Series: Occasional Papers. 0301.
Author: Gianluca Caporello y Agustín Maravall.
Full document
Abstract
A program is presented for error detection in incoming data for data bases of time series. It provides a sensible answer to the question: when new data is reported, which of the new observations are likely to contain an error? The observation is suspicious if it is far from what could have been expected from the past history of the variable. Without considering the new observation, and using a fully automatic procedure, a REGARIMA model is identified for the series, which may have missing observations and be contaminated by outliers. Calendar and, more generally, regression effects can also be present. The new observation is compared to the (out-of-sample) forecast obtained with the model and, if the forecast error is clearly unreasonable, the series is considered suspicious The program, named TERROR, is a particular application of program TRAMO, freely available at the Bank of Spain web site (www.bde.es). The program is reliable and efficient for very large scale applications.