Abstract
The paper introduces a new class of models, named dynamic quantile linear models, which combines dynamic linear models with distribution-free quantile regression producing a robust statistical method. Bayesian estimation for the dynamic quantile linear model is performed using an efficient Markov chain Monte Carlo algorithm. The paper also proposes a fast sequential procedure suited for high-dimensional predictive modeling with massive data, where the generating process is changing over time. The proposed model is evaluated using synthetic and well-known time series data. The model is also applied to predict annual incidence of tuberculosis in the state of Rio de Janeiro and compared with global targets set by the World Health Organization.
Highlights
The paper proposes a broad new class of models by combining two innovative areas developed during the last quarter of the twentieth century, namely dynamic linear models and quantile regression
We extend the dynamic linear models to a new class, named dynamic quantile linear models, where a linear function of the state parameters is set equal to a quantile of the response variable at time t, yt, similar to the quantile regression of Koenker (2005)
We introduce the inference via Markov chain Monte Carlo (MCMC) methods, and via an alternative approach based on normal approximations and Bayes linear estimation (West et al, 1985), which besides being computationally faster than MCMC, recovers the sequential analysis of the data
Summary
The paper proposes a broad new class of models by combining two innovative areas developed during the last quarter of the twentieth century, namely dynamic linear models and quantile regression. We extend the dynamic linear models to a new class, named dynamic quantile linear models, where a linear function of the state parameters is set equal to a quantile of the response variable at time t, yt, similar to the quantile regression of Koenker (2005). This method is suited to high-dimensional predictive modeling applications with massive data in which the generating process itself changes over time.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
