Bayesian Estimation of Total Investment Expenditures For Romanian Economy using DYNARE

In this paper we present the first estimation results of total investment expenditure for the Romanian economy, applying the Bayesian estimation approach of DSGE models presented in a number of various papers appeared recently in the open literature. The procedure requires the linear approximation of the original non-linear model, obtaining a LRE system, which is then solved for the reduced form state equation in its predetermined variables. Subsequently, standard Kalman recursions are applied to compute the likelihood which, combined with the prior assumptions, allows to evaluate the posterior probability. First the posterior mode is estimated, followed by a posterior simulation applying Metropolis Markov Chain Monte Carlo methods. The estimation procedure is implemented using the DYNARE software (Juillard, 1996-2003), a free available and open source software. After the model is defined and the first order conditions computed, the DSGE model is "decoded" by a parser embedded in DYNARE. All subsequent steps (linearisation, LRE system solution via generalized Schur decomposition, likelihood computation, Metropolis posterior simulation) are implemented by the software. DYNARE proved to be an extremely flexible and powerful tool, which allows to easily implement advanced Bayesian estimation techniques of DSGE models, with a considerable spare of coding and debugging time.