Current Working Papers
A New Sampling Method via Approximate Bayesian Computation with Correlated Random Numbers
Publications
Maximum Likelihood Estimation of Regression Effects in State Space Models
In short: It is typically difficult to simultaneously estimate regression coefficients and hidden states in state space models. This paper proposes a novel method that cleverly combines the classic Kalman filter with maximum likelihood estimation, yielding accurate parameter estimates without increasing computational complexity.
Bayesian Estimation of Fixed Effects Models with Large Datasets
In short: When dealing with fixed effects models for massive datasets, the traditional "dummy variable" approach easily leads to out-of-memory errors. This paper proposes a new algorithm based on Bayesian sampling that estimates parameters directly without creating dummy variables, completely resolving the memory burden.
Divide-and-conquer Metropolis–Hastings samplers with matched samples
In short: Traditional Bayesian analysis can be extremely slow when facing massive datasets. This paper designs a "divide-and-conquer" algorithm that splits large data into smaller chunks for separate computation, and then seamlessly merges the results using a clever "matched sample" technique to significantly boost efficiency.
Bayesian Inference in Common Microeconometric Models With Massive Datasets by Double Marginalized Subsampling
In short: To accelerate the analysis of massive microeconomic datasets, this paper develops a "double marginalized subsampling" technique. It allows the computer to evaluate only a tiny fraction of data points in each iteration while accurately reconstructing overall data characteristics, vastly improving processing speed.
Estimating MIDAS regressions via OLS with polynomial parameter profiling
In short: Traditional methods often require complex nonlinear calculations when handling mixed-frequency data regressions. This paper proposes a new framework combining Ordinary Least Squares (OLS) with polynomial parameter profiling, making the analysis of such data simpler and more stable.
Big Data Bayesian Linear Regression and Variable Selection by Normal-Inverse-Gamma Summation
In short: This paper transforms complex Bayesian big data analysis into simple "addition" operations. By introducing a novel summation operator, the algorithm directly pieces together computation results from different data sources, offering a highly efficient solution for variable selection.
Inequality Constrained State Space Models
In short: The classic Kalman filter often fails when dealing with data containing "inequality constraints" (e.g., variables must be strictly positive). This paper introduces an optimized particle filter algorithm that accurately tracks hidden data states even under these complex constraints.
The effects of exports on facility environmental performance: Evidence from a matching approach
In short: Are exporting firms necessarily more environmentally friendly than non-exporters? Using U.S. manufacturing data, this paper finds that this "export-induced environmental effect" is highly dependent on the specific characteristics of the firm's industry, such as pollution abatement costs.
A computationally efficient method for vector autoregression with mixed frequency data
In short: Macroeconomic data often suffers from mixed frequencies (e.g., a mix of monthly and quarterly data). This paper proposes a computationally efficient Bayesian Vector Autoregression (VAR) method that can directly integrate and process these data while preserving economic intuition.
Income distribution in urban China: An overlooked data inconsistency issue
In short: This paper points out that the 2002 expansion of the Chinese Urban Household Survey sample led to a structural break in the data. The study uses a dynamic model to reconstruct a coherent time series, providing a more reliable foundation for measuring urban China's income distribution.
A FLEXIBLE STATE SPACE MODEL AND ITS APPLICATIONS
In short: Traditional models typically treat observed data and unknown hidden states differently. This paper proposes a novel "flexible state space model" that treats both symmetrically, making the model structure much more intuitive and parsimonious for analyzing dynamic factors.