In this paper, the utility of power spectral density (PSD) in ground motion prediction for hazard analysis is examined. The main advantage of PSD is that the area under PSD of excitation gives the energy of the excitation and it is structure independent. The PSD of ground motion from past earthquakes and properties of PSD in terms of spectral moments are studied. 1487 accelerograms from 25 major past earthquakes with magnitude ranging from 4.9 to 7.9 taken from the ‘Pacific Earthquake Engineering Research Centre – Next Generation Attenuation’ database are considered. PSD of past ground motion data is used to form a hybrid genetic algorithm-neural network based attenuation relationship with magnitude of earthquake, Joyner-Boore distance and shear-wave velocity. PSD compatible time histories are generated using the proposed model. An application of the proposed model is demonstrated by developing a uniform hazard power spectral density for Delhi with a return period of 2500 years. © 2020 Elsevier Ltd