RUOCHEN WU

Research

PUBLICATIONS

• Oliver Linton, Ruochen Wu, 2016. Discussion of A. Ronald Gallant: “Reflections on the Probability Space Induced by Moment Conditions with Implications for Bayesian Inference”." Journal of Financial Econometrics, 14(2), 261-264.

​

• Ruochen Wu, Oliver Linton, 2015. Discussion on “Sequential Quasi Monte Carlo Sampling” by Gerber and Chopin. Journal of Royal Statistical Society (B), 77(3), 573.

​

​

RESEARCH PAPER

JOB MARKET PAPER: “A Semi-parametric Bayesian Approach to Equation Systems with Heterogeneous Errors” (Ruochen Wu, Melvyn Weeks)

Abstract: Heterogeneity in the errors is one of the major concerns in empirical economic researches, especially when micro data (e.g. individual or household level) are used. Observations are likely to belong to heterogeneous groups that follow various distributions in this situation. Such heterogeneity posts a challenge for analysts who wish to make reliable inference with the data, as it requires a method to capture the heterogeneity, i.e. how many different groups there are, and which group a particular observation belongs to. In this paper we propose a semi-parametric Bayesian method to cope with heterogeneous errors in equation systems. Our method uses the widely applied seemingly unrelated regression as a starting point that could gain efficiency by exploring the correlation between errors in the system. A Dirichlet process prior is put on the distribution of the errors, leading to a model that could be interpreted as the mixture of a variable number of normal distributions. Our method let the data and prior to determine the number of heterogeneous groups, as well as the group membership of the observations. A series of simulation experiment is designed to explore the performance of our method, and demonstrates that it provides more reliable inference than the parametric Bayesian seemingly unrelated regression. We then apply the method to two empirical examples in demands for production factors.

“Bootstrapped Semi-parametric Bayesian Inference: An Application to Demand Systems”

Abstract: This research compares bootstrapped Bayesian multivariate regression (BBMR) introduced by Heckelei and Mittelhammer (2003) with Monte Carlo integration (MCI). BBMR uses a bootstrapped likelihood, while MCI uses a normally distributed one. This comparison is carried out with the Rotterdam demand model and the linear approximation to the almost ideal demand system (LA/AIDS) with two datasets. The results show that when the normal distribution assumptions are not rejected, BBMR has similar performance to MCI, however, when the normal assumptions are rejected, BBMR has several advantages over MCI, including unbiased posterior means of coefficients and elasticities, tighter posterior distributions of the elasticities, and lower root mean square errors.

WORK IN PROGRESS

• “Instrumental Variables with Predictive Likelihood: An Application to the Return of Education” (Co-authored with Herman van Dijk and Melvyn Weeks)

​

• “Evolution of Carbon Emissions of Chinese Industries: Combining Micro and Macro Data”

​

• "Multiplicity in Firm Efficiencies: A Parametric Bayesian Approach" (Co-authored with Melvyn Weeks)