No. of Recommendations: 0
Thanks, Tim. ISRG also looks very good on the eye-ball test, and I see it too has a very low Sigma.

DB2
Print the post Back To Top
No. of Recommendations: 0
DB2,

How do you screen for those characteristics?
Print the post Back To Top
No. of Recommendations: 0
PYPL showed up in NYUA's Overlaps last month: http://boards.fool.com/overlaps-1013-32868805.aspx?sort=whol...

- zol
Print the post Back To Top
No. of Recommendations: 4
How do you screen for those characteristics?

Good question. I just keep the general idea in the back of my mind whenever I look at the chart of a stock. However, this being the MI board, there are more analytical approaches. :-)

I believe the first cut was taken by LorenCobb in 2000 with this post:
http://boards.fool.com/projected-growth-of-rs-picks-12684077...
Loren later corrected the spreadsheet errors here:
http://boards.fool.com/projected-rs-growth-revise-12699971.a...
and had a download available:
http://boards.fool.com/exponential-growth-ytd-backtest-12981...

He next tested the concepts with a few stocks from the RS screens:
http://boards.fool.com/bestworst-backtest-of-exponential-scr...
http://boards.fool.com/exponential-growth-ytd-backtest-12981...

areas looked at changes in volatility:
http://boards.fool.com/vlvolatility-13076424.aspx?sort=whole...

Et cetera. Some searches will turn up all sorts of dormant information.

DB2
Print the post Back To Top
No. of Recommendations: 0
PYPL showed up in NYUA's Overlaps last month

True, but AFAIK it hasn't been picked by our MI screens.

DB2
Print the post Back To Top
No. of Recommendations: 5
... PYPL doesn't seem to show up on any of our stock screens. I would think that it would.

Back then, Loren's (and others') approaches started with VL data, primarily T1 stocks. I don't have access to VL data, so I don't know PYPL's "timeliness" or any underlying data that helps define that (e.g., PE, projected growth, or whatever).

But, looking at SIPro data ...

-- PYPL has a high total return (and RRS slope, which is what I would use) and a relatively low volatility, both over the last year and last six months. Those are the measures that define what you're looking for.

-- But, it has a PE (as of 11/17 data) of 59. That may not exclude it from being a "T1" (or perhaps "T2") stock, but it could. And PE (or earnings yield) play a role in many VL and SIPro screens used here. Also, I recall that Elan used a T1/T2 screen of this type for his ongoing 6/3 option picks, so maybe he can comment if PYPL (or other stocks below) show up on his option screens.

-- As an FYI ... I looked at the SIPro stocks I started with last week (about 2,500) using daily data and measures. I see PYPL as the 17th highest on the "RRS 252 - 2 Sigma" sort. Here's the list:

  Sym         P/E         Slope        Sigma       RRS - 2 Sigma
-------- -------- ---------- ------- ----------------
WTW 26.90 1.85 60.1% 90.6%
EVRI 1.51 53.0% 56.7%
RACE 34.40 0.89 22.5% 55.3%
SGMS 1.26 43.0% 49.9%
SQ 1.07 33.8% 49.0%
GDOT 44.90 0.95 28.3% 47.0%
ALGN 77.10 0.98 30.7% 44.9%
TTWO 107.70 0.93 29.1% 41.7%
UPLD 1.07 36.4% 40.0%
NBN 14.20 0.87 27.1% 39.0%
IPGP 34.90 0.83 25.2% 38.9%
APPF 264.00 0.91 29.5% 38.4%
BA 24.10 0.65 16.7% 36.7%
TREE 114.60 1.14 42.1% 34.7%
ISRG 50.70 0.65 17.4% 34.7%
CTRL 62.20 1.18 44.4% 33.9%
PYPL 59.40 0.68 20.0% 33.0%

I believe Loren looked at 126-day return (rather than the 252 days above). When I use RRS 126 - 2 Sigma, PYPL drops to 53rd on the list.

Regards,
Tim
Print the post Back To Top
No. of Recommendations: 0
Thanks, Tim. ISRG also looks very good on the eye-ball test, and I see it too has a very low Sigma.

DB2
Print the post Back To Top
No. of Recommendations: 0
Back then, Loren's (and others') approaches started with VL data, primarily T1 stocks. I don't have access to VL data, so I don't know PYPL's "timeliness" or any underlying data that helps define that (e.g., PE, projected growth, or whatever).

PYPL is currently listed by VL with Tim=3.

Elan
Print the post Back To Top
No. of Recommendations: 0
pe: got it.
slope: slope of what to what? is this rrs?
sigma: ?

thanks for any explanation. i think the fundamental idea is likely fruitful.
Print the post Back To Top
No. of Recommendations: 11
slope: slope of what to what? is this rrs?
sigma:


"slope" as shown was RRS over 252 days.

"sigma" is volatility measured over 252 days, using daily data.

A blend of Loren's terminology (sigma) and mine, from when I introduced RRS in 2000.

Although I used my own data and tools to obtain the list of stocks and measures I showed, the following shows in GTR1 the essence of what I did. Which, in hindsight, I should have used in the first place.

http://gtr1.net/2013/?i1f0.4::styp.a:ne30!31!14!15!44!45!73!...

You can run the backtest back to 1926 (nothing special when you pull from all stocks); you can click on "Run Screen" to obtain a list similar to (but not exactly the same) as what I showed. No password is needed because this screen uses no SIPro or other restricted data fields.

Regards,
Tim
Print the post Back To Top
No. of Recommendations: 8
A very early on Posting of work by Tim that I had saved.

My overall conclusion for using RS screens is to combine both improvements: Exclude the top 10 RS-5 day stocks, and measure relative "strength" based on the slope of the regression line. I would not bother with the +/- sigma approaches based on the above evidence.

In addition, I would consider combining at least two RS periods. Although the high results for any particular RS period may be (are?) spurious, I would use RS 189 and RS 42.

Other Thoughts

1. I have said I would also look at using the regression approach where returns are net of the S&P500 index. I have done some preliminary analyses, but the early indications are that results are very similar to those above. I need to drill down further to ensure I'm not doing something wrong, and to try an alternate approach (Log-returns regressed on log S&P500 returns, as suggested by LorenCobb). I will report back.

2. Someone posted a response to a prior Daily Data post asking if I had any spreadsheets they could look at. In a reply to this post, I will describe what I will put out on my Yahoo briefcase for others to look at. This will include a list of individual stocks and monthly returns for one of the RS periods above, and the detailed spreadsheet I used for one of the 180 months to do the calculations and sorts to select the 5 stocks for that month. This will show what was done for each of the 180 months to generate the overall results.


Regression and Sigma Calculation Methods

A brief documentation of how the regression was performed:

o For each stock in each month, daily returns were obtained from Sux's data so that data was available back to 260 days or so prior to the evaluation date and 70 days after. The natural logs of these returns were stored.

o "Pseudo" log-prices were derived as the cumulative sum of the log-returns, with the first log-price simply equal to the earliest log-return. This was done because the returns are the best item to use from Sux's data. They include the effect of dividends and splits, but the prices don't. (The data base tells you when a split occurs, but the price is then cut in half, for example. The returns correctly reflect splits.)

o A column of integers is associated with the log-prices, starting at 1 and going to 330 or so.

o Using the excel "Slope" function, the slope of the regression is determined by regressing the log-prices on the integer trading days. The cells selected are parameterized to choose the designated "evaluation date" in Sux's Value Line files and the designated number of days to look back (e.g., 126).

At the same time, the total return and historical volatility are measured for the correct look-back period from the log-returns.

To determine the "tightness of fit" of the regression, the standard error of the estimate (s) is probably the most useful value, but I chose not to use it when addressing the "+/- Sigma" approaches.

o "s" did not seem to translate well into an annualized value that matched the general range of historical volatility (HV).

o I tried using an adjusted "s", which essentially used the ratio of average HV to average "s" for the 100 stocks, times "s" for each stock. Results differed slightly from above, but showed similar patterns and led to the same conclusions.

o I examined using the formulas for predicting values of Y for given X (Y_hat). Since we are predicting so far outside of the range of the X's used in the regression, this error was not useful as a standard deviation tool. It was much smaller than the HV used above.

So, there may be some way to use a more appropriate tool for measuring tightness of fit in adjusting b, but I'll leave that for other interested readers.


My apologies for the length of the post, but I thought this one was worth it and deserved the documentation.

Regards,

Tim
Print the post Back To Top
Advertisement