Abstract:
|
Tracing short-term changes in monitored data is an important component of analytic modeling for interventions. Identifying change-point in time-series involves sensitivity to the possible, and often existing, autocorrelation between adjacent observations. This paper presents an approach which introduces the feasibility to evaluate successive data along a time series. A five-category interval--mid-range at state "zero" and two on each side are used, reminiscent of control charts. If seasonality is apparent, it is integrated into the established "control bands." An application to petroleum stock monitoring including Motor Gasoline and Crude Oil is presented to graphically depict differences between observed data and pre-established bands. The index developed by the author, CTSS, is superimposed on the charts; 6-month intervals are compared to yield directionality of change.
|