PID-simulations sometimes trail off from measured value - deadband issue?

sometimes pid-simulated u will drift off from measured u

this coincides with real-world data where the PV seems to spend more time below than obove the setpoint (or vice-versa)

observed for LICs

could this be related to deadband in the actual controller, or in the valve it controls? how to deal with this during estimation?