No. of Recommendations: 0
I am interested in exploring the idea of weighting the PEG5 based on it position in the screen. This would be similar to the foolish 4 in Robert Sheard's book where two of the stocks are 33% and two are 17%. If with backtesting we could discover that for example the PEG stock in position 3 has a better historical performance than the other positions, we could weight that stock heavier in our portfolio. Has this ever been researched? I would love to look into this idea, but I don't want to reinvent the wheel. Also, any good tips on finding historical quoutes of stock prices back to 1986?

centralbev
Print the post Back To Top
No. of Recommendations: 0
Not that I am recommending it, but post #78079 just took a look at the impact of PEG by position.

-Ernie
Print the post Back To Top
No. of Recommendations: 5
I am interested in exploring the idea of weighting the PEG5 based on it position in the screen. This would be similar to the foolish 4 in Robert Sheard's book where two of the stocks are 33% and two are 17%. If with backtesting we could discover that for example the PEG stock in position 3 has a better historical performance than the other positions, we could weight that stock heavier in our portfolio. Has this ever been researched?

Interesting idea I fiddled with as well, then dropped. (moral is, share negative results, so other folks don't pursue the same dead-ends!) Here's what I've got:

If you datamine and optimize on CAGR, the weights for each of the top 5 of PEG 13 should be

Position 1: 3%
Position 2: 97%
Position 3-5: 0% (CAGR 84, GSD(M) 56)

Optimizing on Sharpe Ratio instead, you get:

Position 1: 14%
Position 2: 35%
Position 3: 15%
Position 4: 4%
Position 5: 6%
Remainder in cash (CAGR 53, GSD(M) 27).

The preference for position 2 and for the top 3 positions are well established in the screen selections chosen by most folks on the board for blends & switches.

You could try these weightings, I suppose. You also might use them to guide rebalance decisions (ex: let your winners run as long as they are moving up toward position 2, trim them when they hit position 1 or drop back 1 position). I wouldn't.

My basic feeling is that while there maybe some underlying structure here, it's mostly datamined nonsense. Guess I feel that way about Sheard's stuff, too. From the asset allocation thread, we know that proper weighting depends a great deal on correlation, and there seems like no theoretical ground for assuming anything about the correlation between stocks in different screen rank positions, regardless of the accidents of history.

I tried extending to more tests: If you run the same Sharpe optimization on the 10-stock PEG13, you get a similar weighting of positions 1-5, 0% on position 6, but position 7 gets weighted the same as position 1. Huh?

PEG13:10 (optimize on Sharpe)
Pos1:13%, Pos2:33%, Pos3:14%, Pos4:3%, Pos5:5%, Pos6:0%, Pos7:13%, Pos8-10: 0%. (CAGR 53, GSD 27)

I also tried similar analysis on RS13:10.

RS13:10 (optimize on Sharpe)
Pos1:7%, Pos2:16%, Pos3:0%, Pos4:16%, Pos5:15%, Pos6:0%, Pos7:0%, Pos8:7%, Pos9:0%, Pos10:6% (CAGR 39, GSD(M) 25).

RS13:10 (optimize on CAGR)
Pos2: 25%, Pos4:13%, Pos5:62%, all other positions 0%. (CAGR 68, GSD(M) 50)

You can see how weird it gets, and how much it shifts around. That's just random statistical noise. The likely underlying structure may be that the top half does better (ie. overlaps and SOS help), and perhaps the very top stock might have a greater exposure to a correction/fall. Otherwise, it's not useful.

Things like LorenCobb's Exp. Growth and BarryDTO's individual stock efficient frontier models strike me as much more robust and theoretically sound mechanisms for applying asset allocation weighting to individual issues. But it's certainly good to pursue all ideas!

-gelasmus

P.S. I think Sparfarkle is working on a money management model to pursue a portfolio of "bullets" like the position 2 stocks above; there might be something to that.

Print the post Back To Top
No. of Recommendations: 2
>>I think Sparfarkle is working on a money management model to pursue a portfolio of "bullets" like the position 2 stocks above; there might be something to that.

Funny you should mention that. I've been thinking a lot about portfolios of bullets lately. I have found seven annual screens I like with bullets from 40-68% (mean, not cagr--mean is a lot more useful when you're thinking of a portfolio of bullets). I've been considering a dozens approach using these screens for my taxable account.

I like the idea of having the top of several strategies. It's not something I'm committed to yet, but it's something I'm thinking about. Has anyone else thought much about this?
Print the post Back To Top
No. of Recommendations: 0
centralbev.

First, most of us read both the FW and the MI boards. There is no reason to double post your messages on both boards.

If with backtesting we could discover that for example the PEG stock in position 3 has a better historical performance than the other positions, we could weight that stock heavier in our portfolio

That's the kind of activity that kills canaries..:>)

OxBeaux
Print the post Back To Top
No. of Recommendations: 0
The preference for position 2 and for the top 3 positions are well established in the screen selections chosen by most folks on the board for blends & switches.

This inspires an interesting idea for Jamie's backtester. Oh Jamie ...
Print the post Back To Top
No. of Recommendations: 1
I recently posted an analysis of cagr by screen position (#75852). I also compiled the same data for gsd, if anyone is interested.

http://boards.fool.com/Message.asp?mid=13039263

-eremon
Print the post Back To Top
No. of Recommendations: 3
I like the idea of having the top of several strategies. It's not something I'm committed to yet, but it's something I'm thinking about. Has anyone else thought much about this?

Yes. I'm using the MIScreenBrowser as a lab for developing and testing a composite screen index. Right now the index can express any mix of all screens, with weightings based upon a combination of:

• screen rankings
• historical CAGRs for each screen
• smoothed attenuation of returns according to rank for each screen
• a user-specified ranking cutoff (i.e., 1-5, 1-10, 1-25, etc.)
• sledge-hammer weighting applied by the user

With these variables as inputs, the browser produces a list of stocks ranked by according to this customized index for each weekly period going back to 5/28/99. Returns for the index may then be viewed in another table that may be customized for #stocks held, holding period, beginning date, and friction.

I intend to add in the exponential screens and screen switching strategies at some point, but haven't yet figured out how to do that.

I would also like to use some method, such as the recently discussed impact analysis, to refine the weightings among the various screens. This, again, is in the future.

So far, the results are quite encouraging. The TTM return for a 5-stock/1-week hold using the entire battery of MI screens is 436%, with 52 trades required.

I'm also looking into the impact of reducing the number of trades through various holding rules, such as holding for up to two weeks if a stock stays below a specified rank, such as 7, for example, in a five-stock screen.

-eremon
Print the post Back To Top
No. of Recommendations: 1
Eremon wrote:
I also compiled the same data for gsd, if anyone is interested.

Yes, please! I found your tables of CAGR to be very useful for planning blends, etc., and the GSD tables would be helpful for this too.

Thanks for your hard work.

-Jim
Print the post Back To Top
No. of Recommendations: 0
I found your tables of CAGR to be very useful for planning blends, etc., and the GSD tables would be helpful for this too.


OK, Jim. The GSD table has been posted as message #78284, with the subject of "GSD by screen position for backtested screens."

-eremon
Print the post Back To Top