Spatial modulation utilizes the diversity of multiple-input multiple-output channel to improve the spectral efficiency. In time-variant wireless channels, its performance degrades due to the inaccuracy of channel estimation (CE). In this paper, a data-aided channel tracking (DACT) method is proposed to improve its CE accuracy in time-variant wireless channels. Specifically, the demodulated data at current time slot are used to update the subsequent channel state information. Two different DACT methods with different pilot insertion patterns are proposed: 1) feedback DACT (FDACT) where when a symbol error is detected, the receiver will inform the transmitter to insert pilots and estimate the channel again; 2) periodical DACT (PDACT) where pilots are inserted periodically. Theoretical analysis of the CE error in time-variant channels is presented, which indicates that the CE accuracy could be significantly improved in comparison to the conventional counterpart. Moreover, the closed-form expressions of bit error rate for FDACT and frame error rate for PDACT are derived. Simulations are conducted to verify the efficiency of the proposed DACT methodologies.