Anders Warne

Overview

Anders Warne is a Swedish economist associated with the European Central Bank (ECB) and Sveriges Riksbank. His research spans both frequentist and Bayesian inference in cointegrated VAR systems, moving from asymptotic distribution theory for Wald tests under rational expectations restrictions (Warne 1997) to a full Bayesian VECM framework with Grassmann-manifold priors (Warne 2006).

Key Contributions

• Warne (1997) — Wald statistics under RE restrictions. Extends Sims-Stock-Watson (1990) from linear to nonlinear cross-equation (NCE) restrictions. Shows that RE model restrictions generically yield a nonstandard limiting distribution for the Wald statistic because they constrain the row space of the long-run impact matrix A(1). Key result: the NCE restrictions supply a lower bound for the cointegration rank r, making simulation of correct critical values tractable. Monte Carlo evidence reveals oversize in small samples; recommends low nominal significance levels.
• Warne (2006) — Bayesian cointegrated VAR. Full Bayesian framework for VECM inference: Villani/Warne prior (uniform on Grassmann manifold for cointegrating space; Minnesota-style shrinkage prior on short-run dynamics); full Gibbs sampler (Proposition 1) with all standard full conditionals; marginal (3-block) Gibbs sampler integrating out $(\Phi,\Gamma)$ analytically (Proposition 5); posterior mode via generalized eigenvalue problem (Proposition 4); Chib (1995) MLI for rank posteriors; closed-form lag-order MLI (Corollary 1) enabling joint $p(r,k|Y)$ over all rank-lag combinations without MCMC.
• Grassmann manifold prior. Identified that a flat prior on the unconstrained cointegrating space parametrization $\Psi$ overweights the normalization breakdown region (following Strachan and van Dijk 2003) and adopted the matrix-$t$ specification that corresponds exactly to a uniform prior over the Grassmann manifold.
• Euro area money demand. Applied the framework to a 6-variable system for euro area M3 (1980Q1–2003Q4) with a broken linear trend from 2001Q4; found income elasticity ≈ 1.37 but data-uninformative opportunity cost semi-elasticity; rank sensitive to prior specification.

Related

• Cointegration
• Gibbs Sampler
• Marginal Data Density
• Lag-Order Selection Criteria
• Mattias Villani
• Luc Bauwens