The engine grades itself
Track record, Jan 2012 to Oct 2025
At every month-end, every company was scored using only the filings and prices public that day. A year later, its total return was compared with the S&P 500 (the SPY fund). That gives 10,771 company-months across 70 companies.
53%of top-fifth scores beat the S&P 500 over the next year
49%of bottom-fifth scores did
0.03rank correlation between score and next-year result (0 is none, 1 is perfect)
0.7t-statistic over yearly windows; below 2 means the pattern could be chance
No reliable signal.
Higher scores did not reliably lead to better results. In 56% of months, higher-scoring companies did better on average; a coin flip would give about 50%. That matches decades of research: public accounting figures are already in the price. Use the score to read a company quickly, not to predict it.
By score fifth
Each month, companies were split into five equal groups by score.
| Group | Beat S&P | Typical result | Middle 80% of results |
|---|---|---|---|
| Bottom fifth | 49% | −0.4 pts | −29 pts to +43 pts |
| Second fifth | 44% | −3.4 pts | −27 pts to +28 pts |
| Middle fifth | 49% | −0.6 pts | −27 pts to +30 pts |
| Fourth fifth | 51% | +0.9 pts | −28 pts to +31 pts |
| Top fifth | 53% | +1.4 pts | −28 pts to +42 pts |
Which measures carried signal
Average monthly rank correlation of each measure with the next-year result. Bars right of the line did better.
Year by year
Scores taken each December, judged against the following twelve months.
What flatters or limits these numbers
- Survivorship. The 70 companies were chosen in 2026 because they are large today. Companies that shrank or failed since 2012 are missing, which makes every company look better than a real 2012 list would have.
- Overlapping years. Twelve-month results that start a month apart share eleven months, so monthly figures are not independent. The t-statistic uses one month a year to avoid counting the same move twelve times.
- No tuning. The five measures and their equal weights were fixed before the first backtest and were not changed after seeing the results.
- Left out. Trading costs, taxes, and the exact timing of dividends. Prices are month-end closes.