A novel scheme for optimum joint bit and power loading for link adaptation in discrete Fourier transform modulated filter bank based transceivers for underlay cognitive radio is proposed aiming to minimize the secondary user average biterror-rate for bit-interleaved coded modulation and to increase the spectral efficiency. The proposed scheme is subject to a stochastic chance-based constraint as well as a colored Gaussian interference caused by a third-party interference. The performance analysis shows that a bit-error rate saturation of schemes employing exclusive power or bit loading can be avoided over a large range of the average expected bit-energy-to-interferenceplus-noise ratio which is mandatory for improving the spectral efficiency of cognitive radio systems.