Robust methods of building regression dependencies in the tasks of valuation

Abstract

Known robust methods focus on situations where the sample may include assets that are not similar to the asset being valued, so that the corresponding deviations from the regression may have an arbitrary distribution. When using such methods, the selected assets are essentially taken into account in calculating with the “weight” the smaller, the more their prices deviate from the regression. However, these methods do not allow comparing different specifications of regression in order to select the “best” of them. We consider “intermediate” situations typical for valuation problems, when the distribution of deviations from regression is close to normal, but has “heavier tails” that exponentially decrease. For such situations, we propose a number of methods for estimating the calibration parameters of regression based on the maximum likelihood principle, and give examples of their application to the valuation of assets.