Title: Nonparametric tilted regression estimation
Authors: Seyed Mahdi Salehi - University of Neyshabur (Iran) [presenting]
Farzaneh Boroumand - Macquarie University (Australia)
Hassan Doosti - Macquarie University (Australia)
Mohammad Taghi Shakeri - Mashhad University of Medical Sciences (Iran)
Abstract: Tilting methods are employed for modifying the empirical distribution by replacing the uniform distribution of weights over data with a multinomial distribution. The tilting approach has also been utilized for minimizing the distance to an infinite-order regression estimator. We propose a tilted Nadaraya-Watson estimator and proved that it achieves a higher level of accuracy and, at the same time, preserves interesting properties of the infinite order estimator. We also showed that the tilted estimators are consistent and have desirable convergence rates. In a simulation study, we illustrated that the tilted Nadaraya-Watson estimator has a better performance than its classical version in terms of Median Integrated Squared Error.