about scoring based on ROC, which is a percent, by ATR, which is a dollar value. As such, I tested across several universes and time frames to test for robustness. I tested 3 versions of the ROC score:

Original – standard ROC
Said – ROC(C,LB) / ATR(LB)
Said_Adj – ROC(C,LB) / (ATR(LB) / MA(C,LB)

The Said ranking, while it makes little logical sense was the best performer across time and universes. It is robust as well.

[Anonymous]

[LEONARDZIR]

Brent,
Looking at your results for SP 500 from 2000-2016 Said’s original ROC in SP500 you have CAR at 18.7% with MDD at 25%.I believe Said’s results were something like CAR/MDD 20%/17% or something close to that.. I get CAR/MDD of 21%/20% in the same universe and timeframe. How do you account for the differences?

[LEONARDZIR]

Correction my backtested results were from 1995. to 2016

[Anonymous]

The results will not be exactly the same. There are other things going on in my system such as a gap filter, a means of screening stocks that have done extremely well in the short term (given the philosophy that performance is mean reverting in the short term), individual stock MA, index filter, position sizing variances, etc.

While I held all variables and rules away from the actual ranking constant across my three tests, my results will no doubt vary from others simply because of variances in our systems.

The point of this exercise was to see how the three different versions for ranking affected results and if the results were robust across time frames and universes. It appears that the Said ranking (unadjusted) is the best option out of the three that I looked.

That said, it doesn’t make logical sense to me as of now. Despite the test results, I do not think I will trade based on the Said ranking. I will however use this as a launching point for further study.

I would be interested to hear other people’s opinion on this. We have test results that show better performance from using the Said ROC and those results are robust, but the Said ROC isn’t calculated in a way that makes sense as to why it should work. Would you trade it without a logical explanation?

[LEONARDZIR]

Brent,
You raise a very interesting question, namely should you trade a system that doesn’t make logical sense even though it tests well historically and is robust. . I don’t have a definitive answer but I believe it is probably yes. Before I joined the mentoring program I traded a number of systems which made sense but didn’t make (lost) money.. Since I joined the program I have been surprised by a number of systems I developed that were based on logical assumptions that didn’t test well.
In addition to my US MOC and US MR I have traded Nick’s US momo since Jan 2016 with satisfactory results. Nick has told me that he has incorporated Said’s ROC/ATR ranking criteria with a significant improvement in CAR and MDD.

[Nick Radge]

Quote:

Nick has told me that he has incorporated Said’s ROC/ATR ranking criteria with a significant improvement in CAR and MDD.

Correction: ‘a variant’

[OmarAouane]

Just browsing through this very interesting thread, a potential logical explanation for the robustness of this ranking method could be that those slow and steady growers would be more mature companies with a higher absolute share price and consequently higher ATR.

[Anonymous]

As Said said that he didn’t see the same type of results with ROC/ATR/C, I would really do some extra testing before putting something like that (without normalizing for share price) into production. But – I am very good at being wrong, so maybe there is merit there!

[Author]

Posts