Global and Asset Class Diversified ETF Portfolio Strategy - A Convex Mean/CVaR Optimization Approach using Dynamic Copulas with Flexible Probabilities

One of the most important factors to control for the achievements of investment portfolio returns is risk. If we only think that a 100% positive return is needed to recover a portfolio loss of 50%, we can understand why. With the advent of the exponential growth of technology usage in markets, systematic correlated risks are more present than ever, with the trading robots inputting another layer of risk through their interaction in a much higher speed and volume than the mechanical markets of the past. In this paper, we make use of quantitative models to create an “all-weather” dynamic portfolio that seeks to deliver to investors an optimum performance given a minimum ex-ante conditional value-at-risk while investing the full budget amount in ETFs. The investable universe is composed of 41 ETFs diversified among regions of the world and different asset classes such stocks, bonds, REITs, commodities and metals. The models are time and state dependent. We tweak the instruments’ distributions using the Fully Flexible approach with two state and one time factor. We fit GARCH (1,1) models to extract the series’ residuals, which are marginally i.i.d, but not jointly. This is a common issue when using static correlation to model joint movements in financial time-series. We address that issue using a Dynamic Conditional Correlation model. Finally, through Monte-Carlo simulations, we assess the market risk of the portfolio and find the mean-CVaR efficient frontier through convex quadratic programming