Towards optimal portfolio strategy to control maximum drawdown

Purpose - The purpose of this paper is to present a discrete quantitative trading strategy to directly control a portfolio's maximum percentage of drawdown losses while trying to maximize the portfolio's long‐term growth rate. Design/methodology/approach - The loss control target is defined through a Rolling Economic Drawdown (REDD) with a constant look‐back time window. The authors specify risk aversion in the power‐law portfolio wealth utility function as the complement of maximum percentage loss limit and assume long‐term stable Sharpe ratios for asset class indexes while updating volatility estimation in dynamic asset allocation implementation. Findings - Over a test period of the past 20 years (1992‐2011), a risk‐based out‐of‐sample dynamic asset allocation among three broad based indexes (equity, fixed income and commodities) and a risk free asset, is robust against variations in capital market expectation inputs, and out‐performs the in‐the‐sample calibrated model and traditional asset allocation significantly. Research limitations/implications - The current proposal can lead to a new mathematical framework for portfolio selection. Besides investors' liquidity and behavioural constraints, macroeconomic and market cycle, and the potential of central bank interventions following a market crash, could be additionally considered for a more rigorous dynamic asset allocation model. Practical implications - Besides the benefit of a clear mandate to construct suitable client portfolios, the portfolio approach can be applied to design invest‐able securities, such as principal‐guaranteed investment products, target risk asset allocation ETFs, and target‐date mutual funds with a glide path, etc. The formulation can also be implemented as a managed futures hedge fund portfolio. Originality/value - The paper introduces the Rolling Economic Drawdown (REDD) concept and specifies risk aversion as the floor of maximum percentage loss tolerance. Dynamic asset allocation is implemented through updating estimation of asset class volatilities.