A0856
Title: Testing for breaks in the conditional mean based on the estimating function approach
Authors: Christian Francq - University of Lille and CREST (France)
Lorenzo Trapani - University of Leicester (United Kingdom)
Jean-Michel Zakoian - CREST (France) [presenting]
Abstract: The estimating function approach is particularly attractive for time series models where the dynamics are not fully specified, but the conditional mean is assumed to be a given parametric function of past observations. In many financial applications, however, the conditional mean may undergo a structural change. A class of cumulative sum, CUSUM, statistics is proposed to detect breaks in the conditional mean under weak assumptions. This procedure depends on the choice of a sequence of weights, leading to a potentially infinite number of consistent tests, and it is shown that the best test is related to Godambe's optimal estimator, also discussing data-driven procedures for this optimal choice of weights. Inference is studied in the presence of a changepoint, and the case is also studied where the conditional mean is misspecified, developing heteroskedasticity and autocorrelation consistent (HAC) versions of the test. Results are illustrated using Monte Carlo experiments and real financial data.
