Channel extrapolation is a promising technique to estimate the Channel State Information (CSI) in multi-input-multi-output (MIMO) systems operating in frequency division duplex (FDD) without relying on costly terminal feedback. In this paper we analyze the limits that can achieve the extrapolation of the frequency response of a propagation channel from its cluster-based representation. This method consists in measuring all the clusters characteristics (angle of departure, gain, delay) on the uplink and then extrapolating the downlink channel from those measurements. We propose a framework in which the limits of a cluster-based extrapolation process are studied taking into account delay and angular spread derived from the Saleh-Valenzuela model. We evaluate the performance of the optimal linear estimator hereby quantifying the extrapolation residual error bound in terms of Mean-Square Error and Reduction of the Beamforming Gain.