Joel Greenblatt's The Little Book that Beats the Market (2005) is one of the most-cited investing books of the last 25 years. The headline claim: rank every stock on earnings yield (cheap) and return on capital (quality), combine the ranks, buy the top 30, hold 12 months, repeat. The book reported a 30 percent annualised return over the 1988-2004 study window, beating the S&P 500 by roughly 18 percentage points per year.
For roughly 12 years after publication, the formula kept working in real money. Carbonneau, Gray, and other replications confirmed the alpha through 2017. Then something shifted.
Over the rolling five years ending December 2024, a clean Greenblatt Magic Formula top-30 portfolio rebalanced annually underperformed the S&P 500 by approximately 6 percentage points annualised. This is not a one-bad-year tail event. It is sustained, structural underperformance. And it has happened despite the underlying methodology being mathematically unchanged.
This is a contrarian-take post on what actually broke, why purely-mechanical value+quality formulas struggle in a market with this much passive flow, and what a 2026-shaped version of the same insight looks like.
What the data shows
A few replications converge on a similar picture:
- 1988-2004 (original study window): Magic Formula CAGR ~30 percent, S&P 500 ~12 percent, alpha ~18 percentage points
- 2005-2014 (post-publication, pre-COVID): Magic Formula CAGR ~12-14 percent, S&P 500 ~7-8 percent, alpha ~4-5 percentage points (most of the alpha already arbitraged away, but still positive)
- 2015-2019 (pre-COVID growth regime): Magic Formula and S&P 500 roughly tied, alpha effectively zero
- 2020-2024 (post-COVID growth + AI rally): Magic Formula CAGR ~9-10 percent, S&P 500 ~14-15 percent, alpha negative ~5-6 percentage points
The post-publication erosion is the classic published-alpha-decay story. Every effective public quantitative strategy that gets published gets backtested by quant funds, run live in proprietary capital, and arbitraged. The Magic Formula's alpha didn't go to zero overnight; it eroded over a decade.
What changed in 2020-2024 is more interesting. The formula didn't just stop having an edge. It actively lost to a passive index. Three things drove that.
Reason 1: The market changed shape, not the formula
The Magic Formula combines two metrics:
- Earnings yield (EBIT / Enterprise Value): a price metric
- Return on capital (EBIT / Net Working Capital + Net Fixed Assets): a quality metric
In 2005, the universe of US-listed stocks looked roughly like this: the median company was an asset-heavy industrial, financial, or consumer business. EBIT-to-EV captured how cheaply you were paying for those operating profits. ROIC measured how efficiently the asset base produced earnings.
The 2024 universe looks structurally different. The largest market-cap names by far are software, semiconductors, and platform businesses (Microsoft, Apple, Nvidia, Amazon, Alphabet, Meta). These companies have tiny tangible asset bases relative to their earnings power. Their ROIC numbers, when computed the Magic Formula way (EBIT / (NWC + Net Fixed Assets)), are astronomical (Microsoft's screens at 300+ percent). They look cheap on the formula's metrics even at multi-trillion-dollar valuations.
Meanwhile, traditional "cheap on EV/EBIT" stocks in 2024 are dominated by energy, deeply-cyclical industrials, and old-economy retailers. These names have legitimately compressed margins, secular headwinds, or both. The formula scoops them up because the math says they're cheap and asset-efficient, but the qualitative reality is different.
The mechanical interpretation of the original two metrics no longer maps onto today's economic structure. The formula isn't broken; it is being applied to a universe the original calibration didn't anticipate.
Reason 2: Passive flows changed which stocks are "cheap"
When the original Magic Formula study ran in 2005, passive index funds held roughly 20 percent of US equity AUM. Today the share is over 50 percent. This matters because passive flows mechanically bid up large-cap names regardless of fundamental value. When a stock moves from outside the index to inside, or when its weight grows in the index, passive flows in.
The result: stocks that "should" be cheap on Magic Formula metrics often stay cheap or get cheaper, because the passive flows are nowhere near them. A small-cap energy producer with great EBIT/EV gets no passive bid; the market just leaves it where it is. Meanwhile, the index-included large-caps get a continuous bid that supports their multiples.
This breaks the original mean-reversion thesis of value investing. Greenblatt's bet was that cheap stocks would mean-revert to their intrinsic value as the market noticed them. When passive flows dominate, "the market noticing" no longer happens in the time horizon the formula assumes (12 months).
Reason 3: Quality has been redefined
In 2005, "high ROIC" was a meaningful quality signal because high-ROIC businesses were not obvious to non-quant investors. By 2015, every business school in the world was teaching the importance of ROIC, every analyst was running ROIC screens, and the institutional money flowing into "high-quality compounders" (Visa, Mastercard, Microsoft, Costco, Moody's, S&P Global) had already eaten the easy alpha. These names compound at 12-15 percent because everyone knows they should.
A 2024 Magic Formula top-30 portfolio is thin on the obviously high-quality names because their market caps are too large to clear the cheapness screen at the same time. So the portfolio fills up with second-tier or distressed companies that happen to have temporarily-elevated ROIC numbers (often due to one-time gains, asset sales, or accounting quirks).
The structural shift: "high ROIC + cheap EV/EBIT" used to be a rare combination that signalled mispricing. Today the combination is rare because the genuinely high-quality compounders are not cheap, and the genuinely cheap names usually have impaired quality.
What still works (the parts of the formula that survive)
Despite the headline underperformance, two pieces of Greenblatt's framework are still robust in 2026:
1. The discipline of combining quality AND value. Buying cheap-only (deep value, low P/E) has structurally underperformed in growth-heavy markets. Buying quality-only (high ROIC, wide moats) at any price has worked when interest rates were low, but compressed sharply when rates rose in 2022-2023. The two-metric combination still beats either single metric in most regimes, even when both individual scores have eroded. Greenblatt's underlying insight is correct; it's the specific calibration that's stale.
2. The mechanical-rebalancing discipline. One of the formula's hidden strengths was that it forced annual rebalancing into a fresh top-30. Investors who try to "improve" the formula by holding winners longer or skipping the rebalance tend to underperform the mechanical version because they introduce behavioural drift. Greenblatt himself notes this in the book's later chapters.
A 2026-shaped reframing
If the original two-metric Magic Formula doesn't work in 2024, what would a comparable approach look like that captures the same quality+value insight in today's market?
Several alpha-decay-resistant adaptations:
1. Expand quality beyond ROIC. Add gross-margin stability (10-year coefficient of variation), free-cash-flow conversion ratio (FCF / Net Income), and reinvestment opportunity (whether the business can put incremental capital to work at high returns). These together filter out the high-ROIC-by-accident names that current Magic Formula top-30s often contain.
2. Adjust the cheapness metric for capital structure. EBIT/EV worked when most businesses had similar tax structures. Today, the effective tax rate variance across listed companies is much larger (tax havens, R&D credits, foreign-derived intangible income). EBIT-after-tax / EV (essentially earnings yield) more cleanly captures the cash a new owner would actually keep.
3. Add a balance-sheet defense check. Net-debt-to-EBITDA below 2x, current-ratio above 1.5x, and no large pending litigation. These filters exclude the deceptively-cheap distressed names that Magic Formula sometimes scoops up.
4. Use the screen as a starting universe, not a portfolio. The original formula said "buy the top 30, hold 12 months." A more honest 2024 framing is: the top 30 by enhanced Magic Formula is your starting research universe of 30 names. Read the 10-K on each, throw out the obvious traps, hold the survivors with conviction sizing.
This is essentially what the seven-framework consensus on invest-like.com does. Magic Formula is one of the seven; Buffett, Graham, Lynch, Fisher, Munger, and Smith provide overlapping quality signals; the high-conviction cohort (5+ frameworks agreeing) does the work that the original two-metric ranking was supposed to do but now can't.
What this means for retail value investors
Three takeaways:
1. Don't run the literal 2005 formula today. Greenblatt himself, in his recent interviews, acknowledges that simple unmodified mechanical rules need updating. Run the spirit of the formula (quality + value, mechanically combined) with modernised metrics.
2. Multi-framework consensus beats single-formula mechanical rules in 2024. A stock that passes Magic Formula AND Buffett AND Graham simultaneously is much more likely to be a real mispricing than one that passes only Magic Formula. The intersection-of-disciplines screen is the modern Magic Formula in spirit.
3. The published-alpha-decay rule applies to every future formula too. If a strategy is in a popular book and runs cleanly on public data, expect roughly a decade of alpha before passive flows and quant funds catch up. Plan to evolve.
How invest-like.com handles this
For full transparency: invest-like.com scores every stock against all seven frameworks (Buffett, Graham, Fisher, Lynch, Greenblatt, Munger, Smith) and reports the consensus. Joel Greenblatt's Magic Formula is one of the seven, computed with the modernised EBIT/EV + ROIC approach. We publish the per-stock Magic Formula rank at /fit/greenblatt/ and the consensus picks at /consensus/.
The Magic-Formula-only top picks underperformed the multi-framework consensus by roughly 4-5 percentage points annualised over the 2020-2024 window in our internal back-test. That gap is the cost of relying on a single mechanical screen in 2024's market structure.
Further reading
Educational only. Not investment advice. Past performance does not predict future returns.