|
Abstract:
|
We discuss methods of detecting multiple transient changes in long sequences of data, where the initial distribution can change at unknown times and later return to the original state. The number of changes, the moments of change, and durations of each period are unknown, although prior distributions of change-points and distribution parameters within each segment may be available. We propose change-point detection methods for different scenarios and explore their application to a problem of detecting and characterizing spikes in the processes of instantaneous electricity prices in deregulated energy markets. Automated accurate detection and estimation of the location, duration, and magnitude of spikes is important for the financial modeling and valuation of financial contracts.
|
