Campbell (1996)  proposed a methodology based on a VAR(1) process to test the Intertemporal CAPM. Innovations in predictors of portfolio returns are estimated and used as risk factors in an asset pricing model. We show that his triangularization procedure makes the cross-sectional prices of risk associated with the predictors non identifiable. This is because they depend on the arbitrary ordering of the variables in the VAR. Moreover, since the factors are orthogonal to the market and to one another, the comparison with alternative multi-factor models is problematic, e.g. with the Fama-French HML and SMB factors. Link to the article
LIOUI, A. and PONCET, P. (2011). Misunderstanding Risk and Return. Finance, 32(1), pp. 91-136.