Multi-class Investment Strategy Design Sensitivities

How sensitive are multi-class asset allocation strategies to variations in backtesting choices? In their December 2025 paper entitled “The Multiverse Across Asset Classes: Design Uncertainty in Asset Allocations”, Arnaud Battistella, Jean-Charles Bertrand, Guillaume Coqueret and Nicholas McLoughlin explore net annualized Sharpe ratio sensitivities of asset class allocation methods with respect to five backtest design choices:

• Alternative investor utility functions (three levels of risk aversion).
• Choice of signal-generating type (eight ranging across economic indicators, trend-following/momentum, risk, valuation, equal return, machine learning).
• Different sample periods (rolling 10-year intervals, each time shifting the start by one year for a total of 16 intervals during May 1999 through May 2025).
• Different rebalancing frequencies (monthly, quarterly, annually).
• Leeway with respect to benchmark tracking (21 error constraints relative to an equal-weighted benchmark).

Their investment universe consists of the S&P 500 Index, commodities, gold, a 70%/30% U.S. Treasuries/U.S. investment grade bonds and U.S. high-yield bonds. They express sensitivities by providing ranges of Sharpe ratios generated by ranges of design choice values (23,040 total outcomes). Using monthly data for the five asset classes and for 19 economic indicators, a 2.5% risk-free rate and 0.1% 2-way frictions for portfolio turnover, they find that:

(more…)