A Non-Linear Approach for Completing Missing Values in Temporal Databases

摘要

The presence of missing data in the underlying time-series is a recurrent problem for market models. Such models impose to deal with cylindrical and complete samples. This paper presents a new procedure for the missing values recovery. The proposed method is based on two projection algorithms: a non-linear one (Self-Organizing Maps) and a linear one (Empirical Orthogonal Functions). The presented global methodology combines the advantages of both methods to get accurate approximations for the missing values. The methods are applied to three financial datasets.