Consistent selection of the number of change-points via sample-splitting (Q99318)

From MaRDI portal
Revision as of 16:39, 3 February 2024 by Daniel (talk | contribs) (‎Created claim: Wikidata QID (P12): Q98196934, #quickstatements; #temporary_batch_1706974296281)
scientific article
Language Label Description Also known as
English
Consistent selection of the number of change-points via sample-splitting
scientific article

    Statements

    48
    0 references
    1
    0 references
    1 February 2020
    0 references
    5 May 2020
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    Consistent selection of the number of change-points via sample-splitting (English)
    0 references
    This paper proposes a unified framework to select the number of change-points in multiple change point models under a unified framework. A general procedure, named CPOSS, selects the number of change-points that minimizes the squared prediction error, which measures the fit of a specified model for a new sample. The sample is divided into one training set and one validation set by the parity of the time order, resulting in a 2-fold cross-validation with order-preserved sample-splitting which is tailored for the change-point problem. Asymptotic consistency is established, ensuring that the resulting estimated number of change-points equals the true one with probability approaching one under mild conditions. Effectiveness of the proposed selection strategy is demonstrated on a variety of numerical experiments, comparing with the BIC or its variants obtained by modifying the loss function and associated penalization term, on an ample range of simulated examples. Models are considered with changes in different aspects such as the location, scale, distribution and regression relationship. The data can be univariate, multivariate or in linear model structure, either independent or correlated. Two real-data examples are also presented. In the first, CPOSS is used to detect changes in the variance of daily returns of the UK FTSE 100 index. In the second, changes in the proportion of the G+C composition of a human chromosome sequence are considered.
    0 references
    model selection
    0 references
    multiple change-point model
    0 references
    selection consistency
    0 references
    cross-validation
    0 references
    least-squares
    0 references
    dynamic programming
    0 references

    Identifiers

    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references