Sharpe-Weighted vs Equal-Weighted Strategy Blends

When combining several trading strategies into one portfolio, a natural instinct is to give more capital to the strategies with the best historical Sharpe ratios and less to the weaker ones. This is exactly the same idea as mean-variance portfolio optimization applied to strategies instead of stocks, and it has the same weakness: historical Sharpe ratios are estimated with a lot of noise, especially over the short backtests most strategies are evaluated on, so the "best" strategy in-sample is partly the strategy that got lucky. Equal weighting sidesteps this by giving every strategy the same allocation regardless of its measured Sharpe ratio, deliberately refusing to bet on which historical performance was luck and which was skill.

In practice the right answer is somewhere between the two extremes: heavily Sharpe-weighting a small set of strategies estimated from a short history tends to produce a fragile, overfit blend, while pure equal weighting ignores real, persistent differences in strategy quality when there is genuine evidence of one. Shrinking estimated Sharpe ratios toward their average before weighting is a common middle ground.

A worked example

Given three strategies with backtested Sharpe ratios of 1.5, 1.0 and 0.5 estimated from only two years of daily data, a naive Sharpe-weighted blend might allocate roughly 50% to the first strategy alone. But with only two years of data, the standard error on a Sharpe ratio estimate is large enough that the "true" ranking of these three strategies is genuinely uncertain — an equal-weighted blend (a third each) would have historically produced a smoother combined equity curve in similar cases, precisely because it doesn't over-commit to a ranking that noise, not skill, may have produced.