This paper presents a new method for identifying triangular systems of time-series data. Identification is the product of a bivariate GARCH process. Relative to the literature on GARCH-based identification, this method distinguishes itself both by allowing for a timevarying covariance and by not requiring a complete estimation of the GARCH parameters. Estimation follows OLS and standard univariate GARCH and ARMA techniques, or GMM. A Monte Carlo study of the GMM estimator is provided. The identification method is then applied in testing a conditional version of the CAPM.

This paper was revised in March 2007.