CMStatistics 2016: Start Registration
View Submission - CMStatistics
B0368
Title: A robust theoretical regression tree to detect structural breaks in financial time series Authors:  Carmela Cappelli - University of Naples Federico II (Italy) [presenting]
Francesca Di Iorio - University of Naples Federico II (Italy)
Abstract: In several real life and research situations data are naturally bounded by intervals, but usually they are analyzed summarizing the original data into single values. By doing so, some relevant information in the original data is lost. In the last years, efforts have been done either to extend classical methods or to develop new approaches to deal with interval-valued data. We address the problem of locating multiple structural breaks in financial Interval-Valued Time Series (IVTS), which represent a special case of IVTS because, for a given time unit, it is available, besides the extremes values (lowest and highest) of the corresponding interval, also the last value (closing) that is the most widely used in standard analysis. Moreover, financial time series are characterized by high variability and the presence of outliers that can affect the correct identification of break dates. In order to cope with the above issues, we define the lower and upper bound of the intervals as a function of the closing value, and then we employ, in the framework of a regression tree based approach, a robust distance that is able to neutralize the impact of outliers. We present the results of an empirical application to the prices of the American International Group (AIG) that shows the usefulness of the proposed procedure.