question about simulation of mixer conversion gain

I am using Cadence spectre. I simulate LNA+mixer. I found the PAC gain is different with the gain, which simulated by the transient. the gain difference is larger than 3dB. Anyone knows the reason? thanks.

From my experience, I found gain for transient analysis is not accurate. But I am not sure.
but if I do "transient+fft" in hspice ,the result is similar to PAC.

Whether or not your source is small signal in trans?
when you use PSS and PXF, c gain is calulated in DC source stimuli case.