Backfill Bias in Fund Databases

A new hedge fund launches quietly, trades for two years, does very well, and only then decides to start reporting its numbers to a commercial database and actively seek outside investors. The database now shows two full years of strong returns for that fund, backdated to its actual launch date, even though during those two years nobody outside the fund could see those numbers or invest based on them. Multiply this across thousands of funds and the effect on a database's average historical return is systematic: funds that had a bad start rarely bother to start reporting at all, so the "backfilled" history a database shows is disproportionately made of good starts.

Why the selection is one-sided

A fund manager decides whether and when to submit performance data to a database like a hedge fund index provider. Nobody is forced to report, and reporting is mostly done to attract capital. A fund with a mediocre or poor first year has every incentive to wait, keep trading privately, and only start reporting once results look good enough to market — or to quietly shut down and never report at all. A fund with a strong first year has every incentive to start reporting immediately and backfill that strong year into the database. The mechanism filters almost exclusively in one direction.

Worked example

A database contains 100 new funds each year. Historically, of funds with a poor first year (bottom half of first-year performance), only 20% ever choose to start reporting, doing so an average of 18 months late with that early period backfilled. Of funds with a strong first year (top half), 80% start reporting almost immediately.

1. Reported sample composition. Roughly $20 \times 0.20 + 80$... more precisely, of the 100 funds, about $50 \times 0.20 = 10$ poor-starters and $50 \times 0.80 = 40$ strong-starters end up reporting, a visible sample of 50 funds that's 80% drawn from the strong-starting half versus a true population that was 50/50.
2. Effect on average backfilled return. If poor-starters averaged 2% in their first 18 months and strong-starters averaged 18%, the database's reported average for that cohort is $\frac{10 \times 2\% + 40 \times 18\%}{50} = \frac{20\% + 720\%}{50} = \frac{740\%}{50} = 14.8\%$, versus a true population average across all 100 original funds that would have been much lower had every fund's actual first 18 months been counted.

The database isn't lying about any individual fund's number, but its aggregate statistics are built almost entirely from funds selected for having started well.

What this means in practice

Academic studies estimate backfill bias adds somewhere on the order of 1-4 percentage points a year to reported hedge-fund index returns, a large enough number to change conclusions about whether the asset class outperforms after fees. Serious researchers either strip out each fund's backfilled period (using only data reported from the fund's actual database entry date forward) or explicitly flag it as a limitation, and allocators comparing a prospective manager's early-year numbers against a database average should ask whether that average itself is backfill-inflated before treating it as a fair bar.