Widened Learning of Portfolio Selection for Index Tracking

Abstract

This master’s thesis considers index tracking from the perspective of solution space exploration. Several search space heuristics are used in combination with different portfolio optimization models in order to select a tracking portfolio with returns that mimic a benchmark index. Even with the fastest hardware and the most massively parallel systems available today, it is infeasible to conduct an exhaustive search for the large solution space in a reasonable time. Instead of increasing the number of parallel resources with the aim to traverse as much solution space as possible, we try to obtain the best use of every parallel resource. With this aim we introduce several portfolio diversity measures. Experimental results conducted on real-world datasets show that adding diversity to the set of parallel search paths can provide a better solution (tracking portfolio) due to exploration of disparate solution space regions. However, the choice of the diversity measure plays an important role. Poor path diversification can hinder the progress towards a better solution.