Upcoming LTE networks have basically two different modes for scheduling data in the down-link by the base station. Dynamic scheduling brings the advantage of exploiting instantaneous channel state information while it puts on the other hand a significant burden on the system in terms of overhead and computation requirements. Especially for small packets that show up periodically, the overhead is typically too high. Therefore, the base station can serve such packet flows by the semi-persistent scheduling mode. In this mode, a certain resource allocation is fixed to a periodic schedule. While this does not allow any longer to exploit instantaneous channel states, it requires much less overhead. In this paper, we address the problem of selecting a modulation and coding scheme for such semi-persistent scheduling grants. The problem lies here in the stochastic characterization of the resource blocks over the next few seconds while on the other hand estimating based on such a characterization the blok error rate (and hence the average goodput). We provide a novel scheme, which outperforms all previously presented schemes significantly. The underlying model that we provide can also be used for any other long-term decision in an LTE system with semi-persistent scheduling such as interference coordination, handover decision etc. ; QC 20131212