BAYESIAN GIBBS SLICE SAMPLER: A NOVEL APPROACH TO EFFICIENT MCMC AND ITS APPLICATION TO SOVEREIGN CREDIT RATING DETERMINANTS
This paper introduces the Bayesian Gibbs Slice Sampler (BGSS), an MCMC algorithm based on the Latent Slice Sampling framework. BGSS employs Bayesian inference during chain adaptation to refine its proposal distribution, eliminating the need for gradient calculations or complex adaptive proposals. It generates nearly independent proposals using a conditionally univariate factorization combined with a QR decomposition, which enhances exploration efficiency. BGSS matches the speed of LSS and performs comparably to advanced methods like the No-U-Turn Sampler (NUTS). Its effectiveness is demonstrated through both simulated data and real-world applications, including sovereign credit ratings analysis and modeling macroeconomic impacts over various time horizons, making it a robust and computationally efficient tool for Bayesian inference in econometrics.
Palabras clave: Bayesian Methods MCMC Slice Sampling Sovereign Risk