Signal Staleness and Refresh Frequency

A signal computed from last quarter's earnings data is not the same signal by the time next quarter's numbers are due — the information it was based on has aged, and whatever short-lived mispricing it flagged has often already been arbitraged away or simply evolved into something else. Staleness describes how old the underlying data feeding a signal is at the moment it's used to trade, and refresh frequency describes how often the signal is recomputed with newer data. A fast-decaying signal — say, one based on order-book imbalance — is stale within seconds and needs constant refreshing, while a slow-moving signal based on annual balance-sheet quality might stay useful for months.

The practical risk is mismatch: refreshing a fast-decaying signal only once a day means it is trading on information that's effectively obsolete for most of the day, quietly turning a good idea into noise before it ever reaches the portfolio. Refreshing a signal far more often than its underlying data actually changes wastes computation and can introduce spurious turnover from data noise rather than real information, without adding any genuine edge. Matching refresh frequency to a signal's natural decay rate — determined empirically by measuring how fast its correlation with future returns fades — is what keeps a backtest's assumptions honest about how the signal will behave once traded live.