You are here

Global Multi-Asset Allocation – An estimation of the global market portfolio and a back‐test analysis in the Black–Litterman framework

Master's Thesis
Corporate Partner: 
Swisscanto Invest by Zürcher Kantonalbank (ZKB)
Date Published: 
November 28, 2016
The thesis “Global Multi-Asset Allocation” proposes a solution to the overwhelming lack of multi-asset benchmarks in the industry. Indeed, this deficiency is a challenge for multi‐asset portfolio managers when they are carrying out asset allocation or attempting to measure their performance. Estimating a global market portfolio (GMP) is useful for strategic asset allocation purposes, and as a starting point for the Black–Litterman (B–L) model (1991). The model assumes that, starting from the GMP and applying reverse engineering, one can deduce the market’s beliefs with regard to the future returns of each asset. This thesis estimates the value of the GMP and then implements the B–L model. One question arises: Is the market portfolio the method that performs best as a starting point for B–L? In a following step, the thesis investigates the most efficient portfolio allocation from five different methods of portfolio construction—minimum variance, equal risk contribution, naïve diversification, maximum diversification, and inverse volatility—and the previously estimated GMP in two different case studies. The results highlights the fact that the risk‐based models deliver the best performing portfolios in terms of returns, risks, Sharpe ratios, alpha, information ratios, and maximum drawdowns. In addition, when additional information, such as a variance–covariance matrix and an informative lambda that reacts promptly to market changes, is added in the construction of the portfolio the results are higher performance, lower risk, and half the maximum drawdown of the risk-based methods. The final part of the research project introduces the estimation of a factor‐based multi-asset portfolio as an extension of the estimated GMP and the role of transaction costs in the back‐test analysis.