RT Dissertation/Thesis
T1 Predicción en modelos de componentes inobservables condicionalmente heteroscedásticos
A1 Pellegrini, Santiago
AB During the last two decades, there has been an increasing interest in the academic and practitionerworld on modelling the volatility clustering observed in many economic and financialseries. This research was pioneered by Engle (1982) and Bollerslev (1986), with the introductionof GARCH models. It is also common to observe stochastic trends in many economic andfinancial time series. In this case, a popular practice is to take differences in order to obtaina stationary transformation. Then, an ARMA model is fitted to this transformation to representthe transitory dependence. Alternatively, the dynamic properties of series with stochastictrends may be represented by unobserved component models. It is well known that both modelsare equivalent when the disturbances are Gaussian. In this case, the reduced form of anunobserved component model is an ARIMA model with restrictions on the parameters; see,for example, Harvey (1989). The main difference between both specifications is that while theARIMA model includes only one disturbance, the corresponding unobserved component modelincorporates several disturbances. Consequently, working with the ARIMA specification is usuallysimpler. However, using the unobserved components model may lead to discover features ofthe series that are not apparent in the reduced form model because they arise when estimatingthe components.When combining both, stochastic trends and volatility clustering, the ARIMA and unobservedcomponent models are not in general Gaussian. This implies that they are no longerequivalent when allowing for conditional heteroscedasticity in the noises. Among the large numberof works devoted to studying and applying models that combine these features, almostnone of them made a comparative analysis between the two alternatives. This important issueremains somewhat unexplored. Therefore, we think that more effort should be placed in thisrespect, specially in what regards to forecasting performance. The study of this issue representsthe main goal of this thesis
YR 2009
FD 2009-04
LK https://hdl.handle.net/10016/7385
UL https://hdl.handle.net/10016/7385
LA eng
DS e-Archivo
RD 4 may. 2024