A MATLAB toolbox for reliable time series modeling and forecasting in State-Space

Abstract

Software reliability is a wide issue, depending not only on the use of stable implementations of well-reputed algorithms, but also on software design aspects. This philosophy is implemented in E4, a MATLAB Toolbox which uses state-space methods to achieve both, flexibility and reliability. E4 supports many standard formulations such as VARMAX, econometric models in structural form, transfer functions or general linear state-space models. These models are estimated by exact maximum likelihood, under standard conditions, or in an extended framework that allows for measurement errors, missing data, vector GARCH errors and constraints on the parameters. Ready-to-use functions are provided for model specification, preliminary estimation by subspace methods, analytic computation of the likelihood gradient and information matrix, simulation, forecasting and signal extraction. The core algorithms have been optimized for stability and numerical accuracy. In this aspect, the use of a computational platfonn such as MATLAB guarantees computational accuracy, portability and consistency between different hardware-software platforms. Several examples illustrate the mainfeatures of the Toolbox.