The simultaneous work of Brosamler (1988) and Schatte (1988) opened a new understanding of a classical distribution limit, the Central Limit Theorem (CLT). Under the name of Almost Sure Central Limit Theorem (ASCLT), many new theoretical results and methods as well as applications quickly appeared. In this thesis, the ASCLT is studied, emphasizing in the findings of universal character found in Berkes and Csáki (2001). The notion of associativity is presented, providing a method to prove ASCLT-like results for time series. For the first time, the squared GARCH(1,1) processes are proven to be associated, leading to new almost-sure limits. Statistical applications of the ASCLT in construction of confidence intervals are examined. Some of these confidence intervals, for heavy-tailed distributions and for squared GARCH(1, 1) processes, are shown to far outperform their classical CLT relatives. The empirical performance of the ASCLT is deeply studied, providing methods to improve the convergence speed of the limit.
«
The simultaneous work of Brosamler (1988) and Schatte (1988) opened a new understanding of a classical distribution limit, the Central Limit Theorem (CLT). Under the name of Almost Sure Central Limit Theorem (ASCLT), many new theoretical results and methods as well as applications quickly appeared. In this thesis, the ASCLT is studied, emphasizing in the findings of universal character found in Berkes and Csáki (2001). The notion of associativity is presented, providing a method to prove ASCLT-l...
»