O'Shaughnessy formula

Research
1

In the financial press today is an article from James O’Shaugnessy and what he describes as the best quant formula for the past 50 years.

In a nutshell

Trend Value strategy

• Price-to-Sales

• Price-to-Earnings

• Price-to-Book

• Price-to-Cash Flow

• EBITDA/Enterprise Value

• Shareholder yield (dividend yield + rate of share repurchases)

Each stock in the universe gets a score of 1 to 100 for each of these factors. The final value score is an average of these scores. The Trending Value portfolio narrows the investable universe to the 10% of stocks with the best score based on the value composite, and then selects a concentrated portfolio of 25 stocks based on trailing six-month momentum.

This simple combination builds a portfolio of extremely cheap stocks that are on the mend. The combination of value and momentum works better than either of these factors on its own.

Examples of stocks currently passing this strategy are Advance America Cash Advance Centers Inc. AEA -1.15% , Tesoro Corp. TSO -3.98% , and Smithfield Foods Inc. SFD +3.15% .

Of course, we certainly expect there to be stocks in our strategies that underperform, and these three stocks may not perform well in the coming year. But we also know that a diversified portfolio of stocks with the characteristics these stocks currently posses has fared extremely well over time. By focusing on the long term, and taking advantage of market volatility rather than being scared off by it, investors can beat the market in the long term and achieve their investing goals.

James O’Shaughnessy is Chairman and CEO of O’Shaughnessy Asset Management, and author of “What Works on Wall Street.” Patrick O’Shaughnessy is a Research Analyst at O’Shaughnessy Asset Management.

2

This is mentioned on the web in several places:

Marco - you guys should create a public sim with this ranking system and post comments on all the sites that discuss it with links to the P123 sim. This will be a popular book so dont forget to write a review on Amazon and put the link there also. A good opportunity for free advertising.

3

I took a shot at recreating the system in P123:
http://www.portfolio123.com/port_summary.jsp?portid=854248

I dont have the book, so I had to make some assumptions regarding rebalance period, turnover goals, liquidity, etc. Also, I dont know if the ranking factors were vs Industry or not. I split the p/s factor into 2 - one is by industry and one is not.

The sim and rankings system are public, but let me know if you have trouble viewing it. The results are nothing to get excited about, but I’m sure it can be greatly improved with some tweaks.

4

Take a look at Denny’s “Advanced Momentum Value” port it’s comparible

5

Dan: [quote]
The sim and rankings system are public, but let me know if you have trouble viewing it.
[/quote]
The public ranking I see has only 1 factor.

Glenn

6

Hi Glenn,
The sim uses 2 ranking systems. The one you are looking at has the 26 week RS. The sim will pick the top ranked stocks based on that RS.

But it also has a buy rule: Rating(“Oshaugnessy Trend Value_value”)>=90
This rule means that the list of stocks is first filtered down to only those that have a rank >= 90 for this 2nd ranking systems called “Oshaugnessy Trend Value_value”. That RS is public also.

I believe this setup replicates the system based on the articles I read. If anybody has more details on it, or if there is something I did wrong, please post info.

7

Dan:

OK. Thanks. Copying the second RS fixed things.

Glenn

8

All of the factors are relative to the universe, not industry. Minimum market cap is $200mm.

9

thanks for the info Chris. I updated the RS and the sim. The mktcap>=200 rule affected the performance a bit.

10

All:

The Market Watch article above claims O’Shaughnessy’s Trend Value " is the best performing strategy since 1963" and “its annualized return of 20.58% … crushes the All Stocks benchmark, which has a return of 10.71%.” I think Dan’s sim and a P123 public sim crush the crusher!

Below are three sims. The first (TV1) is Dan’s sim which I think is close to O’Shaughnessy’s Trend Value. The second sim (TV2) adds timing and crushes O’Shaughnessy’s 20% returns. The third sim (TV3) is Denny’s public sim Advanced momentum w/value 10 Stks b3s5c1 (with some liquidity changes to make it comparable). This is best and uses no timing.

Kind of interesting that a 5 year old P123 strategy is better than the current “best performing strategy since 1963”. Also, interesting that all of these strategies have not done much over the last two years.

Glenn

TV1.png

TV2.png

TV3.png

11

One of the key takeaways from O’Shaughnessy’s book (yes, I read all 600+ pages of it) is that even a ten-year backtest period for a strategy is relatively meaningless in the big picture. There are strategies that perform wonderfully well for 5-10 years and stink for the next five-ten years.

I’ve seen the same effect with many of my ports/strategies over the years… some work great for a couple years then underperform going forward. Personally I don’t have the intestinal fortitude to ride out a 30, 40 or 50% drawdown and still have confidence to stick with a strategy. But that’s precisely what O’Shaughnessy says is the key to long-term outperformance… having the confidence and discipline to stick with your strategy through thick and thin.

He tests the strategies on two different data sources… one going back to 1963 and one going back to the 1920s. I’d be interested in seeing what some of the stronger strats, like Denny’s Momentum/Value, have done over the same time period for a true apples:apples comparison. (But, I also know my long-term time frame isn’t 40 years, barring some major developments in the medical industry that extend lifespans.)

One other tidbit he discusses concerns microcaps. He mentioned that it’s easy to build a simulated strategy with terrific returns… it probably would be oriented to microcaps. However, it is all but impossible to actually trade that strat live due to the market impact the trading would have on the price. That bit of wisdom comes as no surprise to P123 members.

1 Like
12

Toni,
wise words!

I have the same overall feeling. J just got the O’Shaughnessy book and have only glimpsed at it. But it is already clear to me that no real human being can stand a 50% drawdown and still be a happy camper. The simple fact is: Strategies and their effectiveness come and go, what works today stops working tomorrow.
I too have a few strategies that work well with microcaps but are not tradeable in the real world.
Have yet to come across a sim that works well with good volume (> 1 mill. daily liquidity).
The only thing that would really make a difference in my humble opinion would be a reasonably reliable Timing indicator. But Marco has said here somewhere that even he does not have one or cannot come up with one.
So, what is left???

13

Hi Koronbock,

I believe you are totally correct when it comes to identifying market turn points to improve portfolio performance. Using carefully created exposure lists, I have found that total return of a port/sim can be increased dramatically, sometimes by as much as 50%, while drawdown can be cut in half or better, as a result of avoiding performance-killing drawdowns.

To use these dates, create a new Exposure list under “My 123/Lists” that includes these short-term (daily) dates for market down/market up exposure. The list makes its first purchases on or after 3/15/2001:

01/01/01 03/15/01
06/01/01 09/15/01
12/31/01 02/28/02
04/15/02 10/15/02
11/30/02 03/24/03
01/28/04 03/28/04
04/09/04 05/25/04
07/01/04 08/12/04
10/02/04 10/28/04
01/05/05 02/03/05
03/08/05 04/28/05
08/02/05 08/30/05
09/09/05 10/22/05
12/27/05 02/07/06
05/11/06 07/14/06
12/27/06 01/27/07
02/24/07 03/19/07
06/04/07 08/21/07
10/18/07 11/26/07
12/26/07 01/28/08
05/20/08 07/17/08
09/01/08 12/01/08
01/05/09 03/10/09
06/15/09 07/13/09
04/26/10 07/06/10
11/16/10 12/02/10
02/22/11 03/25/11
05/02/11 06/28/11
07/26/11 10/06/11
11/06/11 12/23/11

Want less frequent trading? Here are the medium-term (weekly) timing dates I use:

08/25/00 04/01/03
07/26/07 04/02/09
05/02/10 09/05/10
07/26/11 01/01/12

Notice that these weekly dates would have put you in cash from July 26 to present, which would have allowed you to miss the August cliff and the volatility since. The last date (market up) is set for next Sunday. P123 requires a set of dates in the exposure list, so you can’t leave the final right-side date blank. Therefore, I put in a date in the future since the timing indicators are not yet signaling ‘market up.’

If the market continues to rise this coming week (as I expect), then the 01/01/12 date is correct as a weekly view of the indicators should signal ‘market up’ next Sunday.

BTW: The last time I shared my exposure dates on P123 there were comments about how I had revised the dates that had previously been posted and I was “curve fitting” or “data mining.” It is correct that the dates I posted on p123 several months ago were slightly modified from the first set of dates created. These dates are also updated, as there have been two revisions to the original dates.

But like almost any human creation, the indicators were improved when an improvement appeared to be necessary and highly beneficial. Just as in many product revisions, significant improvements have been made. If we were forever forced to commit to the first things we created, we would still be driving Model T’s and waiting for our paper stock certificates to come in the mail.

I’ll say it first; these dates are revised and are ‘version 2.0.’ I share this list with the hope they will help P123 members improve their sim performance and perhaps inspire new innovation. Please let me know if that’s the case.

Because of respect for P123’s rules, I can’t tell you in this forum how these timing dates were derived or how I use them, but if you send me a private e-mail, I’ll let you know about the system’s details and which indicators I use to derive the signal dates.

Chris

14

Hi All,

I have spent some time trying various tweeks to get the O’Shaungessy Sim to work better, but have been unsuccessful. So I decided to take a look the initial Ranking System in Dan’s Sim. I was astounded to learn how terrible the results are.
In fact, in the 26-wk RS Ranking, the last buckets performed the best. I tested the performance of the Ranking system with settings of 100 bars, No OTC, minimum price of $1, and NA’s from previous quarter.

I tested the Ranking system with a 1) Max date setting (from 2001 to present), 2) the rally from March 2009 to present, and 3) the last year (from 12/27/2010 to 12/27/2011). The charts below/attached show how bad these runs were. There’s no way that a Sim can fight back from this disadvantage from the start. See the attached charts for details.

Please understand that these results are no reflection on Dan. He was simply following the published OShaugnessy Factors and Formulas to create the Sim. At this point, it’s unclear how Mr. O’Shaugnessy was able to determine that this approach was the ‘Holy Grail’ of stock picking over the last 50 years. Maybe there’s more to his system that we don’t know about.

Chris

111228_OShaungessy_26-wk_RS_Max.gif
111228_OShaungessy_26-wk_RS_Max.gif507×502 19.7 KB

111228_OShaungessy_26-wk_RS_2009-Present.gif
111228_OShaungessy_26-wk_RS_2009-Present.gif508×505 17.5 KB

111228_OShaungessy_26-wk_RS_Dec-2010-Dec-2011.gif
111228_OShaungessy_26-wk_RS_Dec-2010-Dec-2011.gif514×501 18.9 KB

15

O’Shaughnessy never claimed he had found a “Holy Grail” strategy. Nor did he claim his “Trending Value” model was the best quant strategy of all time… just the best one that he tested. He was primarily saying using his five-factor ranking system plus six-month relative strength model (Trending Value) is superior to a buy-and-hold index approach.

16

O’Shaughnessy is in my view pretty much a complete failure with lots of hot air.
Yes, he did the number crunching and wrote a book about it, more than what most of us have done. But practical, useable market strategies…nothing, niente, nada, nichts.

If his numbers are so good (and maybe he has some strategies he won’t reveal), why is his fund performing so badly? He does not seem to be able to put his numbers to work for himself or his fund. That says it all imho.
Here is a link to his results, it is pretty disillusioning.

17

I have the O’Shaughnessy book as well and found myself scratching my head at the top holdings in his funds. In no way can these holdings been considered Trend/Value or even Small Cap Value:

Lulumon
Green Mtn Coffee Roasters
Fossil
Cardtronics
Herbalife

These are Growth names not Value, at best they could be considered GARP but I see a contradiction between his book and his actual investing portfolios.

18

Not sure how good they actually are, but I’ll say this much, I’m flat on the year and it felt like a 15 round boxing match.

“You should see the other guy” - CTA index, Hedge Fund Indices, Mutual Fund league tables, everyone is significantly negative on average.

This is the most brutal sideways action I’ve seen in my 12 years trading, it is chopping people up left and right.

19

An evaluation of the 1990s version of what worked on Wall Street is here for those interested.

20

This is an interesting thread on the premier and perhaps most unrepentant practitioner of data mining around today (I attended a talk given by him last winter and he’s still very much into his approach). But by now, leading quants have left that in the dust as they explore what they refer to as “regime switching models.” Unfortunately, though, the literature is, as far as I’ve been able to see, largely incomprehensible to anyone without a PhD in mathematics. But from what I can glean (admittedly limited), it does not look like they’ve yet found Nirvana, and my own first attempt (the p123 multi-market model, is not there either).

What I have been doing, though, is, actually, going in the opposite direction from those who look to present backtests that stretch as far back as possible. I’ve spend a lot more time lately carving up the available testing periods and trying to focus on those I think will most likely resemble the kind of market I expect in the near future. I often find myself separately evaluating the 2005-07 period, one that was OK but showing signs of running out of steam, and 2010 (in the event I decide the small stocks upon which I like to focus are ready to leave the garbage pile to which the market has lately consigned them). Even if we were to get data that would allow us to stretch backtests into the 1990s, I doubt I’d ever use it because I think there have been too many structural changes in the economy and the market to make the results relevant.

I think we also need to test to identify periods where the market completely ignored fundamentals. For example, I barely think about 2008, when the best way to pick good stocks was to figure out which institutions owned which stocks and how desperate each was to raise cash. I accept that if we have a recurrence of that, I’m just going to have to recognize it and jump ship or if uncertain, hedge a bit with leveraged short ETFs (I have been making some good use of TZA). In other words, there’s no way I’d jettison a model just because it bombed in 2008. For those who focus on the smaller end of the micro-cap universe, 2011 had similar characteristics.

I guess my takeaways from what we’ve been through in recent years is the need to de-emphasize the “Max” backtests and start thinking as creatively about test design as we do about the models we create, almost to the point where the most sensible test might, when real money is put to work, trump the best model.

As to test design, besides looking at smaller time chunks, you may want to check the advanced backtest portion of the screener. Sometimes, too, I’ll run separate tests for my models after limiting results to stocks whose market caps, revenues or any such figure is even, and another test for companies whose metrics are odd numbers; I’d hope to see tolerable results from each of the randomly configured sub-sets. The idea is to try to blow up our test results, and judge our models by how they withstand the assaults we come up with. Under this approach, we wouldn’t aim for Rank performance charts that slope 45 degrees upward as we look from left to right. Instead, we’d aim for results that are as tolerable as possible under as many different scenarios as possible. O’Shaugnessy, with his homage to the past results, is, in my opinion. what we should be running away from.