We consider the design of the radio-frequency (RF) precoder for minimum outage probability transmission for millimeter wave (mmWave) MISO-OFDM systems under the assumption of partial channel knowledge at the transmitter and full channel knowledge at the receiver. We formulate the problem as minimizing the cumulative distribution function (CDF) of an L-stage hypo-exponentially distributed random variable, where L is the number of propagation paths. To solve the optimization problem of the RF precoder, we propose an iterative gradient ascent algorithm. Simulation results show that the proposed algorithm converges in a small number of iterations and achieves lower outage probability, with full diversity order, than the conventional eigenvector beamforming. Moreover, the proposed scheme is more robust against human blockage than the conventional eigenvector beamforming and the maximum capacity scheme with full channel knowledge at the transmitter.