Strategy Diversification and Alpha Correlation

Two strategies, each with a Sharpe ratio of 1.0, combined in equal risk, produce a combined Sharpe higher than 1.0 if their returns aren't perfectly correlated — and how much higher depends entirely on that correlation. This one fact is the entire business model of a multi-strategy hedge fund platform: run dozens of strategies, none individually spectacular, and let the correlation structure between them do the work that no single strategy's edge could do alone.

The arithmetic of combining two Sharpe ratios

For two strategies with Sharpe ratios $SR_1$ and $SR_2$, combined with weights $w_1$ and $w_2$ and return correlation $\rho$, the combined Sharpe ratio is

$SR_{\text{combined}} = \frac{w_1 SR_1 + w_2 SR_2}{\sqrt{w_1^2 + w_2^2 + 2w_1 w_2 \rho}}$

In words: the numerator averages the two strategies' Sharpes (weighted); the denominator is how much total volatility the combination carries, which shrinks as the correlation between the two return streams falls, because uncorrelated risks partially cancel rather than stacking.

Worked example. Two strategies, each Sharpe 1.0, combined equally ($w_1 = w_2 = 0.5$). If they're perfectly correlated ($\rho = 1$), the denominator is $\sqrt{0.25 + 0.25 + 0.5} = 1.0$, and combined Sharpe is $1.0$ — no benefit, you've just built a bigger version of the same bet. If instead $\rho = 0$, the denominator is $\sqrt{0.25 + 0.25 + 0} = \sqrt{0.5} \approx 0.707$, and combined Sharpe is $1.0 / 0.707 \approx 1.41$. Two strategies with Sharpe 1.0 each, run uncorrelated, combine into a book with Sharpe 1.41 — a genuine, structural improvement that came from nothing except the correlation being zero, not from either strategy getting any better.

Drag the correlation slider up from near zero toward one and watch the scatter tighten into a line — that tightening is exactly what erodes the diversification benefit in the Sharpe formula above. A wide, uncorrelated cloud of points is what a multi-strategy allocator is trying to build across their book; a tight, correlated line is what it looks like when two "different" strategies turn out to be making the same bet.

Why platforms obsess over this number

A multi-manager platform (Citadel, Millennium, Point72's satellite model) allocates capital across dozens to hundreds of independent trading teams specifically because most individual teams' strategies have low pairwise correlation with each other — a statistical arbitrage book, a credit relative-value book, and a commodities trend book are betting on genuinely different things. The platform's overall Sharpe ratio can exceed any individual team's by a wide margin purely from this diversification arithmetic, which is why these platforms can profitably run strategies that, standing alone, wouldn't clear the bar for a single-strategy fund's investors. It's also why platforms actively manage correlation as a first-class risk, not just each book's standalone risk — a new team is evaluated partly on how correlated its returns would be with the existing book, not just on its Sharpe in isolation.

Where correlation hides and reappears

The trap is that correlations measured in calm markets understate the correlation that shows up in a crisis. Strategies that look uncorrelated most of the time — merger arb, stat arb, credit relative value — can all suddenly correlate through a shared channel that doesn't show up in a normal-times covariance matrix: shared prime broker financing, shared leverage constraints, or a shared pool of crowded positions that several "independent" strategies happen to hold for unrelated reasons. August 2007's quant quake is the standard example — many statistically distinct-looking quant equity strategies turned out to be holding similar factor tilts, and when one large fund deleveraged, the correlation everyone had modeled as near zero spiked toward one, in the one week it mattered most.

In interviews

Derive the combined Sharpe formula and be ready to plug in numbers — this is a standard quantitative interview question and the two-line derivation (numerator averages Sharpes, denominator shrinks with correlation) should come without hesitation. Then pivot immediately to the caveat: normal-times correlation understates crisis correlation, and a platform or portfolio manager's real job is stress-testing what happens to the correlation matrix under a liquidity shock, not just optimizing against its historical average.