...son to the most plausible model of at most 20. The Bayes factor is equal to the ratio of posterior model probabilities, and a value larger than 20 can be considered strong evidence against the model (=-=Kass and Raftery, 1995-=-). This leads to inclusion of six models, which together account for a total posterior model probability of 75.8%. To be clear: of the 29 = 512 possible specifications in the model space of the GVS sa...

...ement of the first row of λ1 being non-zero. 36The first approximation, Ĉ∗B in those authors’ notation, is a localized (i.e. volume-corrected) version of the Bartlett-adjusted Laplace estimator (see =-=DiCiccio et al., 1997-=-, Section 2.2). The second approximation, ĈC , a Candidate’s estimator, is based on a simple Kernel density estimate of the posterior distribution (see DiCiccio et al., 1997, Section 2.5). For the vo...

...results from inference about both parameters and latent factors is not possible with classical methods, since latent factors are inferred conditional on given point estimates of the model parameters (=-=Kim and Nelson, 1999-=-, chap. 8).20 A Bayesian approach employing Markov Chain Monte Carlo (MCMC) methods overcomes these challenges.21 It has numerous advantages: Computationally it is less challenging since it amounts to...

...unction generally has many dimensions (there are 20 parameters in our case) and is highly non-linear, which makes numerical optimization expensive and finding the global maximum difficult (Duffee and =-=Stanton, 2004-=-; Duffee, 2009). More specifically, the likelihood function of a DTSM oftentimes has “multiple inequivalent local maxima which have similar likelihood values but substantially different implications f...

...ihood function of a DTSM oftentimes has “multiple inequivalent local maxima which have similar likelihood values but substantially different implications for economic quantities of interest” (Kim and =-=Orphanides, 2005-=-, p.10). Attempts to estimate the 14 model with ML confirmed this: the likelihood function has several local maxima with different prices of risk, which confirms that the physical dynamics are hard to...

...e of BMA solves the “discontinuity problem”, the stark difference between forecasts for the short rate depending on whether its largest root is less than or equal to one (Cochrane and Piazzesi, 2008; =-=Jardet et al., 2009-=-), since estimates effectively are averages of stationary and non-stationary specifications. The paper is structured as follows: Section 2 describes the DTSM and shows how it can be used to decompose ...