SledgeKey Backtests
Does the Magic Formula actually work? We backtested it honestly.
The Magic Formula comes from Joel Greenblatt's The Little Book That Beats the Market. The idea is to buy good companies at cheap prices by ranking every stock on two things at once, how cheap it is against its operating earnings and how much it earns on the capital it puts to work, then holding the names that score best on the two measures combined. The book made it famous by reporting that this simple, mechanical portfolio beat the market by a wide margin over its test years, which is exactly the kind of claim worth checking against honest data.
We ran a version of the Magic Formula over the last nine and a half years. Our screen keeps Greenblatt's two pillars, buying companies that are cheap on enterprise value relative to EBITDA and that earn a high return on their assets, and it holds twenty of them in equal weight, rebalanced once a year. The part that matters is how the universe was built. Every company listed on each rebalance date was eligible, including the ones that were later bought out, sanctioned off the exchange, or delisted for any other reason, and any holding that left the market was booked out at its last traded price rather than erased from the record. That is the opposite of how most free backtesters work, and it is the difference between a flattering number and an honest one.
Over this window the screen turned $100,000 into $346,718, a return of just under 14% a year. A plain S&P 500 index fund did better, and the gap between them is the whole point of running it this way.
Growth of $100,000 · 2017–2026
The Magic Formula screen against the S&P 500, both starting from $100,000 in January 2017, with delisted companies kept in the strategy the whole way through.
View the data as a table (calendar year-end values)
| Year | Magic Formula | S&P 500 |
|---|---|---|
| 2017 | $124,988 | $119,085 |
| 2018 | $125,002 | $126,396 |
| 2019 | $132,369 | $146,987 |
| 2020 | $139,463 | $174,481 |
| 2021 | $156,114 | $217,771 |
| 2022 | $176,194 | $199,907 |
| 2023 | $198,850 | $228,834 |
| 2024 | $265,197 | $304,392 |
| 2025 | $289,694 | $347,832 |
| 2026 | $346,718 | $380,564 |
The exact screen
The full configuration is below exactly as it ran, so anyone who wants to reproduce the result or argue with it can start from the same table we did.
| Value filter | EV/EBITDA of 16 or lower (trailing twelve months) |
| Quality filter | Return on assets of 5% or higher (trailing twelve months) |
| Market cap | $1 billion and up |
| Selection | The 20 qualifying companies with the largest market cap at each rebalance |
| Weighting | Equal weight, 10% maximum position size |
| Rebalance | Once a year (10 rebalances over the window) |
| Transaction cost | 0.10% per trade ($1,751 in modeled costs over the run) |
| Initial capital | $100,000 |
| Benchmark | SPY, the S&P 500, over the same window |
| Universe | NYSE + NASDAQ operating companies (no SPACs, REITs, ETFs, or funds). Point-in-time: eligibility at each rebalance reflects the companies listed on that date. |
| Delisting treatment | Any holding that later delisted was booked out at its frozen last traded price, never dropped from the history. |
Two things separate this from the letter of Greenblatt's method. He ranks the whole universe on both measures and buys the best combined scores, where our screen sets a threshold on each measure and then takes the twenty largest companies that clear both. He also defines cheapness as earnings before interest and taxes over enterprise value and pairs it with a specific return-on-capital formula, where we use EV/EBITDA and return on assets, which are close relatives of those measures rather than exact matches. The spirit is the same, cheap companies that earn well on their capital, and the shape of the result carries over, though a strict rank-based build of the formula could land at a different final number.
The result against the index
| Metric | Magic Formula | S&P 500 |
|---|---|---|
| Total return | 246.72% | 280.56% |
| Annual return (CAGR) | 13.89% | 15.01% |
| Final value | $346,718 | $380,564 |
| Sharpe ratio | 0.71 | 0.79 |
| Volatility (ann.) | 16.67% | 16.26% |
| Max drawdown | -25.41% | -23.93% |
| Calmar ratio | 0.55 | n/a |
| Winning months | 65 of 115 | n/a |
| Total trades | 256 | n/a |
| Avg holding period | 653 days | n/a |
| Annual turnover | 106.3% | n/a |
A note on survivorship. Run this same screen the way most free backtesters quietly do, against only the companies still listed today, and it returns 15.09% a year and edges past the index. Kept honest, with the companies that later delisted still eligible and still held to their last price, it makes 13.89% and finishes behind. That gap of a little over a point a year is survivorship bias, and we pulled it apart in detail on the midcap value backtest.
The companies it held that later delisted
These eleven names were in the portfolio at some point and then left the exchange, positions a survivor-only backtest never gets to take. Most of them were acquisitions at healthy premiums rather than failures, which is common for cheap, profitable companies: they get bought. A few are foreign issuers whose US listings ended, and one, Yandex, was a real loss for US holders after its shares were frozen in the wake of sanctions. Whatever the exit, the honest run had to live through it, because the companies were in the book on the day they delisted.
What this backtest does not prove
A page like this earns its credibility in this section, so here is what the numbers above do not establish.
The strategy lost to the index. Fourteen percent a year over a decade is real money, and it still fell short of simply owning the S&P 500, which returned more with a slightly shallower drawdown. Anyone reading this as a market-beating system should keep looking. What the run shows is that a sensible, disciplined value screen made good absolute returns and trailed a plain index fund over this particular decade.
This is our reading of the Magic Formula, not Greenblatt's exact recipe. As described above, the classic method ranks the universe on earnings yield and return on capital and buys the top scorers, where we set thresholds on EV/EBITDA and return on assets and hold the twenty largest that qualify. A stricter rank-based build would pick a different set of names and could post a different number, higher or lower.
One window, one configuration. This is a single nine-and-a-half-year run across an era that was hard on value and kind to megacap growth, the exact stretch where the S&P was hardest to beat. Different dates, market-cap bands, or thresholds will produce different results, and the size of any survivorship gap shifts with them.
Some of the delistings are administrative. Several of the names above are foreign issuers whose US listings ended, where the delisting is a paperwork event rather than a verdict on the business. A couple of the 2025 and 2026 exits are recent enough that the reason behind them is still being confirmed, so those are labeled simply as delisted.
Frozen-price booking is conservative but imperfect. When a company delists, the run books the position out at its last traded price. For acquisitions that lands near the deal price, and for a forced delisting like Yandex the real proceeds to a retail holder could be worse than the frozen mark. The modeled 0.10% per trade also covers commissions and typical slippage at large-cap liquidity, and real execution in a stressed market runs worse than any flat assumption.
You can run this exact screen, or your own version of it, on the same survivorship-free point-in-time data and see how it holds up with the delisted names left in.
Run your own backtest · free, no card required