On Information Criteria for Estimating the Number of Knots in Splines with Fractional Residuals
Jan Beran, Arno Weiershäuser
Department of Mathematics and Statistics, University of Konstanz, Konstanz, Germany

We consider spline regression with fractional residuals and an unknown number of knots. An information criterion that allows for consistent estimation of the number of knots is introduced, with residuals being of FARIMA (long memory, short memory, antipersistence) or Hermite type (long memory). Asymptotic results are derived. Simulations illustrate finite sample properties. The method is applied to data examples from neurobiology.

Keywords: Spline regression; Fractional residuals; Information criteria

Biography: PhD student at the University of Konstanz (Germany) since 2008. Studied mathematics at the University of Marburg (Germany, master's degree in 2007) and statistics at Stellenbosch University (South Africa, bachelor's degree in 2005).