Proceeding from the interpretation of a given sample as a realization from some Haavelmo distribution,discussed the role of probability models as a basic tool for modelling this unknown distribution. To be empirically acceptable, such a probability model should summarize all relevant sample information, i.e., account for important statistical properties of the data. A class of models suitable for the dynamic trending properties typically found in macroeconomic data is that of autoregressive (AR) probability models. The present chapter discusses these models, basically linear stochastic difference equations. deals with the scalar case, then generalizes to the vector case.