Conventional long-range weather prediction is an initial value problem that uses the current state of the atmosphere to produce ensemble forecasts. Purely stochastic predictions for long-memory processes are “past value” problems that use historical data to provide conditional forecasts. Teleconnection patterns, defined from cross-correlations, are important for identifying possible dynamical interactions, but they do not necessarily imply causation. Using the precise notion of Granger causality, we show that for long-range stochastic temperature forecasts, the cross-correlations are only relevant at the level of the innovations - not temperatures. This justifies the Stochastic Seasonal to Interannual Prediction System (StocSIPS) that is based on a (long memory) fractional Gaussian noise model. Extended here to the multivariate case, (m-StocSIPS) produces realistic space-time temperature simulations. Although it has no Granger causality, emergent properties include realistic teleconnection networks and El Niño events and indices.