Robust portfolio decision analysis: A multicriteria constrained clustering approach
Kandakoglu, Makbule; Walther, Grit (Thesis advisor); Ben Amor, Sarah (Thesis advisor)
Aachen : RWTH Aachen University (2022)
Dissertation / PhD Thesis
Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2022
This dissertation consists of three publications. All publications are currently under review at international journals with high reputation. They contribute to the field of operations research and management focusing on portfolio decision analysis. The dissertation is structured into an introductory, a methodological and a conclusion part.In the introduction part, a detailed description of the portfolio management framework is presented and the underlying portfolio decision problem that has received increasing attention in recent years both in practice and as a field of academic research is motivated. The key characteristics and requirements for an advanced analytics solution approach for portfolio decision problems with multiple evaluation criteria regarding decision makers' preferences and real-life constraints are derived. A comprehensive literature review on the use of different multicriteria decision analysis methods to support project portfolio decision problems provides an overview on recent practice and research. Additionally, it introduces a guideline to assist researchers and practitioners in the use of approaches and methods in a specific problem context, and outlines avenues for further research.The methodological part of this dissertation contains two articles that aim at developing a comprehensive methodology for portfolio decision problems with multiple evaluation criteria, decision makers' preferences and real-life constraints covering a wide range of application cases. For this purpose, a new multicriteria constrained clustering model is developed in order to generate a portfolio of similar best projects with respect to preferences among each other. This model includes a two-phased procedure that first performs a multicriteria analysis method to evaluate the individual projects and then it applies a novel clustering model to generate a cluster of best projects as the optimal portfolio. The clustering model employs a unique mixed-integer linear programming formulation to introduce resource and/or other constraints into a new kind of cluster (so-called optimal portfolio).The model developed in this dissertation considers several relevant portfolio features such as multiple qualitative and quantitative criteria, preferences associated with the criteria and evaluations, uncertain and incomplete information, predetermined minimum and/or maximum number of alternatives to include in the portfolio, exhaustive use of the available resources, compatibility or incompatibility between the alternatives, and different types of decision rules related to the portfolio quality. In addition to these, an iterative approach to select or reject projects to gradually construct the final robust portfolio in multiple decision rounds strengthens the methodology. The contribution and the applicability of the methodology are illustrated through two case studies in different applications areas. The underlying models are able to generate better solutions than the traditional models in a short execution time. In this sense, this efficient and flexible methodology provides a basis for upcoming application cases that have not been explored so far. Finally, the conclusion part summarizes the main findings of each chapter of this dissertation. Furthermore, the main contributions of the thesis are discussed focusing on the impact for both researchers and practitioners. This part is completed by an outlook on future research.
- Chair of Operations Management