A1211
Title: Joint quantile shrinkage: Toward non-crossing Bayesian quantile models
Authors: Tibor Szendrei - Heriot-Watt University (United Kingdom) [presenting]
David Kohns - Heriot-Watt University (United Kingdom)
Abstract: The prevalence of crossing quantiles has led to various methods for estimating monotonically increasing quantiles. Recent research has shown that one can achieve non-crossing quantiles by jointly estimating the quantiles and imposing fused shrinkage with quantile-specific hyperparameters. This approach is extended to the Bayesian realm. To achieve this, the Bayesian Quantile norm of using the Asymmetric Laplace distribution is deviated from, and instead, a general Bayes method is opted for to minimize a loss function. Additionally, instead of directly estimating quantiles, the focus is on estimating and shrinking the differences between quantiles. The formulation aligns with the time-varying parameter (TVP) models common in macroeconometrics, allowing us to use efficient TVP samplers for estimation. It is demonstrated that the proposed framework provides superior fits compared to conventional Bayesian and frequentist quantile regression estimators.
