Greenblatt system question (again sorry)

Hi everybody.

I recently read the Greenblatt’s book, as all of you I guess, and despite there are a lot of thread in this forum about this system I amb not sure if the ranking Porfolio 123 offers us it is enough to replicate the Greenblatt’s system.

I tried to build a model, just with the ranking, one rebalance a year and 30 holdings and nothing else, and this give me a crazy result:

But in the other hand, there is a blog on the net, from a guy who is following the original Greenblatt system, with the screener Grenblatt itself offers in his web page, and the guy AFTER FIVE YEARS is losing money.

Here the blog, if you like to have a look : http://www.magicformulaexperience.com/

Then, I’m little bit confuse:

It’s the blog a fake?

It’s the system over?

The ranking system Porfolio 123 offers just do not work properly?

If someone can give their opinion would be appreciated.

Thanks a lot. :slight_smile:


My Greenblatt systems responds quite well. No one has been able to reproduce the 30% CAGR of the books, but I can get 18% pretty easily. Remember to use the following buy rules:

Frank(“ROI%5YAvg”)>=65
Universe(NOOTC)
MktCap>=50
Universe($ADR)=false
sector!=Financ
sector!=Utilit
industry!=reoper

Sell when rank below 75 … and of course hold a year.

Beyond that, remember that the EV/EBITDA is doing most of the work in your ranker.

Epimethius is right . . . the buy rules are important and most match up with what Greenblatt prescribes. The exception is the first one involving ROI. This is a fudge I added to account for an important part of Greenblatt’s work that can’t be replicated from his book - he wrote there that in his asset management practice, he forecasts into the future; it’s been a while and I don;t recall the details and it seems the blog I wrote on it back in the day is no longer on the site (I’ll try to hunt it down). also, I’m not a fan of 1-year holding periods regardless of what gurus say (advice they may or may not actually apply with their on money or with clients’ money); my default choice is 4-weeks or 13 weeks. Here is the performance of the pre-set Greenblatt screen: https://www.portfolio123.com/app/screen/summary/30542?st=1&mt=1

Thanks a lot for your answer Epimetheus, I’m going to try all what you suggest.

But, then, what’s your oppinion about the blog http://www.magicformulaexperience.com/ ?

Maybe the screener Greemblatt offers is broken? Or the blog is a fake?

Hello Mgerstein.

Thanks a lof for your answer.

Just I would like to ask you, why do you prefer the 4weeks or 13weeks periods? it is because some technical rules in your models maybe?

It’s only been one year according to the source.

Rather than assume malicious intent (i.e., “it’s a fake!”) and/or presume that under-performance can be attributed to parameter mis-specification (i.e., “the rules are wrong!”), I would suggest that the Magic Formula’s under-performance can be more easily explained by style drift and alpha decay.

  1. Style drift
    Statistically, his 24.7% underperformance after one-year is significant (i.e., roughly speaking, a 2 sigma event still happens 4.6% of the time). However, one year is actually not a very enough time frame to judge whether or not a strategy is working as intended due to the well-documented phenomenon known as “style drift”. Style drift is simply an observation that different factors tend to out and under-perform over extended time periods. For example, the size-anomaly (small outperforms big) didn’t work at all during the 80s. More to the point, value and quality can go out of style for years at time.

  2. Alpha decay
    You’re also fighting “alpha decay”, which is the phenomenon whereby the predictive power of a strategy loses its competitive advantage over time. While the magic formula may have worked when it wasn’t widely known, if everyone started using it, it would then not offer any advantage. So even if value and quality remain “in style”, the ways which we measured these things in the past are not likely to remain relevant in the future. Alpha decay is simply the intended product of market efficiency, detailed in Andrew Lo’s Adaptive Market Hypothesis. It is also the primary reason why 95%+ of active money managers underperform their benchmarks after a long enough time frame–they are executing beta-driven strategies with high fees. Also worth checking out: The Unbearable Transience of Alpha.

I don’t mean to discourage. I just think it’s important to be realistic with expectations. My advice is to ask yourself whether a given strategy does anything to differentiate itself from peers’ in terms of either proprietary data, innovative metrics, or exploiting persistent cognitive errors. If the strategy does not do any of these things, it is categorically beta.

Forgetting about the “Greenblatt System”, the ranking system is the best overall RS at P123. Sector Inspector’s Aerospace & Defense Designer Model https://www.portfolio123.com/app/r2g/summary?id=1445554 is based on the Breenblatt RS.

Steve

Hello Primus.

Thanks a lot for your complete answer.

Yes, I re-read the blog, and you are right.

I’m not a native english speaker and sometimes I read english too fast and misunderstood things.

When we invest based on data, potential staleness becomes a problem. But it’s a matter of balance: The ultimate in freshness would require a daily rebalance but very short holding periods don’t give stories time to develop (these are companies and people and don’t act with the speed of the nano-second algorithm processors).

There is no single correct answer: It’s an effort to strike a balance between giving things time to develop (different types of models require different amounts of time) and keeping things reasonably fresh. I haven’t yet found anything where that balance is found in a one-year holding period.

For academic research, I really like the work that Lu Zhang at Ohio State does.

To summarize his research in a few words:

There’s a thousand ways to cover that from a factor standpoint, but I tend to think P123’s Greenblatt’s ranking is a pretty obvious one combining ROC and Earnings Yield. Maybe there are more eloquent and sophisticated ways to do it, but that’s the meat and potatoes. It may drift in and out of favor, but you can sleep at night knowing that the theory is sound.

That sums it up pretty well, especially ImanRoshi’s last sentence: “It may drift in and out of favor, but you can sleep at night knowing that the theory is sound.”

Doing the right thing (stocks that are reasonably valued relative to quality and growth prospects), you won’t succeed month after month, but at the end of the day, you’re a lot more likely to wind up happy with what you did.

A quick look at recent holdings produced in a backtest of the Greenblatt screen found that most of them were small-caps. And of course, like all screens, the holdings are equal-weighted at purchase. So the capitalization-weighted S&P 500 is not a good benchmark.

To get a better idea of the system’s alpha potential, I re-ran it using the S&P 500 as the universe and the RSP as the benchmark. The alpha was -1.37% for the past five years and +3.79% for the past 10 (both rebalanced every four weeks). So this does look like yet another strategy that used to work.

Thanks for bringing a stronger semblance of objectivity to this discussion. Your findings conform to my intuitions.

Hi, I’ll just add to Marc’s reply on this question. Path dependence can be strong with long holding periods. With long holding periods different start dates can have widely different results. You can test this by changing the starting dates on test models. Shortening the holding periods can possibly help eliminate this potential risk and make sure you get closer to the true expected return for the strategy.

Thanks to all of you guys for your comments, they help me a lot. :slight_smile:

I tried to implement the Greenblatt system, but I do not get the 18% Epimetheus said;

Do you think it’s all right? Is there room for improvement? Any suggestions?



What is a “stocks with low investment” for you? How do you mesure that?

Thanks.

15% per year is pretty good.
Don’t get greedy. :slight_smile:

One suggestion is to raise your minimum price you are willing to buy at to say at least $3-5.
The second is some kind of minimum trading liquidity number like Average Daily Total of at least $250,000 per day.

The third is to constrain the stocks bought to the kinds of companies where comparing across your universe for ROIC and EV makes sense. That is why Greenblatt excluded real estate companies, financial companies and utilities. Don’t want to compare apples and oranges for ranking purposes.

The last thing to look at is your re-balance frequency. Having a long, one year re-balance, can give you very odd results on your simulation because the performance becomes very dependent on the days that re-balance occurs. I would do monthly or quarterly.

This is a follow-up to my previous post, in which I found that the Greenblatt screen produces an alpha of -1.37% over the past five years and +3.79% over the past 10 if the universe is limited to the S&P 500 and the benchmark is set to the RSP (to eliminate bias from equal weighting).

I decided to look at the effectiveness of the 7th rule in the screen: Frank(" ROI%5YAvg")>=65. To do this, I re-set the maximum number of stocks in Settings from 15 to 0 to include all stocks that passed the screen rules, irrespective of their rank in the Greenblatt ranking system. The result was an alpha of +1.48% over the past five years and +2.88% over the past 10.

Then I looked at the effectiveness of the ranking system by itself by turning off the 7th rule in the screen. The result was an alpha of +3.27% over the past five years and +4.84% over the past 10.

So the Greenblatt ranking system and the ROI screen both added value in the two periods tested. But the combination of the two detracted in the past five years. It’s my view that ranking systems are vastly superior to screens as stock selection tools, so I’m tempted to take this as evidence that adding screening rules to ranking systems is a bad way to try to increase alpha. But I’ll admit that this is just one datapoint.

Note that both of the factors in the Greenblatt ranking system favor asset-light companies. Both use operating income AFTER depreciation in the numerator, and ROC has net plant in the denominator. So I probably should have been more skeptical when I found that the screen didn’t work over the past five years, when the market has been dominated by asset-light companies. Maybe that will continue.

Finally, someone should fix the 6th rule in the screen, which is supposed to eliminate real estate companies. It looks like the industry code used is out of date or otherwise erroneous.

See Fama-French’s CMA factor (“conservative minus aggressive”). The general observation is that firms which invest aggressively tend to have stocks which underperform.

The prevailing theory is that investors over-value companies which post strong earnings in the short term and correct that error over time as additional information is revealed regarding firms’ actual earnings potentials. Persistent anecdotal evidence corroborates that this mis-valuation can be attributed to investors’ cognitive errors.

[]Stocks of firms with high level of accruals (i.e., the non-cash portion of earnings) tend to under-perform while those with low level of accruals tend to out-perform
[
]Stocks of firms which raise equity tend to under-perform while those which decrease it tend to out-perform
[]Stocks of firms with growing assets tend to under-perform while those with shrinking assets tend to out-perform
[
]Stocks of firms with high levels of operating assets (relative tend to total assets) tend to under-perform while those with high level of non-operating assets tend to out-perform

These observations are generally true over the past 20 years (let me know if you’d like some references). But I am not sure they will persist very far into the future.

The “underinvestment” anomaly is partly a result of the market’s propensity to punish companies which have invested in growth while reward those which have emphasized cash flows. However, prior to the 1990’s, the marketplace rewarded out-investment. This can be observed through long-term trends of corporate conglomeration from the 1960s-to 1980s, followed by a period of deconglomeration through the 2000s. But the pendulum will swing. I think what we’re witnessing in the tech space–where investments are rewarded moreso than cash flows–may presage the shape of things to come.

Also, part of the anomaly is attributable to inefficiency. The purpose of accrual accounting methods (i.e., those which are endorsed by all major accounting regulators) is to provide stakeholders (e.g., owners, managers, investors, regulators, tax-collectors, etc.) with an accurate estimate of a firm’s cash flow generating potential. Historically, an over-emphasis on earnings led companies to “cook” the books by making the earnings look better than actual cash-flow generating potential–there are many ways that this can play out on the books. Therefore, investors which looked at other sources of information (e.g., cash flows) had an edge in determining the true cash generating potential of a firm. But now that the knowledge of this cognitive error has diffused, so have its potential advantages.

Thanks for your answer Primus, it’s very very interesting.