non-real components indicate reactive power - capacitive or inductive load components that will 'feed' part of the power back.

Or - in other words - voltage and current are out of phase.

I have no experience with cpdsd(), but I would not expect the cpsd() funcction to introduce such 'artifacts'. Might be a side effect of the hann() (like 'hanning window' ?) function - windowing is capable to introduce artifacts. But you may first want to check the load: many loads are not purely ohmic, thus showing reactive components.

Regards

Thanks for your response @Dreher. 

I was able to figure out since my two point correlation is not an even function, DFT (Discrete fourier transform)  of it provides complex values for Power. Hanning window doesn't seem to have a role in this. 

Thank you. 

OK - in general non-real components are to be interpreted as Phase Shirts. Could be that your algorithm introduced them (should be avoided), could be from the signal (indicating reactive Power).

Wish you success