Skip to content

CGQI

Menu

Perspectives / Strategy testing

Would the factor have been visible at the time?

A backtest can use the right formula and still have the wrong information. Start by rebuilding what a researcher could actually have known on each test date.

Home   /   Perspectives   /   Strategy testing

CGQI editorial · November 16, 2026

Two paper releases compared side by side in an archive

Rebuild the information set

Consider a strategy that ranks companies after an earnings release. A quarter may end in March, but a result published in May first becomes available in May. Store both the period end and the release timestamp. At each rebalance, only use records released by that moment.

Keep revisions visible

Macroeconomic series can change after their original release. The St. Louis Fed’s ALFRED archive preserves successive versions of FRED observations. A model that used an economic release should be tested with the version available on the decision date. Keep a record of the selected vintage and its release date.

Reserve a later window

For an illustrative monthly ranking rule, choose the ranking inputs and lookback length in an earlier sample. Freeze those choices before evaluating the next period. If you change the rule after seeing that period, record the change and reserve a new period for evaluation.

Read beyond the headline return

Record turnover, the largest peak-to-trough decline and sensitivity to trading costs. Repeat the calculation with an extra day of reporting delay. Large changes under a small timing adjustment are a reason to revisit the timestamps or signal construction. Historical results alone cannot establish future performance.

Capital Quant

Build, test and refine quantitative strategies for equities and digital assets.

Explore the product