CMStatistics 2016: Start Registration
View Submission - CFE
Title: A robust LM test for long memory Authors:  Michael Will - Leibniz University Hannover (Germany) [presenting]
Matei Demetrescu - CAU Kiel (Germany)
Philipp Sibbertsen - University of Hannover (Germany)
Abstract: Existing statistical tests for long memory in general require the assumption of at least conditional homoskedasticity in the error terms of the data generating process. Furthermore many parametric or frequency-domain based procedures display heavy size distortions if the analyzed model is misspecified. Therefore, we propose a nonparametric time-domain based test for long memory. Our test is of the Lagrange Multiplier type and therefore does not require any estimation of the memory parameter under the null. Due to its nonparametric nature, it does not require any specification of a model, making it robust to the influence of short-run dynamics. Moreover it displays robustness to conditional and unconditional heteroskedasticity (nonstationary volatility) in the error terms of the data generating process of quite general form. We illustrate the performance of our test by comparing it to other existing testing procedures in the literature via Monte Carlo simulation.