multivariate robust filtering

abstract: this paper presents an extension of an idea by kleiner, martin & thomson (1979) to multivariate autoregressive processes. the properties of the procedures are reported by some examples with economic data. starting from the assumption that observations obey an autoregressive vector process but are contaminated by additive disturbances, it is endeavoured to eliminate the disturbances to regain the "true" process and its law of generation. this is done iteratively by the multivariate robust filter set forth in chapter 2. chapter 3 presents the basic experiment. the procedure is applied to quarterly data for the austrian monetary base and gross domestic product after stationarizing the sample in two different ways. chapter 4 is devoted to small experiments with alternative methods. chapter 5 investigates into how the results are affected by artificially generated outliers and whether these outliers are safely detected. in chapter 6 it is shown that the results depend on the width of the band within which data are still judged to be "good". chapter 7 reports an experiment with stock price data where the outliers are supposed to obey a law different from the basic assumption of robust filtering. consequently, the results are poor. in chapter 8 the effects of changing the weighting function in the filter are illustrated. finally, chapter 9 summarizes suggestions for measuring the "outlier-ness" of single observations based on the application of the filtering procedure.;