Abstract
In this article, we examine whether traditional linear models are suitable to assess financial samples, because financial data usually present nonnormality or nonlinear patterns, therefore linear models do not always adequately capture them. For this reason, as returns series usually follow autoregressive patterns, nonlinear models such as Self-Exciting Threshold Autoregressive (SETAR), Logistic STAR (LSTAR), Additive Autoregressive (AAR) or Neural Network (NNET) could provide a good fit. We study whether two samples of pension funds' returns in Spain and the United Kingdom present these features, and we find that the most appropriate model for the Spanish sample is an autoregressive model, but in the United Kingdom sample, we fit a neural network.