Nicholas G. Polson is a professor of econometrics and statistics at the University of Chicago Booth School of Business. His research spans Bayesian computation, data augmentation, particle filtering, and applications to finance and machine learning.
Key Contributions / Features
McCulloch, Polson, and Rossi (2000): ID prior for the multinomial probit — reparameterises Σ as [[1,γ'],[γ,Φ+γγ']] with γN, Φ⁻¹Wishart, enforcing σ₁₁=1 with probability one; four-block conjugate Gibbs; analytical results characterising NID vs. ID prior marginals. Journal of Econometrics 99: 173–193. See Multinomial Probit.
Jacquier, Polson, and Rossi (1994): Introduced Bayesian MCMC for discrete-time SV models; single-move independence Metropolis with IG blanketing density; proved Bayesian RMSE dominance over MM and QML.
Carlin, Polson, and Stoffer (1992): Monte Carlo approach to nonnormal and nonlinear state-space models; early particle-filter-type algorithm for non-Gaussian state spaces.
Applegate, Kannan, and Polson (1991): Random polynomial-time algorithms for sampling from joint distributions; theoretical foundations for MCMC mixing.
Jacquier, Polson, and Rossi (2004): Extended JPR (1994) to fat-tailed SV (Student-t via scale-mixing augmentation) and leverage (ASV2: corr(ut,vt)=ρ); Bayes Factors favor both extensions for equity; adaptive rejection sampling applicable due to log-concavity of full conditional. See Stochastic Volatility.
Jacquier, Johannes, and Polson (2007): MCMC-ML algorithm — J-copy data augmentation; J(draws−MLE)→N(0,I−1); SV and multivariate jump-diffusion applications. See MCMC Maximum Likelihood.
Jacquier and Polson (2010): 92-page handbook survey of Bayesian methods in finance covering portfolio optimization, return predictability, APT, volatility models, options, and particle filtering with parameter learning via sufficient statistics (CJLP 2010). See Jacquier and Polson (2010).