CMStatistics 2016: Start Registration
View Submission - CMStatistics
B1200
Title: The method of simulated quantiles for regression parameters estimation. Authors:  Paola Stolfi - CNR - Institute for Applied Mathematics (Italy) [presenting]
Lea Petrella - Sapienza University of Rome (Italy)
Abstract: A method of simulated quantile (MSQ) is developed which generalise the matching sample quantile methods and the method of simulated moments, in order to estimate parameters in all the situations where the standard methods like ML, GMM and SMM may be difficult or impossible to apply. We will consider regression models $y|x\thicksim F(\theta)$ where $F$ may not have an analytic form and/or moments in which $x$ can be both a vector of exogenous or lagged variables. In order to implement the MSQ method we introduce a statistic $T(y,x)$ which is informative for the parameters of the models considered, and we match its sample quantiles and theoretical ones, the last simulated from $F(\theta)$, to find the values of the unknown parameters. To show the performances of the method we will consider regression and autoregressive models when different error term distributions are assumed. In addition to test the robustness of the method we also consider the case when contaminated data are observed. We extend the results to conditional variance models in order to take into account for more general non linear and heroscedasticity situations.