Investment Strategies / Index Investing
No. of Recommendations: 10
No. of Recommendations: 19
3 steps? Ha! We laugh at danger.
Let's try one step:
I still use the VL databases. Call me a dinosaur.
The one step:
([Sales Growth 1-Year] + [Sales Growth 5-Year]) * [Cash] / [Shareholders Equity] top 20
Monthly after friction 1997-2023 inclusive beat S&P 500 by 8.3%/year in backtest.
It beat the S&P in each of calendar 2020, 2021, 2022, 2023. That has been a tough bogey lately if you didn't own the Magnificent Faangs.
If you're willing to countenance one more step, adding a final "bottom 15 by beta" is nice.
Brings the advantage over SPY to 10.8%/year and it's a very much smoother ride. Lower rolling year downside deviation than the S&P, rather than higher.
Almost as good returns at 6 month holds, too, still a double digit advantage.
In backtest, anyway : )
What, you still want a third step?
Top 10 by dividend yield is even higher returns. Advantage over spy 12.5%/year. (but dodgy result because there aren't always 10 dividend paying stocks among the low 15 by beta)
Jim
No. of Recommendations: 0
Let's try one step:
I still use the VL databases. Call me a dinosaur.
The one step:
([Sales Growth 1-Year] + [Sales Growth 5-Year]) * [Cash] / [Shareholders Equity] top 20
Is that the VL 1500 stocks with timeliness?
No. of Recommendations: 0
Do you use it yourself?
If not: Why?
No. of Recommendations: 0
I get SPY as 9% per year since the beginning of 1997 to the end of 2023.
So CAGR for 1 step is 9+8.3=17 3% Nice! Although I might quibble that this is actually 2 steps. Step 1, stocks in Valuline, step 2 sales growth.
2 Step 19.8%
3 Step 21.5%
Craig
No. of Recommendations: 0
What does it look like if you use
[Sales Growth 5-Year]
instead of
[Sales Growth 1-Year] + [Sales Growth 5-Year]
Asked because I can find 5yrSG but not 1yrSG.
No. of Recommendations: 0
Jim
When you say "If you're willing to countenance one more step, adding a final "bottom 15 by beta" is nice."
Are bottem Beta the most negative, or the closest to zero?
Craig
No. of Recommendations: 1
Jim, Could you list current stocks in your screen?
-----------------------------------------------------------------------------------------
Let's try one step:
I still use the VL databases. Call me a dinosaur.
The one step:
([Sales Growth 1-Year] + [Sales Growth 5-Year]) * [Cash] / [Shareholders Equity] top 20
Monthly after friction 1997-2023 inclusive beat S&P 500 by 8.3%/year in backtest.
It beat the S&P in each of calendar 2020, 2021, 2022, 2023. That has been a tough bogey lately if you didn't own the Magnificent Faangs.
If you're willing to countenance one more step, adding a final "bottom 15 by beta" is nice.
Brings the advantage over SPY to 10.8%/year and it's a very much smoother ride. Lower rolling year downside deviation than the S&P, rather than higher.
Almost as good returns at 6 month holds, too, still a double digit advantage.
In backtest, anyway : )
What, you still want a third step?
Top 10 by dividend yield is even higher returns. Advantage over spy 12.5%/year. (but dodgy result because there aren't always 10 dividend paying stocks among the low 15 by beta)
Jim
No. of Recommendations: 7
Jim, Could you list current stocks in your screen?
I think they would be these
HRB SRPT BKNG NSP RCL CNK IBKR ORCL CCL MELI VVV MSI UAL DVAX ENPH STX SE DK LNTH CELH
[See recent post on garbage data from VL this week on prices...with luck it doesn't affect these specific fields]
However, arguably the screen definition has a bug: it really should start with (yes) another starting step to require the ROE and cash fields both to be positive before doing the math.
Otherwise you could pick a firm with a negative equity and negative cash, but the quotient is positive.
(with those two both positive, the sales growth figures will take care of themselves in the sort)
This makes a difference at rank 22 this week (which has negative equity and negative sales growth) so at least it doesn't affect the tickers above.
In the backtest there is a difference in the top 20 picks in 80% of months by adding the positive field check, but the CAGR difference is within statistical error at 0.64%. Unsurprisingly the win rate goes up a bit. Who wants to buy the "wrong" stock?
Jim
No. of Recommendations: 4
Are bottom Beta the most negative, or the closest to zero?
VL has only positive betas in their database, so it's closest to zero from the positive side: the smallest positive numbers.
To overgeneralize, as beta falls down towards 1 you get stocks with lower volatility, and as it falls even further you start to get those that [also] don't have much market correlation.
I suppose it's theoretically possible to have a common stock that has negative beta, but I think in the real world you would only see that in an inverse fund or similar. In any case, VL doesn't have any.
Jim
No. of Recommendations: 3
Is that the VL 1500 stocks with timeliness?
No timeliness requirement, just whatever is in their standard "1700 stock" database subscription.
About 1603 stocks lately...it varies.
Once you eliminate those that have one or more of the four required fields not populated, there are currently 1210 stocks that go into the sort step.
If you add the sensible step of requiring that the cash and equity fields both be positive (which I just thought of, oops), there would be 1145 going into the sort.
Jim
No. of Recommendations: 7
Do you use it yourself?
If not: Why?
Because it's new.
I like to have screens that have at least some out-of-sample validation. A screen which at least did not take the first opportunity to blow up spectacularly after it was created.
Also I haven't been doing much MI lately, other than creating, tracking, and analyzing screens. For reasons which are not "I don't believe in it any more".
Jim
No. of Recommendations: 1
I suppose it's theoretically possible to have a common stock that has negative beta, but I think in the real world you would only see that in an inverse fund or similar. In any case, VL doesn't have any.
I am not sure how GTR1 calculates beta, but if you limit the universe to SP500 stocks and use beta(1,253), there are up to 42 stocks that have negative beta. For instance, 2001 Feb to October had a lot of negative Beta stocks. At the moment, there is zero SP500 stock with negative beta.
My understanding of beta from Investopedia is Beta = 1, stock moves in line with market, beta = 0, stock moves don't correlate with the market and beta = -1, stocks move in opposite direction of market (I seem to have a lot of these at the moment :) ).
Craig
No. of Recommendations: 7
What, you still want a third step?
Top 10 by dividend yield is even higher returns. Advantage over spy 12.5%/year. (but dodgy result because there aren't always 10 dividend paying stocks among the low 15 by beta)
I created a paper portfolio of this screen on 3/9/2024. The stock SWAV was in the 1st & 2nd steps, but not in the third. It was taken over at a 32% gain.
As of 6/6/2024 the gains (including SWAV) for 1, 2, and 3 step are:
4.95%
8.88%
12.57%
SPY gain: 4.48%
When there are fewer than 10 dividend paying stocks for step 3, the remaining slots are filled in with the top stocks of step 2.
Quite a range of returns:
zdiv-10 Gain
NSP -4.40%
IBKR 13.62%
MRNA 50.14%
ZS -9.95%
CRWD 5.46%
DVAX -6.62%
TMDX 58.83%
BILL 25.44%
CPRX -3.02%
GH 47.09%
Avg 12.57%
No. of Recommendations: 0
Rayvt
I created a paper portfolio of this screen on 3/9/2024. The stock SWAV was in the 1st & 2nd steps, but not in the third.
When you use low Beta in step 2, do you use negative Beta stocks in the calculation or just positive. At the moment it looks like there are only a few stocks with negative beta, but sometimes there can be quite a few (in the SP1500, not sure about Valueline).
Aussi
No. of Recommendations: 1
When you use low Beta in step 2, do you use negative Beta stocks in the calculation or just positive.
A numeric sort, so any negative beta stocks would be included. But none of the VL Timeliness 1-5 stocks are negative. The lowest beta is 0.4.
The full ~5000 VL stocks has only 8 with negative, and none of them are T1-5.
No. of Recommendations: 6
If you're willing to countenance one more step, adding a final "bottom 15 by beta" is nice...
Another good final sort would be Top N by [Sales Growth 5-Year].
Yes, it's already in the main step, but it seems to give a nice final sort...good apparent falloff in returns by depth down to "top 4".
I noted this at the time but neglected to mention it in the first post about the screen.
Off to a nice little start--
First four months of this year, top 4 with two month holds after friction, up 26.5% versus 10.3% for the S&P.
Not that I would recommend running a 4-stock screen. Or placing any weight on a four month anecdote. But it works nicely in a long backtest too, e.g. top 15 2 months after friction beat S&P by 6.5%/year 2000-2023. In backtest.
Note, it's a screen that goes down along with the market in bears.
Jim
No. of Recommendations: 2
How about this as a combined twist?
Step 2 is bottom 15 by beta
Current step 3 is top 10 by dividend yield.
Step 4: Using the 15 by beta, sort by [Sales Growth 5-Year]. Take the top 4 (or 5) plus the 10 from step 3, giving a possible 14 or 15 stocks.
Using my most recent data, there is quite a few stocks in common between steps 3 & 4.
Top 5, 3 in common and 2 additional.
Top 4, 3 in common and 1 additional.
Top 2, 1 in common and 1 additional.
Top 10, 7 in common and 3 additional.
I guess you could buy a double weight of the ones in common. Or just buy the new ones, for a varying size portfolio of 10 to 15 stocks.
No. of Recommendations: 6
two steps:
Nasdaq100 stocks
lowest 5 by ratio of 1 week/1 year return
19850201 to 20240605
trade every 60 days
assume .2% friction
cagr 23.9%
sharpe .69
beta 1.4
You can improve drawdown with the timing screen often used here.
The data is all free at sites online that list the 101 stocks (Google is in the list twice) and the 1 week and 1 year return.
200 days instead of 253 works also.
https://gtr1.net/2013/?h60f0.20000::nas100.a:nenul...
No. of Recommendations: 1
two steps:
Nasdaq100 stocks
lowest 5 by ratio of 1 week/1 year return
What is the reasoning behind this? Is it written about anywhere?
Some comments:
60 market days is more than 2 months. When the parameters are changed to 42 market days and the period of 20000103 to 20240530 the CAGR is 12.5% with MDD -79.9%.
21 market days (1 month) is CAGR 13.6%
SPY is CAGR 7.3%, MDD -55.2%
QQQ is CAGR 7.3%, MDD -83%
Interesting that SPY and QQQ are so close.
No. of Recommendations: 2
two steps:
Nasdaq100 stocks
lowest 5 by ratio of 1 week/1 year return
What is the reasoning behind this? Is it written about anywhere?
Pure momentum screen with a penalty for recent good performance and reward for recent poor performance
last step: (5 day return) divided by (1 year plus 10 or 12 days return) bottom 5. So the lower the recent return and the higher the 1 year return the better.
Similar returns with a 15 day hold.
The theory, pick a group of stocks that has generally performed well over the years and then pick "x" number of the top performing stocks within that group.
Aussi
No. of Recommendations: 3
I get same return at 20,30,40,50, 60 days.
To reduce friction infrequent trading is better so if you get near the same return,
trade less.
The idea is to start with a good universe of stocks, like WER does.
And plenty of studies show reversal at the week level, so a bad last week is good.
And others that intermediate positive momentum periods are good.
So you are getting stocks that have done well this year except for last week, and expecting them
to bounce back from 1 bad week.
No. of Recommendations: 3
Pure momentum screen with a penalty for recent good performance and reward for recent poor performance
Ah.
These was a similar screen back in the late 90's called FlareOutGrowth, from a writer on MSN Money. IIRC, that was 12 month momentum + 6 month momentum MINUS 3 (or maybe 1) month momentum. Same rationale.
The theory is a high-moving stock(s) that simply had a recent stumble, presuming that the short term pullback will be reversed.
It worked until it didn't. Like, sadly, so many MI screens.
No. of Recommendations: 2
200 days instead of 253 works also.
If I understand the backtester right, the screen is actually set up for 263 days, instead of 253. When I set it to 253, the screen works just as well, and as you note, it works just as well at 200 days.
Going back to the 263 day set up, and resetting the hold period to 42 days and then to 63 days (~3 months), the screen works just as well.
Seems like a pretty good mound of toast. Of course, how it works into the future might vary somewhat from the backtest results. Still looks worth serious consideration: "works until it doesn't" is the nature of life.
Eric Hines
No. of Recommendations: 9
And plenty of studies show reversal at the week level, so a bad last week is good.
True enough.
But, anecdotally: though this may be true on average, I find it is so unreliable that I have removed the negative one-week factor from all my screens. There seem to be multi-year stretches that good one week performance is good, not bad, for the next month.
The better improvements to momentum I found were (again, anecdotally from trading and backtesting)--
* the optimum length of lookback seems to vary with a stock's turnover ratio. i.e., the optimum look back seems to be some constant K times the average trader's hold period.
* lookbacks work better for some classes of stock with a LOT of lag than they do with fresh data, especially long lookbacks. i.e., instead of a negative factor on the most recent week (or more), just ignore the most recent week (or more). Even up to six weeks of lag for some boring value/dividend stuff.
I don't know if those comments have any merit, but I thought I'd share 'em.
Jim
No. of Recommendations: 4
Of course, how it works into the future might vary somewhat from the backtest results. Still looks worth serious consideration: "works until it doesn't" is the nature of life.
With my luck, it stops working as soon as I commit significant money to it. ::sigh::
No. of Recommendations: 5
In addition to look back duration, for smaller stocks like the SP400 universe, using GTR1, eliminating biotech and mining stocks improves the backtest results. My thoughts are that both groups are more prone to sudden jumps and drops than the average stock due to news releases rather than long term trends. I did not find any benefit eliminating mining and biotech stocks from the SP500 universe when checking momentum screens.
Antidotally, I think each group of stocks, SP500, NAS100, SP400, etc., have different momentum characteristics (look back duration, hold period and type of stock to exclude from constituents).
Aussi
No. of Recommendations: 9
Pure momentum screen with a penalty for recent good performance and reward for recent poor performance
Ah.
These was a similar screen back in the late 90's called FlareOutGrowth, from a writer on MSN Money. IIRC, that was 12 month momentum + 6 month momentum MINUS 3 (or maybe 1) month momentum. Same rationale.
It's more like what I do with the 6/3 options. Eliminate the top 10% of stocks with the highest 5 day return, then sort the remaining stocks by Regression Relative Strength. So it's eliminating the very short term price pops, not the recent 3 or 1 month top performers.
Elan
No. of Recommendations: 0
Regression Relative Strength
How is this calculated?
I think I've been here before, but don't recall.
Thanks.
No. of Recommendations: 10
Regression Relative Strength
How is this calculated?
Conventional relative strength for 26 weeks, for example, is calculated by ranking the group of stocks by their 26 week total return.
Regression relative strength uses a linear regression (least mean squares) through the (log) prices over the last 26 weeks, and ranking the stocks by the slopes of their regression lines. The effect is to give less weight to the end points of the total price trend, and looking instead at the price trend line.
Elan
No. of Recommendations: 1
Thanks Elan.
Regression relative strength uses a linear regression (least mean squares) through the (log) prices over the last 26 weeks, and ranking the stocks by the slopes of their regression lines. The effect is to give less weight to the end points of the total price trend, and looking instead at the price trend line.
No. of Recommendations: 4
Just noted today that for my 2 step Nasdaq100 strategy you need to watch out for the fact that 50/ -1 = -50/ 1 so you can get anomalies if you don't exclude 1 year returns that are negative, since we don't want a low ratio of 1 wk/ 1 yr based on last week being a big positive return divided by a negative annual return.
If you require 1 yr>0 it is better to avoid this. That way a low ratio of 1 wk/ 1 yr will mean "bad last week, good in the last year".
No. of Recommendations: 2
Just noted today that for my 2 step Nasdaq100 strategy you need to watch out for the fact that 50/ -1 = -50/ 1 so you can get anomalies if you don't exclude 1 year returns that are negative, since we don't want a low ratio of 1 wk/ 1 yr based on last week being a big positive return divided by a negative annual return.
If you require 1 yr>0 it is better to avoid this. That way a low ratio of 1 wk/ 1 yr will mean "bad last week, good in the last year".
I don't think the GTR1 screen you referenced gets negative returns. I think the return is just less than 1. (tr(1,265) varies from 0 to infinity).
Aussi
