Some new model selection criteria in simple regression

Abstract

Two new model selection procedures based on a measure of roughness of the residuals in simple regression are proposed and studied. The first criterion utilises a certain loss function and the second comprises the application of hypotheses tests, using the bootstrap methodology. The performances of these selection rules are illustrated and comparisons are made with traditional criteria using real and artificial data, and it is found that the new selection methods perform more satisfactorily.

Keywords:

• Bootstrap
• model selection criteria
• regression
• roughness