Overview
Dongchu Sun is a statistician at the University of Missouri, Columbia. His research focuses on Bayesian methodology, reference prior theory, and applications to econometric models including VARs.
Key Contributions
- Sun-Ni (2005): Proved propriety of the constant-Jeffreys and constant-reference posteriors for VAR models, identified the over-estimation bias of the Jeffreys prior for Σ, and developed the hit-and-run MCMC algorithm for the Yang-Berger reference prior. See Noninformative Prior for VAR.
- Contributes to reference prior theory for multiparameter statistical models (Berger-Bernardo framework).
- Ni-Sun (2005): Extended the Bayesian VAR prior comparison with a shrinkage prior on Φ and LINEX loss; established frequentist risk dominance of the shrinkage+reference combination over constant priors; derived the Student-t VAR Gibbs sampler with Gilks-Wild adaptive rejection for ν. See Noninformative Prior for VAR.
- George-Sun-Ni (2008): Co-developed Bayesian stochastic search variable selection for VAR models, extending SSVS to simultaneously identify restrictions on the coefficient matrix Φ and the Cholesky precision factor Ψ. Five-step all-standard-form Gibbs sampler; Rao-Blackwell forecasts improve over MLE by 20–67% in simulation; empirical application to 7-variable PPI→CPI VAR. Journal of Econometrics 142 (2008): 553–580.
- Sun-Tsutakawa-Speckman (1999): Established necessary and sufficient conditions for the posterior to be proper in hierarchical models with CAR(1) spatial random effects (Theorem 2: rank condition rank(X2′R1X2+B)=q; Theorem 3: necessity; Theorem 4: GLMM extension to Poisson log-linear disease mapping). Extended Hobert-Casella (1996) to the singular B case arising in the Besag-York-Mollié intrinsic CAR model. See Conditional Autoregressive Model. Biometrika 86(2): 341–350.
Related