Robust portfolios
In backtests, 10 to 13% less volatility than equal weighting in every market tested, with similar risk-adjusted returns.
Backtest
- years of data
- 53years of data
- markets
- 3markets
- less volatility than equal weighting
- 10–13%
- trading cost on every dollar traded
- 0.1%
Overview
Robust portfolios are designed to account for errors in their forecasts. Can they lower risk without giving up risk-adjusted return?
A portfolio optimiser relies on forecasts of how assets will behave, and those forecasts are always partly wrong. Standard optimisers treat them as exact, which tends to produce concentrated portfolios that perform poorly once the forecasts miss.
A robust portfolio is built to remain sound when its forecasts are wrong. This study asks whether that design delivers a measurable benefit over two simple alternatives: holding every asset in equal weight, and holding the mix with the lowest forecast volatility.
What the backtests found
- 01
The robust portfolio was less volatile than equal weighting in all three markets, by 10 to 13%. The difference is statistically significant in each market (p < 0.001).
- 02
Its largest loss was smaller in every market, by 6 to 7 percentage points.
- 03
Its Sharpe ratio was slightly higher than equal weighting in all three markets. These differences are not statistically significant.
- 04
The lowest-volatility portfolio reduced volatility further, by 17 to 26%, but held fewer assets: in sector ETFs its weight was spread across the equivalent of 3.4 funds, against 6.3 for the robust portfolio.
Backtested volatility relative to equal weighting
- Robust
- Lowest volatility
- Equal weighting
10 U.S. industries
Jul 1973 – Aug 2026
48 U.S. industries
Jul 1973 – Aug 2026
9 U.S. sector ETFs
Jan 2004 – Aug 2026
Backtested growth of $10,000 in nine U.S. sector ETFs
- Robust$94,500
- Lowest volatility$79,300
- Equal weighting$98,300
Decline from previous high
Backtest results by market
10 U.S. industriesJul 1973 – Aug 2026
| Strategy | Volatility | Largest drawdown | Sharpe ratio |
|---|---|---|---|
| Robust | 13.6% | 41.2% | 0.57 |
| Lowest volatility | 12.4% | 34.4% | 0.59 |
| Equal weighting | 15.0% | 48.1% | 0.55 |
48 U.S. industriesJul 1973 – Aug 2026
| Strategy | Volatility | Largest drawdown | Sharpe ratio |
|---|---|---|---|
| Robust | 14.9% | 46.5% | 0.52 |
| Lowest volatility | 12.6% | 38.3% | 0.56 |
| Equal weighting | 16.9% | 52.7% | 0.51 |
9 U.S. sector ETFsJan 2004 – Aug 2026
| Strategy | Volatility | Largest drawdown | Sharpe ratio |
|---|---|---|---|
| Robust | 12.4% | 42.1% | 0.73 |
| Lowest volatility | 11.5% | 37.2% | 0.71 |
| Equal weighting | 14.3% | 49.2% | 0.66 |
Paper trading
Following historical testing, the study is now forward testing in an Alpaca paper trading account. The first monthly rebalance was placed on 2 November 2026.
Live since 2 November 2026
- Broker
- Alpaca paper trading
- Assets
- Nine U.S. sector ETFs
- Capital
- $100,000 simulated per strategy
- Rebalancing
- Monthly
References
- [9]
Gotoh, J., Kim, M. J. and Lim, A. E. B. (2020). Calibration of distributionally robust empirical optimization models. arXiv:1711.06565.
- [4]
DeMiguel, V., Garlappi, L. and Uppal, R. (2009). Optimal versus naive diversification: How inefficient is the 1/N portfolio strategy? Review of Financial Studies 22(5), 1915–1953.
- [7]
Scherer, B. (2007). Can robust portfolio optimisation help to build better portfolios? Journal of Asset Management 7(6).