This thesis considers the estimation of missing values in incomplete data. First, we will theoretically derive the maximum likelihood estimator for missing values of multivariate normally distributed data with known distribution parameters. In the second step, we propose a two step estimation procedure when parameters of the normal distribution are unknown and give an outline of the estimation procedure in a general case, i.e. for not necessarily normally distributed data. Further, a Monte Carlo Analysis is carried out for the normal case with unknown distribution parameters to estimate missing values by numerically maximizing the likelihood function. Later in this thesis, this estimation procedure will be extended to arbitrary distributions by ignoring marginal distributions. Therefore the concept of copulas will be introduced. In this thesis, we will focus on the D-Vine copula, a specific pair copula construction. In the next step, a Monte Carlo Simulation is implemented for the D-Vine copula. We will survey five different copula families and estimate the missing values by maximizing the likelihood of the D-Vine on two different scales. In the last part of this thesis, an empirical application of the proposed methodology to forecast the DAX 30 index is carried out and is compared with a classical linear regression approach.     «

This thesis considers the estimation of missing values in incomplete data. First, we will theoretically derive the maximum likelihood estimator for missing values of multivariate normally distributed data with known distribution parameters. In the second step, we propose a two step estimation procedure when parameters of the normal distribution are unknown and give an outline of the estimation procedure in a general case, i.e. for not necessarily normally distributed data. Further, a Monte Carlo...     »