Empirical comparison of portfolio risk diversification algorithms

---

Tezin Türü: Yüksek Lisans

Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Uygulamalı Matematik Enstitüsü, Finansal Matematik Anabilim Dalı, Türkiye

Tezin Onay Tarihi: 2018

Tezin Dili: İngilizce

Öğrenci: ÇİĞDEM YERLİ

Asıl Danışman (Eş Danışmanlı Tezler İçin): Sevtap Ayşe Kestel

Eş Danışman: Nilüfer Çalışkan Schindler

Özet:

The enhanced correlations during global financial crisis has revealed that simple asset allocation portfolios prove to be not well-diversified across different risk factors, which makes the risk based asset allocation strategies popular. However, the strategies still construct the risk concentrated portfolios due to the correlation among the asset classes. As a result, risk allocation among uncorrelated risk factors instead of risk allocation among asset classes have become widely used. This thesis aims to distribute the risk among uncorrelated risk factors in a portfolio to prevent constructing risk concentrated portfolio. We employ “diversified risk parity strategy”. The first step in this approach is the construction of the uncorrelated portfolios. To construct uncorrelated portfolios, we follow two different approaches: principal component analysis and minimum linear torsion model. These uncorrelated portfolios are also known as uncorrelated risk factors in the literature. In the second step, we apply the risk parity strategy to these uncorrelated risk factors to obtain equal risk budget from each risk source. While the literature evaluates each uncorrelated portfolio as one kind of risk factor, we focus on three main risk sources, namely equity risk, inflation rate risk and inflation risk. In this work, we give the background of diversified risk parity strategy and traditional risk based asset allocation strategies and explain how uncorrelated portfolios constructed based on principal component analysis and minimum linear torsion model with examining their return and risk properties. Then we provide an application of the strategies to selected asset classes. The poor performance of mean-variance strategy due to large estimation errors in estimated mean leads to risk-based strategies popular. Therefore, to make clear comparison, we also include mean-variance optimization and compare the out-of-sample performance with both risk-based strategies and diversified risk parity strategies in the empirical analysis.