Discrete mixtures of normals pseudo maximum likelihood estimators of structural vector autoregressions

The importance of supply and demand for oil prices: Evidence from non‐Gaussianity

“…This is particularly important in non‐Gaussian SVAR models. As highlighted in Fiorentini and Sentana (2022), specifying the wrong marginal distribution risks inconsistent estimation of shocks standard deviations, thereby invalidating inference on forecast error variance decompositions. Besides robustness, the estimators flexibility to adapt to the unknown shock distribution can also offer efficiency gains.…”

“…Previous methods either required specifying a functional form for the error distribution (Lanne, Meitz, and Saikkonen (2017), Lanne and Luoto (2020), Anttonen, Lanne, and Luoto (2021)) or a selection of suitable moments or criteria function (Lanne and Luoto (2021), Herwartz (2018)). Large‐sample arguments for pseudolikelihood inference are used to discuss robustness to model mis‐specification (Gourieroux, Monfort, and Renne (2017), Fiorentini and Sentana (2022)), but without the focus on providing an efficient procedure that adapts flexibly to the error distribution.…”

Identification of Structural VAR Models Via Independent Component Analysis: A Performance Evaluation Study

“…Another stream of literature has exploited non-Gaussianity, instead of heteroskedasticity, as a statistical property sufficient for local identification (see, among others, Fiorentini and Sentana, 2020;Gouriéroux et al, 2020;Guay, 2021;Lanne and Luoto, 2021;Lanne and Lütkepohl, 2010;Lanne et al, 2017 ). The idea of some of these studies is to recover the SVAR shocks as linear combinations of reduced-form VAR residuals, under the assumption that they are not just uncorrelated, but mutually statistical independent.…”

Identification of Structural VAR Models via Independent Component Analysis: A Performance Evaluation Study

“…In this respect, a recent stream of literature exploits directly specific statistical properties of the data -non-Gaussianity -for identification (see, among others, Lanne and Lütkepohl, 2010;Lanne et al, 2017;Gouriéroux et al, 2020;Lanne and Luoto, 2021;Fiorentini and Sentana, 2020;Guay, 2021). The idea of some of these studies is to recover the SVAR shocks as linear combinations of reduced-form VAR residuals disturbances, under the assumption that they are not just uncorrelated, but mutually statistical independent.…”