Monte Carlo inference in econometric models with symmetric stable disturbances (Q1305675)

The paper formulates a general framework for Monte Carlo posterior inference in models with symmetric stable errors. The models can be linear, nonlinear or of seemingly unrelated regression type. The method is based on the representation of symmetric stable variates as normal scale mixtures. It is shown how to compute posterior marginal moments and densities of functions of parameters as well as how to compare the normal specification with other members of the symmetric stable Paretian distributions. A Markov chain Monte Carlo technique is proposed to perform the computations. The method is based on a combination of the Gibbs sampler with data augmentation and Metropolis-Hastings chains. The proposed new method is applied to constructed examples and is shown to perform satisfactorily. Subsequently inference is taken up in a model with stable desturbances that allows to study the question of trend versus difference stationarity. The method is also applied to a set of monthly real exchange rates of seven currencies versus US dollar and the questions of Paretian stability and trend-difference stationarity or non-stationarity are jointly examined.