Sticky HDP-HMM for Prediction of Economic Events

This research seeks to generate temporal event predictions using the sticky Hierarchical Dirichlet Process Hidden Markov Model (sticky HDP-HMM) [2], a generalization of the infinite HMM [1]. Hidden Markov Models (HMMs) are one of the most widely used machine learning techniques for analyzing temporal data. One significant limitation of this traditional approach is that the number of states in the HMM, N , has to be decided a priori, but for a number of applications it is not possible to hypothesize this accurately. The nonparametric Bayesian solution [3] to this is to remove the dependence on N by effectively making it infinite and specifying a prior over it; such as done in the HDP-HMM model [4]. An extension to the HDP-HMM model, known as the sticky HDP-HMM model [2] additionally contains a bias towards self-transitions.