Calendar effects as conditioning variables, not signals

Calendar effects are among the most replicated anomalies in the equity literature. They are also among the most misused. Read as stand-alone signals, they typically deliver weak Sharpe ratios and noisy out-of-sample behaviour. Read as conditioning variables — switches that gate when other signals are allowed to trade — they recover much of their economic content.

Why stand-alone framing fails

The classic studies on turn-of-month, FOMC drift, and end-of-quarter rebalancing report unconditional means computed across decades of data. Those means are real but small relative to the daily volatility of the underlying instrument. A strategy that goes long on the conditioning day and flat otherwise spends most of its risk budget out of market, never compounding.

Conditioning, not trading

The more honest use is to treat the calendar as an exogenous regime variable. We do not trade the FOMC drift directly; we permit our trend and carry signals to size up only on days where the calendar regime is consistent with their historical edge. The calendar adds a binary mask, not a position.

“When in doubt about the meaning of a marginal effect, ask what role it would play inside the joint distribution.”

An out-of-sample discipline

We require that the conditioning improvement is meaningful on data the model has never seen, that it survives transaction-cost assumptions calibrated to live execution, and that it does not collapse when the gating threshold is moved within a reasonable neighbourhood. Two of those three checks are routinely failed by candidate signals from the literature.

Practical takeaway

• Treat calendar effects as masks on existing signals, not as signals themselves.
• Validate the conditional version on held-out windows; the unconditional version is a sanity check, not a result.
• Penalise low-activity strategies — a mask that gates out 95% of days has to be very right when it does fire.