Reservoir Computing (RC) is a framework within recurrent neural networks (RNNs) for temporal information processing, proposed as an efficient alternative to conventional deep neural network architectures. In RC, the intrinsic dynamics of a fixed high-dimensional system are exploited as a reservoir, while training is limited to a linear readout layer. Resistive memory networks based on self-organizing memristive nanowire systems represent a promising platform for Physical Reservoir Computing, owing to their nonlinear conductance dynamics, fading memory, and complex recurrent connectivity. The computational behavior of these networks is investigated through a physics-based model of memristive junctions coupled with circuit-level simulations based on Modified Voltage Nodal Analysis.