Mmmh. My idea is that the solver cannot calculate the transient response. I assume, the solver works by calculating every time step after another based on the source signals and current currents/voltages.
Positive (or negative, not sure about the definition) phase shifts would be impossible, as the solver would need to predict the future.
What would be the response of a step function? The solver would need to transform the input into frequency domain, apply the phase shift and transform it back to time domain. My guess is that this does not comply with the time-step-by-time-step simulation procedure