We consider the downlink of a multiuser multiple-input multiple-output (MIMO) system operating in the time-division duplexing (TDD) mode. In this mode, assuming reciprocity, the channel coefficients estimated during the uplink channel training are utilized by the base station (BS) in the downlink data transmission. However, due to hardware mismatches, the uplink and downlink channels are not exactly the same, and therefore, there are calibration errors, which degrade the system performance. In this paper, our goal is to provide a transceiver design which has a robust performance under imperfect channel reciprocity. To this end, we first formulate a robust joint precoder and combiner design as a stochastic minimum mean square error (MMSE) optimization problem. Then, employing an alternating optimization approach, we propose an algorithm to obtain the precoding and combining matrices assuming imperfect CSI and calibration errors at both the BS and user sides. Extensive simulation results show that the proposed robust joint precoder/combiner outperforms the existing solutions in the literature.