In the context of large factor models, the classical tools for model selection such as the Akaike and Bayesian Information Criteria (AIC and BIC), deliver, in general, biased estimators for the true number of factors in the model. The present thesis deals with asymptotic consistent model selection criteria for the true number of factors that were presented by Bai and Ng in 2002. Thus, the thesis covers the process of estimating the factors by principal components and it is shown that these estimates are asymptotically consistent. Then, two classes of model selection criteria are presented, and it is proven that they correctly estimate the true number of factors in an asymptotic framework. The performance of the criteria on finite data sets is profoundly studied. A Monte Carlo study evaluates an estimation precision of the considered model selection criteria for different models. Practical application of factor models includes forecasting econometric time series. Here, an extensive case study on forecasting the SP500 index return is provided. Finally, an investment strategy based on factor models is proposed and compared to existing ones.      «

In the context of large factor models, the classical tools for model selection such as the Akaike and Bayesian Information Criteria (AIC and BIC), deliver, in general, biased estimators for the true number of factors in the model. The present thesis deals with asymptotic consistent model selection criteria for the true number of factors that were presented by Bai and Ng in 2002. Thus, the thesis covers the process of estimating the factors by principal components and it is shown that these estim...     »