Apparatus and methods are described for a video encoder that uses reference motion vectors (MVs) with reduced precision to generate predicted MVs. The video encoder receives a block of pixels of a video frame and generates a MV for encoding the block of pixels by motion compensation. The video encoder creates a reduced-precision MV by reducing a precision of the generated MV based on a precision reduction scaling factor. The video encoder also stores the reduced-precision MV in a MV buffer as a reference MV for temporal or spatial MV prediction.