Not yet reviewed by Pith; the record is open.

State-of-the-art searches for gravitational waves (GWs) in pulsar timing array (PTA) datasets model the signal as an isotropic, Gaussian and stationary process described by a power-law. In practice, none of these properties are expected to hold for an incoherent superposition of GWs generated by a cosmic ensemble of supermassive black hole binaries (SMBHBs), which is expected to be the primary signal in the PTA band. We perform a systematic investigation of the performance of current search algorithms, using a simple power law model to characterize GW signals in realistic datasets. We use, as the baseline dataset, synthetic realisations of timing residuals mimicking the European PTA (EPTA) second data release (DR2). Thus, we include in the dataset uneven time stamps, achromatic and chromatic red noise and multi-frequency observations. We then inject timing residuals from an ideal isotropic, Gaussian, single power-law stochastic process and from a realistic population of SMBHBs, performing a methodical investigation of the recovered signal. We find that current search models are efficient at recovering the GW signal, but several biases can be identified due to the signal-template mismatch, which we identify via probability-probability (P-P) plots and quantify using Kolmogorov-Smirnov (KS) statistics. We discuss our findings in light of the signal observed in the EPTA DR2 and corroborate its consistency with an SMBHB origin.