Rodrigues, Paulo Manuel MarquesSalish, Nazarii2018-04-112018-04-1120102010http://hdl.handle.net/10400.1/10602Dissertação de Mestrado, Economia, Universidade do Algarve, Faculdade de Economia, 2010The analysis of high-frequency economic and financial data has recently received considerable attention and requires the development of new sophisticated tools for processing information. This dissertation investigates interval-valued data approaches as an alternative to the classical single-valued methods. Several important theoretical issues were explored and developed, such as for instance, i) metric on interval space and quality measures of forecast performance and model fitting; ii) basic statistical analysis of intervalvalued time series (range descriptive statistics); and iii) a review and extensions of existing modelling methods of interval-valued data. The most important issue when modelling financial data are its non-linear properties. As a result of this research a new class of non-linear threshold models for interval-valued time series that are capable to modelling different types of asymmetry in highfrequency data (e.g., burst of speculative bubbles, business cycles, crisis, etc.) are proposed. These techniques were implemented to the Portuguese stock market index (PSI20 index). The results obtained are very encouraging and compare very favourable to available procedures (K-NN and ARIMA-GARCH methods).engDados com valores de intervaloEstatística descritiva de intervaloDistância de intervaloMedidas de qualidade do intervaloModelos de limiares não linearesMedidasmaster thesis