Joel Greenblatt published "The Little Book That Beats the Market" in 2005 and introduced retail investors to what he called the Magic Formula. The headline backtest was unforgettable: from 1988 to 2004, a portfolio of stocks ranked by the Magic Formula returned 17.4 percent CAGR versus 12.4 percent for the S and P 500. That is a 5-percentage-point annual edge over 17 years, on a public market, with a simple two-factor screen. The book sold over 300,000 copies and the formula became one of the most-discussed quantitative value strategies in the world.
Twenty years later, the legitimate question is: does the Magic Formula still work? The answer is nuanced. The published backtest is real. The out-of-sample performance since 2005 is positive but meaningfully degraded. And the original 2-factor framework has been extended by modern multi-factor approaches that incorporate more of what Buffett, Munger, Fisher, and others actually look for. This post walks through the numbers.
We have a complementary piece at /blog/joel-greenblatt-magic-formula-explained/ that walks through the construction of the formula in detail. This post focuses on whether it still works.
What the Magic Formula actually is
Greenblatt's formula combines two factors:
Factor 1: Earnings Yield (a quality-of-price metric)
Earnings Yield = EBIT / Enterprise Value
Higher is cheaper. The formula uses EBIT (earnings before interest and tax) and enterprise value (market cap plus debt minus cash) rather than the more common P/E and market cap, because EBIT-to-EV is structurally cleaner across companies with different capital structures.
Factor 2: Return on Capital (a quality metric)
Return on Capital = EBIT / (Net Working Capital + Net Fixed Assets)
Higher is better. The formula uses a particular formulation of return on capital that excludes goodwill and intangibles, focusing on the tangible productive capital actually being deployed.
The Magic Formula ranks every stock above a certain market cap on both factors separately, sums the two ranks, and selects the stocks with the lowest combined rank (best combined rank). The portfolio holds 20-30 stocks, rebalanced annually.
The original 1988-2004 backtest
Greenblatt's original backtest, published in the 2005 book and replicated by multiple academic researchers since:
- Universe: top 3,500 US stocks by market cap (above roughly 50 million USD market cap at the time)
- Holding period: 1 year per stock
- Rebalancing: monthly cohort rolling
- Portfolio: 30 stocks
Annualised returns, 1988-2004 (17 years):
| Strategy | CAGR |
|---|
| Magic Formula portfolio | 17.4% |
| S and P 500 | 12.4% |
| Excess return | +5.0% per year |
A 5-percentage-point annual edge over 17 years compounds enormously. A 10,000 USD investment grew to:
- Magic Formula: 165,000 USD
- S and P 500: 73,000 USD
The edge is real. The backtest is real. The question is what has happened since.
The out-of-sample 2005-2025 performance
Multiple researchers have replicated and extended the Greenblatt backtest into the period since the book was published. The findings converge on a similar story:
| Period | Magic Formula CAGR | S and P 500 CAGR | Excess return |
|---|
| 1988-2004 (original) | 17.4% | 12.4% | +5.0% |
| 2005-2014 | roughly 9% | roughly 8% | +1% |
| 2015-2024 | roughly 7% | roughly 13% | -6% |
| Full 2005-2024 | roughly 8% | roughly 10.5% | -2.5% |
The headline: the Magic Formula has not just decayed; in the last decade specifically, it has underperformed the index. The 5-percentage-point original edge has flipped to a 6-percentage-point deficit in the most recent decade.
Why?
Three reasons the original edge degraded
Reason 1: the Magic Formula systematically excludes the largest tech compounders. In the 2015-2024 decade, the S and P 500's returns were heavily driven by the mega-cap tech names (Apple, Microsoft, Nvidia, Alphabet, Meta, Amazon). The Magic Formula tends not to select these names because they trade at premium multiples on the earnings yield factor. A strategy that systematically excluded the largest contributors to index returns was going to underperform.
Reason 2: the formula has been arbitraged. Once a published quantitative strategy has the volume of attention the Magic Formula received post-2005, sophisticated investors begin to front-run the rebalancing flows, narrowing the spreads and reducing the edge. This is the classic alpha-decay-after-publication pattern that Fama and others have documented for other published factors.
Reason 3: the formula misses qualitative factors that matter. The 2-factor formula does not capture moat durability, management quality, balance-sheet resilience, or growth runway. In a market where these factors increasingly differentiate winners from losers (especially in software and platform businesses), a 2-factor formula is structurally undermatched.
The 2025 view: is the Magic Formula dead?
Not exactly. Three nuances matter:
Nuance 1: rolling 5-year and 10-year periods still show the Magic Formula passing more often than failing in the historical record. The 2015-2024 underperformance is one extreme decade, not a permanent regime change. The next decade may revert.
Nuance 2: the Magic Formula's worst underperformance has historically come during heavily concentrated mega-cap rallies (1999, 2020-2021, 2023-2024). When market leadership broadens back to mid-caps and value, the Magic Formula has historically caught up. We saw a partial version of this in 2022's bear market.
Nuance 3: the formula remains a useful starting screen. A list of 30 stocks ranked by combined earnings yield and return on capital is a defensible value-investing universe. The decay is in the unmodified buy-everything-on-the-list strategy, not in the underlying ranking logic.
How invest-like extends the Magic Formula concept
invest-like's design philosophy is that no single value-investing framework, however well-conceived, captures everything that matters. The Magic Formula's 2-factor approach is one such framework; Buffett's 5-pillar approach is another; Graham's defensive screen is a third; Fisher's growth-quality framework is a fourth.
The 7-framework consensus on invest-like combines:
- Buffett-Fit (moat, durability, management, health, valuation)
- Graham defensive (margin of safety, balance sheet, dividend record)
- Fisher growth-quality (organic growth durability, management depth, R and D efficacy)
- Lynch GARP (PEG, story comprehension, insider activity)
- Greenblatt Magic Formula (earnings yield + return on capital)
- Munger mental models (compounder durability, capital allocation, narrative consistency)
- Smith Fundsmith (high ROCE, low gearing, free-cash-flow quality)
Each framework scores 0-100 on each stock. The cross-framework consensus is the synthesis. The published study at /blog/12500-stocks-7-frameworks-cross-framework-consensus/ walks through the methodology in detail.
The insight is that a stock that passes 6 of 7 frameworks at B-minus or better is structurally more robust than a stock that ranks at the top of any single framework. The single-framework top-rank is more vulnerable to the specific blind spots of that framework. The multi-framework consensus filter catches the blind spots of each individual framework using the strengths of the others.
What the Magic Formula does well today, even after decay
Two specific contexts where the Magic Formula still adds real value:
Context 1: cross-checking other frameworks. If a stock passes Buffett-Fit but fails the Magic Formula, the failure usually points to either (a) the valuation is too rich on EV/EBIT, or (b) the return-on-capital metric is being inflated by intangibles in a way Buffett-Fit does not catch. Either way the failure is informative.
Context 2: starting universe for deep-value research. For investors who want to do bottom-up research on a smaller universe, the Magic Formula's top 100 or top 200 is a defensible starting screen. The stocks that survive into the top quintile across both factors are at minimum reasonably priced relative to their return on capital, which is a better starting point than the unfiltered universe.
What the Magic Formula does poorly today
Three specific contexts where the Magic Formula is structurally weak:
Context 1: software businesses. Software companies have unusual capital structures (low fixed assets, high intangibles). The Magic Formula's return-on-capital metric does not handle this well. Software businesses are systematically under-ranked.
Context 2: cyclical commodities. The Magic Formula picks up cyclicals at peak earnings (high earnings yield, high return on capital in good times) and gets crushed when the cycle turns. The formula does not normalize earnings across the cycle.
Context 3: financials. The Magic Formula's EV definition does not map well to banks, insurers, and asset managers. Most quantitative implementations exclude financials entirely.
Where invest-like fits
Every stock on invest-like has a Magic Formula rank computed and visible at /buffett/[ticker]/ under the Greenblatt tab. The site-wide Magic Formula ranking is at /fit/greenblatt/. The intersection of Magic Formula top quintile AND broader 7-framework consensus is the strongest filter.
For the broader Magic Formula context:
The working papers that formalise invest-like's methodology have been independently published at DOI 10.5281/zenodo.20393518 and 10.5281/zenodo.20393706 for those who want the academic-style citations.
The honest takeaway
Joel Greenblatt's original 1988-2004 backtest is real. The strategy still works as a starting screen and a cross-check. The unmodified mechanical version has degraded against the index in the last decade because the market regime favoured concentrated mega-cap tech that the formula systematically excluded.
The Magic Formula is not dead. It is a 2-factor framework in a world that requires multi-factor frameworks. The 7-framework consensus on invest-like is the natural extension: combine the Magic Formula with Buffett-Fit, Graham, Fisher, Lynch, Munger, and Smith, and select the stocks that pass the consensus, not the stocks that top any single framework.
That is the practical synthesis. The Magic Formula is one of the seven inputs. It is not the whole answer.
Disclosure
Educational tool. The 1988-2004 Magic Formula backtest is from "The Little Book That Beats the Market" by Joel Greenblatt (2005). The out-of-sample figures cited are approximate and based on multiple academic and quantitative-fund replications; specific decade-by-decade numbers vary by methodology choices. Past performance does not predict future returns. Not investment recommendations.
Author: Zaid Ghazal, founder of invest-like, Kiel, Germany. Not a registered investment adviser. Full methodology at /methodology/.