The question whether the preparation process of foreshocks and main shocks is different from other earthquakes is of great interest with regard to earthquake predictability. We show that the most conspicuous properties of earthquake clustering can be explained without assuming any differences in the initiation processes. In particular, the Gutenberg-Richter law as well as the Omori law for foreshock and aftershock sequences can be reproduced by model simulations with the simple assumption that all subsequent events are initiated in the same manner at the edges of the recently ruptured area. In this way, the empirically observed b and p values are reproduced naturally without any parameter tuning as well as their differences with regard to foreshock and aftershock activity. These properties are shown to result from the shrinking of the loaded fault region with time. In the model, foreshocks occur in extended and almost compact fault segments, whereas aftershocks are mostly restricted to recesses left unruptured by the main shock. Our investigations lead to the conclusion that the spatial effects rather than the temporal effects of the initiation mechanism are decisive for earthquake clustering.