Post New

Hi All,

A while back (pre-Shrewd) we had been discussing an ROE Cash to LT Debt screen based on the VL1700 universe. Has anyone been following it, and how did it do post-discovery?

I'm curious if anyone rode it through 2022 and how it went.

--Gabriel

I don't know about the valueline 1700 results but a backtest using the sp500 + sp400 +sp600 (1500 stocks) using SIP data
ROE_CASH 2020-01-03 thru 2023-05-19 Top 12 tilldrop 15 period 28 days slipRndTrp 0.70% , Dividends True , CashForNoPicks True
CAGR Fri 16.9% GSD 28.3% CAGR/GSD 0.6 UlcrIndx 15.4% MxDD 33.1% AT 94%
1.69 X accumulated with 38 Buys, 528 Holds and 100.0% Avg Invested.
The 69% total return compared to a 27% total return of the SP500 over the same period.

I've been using a different variation of ROE_Cash as part of a tactical allocation since 2020. My results are significantly lower,
but my volatility was also much lower.

RAM

I tested a P123 version. Over the past 5 years it backtested well. However, I couldn't figure a way to do the original 39 stock 'dozens' portfolio in P123 where you pick 13 stocks each month and hold them for 3 months, so I just picked 13 and held them for 3 mos.

13 stock, 3 mo. hold version
Total return: 132% vs S&P: 68%
Max DD: -33% vs S&P: -33%
StdDev: 23% vs S&P: 18%
Beta: 1.09

If anything knows how to script the 39 stock rotation in P123, I would love to know it.

Taz

A while back (pre-Shrewd) we had been discussing an ROE Cash to LT Debt screen based on the VL1700 universe. Has anyone been following it, and how did it do post-discovery?

Ha! I was just going to write about this screen. I've been using two variants of ROE_Cash as my sole MI screens for about three years. These were SIPro based using GTR1 to produce the rankings. With the demise of GTR1 I've been up the creek without a paddle for the last two months. I traveled for a month, so just held on to my March picks. Now that I'm back I decided to do something about producing fresh rankings and continuing to use the screens. Using VL is an available option, although to directly address your question I don't have performance stats for the last three years.

After reviewing the discussions from three years ago (datahelper is a godsend) I've reconstructed the Radiscript for my two screen variants and I intend to use these rankings for my next round of portfolio updates.

A few observations:
Way back three years ago I concluded that the ROE numbers coming from VL were extremely unreliable, which was the primary reason I decided to go with the SIPro based screens. I was looking at [ROE Latest QTR] which continues to be infested seemingly with a lot of garbage. Now I've decided to use instead [Return On Shareholders Equity] which is much more reasonable looking even though it may only reflect figures from the annual report, so not always very up to date.
The key observation that I found is that the latest rankings from the VL screens that I've constructed are remarkably similar to those from the old SIPro screens. This gives me a sense of confidence.

The two screens differ in the number of stocks ranked by ROE. The basic version takes Top 500. The second version uses Top 60. This tends to produce very different stock picks. With Top 500 you get very good but not extreme ROEs and the final sort by [CashminusLTD] is dominant. It puts cash rich megacap companies at the top of the rankings. But it not a pure megacap screens. The list of megacaps is pretty much in the top 7 of the screen, and further down the companies aren't necessarily huge, but they are cash rich relative to their size.

The second screen produces companies with spectacular ROEs but not necessarily great cash positions. Roughly the top 10 have a positive [CashminusLTD] and the rest I would say have solid balance sheets but they're not cash behemoths.

Regarding the screening steps:
Timeliness<=5 produced only 1441 stocks last week.
I applied two "crap filters" if you will. [Domicile Code]="US" needs no explanation.
OR([SIC Code]<6000,[SIC Code]>6999) filters out all financial companies (banks, insurance, investment companies) as Mungofitch intended originally.

In my portfolio I buy only the top 12, HTD 20, but I rank the top 45 following Mungo's original suggestion. I'll try to figure out soon how to add these screens to my weekly rankings.

Here are the Radiscripts for the two screens.

	Define	{ROE_Cash}								
	Uses	[Timeliness Rank]	[Return On Shareholders Equity]	[Cash]	[Long-Term Debt Latest Qtr]	[Market Cap $ (Mil)]	[Domicile Code]	[SIC Code]	[ROE Latest Qtr]	
	Deblank	[Timeliness Rank]	[Return On Shareholders Equity]	[Cash]	[Long-Term Debt Latest Qtr]	[Market Cap $ (Mil)]	[Domicile Code]	[SIC Code]	[ROE Latest Qtr]	
	Create	[CashminusLTD]	:([Cash]-[Long-Term Debt Latest Qtr])							
	Keep	:[Timeliness Rank]<=5								
	Keep	:OR([SIC Code]<6000,[SIC Code]>6999)								
	Keep	:[Domicile Code]="US"								
	Sort	Descending	[Return On Shareholders Equity]							
	Top	:500	PlusTies							
	Sort	Descending	[CashminusLTD]							
	Top	:45								
	End									


	Define	{ROE_Cash2}							
	Uses	[Timeliness Rank]	[Long-Term Debt Latest Qtr]	[Market Cap $ (Mil)]	[Domicile Code]	[SIC Code]	[Return On Shareholders Equity]	[ROE Latest Qtr]	[Cash]
	Deblank	[Timeliness Rank]	[Long-Term Debt Latest Qtr]	[Market Cap $ (Mil)]	[Domicile Code]	[SIC Code]	[Return On Shareholders Equity]	[ROE Latest Qtr]	[Cash]
	Create	[CashminusLTD]	:([Cash]-[Long-Term Debt Latest Qtr])						
	Keep	:[Timeliness Rank]<=5							
	Keep	:OR([SIC Code]<6000,[SIC Code]>6999)							
	Keep	:[Domicile Code]="US"							
	Sort	Descending	[Return On Shareholders Equity]						
	Top	:60	PlusTies						
	Sort	Descending	[CashminusLTD]						
	Top	:45							
	End								

Elan

Taz2 . . I tested a P123 version. Over the past 5 years it backtested well.

I too have a P123 subscription with backtesting back to 1999 but haven't tried to construct a ROE_Cash
screen on P123. Would you be willing to post the p123 script you used. Before GTR1 crashed I was using
a GTR1 version of ROE_Cash that added some volatility constraints that improved long term results and
SAWR but hasn't helped post discovery. My Stock Investor Pro data only version I'm currently using
backtests well but as my backtester doesn't give me the same daily start statistical confidance that GTR1
provides. I have tested mungo's basic concept with enough deeper holds, longer periods and variations
of thresholds to satisfy myself that this screen is currently as robust as they get.

RAM

In the past few days since the topic came up, I've been experimenting with the screen. What is included here is not the original, but it has a much higher total return. I've included a Dividend, high operating margins, positive recent earnings, and a final sort on (Cash minus Debt)/ MktCap. I know this isn't consistent with Mungofich's original concept because it selects, in addition to the huge caps, occasional market caps down to $2B. Anyway, the basic structure is here and you can use it as a starting place for experimentation.


5-yr total return: 148%, max DD: -35%, Beta: 1.19.

Universe(NOOTC)=TRUE & Universe($adr)=false & Country("USA")
!GICS(FINANC)//No finance cos.
Yield > 0
FOrderOLD("$Timeliness",#All,#DESC,#Previous,TRUE)<=600
FOrderOLD("ROE%A",#All,#DESC,#Previous,TRUE)<=80
OPMgn%TTM > 25 //Healthy Operating Margin
CurFYEPSLow>0 // Recent and upcoming Earnings positive
EPSExclXorTTM>0 AND EPSExclXorPYQ>0
FOrderOLD("CashTTM - DbtLTTTM",#All,#DESC,#Previous,TRUE)<=40
(CashTTM - DbtLTTTM) / MktCap > 0
FOrderOLD("((CashTTM - DbtLTTTM) / MktCap)",#All,#DESC,#Previous,TRUE)<=13

where $Timeliness is: Beta*TotalReturn26W*(CurFYEPSMean/EPSInclXorA)
(my crude approximation of VL timeliness)

Taz

Taz, I loaded your screen into P123. The only change I made was to change the Universe to
S&P 1500 CompositeCap (IVV+IJH+IJR) and I backtested holding 15 stocks
For a 5 year backtest I get similar results 21.25% vs SP500 11.19% CAGR
And for a 10 year backtest I also get a nice 18.1% return vs the SP500 11.76
But for a 20 year backtest the advantage was still positive but only 13.0% to 9.8%

Removing your OPMgn%TTM > 25 //Healthy Operating Margin increased the 20yr CAGR to 14.35
Removing EPSExclXorTTM>0 AND EPSExclXorPYQ>0 increased the 20yr CAGR to 14.99
Removing (CashTTM - DbtLTTTM) > 0 increased 20 yr to 15.24%
Changing the FOrderOLD("((CashTTM - DbtLTTTM) / MktCap)", #All, #DESC, #Previous,TRUE) <= 13
To <= 8 gives a CAGR of 17.4%

The 5 year return is not quite as good, a CAGR of 19.2% compared to the 21.25 but overall a longer term
consistant outperformance.

RAM

Thanks RAM, your analysis shows why I don't like making decisions based only on 5 years of backtest data.

Taz

I've been using two variants of ROE_Cash as my sole MI screens for about three years. These were SIPro based using GTR1 to produce the rankings.

Elan, how have your SIPro-based variants performed since you started using them? Also curious if you have tracked their max drawdown?

I don't track the performance of individual screens. My portfolio is a combination of MI screens and other holdings. It's really hard to break down.

Elan

Hi, I didn't see this thread for a while.

FWIW
Going back to the thread which introduced the screens, this post has the endpoints of the tuning process that I did back when
http://www.datahelper.com/mi/search.phtml?nofool=y...

That post has one basic version, and one version that requires dividends.
The cutoffs are tuned, but the screens are still very simple.
Those suggested versions used the "ROE latest quarter" field, not the annual figure as in the kickoff post in that thread.

Backtest figures in that post were up to Q1 2020 inclusive.
So, three years out of sample, 2020-Q2 through 2023-Q1 inclusive, tested the same way as in the old post.
Figures below are run as recommended at the top of the original thread: Top-40-hold-till-drop-45, run every two months, with 0.4% friction.


(1) First version as in that post (no requirement of dividend): Absolute return 22.3%, beat SPY by +4.8%/year
(2) Second version as in that post (requires dividend): Absolute return 20.6%, beat SPY by +3.1%/year

I have always had a lot of success with screens using the simpler annual VL field "Return on Shareholders Equity", so I tested those too.
Lower turnover, and it seems to work fine. Returns a bit higher in the first three years out of sample, especially for the dividend version. A lot of that is probably random noise, but maybe not all of it, hard to say.
(3) First version as in that post, but using annual ROE (no requirement of dividend): Absolute return 22.7%, beat SPY by +5.1%/year
(4) Second version as in that post, but using annual ROE (requires dividend): Absolute return 24.3%, beat SPY by +6.8%/year

Some recent picks of version (1) above, quarterly ROE and no requirement of a dividend
MSFT CSCO NVDA HUM ACN RNR MOH COST APO NKE
PANW VEEV ANET TJX ROST EOG EXPD AMP NTAP BBY
MELI NVR PSTG CPRT EQH LOGI LULU FTNT ALNY SPNT
RL NYCB UAA CMG DECK TXN MPWR ULTA WIRE PBF

Some recent picks of version (2) above, quarterly ROE and requiring a dividend
MSFT CSCO NVDA HUM ACN RNR COST APO NKE TJX
ROST EOG EXPD AMP NTAP BBY EQH LOGI RL NYCB
TXN MPWR WIRE PBF MLI RHI IDCC MKTX CNS NSP
DKS WSM WHD LPLA LSTR DDS MGY BKE GGG PAYC

As an aside, another nice way to run it is Dozens: 13 stocks not already held, hold 3 months, so you always have 39 stocks.

A parting thought:
Remember that the goal of this screen is not necessarily the highest possible return in backtest.
It's intended to be large cap and safe alternative to SPY, but with a good chance of a slightly higher average long run return.
A 40 stock equally weighted MI screen has largest single-company risk of 2.5% of portfolio. SPY's risk on that metric is three times as high.
The fact that returns have been so good relative to market in the first three years post discovery probably includes a lot of luck.
Still, all the firms are big, have great returns on equity, and lots of cash. Really bad explosions will presumably be rare.

Jim

Thanks Jim for the data about post discovery performance.

Regarding ROE latest quarter versus ROE annual, I noticed a while back how haphazard the VL data was for ROE latest quarter. I re-checked it recently and found a lot of apparent garbage in the quarterly figures. That's why I decided to use the annual ROE for the screen rankings I've started posting weekly.

Elan

I noticed a while back how haphazard the VL data was for ROE latest quarter. I re-checked it recently and found a lot of apparent garbage in the quarterly figures.

Yeah, I have never quite figured out that field.
If they are calculating the ROE based on a single quarter's earnings, it will be a wildly fluctuating figure, hence not very useful.
Using the trailing four quarters excluding extraordinary items would make some sense, but that's not easily available.
The annual "return on shareholders equity" field has its own issues, no doubt, but in testing it has always given pretty good results on average.
And the turnover is tiny.

Jim

Jim, thanks for your posts. For this ROE_Cash screen, thoughts on buying top 12 stocks and HTD 20, as Elan had suggested a couple of weeks ago?

Jim - very much appreciate your update on the screen performance.

Based on my notes from the original thread(s), this is what I have for the steps:

(1) Keep only those stocks that have a Timeliness ranking (good or bad)

(2) (Optional) Find those stocks with a dividend. The dividend version leans to *slightly* smaller large-caps,
but the average annual return in back-test is a bit higher.

(2.5) (Optional) Take the 50% of stocks closest to their 52 week high.

(3) Of those still in the running, keep the 30% of them that have the highest reported ROE.
Taking 32% might be better if you're obsessive.

(4) For those remaining, calculate their cash minus long term debt.

(5) Sort descending on CASH - LTD

Buy equal dollar amounts of the 40 with the most cash. Hold for a month (or two or three), and repeat.

Jim - were you implementing step (2.5)? If so, does that change the percentages in step (3)? I was never clear how 2.5 might affect 3.

P.S. - Love reading you're posts, grateful you're posting again, and gratitude to Manlobbi for creating these boards.

Addition:

I also have notes that give step 5 as:

(5) CASH - LTD - CURRENT LIABILITIES.

I ran a 23-yr P123 backtest of an 8-stock version. In each 13-wk up market it had a +1.12% excess return over benchmark and for each down market it had a +1.41% excess return -- very good. However, in the 2007-2009 recession, it had a -64% max DD.

Instituting market timing:

close(0,#bench)>sma(200,0,#bench) AND sma(5,0,#SPEPSCY) > sma(21,0,#SPEPSCY)

decreased the max DD to -23%. This kept it completely out of the market through the 2001-2003 internet crash, the 2007-2009 recession, and the 2020 COVID dip.

Taz

Sorry, I forgot to include: the average excess return after instituting market timing did fall to +0.69% for each 13-week period, but still better than the benchmark.

Taz

(2.5) (Optional) Take the 50% of stocks closest to their 52 week high.
Jim - were you implementing step (2.5)? If so, does that change the percentages in step (3)? I was never clear how 2.5 might affect 3.

Not in this post, I was just trying to duplicate precisely what the old primary suggestion was.
I saved 22 variations, so p<.05 isn't much of a hurdle if you don't stick to whatever you consider the original : )
The rest of your steps seem to match my test.
I used ROE top 32% for the with-dividend version and top ROE 26% for the don't-care version.

A quick glance suggests that adding the "off high" test didn't make things better for the with-dividend version post discovery.
The version which did not require a dividend was fine with the "off high" test was improved a hair, but it's a very short post discovery period.
I would be hesitant to draw conclusions one way or the other. There is much to be said for KISS.

Jim

For this ROE_Cash screen, thoughts on buying top 12 stocks and HTD 20, as Elan had suggested a couple of weeks ago?

I'm sure there are lots of good variations.
I tend to stick with the "more stocks" version since it meets the original goal of being a SPY substitute. That requires some diversification.
Sharpening a screwdriver might make it a better awl, but it's intended to be a screwdriver so is that an improvement?

If you need an awl, of course, go ahead. But I was looking for a screwdriver.

Separately, aside from the bad metaphor, the screen doesn't seem to offer better returns with fewer stocks, so I didn't pursue that.
If your plan is to make 12%, it's better to make 12% with 40 stocks than to make 12% with 12 stocks. A surprise blowup hurts less, emotionally and financially.
If dropping 12 stocks indicated you might expect 5%/year more that's a different kettle of fish, but I didn't see that here.

Jim

A couple questions I have.
IS the screen Mungofitch listed similar or same as weekly screens on this board?
Which ROE is used quater, year or 5 year?
Where can I do my own screening, using which screening sites?
Does FT screener work for this, StOck Invetor Pro, Value Line, Portfolio 123 or other.
Which is best?

A couple questions I have.
IS the screen Mungofitch listed similar or same as weekly screens on this board?
Which ROE is used quarter, year or 5 year?
Where can I do my own screening, using which screening sites?
Does FT screener work for this, StOck Invetor Pro, Value Line, Portfolio 123 or other.
Which is best?

The ROE_Cash screen that is posted is similar, but not identical, to the screen I posted about which I call LargeCapCash outlined here
http://www.datahelper.com/mi/search.phtml?nofool=y...
I used the Value Line database to create and test it, and still do.
Their "Return on Shareholders' Equity" field is for one year, from the last full year annual statements available. They might do some adjustments, I'm not sure.
You can also use their figure for ROE latest quarter, which is I think the trailing four quarters (?), with very similar results in backtest.
It tests a bit better over a long backtest, a hair worse lately, close to a wash.
I have done more testing over the years with the annual figure, and there is lower turnover because it only changes once a year for each stock, so I tend to stick with that even though it is a little out of date.

If you wanted to do the screening yourself for precisely the screen I posted, you would need a subscription to Value Line.
It's hard to get the right product on line, usually you have to phone them and ask for the cheapest product that includes the "Analyzer" for the standard 1700 set of stocks.
You get a Windows program, and the subscription lets you download their database updates weekly and export it all to tab-delimited file.
There are multiple tools to proceed from there. I still use Radiscreen, the very old Excel macro.

As for which version (Value Line or SIpro or other) is best, I can't comment.
I have not really looked at the results of the SIpro version which is in the regular postings.
However, I might make the general comment that "translating" a screen to another stock database really does make it a different screen in the general case.

I have to say I am quite pleased with the first three years out of sample for the Value Line version.
(outperforming the S&P by 3-7%/year depending on the variant, when I would conservatively think something more like 1-2% would be expected in real life use)
In fact they have been so good that a couple of below-average years would not be unexpected : )
So far, it seems to meet the originally stated goal: a fairly diversified alternative to the S&P that has lower company-specific risk, and a reasonable expectation of slightly higher long run returns.
And not too much work or too much turnover.
(in a 40 stock portfolio HTD 45 run each two months, you replace under 1.5 stocks per month on average)

Jim

You can also use their figure for ROE latest quarter, which is I think the trailing four quarters (?), with very similar results in backtest.

I think it's just the number for the latest quarter, so typically it's a bout 1/4 of the annual figure. And I've found some inexplicable garbage in the quarterly numbers, so I've gone to using the annual ROE as you have.

As for which version (Value Line or SIpro or other) is best, I can't comment.
I have not really looked at the results of the SIpro version which is in the regular postings.

The SIPro version is no longer being posted or produced, since GTR1 is broken. The rankings that I post weekly now are VL based.

Elan

The SIPro version is no longer being posted or produced

Elan, I'm surprised you don't know this, since one is named after you :)

If I'm not mistaken, lohill is still posting the weekly SIPro Rankings on the MF MI board, including two versions of REO_Cash.
His weekly post is about half the traffic on that board :(

Here are last week's post.
Sorry, I don't know how to format nicely.

ROE_Cash_boris GOOGL MSFT TSLA CSCO BIIB VRTX HAL KLAC ACN MRNA COST MOH CF ADSK NVDA SWN FTNT M ANET ADBE GE BKNG EXPD IEX ROST
ROE_Cash_elann GOOGL MSFT TSM EQNR TSLA PDD CSCO BNTX NTES BIIB ASML VRTX HAL KLAC ABNB ACN MRNA COST MOH CF ENLC ADSK NVDA DQ UMC

https://discussion.fool.com/t/si-pro-rankings-2023...

Elan, I'm surprised you don't know this, since one is named after you :)

I'm surprised too. I don't go to the Fool discussion site any more. It's a ghost town.

Elan

Post New

US Policy FAQ

---

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