Simpler Strategies Survive Markets That Destroy Overfitted Backtests
More parameters in a backtest do not produce more reliable returns. Tactical asset allocation adjusts portfolio weights across asset classes using quantitative signals such as momentum, valuation and volatility rather than holding a fixed strategic mix. The goal is to improve the risk-return profile, particularly by reducing drawdowns during crises. Yet the academic and practitioner literature reveals a structural trade-off: strategies that look most impressive on paper often deteriorate fastest when applied to unseen data.
The institutional multi-factor approach combines trend-following, carry and value across liquid futures in equities, fixed income, currencies and commodities. Moskowitz, Ooi and Pedersen (2012) studied time-series momentum, while Asness, Moskowitz and Pedersen (2013) examined cross-asset value and momentum. These models weight exposures through risk parity and volatility targeting. Hurst, Ooi and Pedersen (2017) tested trend-following on datasets spanning more than a century. Institutional CTAs such as AQR and Man AHL deploy these methods to absorb billions of dollars in capital. The limitation is cost: post-transaction Sharpe ratios compress by an estimated 20 to 30 percent relative to pre-cost results.
A second strand relies on protective momentum strategies developed mainly by Wouter Keller through variants such as HAA, PAA, VAA and BAA. These approaches use sentinel assets to generate a binary risk-on or risk-off signal, often combined with selecting offensive assets by relative momentum. The problem is the high number of free parameters, including breadth thresholds, multiple lookback periods and the choice of canary assets, which increases overfitting risk. Independent replications published on forums such as QuantConnect systematically show worse results than the original papers. More importantly, they fail precisely at the moments protection should activate, such as the March 2020 crash. Publicly available backtests also suggest an extremely limited management capacity, on the order of a few million dollars.
The third strand, single-parameter tactical allocation, offers the strongest empirical validation. Mebane Faber’s work (2007, updated in 2018) compares a month-end price to a ten-month moving average across five asset classes. The robustness stems from the fact that similar results emerge using six-, eight- or twelve-month lookbacks, signaling that the model does not depend on fragile optimization. Researchers have replicated it successfully on markets outside the United States, including South Africa, with consistent results even after transaction costs. This family includes the Ivy Portfolio, a multi-asset extension of the same principle, and Gary Antonacci’s Dual Momentum, which combines relative momentum between US and foreign equities with an absolute momentum filter toward bonds.
Sophisticated strategies promise higher Sharpe ratios in backtests but tend to deteriorate when confronted with new data. Simpler strategies, built on a minimal number of parameters, offer more modest returns on paper but superior resilience over long historical periods and across different markets. For an investor who must trust a model during stress, empirical reliability outweighs theoretical maximum performance.
References
Moskowitz, Ooi, Pedersen (2012)
Time Series Momentum
Asness, Moskowitz, Pedersen (2013)
Value and Momentum Everywhere
Hurst, Ooi, Pedersen (2017)
A Century of Evidence on Trend-Following Investing
Asness, Frazzini, Pedersen (2012)
Leverage Aversion and Risk Parity
Antonacci (2012)
Risk Premia Harvesting Through Momentum
Faber (2018)
Practical Applications of A Quantitative Approach to Tactical Asset Allocation Revisited 10 Years Later
Key takeaways
- Adding parameters to a tactical model increases overfitting risk and degrades performance out of sample, as independent replications of protective momentum strategies consistently demonstrate.
- Institutional multi-factor models backed by decades of research can absorb large capital flows, but transaction costs compress their Sharpe ratios by 20 to 30 percent.
- Single-parameter strategies like comparing prices to a ten-month moving average deliver less spectacular backtests yet maintain their effectiveness across geographies and market regimes.