Model-based clustering of categorical time series with multinomial logit classification

Sylvia Frühwirth-Schnatter*, Christoph Pamminger, Rudolf Winter-Ebmer, Andrea Weber

*Corresponding author for this work

Research output: Contribution to journalConference articlepeer-review

Abstract (may include machine translation)

A common problem in many areas of applied statistics is to identify groups of similar time series in a panel of time series. However, distance-based clustering methods cannot easily be extended to time series data, where an appropriate distance-measure is rather difficult to define, particularly for discrete-valued time series. Markov chain clustering, proposed by Pamminger and Frühwirth-Schnatter [6], is an approach for clustering discrete-valued time series obtained by observing a categorical variable with several states. This model-based clustering method is based on finite mixtures of first-order time-homogeneous Markov chain models. In order to further explain group membership we present an extension to the approach of Pamminger and Frühwirth-Schnatter [6] by formulating a probabilistic model for the latent group indicators within the Bayesian classification rule by using a multinomial logit model. The parameters are estimated for a fixed number of clusters within a Bayesian framework using an Markov chain Monte Carlo (MCMC) sampling scheme representing a (full) Gibbs-type sampler which involves only draws from standard distributions. Finally, an application to a panel of Austrian wage mobility data is presented which leads to an interesting segmentation of the Austrian labour market.

Original languageEnglish
Pages (from-to)1897-1900
Number of pages4
JournalAIP Conference Proceedings
Volume1281
DOIs
StatePublished - 2010
Externally publishedYes
EventInternational Conference on Numerical Analysis and Applied Mathematics 2010, ICNAAM-2010 - Rhodes, Greece
Duration: 19 Sep 201025 Sep 2010

Keywords

  • Auxiliary Mixture Sampler
  • Bayesian Statistics
  • Classification
  • Markov Chain Monte Carlo
  • Multinomial Logit Model
  • Panel Data
  • Random Utility Model
  • Transition Matrices

Fingerprint

Dive into the research topics of 'Model-based clustering of categorical time series with multinomial logit classification'. Together they form a unique fingerprint.

Cite this