I noticed some inconsistencies between transition dipole moments computed directly (function 18 -> 5) and from ESP atomic charges. Am I doing something wrong? I do the following:

First, I run a TD-DFT calculation using gaussian16. In this example I consider para-nitroaniline to get the required .fch and .log files

To compute ESP charges i do the following
1) compute ground state ESP charges using function (7 -> 12 -> 5 -> 1)
2) compute excited state ESP charges: generate natural orbitals (18 -> 13) and compute ESP charges using the natural orbitals (7 -> 12 -> 5 -> 1)
3) compute ground-to-excited state TrESP: generate transition density matrix (18 -> 9 -> 1), generate natural orbitals from transition density matrix (200 -> 16 -> SCF), generate TrESP (7 -> 12 -> 5 -> 3)
4) compute excited-to-excited state TrESP: generate transition density matrix (18 -> 9 -> 2), generate natural orbitals from transition density matrix (200 -> 16 -> SCF), generate TrESP (7 -> 12 -> 5 -> 3)
5) dipole moments are computed as \sum_i q_i*r_i where, q_i is the charge on atom i with position r_i

To compute permanent dipole moments and transition dipole moment between all excited states directly, I use function (18 -> 5)

As you can see from the data attached, there is good agreement between dipole moments computed directly and dipole moments computed from ESP charges, except in the case of permanent dipole moment of the excited states. Is this expected? Is there a problem with the procedure I'm using?

attachments:
pna-data.xlsx

Thank you for your time