We propose a joint statistical model for the received power, mean delay, and rms delay spread, which are derived from the temporal moments of the radio channel responses. Indoor wideband measurements from two different data sets show that the temporal moments are strongly correlated random variables with skewed marginals. Based on the observations, we propose a multivariate log-normal model for the temporal moments, and validate it using the experimental data sets. The proposed model is found to be flexible, as it fits different data sets well. The model can be used to jointly simulate the received power, mean delay, and rms delay spread. We conclude that independent fitting and simulation of these statistical properties is insufficient in capturing the dependencies we observe in the data.