A novel steepest descent (NSD) algorithm and its improved version (NSDM algorithm) for adaptive filters have been suggested and the local convergence analysis of the NSD algorithm has been performed recently. In this paper we present two main results. The first is a performance analysis of the NSDM algorithm for adaptive filters with correlated Gaussian data based on the expected global behaviour approach in nonstationary environments as well as in stationary case. The second is an extension of the previous analysis for the NSD algorithm in stationary case to nonstationary environments. The results from these analyses are verified numerically through computer simulations for an adaptive system identification example with highly correlated input data.