Tuesday, December 1 @ 11:00 - 12:30 PM (ONLINE)
Multi-Anchor Point Shrinkage for Better Betas
Hubeyb Gurdogan, Florida State University
ABSTRACT: The GPS (Goldberg, Papanicolaou, Shkolnik) method shrinks the leading eigenvector of the sample covariance matrix towards the vector of all 1’s by a data driven amount in the low sample-high dimension regime. That creates an estimate of betas that has lower l_2 error and significantly reduces the impact of the estimation error on minimum variance portfolio weights and risk forecasts. We study a generalization of the GPS method and provide a class of estimators of beta that further improve the l_2 error and accuracy of the minimum variance portfolio weights. To do this we utilize additional observable information about the separation of betas into sub-collections where the betas in each sub-collection are close to each other.