Title: One-step online maximum likelihood for linear state-space representations
Authors: Frederic Karame - Le Mans University (France) [presenting]
Alexandre Brouste - Universite du Maine (France)
Abstract: State-space models can be difficult to estimate by maximum likelihood due to the usual numerical problems (size, slow calculations, local solutions, ...). The one-step online approach has the double advantage of circumventing the usual numerical problems and providing efficient estimators. Nevertheless, these properties have been obtained for rather simple models and not for complex models like linear state-space representations. The aim is to extend this online estimation method to linear state-space representations. An efficient and fast estimation of these models represents an important breakthrough, especially if it can be implemented transparently for a user. The first part of the paper presents the theoretical proof. The second part is devoted to some Monte Carlo experiments. In the third part, the method is implemented to macroeconomic or financial issues.