K. Rao Kadiyala

Overview

K. Rao Kadiyala is an econometrician known for his systematic comparison of prior distributions for Bayesian VAR forecasting. His joint work with Sune Karlsson provides the first broad empirical and theoretical critique of the Minnesota prior's forced equation-independence and its treatment of the residual covariance matrix $\Psi$ as known and diagonal — showing that priors allowing inter-equation dependence consistently match or outperform the Minnesota prior, particularly in small samples.

Key Contributions

• Prior comparison (Kadiyala-Karlsson 1993): Compared Minnesota, Normal-Wishart, Diffuse (Jeffreys'), Normal-Diffuse, and ENC priors across three empirical forecasting experiments on Canadian and US data; established that inter-equation dependence in the prior consistently improves forecast accuracy.
• ENC prior application: Applied the Drèze-Morales (1976) Extended Natural Conjugate prior to VARs, overcoming the $\Psi \otimes \Omega$ restriction that constrains $\text{Var}(\gamma)$ under the Normal-Wishart prior.
• Numerical methods (Kadiyala-Karlsson 1997): Extended the 1993 comparison with Gibbs-sampling-based inference and additional numerical methods for Bayesian VAR models.

Related

• Sune Karlsson
• Minnesota Prior
• Noninformative Prior for VAR
• Vector Autoregression