Linear Filtering and Taiwan Stock Return Nonlinearity

The results of the previous chapter demonstrate that the serial dependencies within Taiwanese stock returns that are reflected in the significant autocorrelation and nonlinearity test results of Chapter Four are not constant; rather, they shift in direction and magnitude over time, with a number of brief episodes of very strong dependencies accounting for much of the magnitude of the full sample results. Hinich and Patterson (1996) find similar results for U.S. securities. These results notwithstanding, numerous researchers, such as Hsieh (1992), have found that GARCH-type models seem to be able to account for or capture the full-sample nonlinear serial dependencies found within a number of financial time series.