Multiple breaks detection in financial interval-valued time series
Multiple structural breaks detection for Interval-Valued Time Series (IVTS) is undoubtedly relevant under
practical perspectives and challenging under the point of view of the analysis of expert systems. In this
respect, financial time series usually show high variability and outliers; moreover, they often exhibit the
property of being of high frequency nature; thus, it is naturally advisable to consider them as IVTS type
for a given time unit. Despite this relevance, scarce effort has been spent by scholars to apply the methodological
advancements in breaks detection for IVTS to the crucial environment of financial time series.
This paper contributes to fill this gap. It employs the Atheoretical Regression Trees framework – a very
recent tool that is able to automatically locale multiple breaks occurring to unknown dates – to stock
prices. Such a procedure is able to estimate in an efficient way the structural breaks of the considered
series; at the same time, it keeps into account the main characteristics of the intervals describing the
IVTS. For our purposes, we adopt a theoretical proposal of reading daily stock prices as intervals whose
bounds are defined through the closing prices. Empirical experiments on the American International
Group – whose daily prices have experienced structural breaks in the past – validate the theoretical
model and show the usefulness of the proposed procedure.