Signal Autocorrelation vs Return Autocorrelation

Signal autocorrelation measures how similar a signal's own values are from one day to the next — a value's fundamentals-based signal barely changes day to day, so it's highly autocorrelated; a signal built from overnight order-flow imbalance resets almost completely each day, so it's nearly uncorrelated with yesterday's value. Return autocorrelation is a different, unrelated quantity: whether the asset's returns — not the signal — tend to continue or reverse from one period to the next.

The confusion arises because both numbers can be high or low independently, and only one of them tells you what you actually want to know for trading: a signal can be extremely persistent (autocorrelation near 1, barely changing week to week) while the underlying stock's returns are close to unpredictable white noise — persistence in the signal says nothing by itself about whether the signal predicts those returns, which is a separate correlation entirely (the information coefficient).

Practically, signal autocorrelation mostly determines turnover and trading cost: a highly autocorrelated signal barely trades, a fast-decaying one churns the portfolio constantly. Return autocorrelation instead shapes which strategy family even makes sense — positive return autocorrelation favors momentum, negative favors reversal — independent of any one signal's own persistence.