Hello protocol is the basic technique for neighborhood discovery in wireless ad hoc networks. It requires nodes to claim their existence/ aliveness by periodic 'hello' messages. Central to a hello protocol is the determination of 'hello' message transmission rate. No fixed optimal rate exists in the presence of node mobility. The rate should in fact adapt to it, high for high mobility and low for low mobility. In this paper, we propose a novel mobility prediction based hello protocol, named ARH (Autoregressive Hello protocol). Each node predicts its own position by an ever-updated autoregression-based mobility model, and neighboring nodes predict its position by the same model. The node transmits 'hello' message (for location update) only when the predicted location is too different from the true location (causing topology distortion), triggering mobility model correction on both itself and each of its neighbors. ARH evolves along with network dynamics, and seamlessly tunes itself to the optimal configuration on the fly using local knowledge only. Through simulation, we demonstrate the effectiveness and efficiency of ARH, in comparison with the only competitive protocol TAP (Turnover based Adaptive hello Protocol) [9]. With a small model order, ARH achieves the same high neighborhood discovery performance as TAP, with dramatically reduced message overhead (about 50% lower 'hello' rate).