Title: Portfolio pretest estimation with machine learning
Authors: Ekaterina Kazak - University of Manchetser (United Kingdom) [presenting]
Winfried Pohlmeier - University of Konstanz (Germany)
Abstract: The general idea of the pretest estimation is to choose optimally between various competing estimators based on a testing outcome. However, the optimal decision rule may not rely on conventional choices for the significance level. Under certain circumstances it might be reasonable to select a lower significance level (higher probability of a Type I error) than a conventional one in order to increase the probability of rejecting the benchmark strategy. This problem arises in portfolio analysis, when the investor has to decide between two or more alternative portfolio strategies in the presence of low powered performance tests when truly superior strategies are rejected in favor of the benchmark strategy. The optimal size for the pretesting strategy can often not be derived and may change over time in time-series settings. We develop a data driven approach for choosing an optimal significance level for pretest estimators. We show that the bagged pretest estimator performs exceptionally well, especially when combined with adaptive smoothing. The resulting strategy allows for a flexible and smooth switch between the underlying strategies and is shown to outperform the corresponding stand-alone strategies. Our learning pretest estimation technique is shown be a strong competitor to alternative regularization based portfolio estimators.