Nichtparametrische integrierte Rendite- und Risikoprognosen im Asset Management mit Hilfe von Prädiktorselektionsverfahren (English shop)

Preview

Table of Contents, Datei (84 KB)
Extract, Datei (230 KB)

In asset management, the investor faces the central task of systematically structuring the invested capital. Optimisation calculi such as the Portfolio Selection approach of Nobel laureate Harry M. Markowitz are based in particular on treating the returns of the investment alternatives as stochastic random variables and on the assumption that the future expected values, variances and correlations of the return distributions are known or can at least be determined approximately. Some empirical findings suggest that, for the purposes of asset management, return estimates are more significant than risk estimates. While more or less elaborate models are employed for return forecasts, (time-weighted) historical variance-covariance matrices are still frequently used as risk estimators. However, the trade-off between accepted risk and expected return is one of the fundamental problems in finance, so that the uncertainty associated with an investment plays a major role in financial theory and practice. This work first examines the theoretical suitability, adequate implementability and empirical performance of nonparametric kernel regression for constructing forecasting models for asset allocation. The methodology can be used both for modelling return expectations and for forecasting consistent integrated risk. Promising approaches exist for the nonparametric selection of the relevant explanatory variables, and their empirical performance is analysed. Such a model should not only be statistically sound but also economically interpretable, thus offering the possibility of comprehending and qualitatively assessing the relationship. This work therefore addresses the theoretical suitability and adequate implementability of the procedures taking into account both the financial problem at hand and the requirements of the statistical methods. The empirical performance of the models is investigated on the basis of extensive simulation studies and historical capital market data.