ConclusionΒΆ

This paper describes the use of the Statsmodels Python library for the specification and estimation of state space models. It begins by presenting the notation and equations describing state space models and the filtering, smoothing, and simulation smoothing operations required for estimation. Next, it maps these concepts to programming code using the the technique of object oriented programming and describes a simple method for the specification of state space models. Brief theoretical introductions to maximum likelihood estimation and Bayesian posterior simulation are given and mapped to programming code; the object oriented representation of state space models makes parameter estimation simple and straightforward.

Three examples, an ARMA(1,1) model, the local level model, and a simple real business cycle model are developed throughout, first theoretically and then as models specified in programming code. Classical and Bayesian estimation of the parameters of each model is performed. Finally, four flexible generic time series models provided in Statsmodels are described. Using these built-in classes, two of the example models, the ARMA(1,1) model and the local level model, are re-estimated and then extended to more complex, better fitting models.