Sunday, July 31, 2011

MARSS 2.3 uploaded to CRAN

The MARSS package fits constrained and unconstrained linear multivariate autoregressive state-space (MARSS) models to multivariate time series data via priamarily an Expectation-Maximization (EM) algorithm, although fitting via the BFGS algorithm (using the optim function) is also provided. Functions are provided for parametric and innovations bootstrapping, Kalman filter and smoothing, bootstrap model selection criteria (AICb), confidences intervals via the hessian approximation and via bootstrapping and calculation of auxilliary residuals for detecting outliers and shocks. The user guide shows examples of using MARSS for parameter estimation for a variety of applications, model selection, dynamic factor analysis, outlier and shock detection, and addition of covariates.

http://cran.r-project.org/web/packages/MARSS/index.html

MARSS 2.3 is a big change over 1.1
  • Allows B and Z estimation 
  • Allows models with diagonal elements of R or Q set to 0. So you can have partially deterministic systems.
  • Covariates are allowed (see user guide and look up 'covariates' in the index)
  • Allows many types of Q and R matrices. Few constraints except it needs to be a valid variance-covariance matrix. See section in Derivations.pdf 
  • Allows you to set the initial state at t=0 or t=1
  • Allows diffuse priors if initial state at t=1 and method=BFGS
  • case studies added to the user guide on dynamic factor analysis, B estimation, covariates, and detecting outliers and shocks