Performance and Evidence · Hedge Funds

A track record is evidence to investigate. Backtests, selected histories and incomplete databases can create a much stronger impression than an implementable strategy deserves.

Distinguish simulated and live results

A backtest applies rules to historical data. Live performance reflects decisions and constraints encountered in actual operation. Simulations can omit borrowing difficulty, market impact, execution delays or realistic financing. Labeling a backtest clearly is necessary, but does not resolve those limitations.

Lookahead bias occurs when a test uses information unavailable at the decision time. For example, using a revised earnings figure as though it were known at the original release can improve apparent selection. Point in time data and realistic decision timing help address this problem.

Watch what enters the sample

Survivorship bias appears when failed or closed funds are excluded from the comparison. Backfill bias can arise when a manager adds earlier favorable results after choosing to report to a database. Neither issue can be fixed by looking only at the average return of the remaining funds.

Imagine five funds each begin with $100. Four finish at $110 and one at $40. The surviving four show an average gain of 10%. Including all five gives total ending value of $480 on $500 initial capital, or a 4% loss. This simple example shows how the sample can change the story.

Test whether the process survives scrutiny

Ask how many models or parameter combinations were tried before selecting the published version. A strategy optimized repeatedly on the same history can fit noise. Separate development data from evaluation data and examine whether later live results resemble the original expectation.

For a manager moving firms, establish who made the historical decisions, what assets were included and whether the new vehicle has comparable resources and restrictions. Request complete periods, losses included, and reconcile published results with independent reporting where available. Unverifiable performance should remain an unresolved question rather than become a favorable assumption.

Try it yourself

A manager selects the best of 200 tested signals and shows only its historical curve. What additional evidence would improve the assessment?

Show the worked answer

Request the testing process, selection criteria, realistic cost assumptions, untouched evaluation data and subsequent live results. The best historical curve alone does not show that the signal will generalize.

Further reading

Educational material, not a recommendation to buy or sell a fund. Numerical examples are hypothetical. Fund terms, investor eligibility and tax treatment vary by product, account and jurisdiction.

Continue learning