External links
- Homepage
- Siddhartha Chib publications indexed by Google Scholar
| International | |
|---|---|
| National | |
| Academics | |
| Other | |
Siddhartha Chib | |
|---|---|
| Alma mater | University of California, Santa Barbara Indian Institutes of Management Delhi University |
| Known for | Framework for understanding the Metropolis–Hastings algorithm and its extensions in high-dimensional settings |
| Awards | Fellow of the American Statistical Association |
| Scientific career | |
| Fields | Econometrics, statistics |
| Institutions | Washington University in St. Louis |
| Thesis | Some Contributions to Likelihood Based Prediction Methods (1986) |
| Academic advisors | Sreenivasa Rao Jammalamadaka Thomas F. Cooley |
| Website | siddharthachib |
Siddhartha Chib is an econometrician, statistician, and the Harry C. Hartkopf Professor of Econometrics and Statistics at the Olin Business School at Washington University in St. Louis. His work is primarily in Bayesian statistics, econometrics, and Markov chain Monte Carlo methods. Chib's research spans a wide range of topics in Bayesian statistics, with influential contributions to statistical modeling, computational methods, and Bayesian model comparison techniques.
Chib pioneered a latent variable framework in Albert and Chib (1993) [1] , that greatly simplifies the Bayesian estimation of binary and categorical response models. It is a foundational method in Bayesian statistics. Along with the work in Chib and Greenberg (1998) [2] , the Albert and Chib (1993) latent variable framework provides a unified approach in the Bayesian context for handling univariate and multivariate categorical outcomes.
Another widely cited and influential paper by Chib is Chib and Greenberg (1995), [3] which provides an intuitive framework for understanding the Metropolis–Hastings algorithm and its extensions in high-dimensional settings. Central contributions of this work are the included derivations of the single block and multiple block versions of the algorithm using the principles of global and local reversibility, the first such derivations, and the guidance on the choice of proposal distributions for efficient implementation of the algorithm in practice.
For the problem of comparing Bayesian models, Chib developed a method for calculating marginal likelihoods from the MCMC output in Chib (1995) [4] that has been shown to be applicable to parametric and nonparametric models, and to models estimated by the Gibbs or Metropolis-Hastings algorithm. It is also straightforward to implement. The method is based on an identity that expresses the marginal likelihood as the product of the likelihood and the prior, divided by the posterior ordinate at a fixed point in the parameter space. Chib developed an approach for estimating this ordinate from the MCMC output. For models estimated by the Metropolis-Hastings algorithm, a generalization is given in Chib and Jeliazkov (2001) [5] . Basu and Chib (2003) [6] further extend the method to nonparametric Dirichlet process mixture models.
Chib has also worked on a model jump approach for comparing Bayesian models. The idea, developed in Carlin and Chib (1995) [7] , is to sample models and model-specific parameters by Markov chain Monte Carlo methods on a product of model spaces. The posterior distribution over models emerges from the frequency of visits to each model. This product-space approach has proved useful for comparing complex Bayesian models.
Chib has also written extensively on the problem of estimating stochastic volatility models in time series. The simple to implement and effficent method developed in Kim, Shephard, and Chib (1998) [8] is widely used. Extensions of the basic method to student-t models, covariates and multivariate stochastic volatility models are discussed in Chib, Nardari and Shephard (2002), [9] Chib, Nardari and Shephard (2006) [10] and Omori et al. (2007). [11]
Again, within the time series context, Chib (1998) [12] introduced a reparameterization of the change point model as a unidirectional hidden Markov model (HMM) that simplifies estimation and inference and enables the use of efficient forward-filtering and backward-sampling techniques for HMMs developed in Chib (1996) [13] and Albert and Chib (1993). [14]
Chib has also worked on and developed original methods for Bayesian inference in Tobit censored responses, [15] discretely observed diffusions, [16] univariate and multivariate ARMA processes, [17] [18] multivariate count responses, [19] causal inference, [20] [21] hierarchical models of longitudinal data, [22] nonparametric regression, [23] [24] [25] and tailored randomized block MCMC methods for complex structural models. [26]
In Chib, Shin, and Simoni (2018, 2022) [27] [28] he has developed estimation and model comparison tools for conducting Bayesian inference in models that rely only on moment restrictions and do not specify a parametric or non-parametric data generating process. In this work, he has supplied finite sample computational methods and large sample Bernstein--von Mises and model consistency theory under both correct and mis-specified moment restrictions.
Chib received a bachelor's degree from St. Stephen's College, Delhi, in 1979, an M.B.A. from the Indian Institute of Management, Ahmedabad, in 1982, and a Ph.D. in economics from the University of California, Santa Barbara, in 1986. [29] His advisors were Sreenivasa Rao Jammalamadaka and Thomas F. Cooley.
Chib is a fellow of the American Statistical Association (2001), [30] an inaugural fellow of the International Society of Bayesian Analysis (2012), [31] and a fellow of the Journal of Econometrics (1996). [32]
| International | |
|---|---|
| National | |
| Academics | |
| Other | |
