Dynamic Asset Allocation Using Machine Learning: Seeing the Forest for the Trees
利用随机森林算法识别宏观体制，围绕美国60/40股债组合构建动态资产配置策略，在三种叠加方案下显著提升了1950年以来的风险调整后收益。
High inflation and aggressive monetary policy tightening in 2022 triggered one of the largest return drawdowns for a US 60/40 portfolio in the last 100 years. In this article, the authors develop a dynamic asset allocation framework based on macro regimes using machine learning to improve the risk/reward versus static balanced portfolios with higher macro volatility. Using both macro and market data, they construct indicators for growth, inflation, and policy to track the business cycle and for risk appetite since 1950. They then use a random forest algorithm on those indicators to identify macro regimes that drive tail risks that matter for portfolio construction around a US 60/40 equity/bond portfolio. Based on real-time regime probabilities, they implement one of three dynamic asset allocation overlays: 1) switch between a 60/40 portfolio and cash, 2) rotate between equities and bonds, and 3) allocate to commodities/gold. The overlays materially enhanced risk-adjusted returns compared with a static 60/40 portfolio since 1950, although results were mixed over time.