Authors: Tobias Adrian and Emanuel Moench

We develop an affine term structure model from a conditionally linear pricing kernel, without making distributional assumptions about shocks. Assuming pricing factors to be observable, we estimate the model by way of three-stage ordinary least squares, which can be interpreted as dynamic Fama-MacBeth regressions. We derive cross-equation restrictions for bond yields, which we do not impose in the estimation, but instead test. We can easily estimate specifications with large numbers of pricing factors, including volatility factors. We uncover specifications that give rise to lower pricing errors than do commonly advocated specifications, both in- and out-of-sample. Efficiency can be obtained by way of the generalized method of moments (GMM) estimator.